Early versus late tracheostomy in critically ill COVID‐19 patients

Abstract

Background

The role of early tracheostomy as an intervention for critically ill COVID‐19 patients is unclear. Previous reports have described prolonged intensive care stays and difficulty weaning from mechanical ventilation in critically ill COVID‐19 patients, particularly in those developing acute respiratory distress syndrome. Pre‐pandemic evidence on the benefits of early tracheostomy is conflicting but suggests shorter hospital stays and lower mortality rates compared to late tracheostomy.

Objectives

To assess the benefits and harms of early tracheostomy compared to late tracheostomy in critically ill COVID‐19 patients.

Search methods

We searched the Cochrane COVID‐19 Study Register, which comprises CENTRAL, PubMed, Embase, ClinicalTrials.gov, WHO International Clinical Trials Registry Platform, and medRxiv, as well as Web of Science (Science Citation Index Expanded and Emerging Sources Citation Index) and WHO COVID‐19 Global literature on coronavirus disease to identify completed and ongoing studies without language restrictions. We conducted the searches on 14 June 2022.

Selection criteria

We followed standard Cochrane methodology.

We included randomized controlled trials (RCTs) and non‐randomized studies of interventions (NRSI) evaluating early tracheostomy compared to late tracheostomy during SARS‐CoV‐2 infection in critically ill adults irrespective of gender, ethnicity, or setting.

Data collection and analysis

We followed standard Cochrane methodology.

To assess risk of bias in included studies, we used the Cochrane RoB 2 tool for RCTs and the ROBINS‐I tool for NRSIs. We used the GRADE approach to assess the certainty of evidence for outcomes of our prioritized categories: mortality, clinical status, and intensive care unit (ICU) length of stay. As the timing of tracheostomy was very heterogeneous among the included studies, we applied GRADE only to studies that defined early tracheostomy as 10 days or less, which was chosen according to clinical relevance.

Main results

We included one RCT with 150 participants diagnosed with SARS‐CoV‐2 infection and 24 NRSIs with 6372 participants diagnosed with SARS‐CoV‐2 infection. All participants were admitted to the ICU, orally intubated and mechanically ventilated. The RCT was a multicenter, parallel, single‐blinded study conducted in Sweden. Of the 24 NRSIs, which were mostly conducted in high‐ and middle‐income countries, eight had a prospective design and 16 a retrospective design. We did not find any ongoing studies.

RCT‐based evidence

We judged risk of bias for the RCT to be of low or some concerns regarding randomization and measurement of the outcome.

Early tracheostomy may result in little to no difference in overall mortality (RR 0.82, 95% CI 0.52 to 1.29; RD 67 fewer per 1000, 95% CI 178 fewer to 108 more; 1 study, 150 participants; low‐certainty evidence).

As an indicator of improvement of clinical status, early tracheostomy may result in little to no difference in duration to liberation from invasive mechanical ventilation (MD 1.50 days fewer, 95%, CI 5.74 days fewer to 2.74 days more; 1 study, 150 participants; low‐certainty evidence).

As an indicator of worsening clinical status, early tracheostomy may result in little to no difference in the incidence of adverse events of any grade (RR 0.94, 95% CI 0.79 to 1.13; RD 47 fewer per 1000, 95% CI 164 fewer to 102 more; 1 study, 150 participants; low‐certainty evidence); little to no difference in the incidence of ventilator‐associated pneumonia (RR 1.08, 95% CI 0.23 to 5.20; RD 3 more per 1000, 95% CI 30 fewer to 162 more; 1 study, 150 participants; low‐certainty evidence). None of the studies reported need for renal replacement therapy.

Early tracheostomy may result in little benefit to no difference in ICU length of stay (MD 0.5 days fewer, 95% CI 5.34 days fewer to 4.34 days more; 1 study, 150 participants; low‐certainty evidence).

NRSI‐based evidence

We considered risk of bias for NRSIs to be critical because of possible confounding, study participant enrollment into the studies, intervention classification and potentially systematic errors in the measurement of outcomes.

We are uncertain whether early tracheostomy (≤ 10 days) increases or decreases overall mortality (RR 1.47, 95% CI 0.43 to 5.00; RD 143 more per 1000, 95% CI 174 less to 1218 more; I² = 79%; 2 studies, 719 participants) or duration to liberation from mechanical ventilation (MD 1.98 days fewer, 95% CI 0.16 days fewer to 4.12 more; 1 study, 50 participants), because we graded the certainty of evidence as very low.

Three NRSIs reported ICU length of stay for 519 patients with early tracheostomy (≤ 10 days) as a median value, which we could not include in the meta‐analyses. We are uncertain whether early tracheostomy (≤ 10 days) increases or decreases the ICU length of stay, because we graded the certainty of evidence as very low.

Authors' conclusions

We found low‐certainty evidence that early tracheostomy may result in little to no difference in overall mortality in critically ill COVID‐19 patients requiring prolonged mechanical ventilation compared with late tracheostomy. In terms of clinical improvement, early tracheostomy may result in little to no difference in duration to liberation from mechanical ventilation compared with late tracheostomy. We are not certain about the impact of early tracheostomy on clinical worsening in terms of the incidence of adverse events, need for renal replacement therapy, ventilator‐associated pneumonia, or the length of stay in the ICU.

Future RCTs should provide additional data on the benefits and harms of early tracheostomy for defined main outcomes of COVID‐19 research, as well as of comparable diseases, especially for different population subgroups to reduce clinical heterogeneity, and report a longer observation period. Then it would be possible to draw conclusions regarding which patient groups might benefit from early intervention. Furthermore, validated scoring systems for more accurate predictions of the need for prolonged mechanical ventilation should be developed and used in new RCTs to ensure safer indication and patient safety.

High‐quality (prospectively registered) NRSIs should be conducted in the future to provide valuable answers to clinical questions.

This could enable us to draw more reliable conclusions about the potential benefits and harms of early tracheostomy in critically ill COVID‐19 patients.

Plain language summary

Is early or late tracheostomy more effective in critically ill COVID‐19 patients who are expected to require long‐term artificial ventilation?

Key messages

• For adults hospitalized with COVID‐19 on mechanical ventilators, performing an early tracheostomy (where doctors cut through the skin into the trachea (windpipe) to insert a breathing tube) before 10 days after starting ventilation, may have little or no effect on deaths and the time patients spend on a ventilator compared with late tracheostomy, performed 10 days or more after starting ventilation.

• We are uncertain whether early tracheostomy improves or worsens patients’ condition or shortens their intensive care unit stay.

• Researchers should agree on key outcomes to be used in COVID‐19 research; future research should focus on well‐designed studies with robust methods. We could then draw stronger conclusions about the best timing for tracheostomy in critically ill COVID‐19 patients.

What is a tracheostomy?

A tracheostomy is a procedure where doctors cut through the skin into the trachea (windpipe) to insert a breathing tube. Breathing then takes place completely through this tube. Tracheostomies are performed on patients who require long‐term ventilation in order to make ventilation easier and provide a safe airway access directly to the trachea. Compared to a breathing tube through the mouth, a tracheostomy tube offers less resistance to airflow. This can help to reduce the work of breathing and make weaning from mechanical ventilation easier. However, tracheostomies can also lead to complications. There is a risk of infection at the tracheostomy site. Prolonged placement of a tracheostomy tube can lead to obstruction of the windpipe. This can obstruct the flow of air and lead to breathing difficulties.

Tracheostomies may be performed 'early' or 'late' during ventilation. 'Early' is often defined as during the first 10 days of ventilation and 'late' as 10 days or more after ventilation started.

What is the link between tracheostomy and COVID‐19?

Most patients with severe COVID‐19 need help with breathing. In some cases, this means long‐term mechanical ventilation, so tracheostomy may be advised. In these patients, a tracheostomy can be associated with serious complications for both the patient and the caregiver. Patients with COVID‐19 already have a higher risk of additional infections because their immune system is weakened. The tracheostomy can bring an additional risk of infection. These patients often have a higher risk of bleeding. Bleeding complications can happen during a tracheostomy. Doctors and nursing staff are at increased risk of becoming infected with the virus during the procedure.

To date, there are no universal recommendations for the best time to perform a tracheostomy for these patients.

What did we want to find out?

We wanted to find out the effects of early tracheostomy in very ill COVID‐19 patients on:

• death from any cause;

• whether patients got better after treatment, measured by how long they spent on a ventilator;

• whether patients' condition worsened so that they developed unwanted effects, such as lung infections; and

• how long they stayed in the intensive care unit.

What did we do?

We searched for studies that investigated the performance of early tracheostomy compared to late tracheostomy in hospitalized adults with COVID‐19.

We compared and summarized their results, and rated our confidence in the evidence, based on factors such as study methods and sizes.

What did we find?

We found 1 good‐quality study with 150 people, and 24 lower‐quality studies with 6372 people. Patients’ average age was 62 years. Studies took place around the world, mainly in high‐ and upper‐middle‐income countries. All studies compared early with late tracheostomy but defined early and late differently. Early tracheostomy was defined at 7, 10, 12, 14 and 21 days after the start of mechanical ventilation. We selected up to 10 days for early tracheostomy and after 10 days as late. This was the time used by the good‐quality study and in 6 of the other studies.

Main results

We found the following results from 1 study with 150 people.

Deaths: early tracheostomy may result in little to no difference to deaths from any cause. Of 1000 people, 67 fewer die when a tracheostomy is performed early.

Did patients get better with early tracheostomy? Early tracheostomy may result in little to no effect on how long patients spend on a ventilator.

Did patients get worse with early tracheostomy? Early tracheostomy may result in little to no difference in the number of patients:

• with any unwanted effect; or

• with ventilator‐related lung infections.

How long did patients have to stay in the intensive care unit? Early tracheostomy may result in little benefit to no difference in the length of time patients spend in the intensive care unit.

What are the limitations of the evidence?

Our confidence in the evidence is very limited, because we found only 1 good‐quality study with few participants. The other, less robust studies, performed tracheostomies at very different time points and measured and reported their results inconsistently.

How up to date is this evidence?

The evidence is up‐to‐date to 14 June 2022.

Summary of findings

Summary of findings 1. Early versus late tracheostomy in critically ill COVID‐19 patients.

Early versus late tracheostomy in critically ill COVID‐19 patients
Patient or population: hospitalized, mechanically ventilated adults with confirmed SARS‐CoV‐2 infection Settings: in hospital Intervention: early tracheostomy (≤ 10 days after intubation) Comparator: late tracheostomy (> 10 days after intubation)
Outcomes	Anticipated absolute effects		Relative effect 95% CI	No. of participants (studies)	Certainty of the evidence (GRADE)	Comments
	Assumed risk
	Late (> 10 days) tracheostomy	Risk difference with early (≤ 10 days) tracheostomy
Evidence from RCTs
Mortality
Overall mortality Follow‐up at up to day 90	372 per 1000	67 fewer per 1000 (178 fewer to 108 more)	RR 0.82 (0.52 to 1.29)	150 (1 RCT)	⨁⨁◯◯ Lowa	Early tracheostomy (≤ 10 days) may result in little to no difference in overall mortality compared with late tracheostomy.
Improvement of clinical status
Duration to liberation from mechanical ventilation Follow‐up at up to day 90	The mean duration to liberation from mechanical ventilation was 19.6 days.	MD 1.50 days fewer (5.74 fewer to 2.74 more)	‐	150 (1 RCT)	⨁⨁◯◯ Lowa	Early tracheostomy (≤ 10 days) may result in little to no difference in duration to liberation from mechanical ventilation compared with late tracheostomy.
Worsening of clinical status
Adverse events (any grade) Follow‐up at up to day 90	782 per 1000	47 fewer per 1000 (164 fewer to 102 more)	RR 0.94 (0.79 to 1.13)	150 (1 RCT)	⨁⨁◯◯ Lowa	Early tracheostomy (≤ 10 days) may result in little to no difference in the incidence of adverse events compared with late tracheostomy.
Ventilator‐associated pneumonia Follow‐up at up to day 90	38 per 1000	3 more per 1000 (30 fewer to 162 more)	RR 1.08 (0.23 to 5.20)	150 (1 RCT)	⨁⨁◯◯ Lowa	Early tracheostomy (≤ 10 days) may result in little to no difference in the incidence of ventilatorassociated pneumonia compared with late tracheostomy.
Need for renal replacement therapy Follow‐up at up to day 90	NA	NA	NA	NA	NA	None of the included studies reported need for renal replacement therapy, therefore we do not know whether early tracheostomy has any impact on this outcome.
ICU length of stay	The mean ICU length of stay was 24.2 days	MD 0.5 days fewer (5.34 fewer to 4.34 more)	‐	150 (1 RCT)	⨁⨁◯◯ Lowa	Early tracheostomy (≤ 10 days) may result in little to no difference in ICU length of stay compared with late tracheostomy.
Evidence from NRSIs
We are uncertain whether early tracheostomy (≤ 10 days) increases or decreases overall mortality, because we graded the certainty of evidence as very low (RR 1.47, 95% CI 0.43 to 5.00; RD 143 more per 1000, 95% CI 174 less to 1218 more; 2 studies, 719 participants; I2 = 79%). We are uncertain whether early tracheostomy (≤ 10 days ) increases or decreases duration to liberation from mechanical ventilation, because we graded the certainty of evidence as very low (MD 1.98 days fewer, 95% CI 0.16 days fewer to 4.12 more; 1 study, 50 participants). Three NRSIs reported ICU length of stay for 519 patients with early tracheostomy (≤ 10 days) as a median value, which we could not include in the meta‐analyses. We are uncertain whether early tracheostomy (≤ 10 days) increases or decreases the ICU length of stay, because we graded the certainty of evidence as very low.
CI: confidence interval; ICU: intensive care unit; MD: mean difference; NA: not applicable; NRSIs: non‐randomized studies of interventions; RR: risk ratio; RCT: randomized controlled trial
GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of effect.

^aDowngraded by two levels due to serious imprecision because of wide confidence intervals in the study, the 95% confidence interval includes both benefits and harms, and few participants in only one study.

Background

This review is part of a series of Cochrane Reviews examining treatments and therapies for coronavirus disease 2019 (COVID‐19) as part of the German research project 'CEOsys' (COVID‐19 Evidence Ecosystem; CEOsys 2021). Based on the first published reviews from this research project on monoclonal antibodies (Kreuzberger 2021), convalescent plasma (Chai 2020), and remdesivir (Ansems 2021), the reviews in this series share information in the background and methodology sections.

Description of the condition

In late 2019, a novel severe acute respiratory syndrome coronavirus called SARS‐CoV‐2 appeared in China and caused an outbreak of a novel viral disease, also known as COVID‐19. Because of the highly transmissible characteristics of the virus, it quickly led to a worldwide pandemic (Chen 2020; Huang 2020; Zhu 2020). According to current studies, the disease is severe with signs of hypoxic respiratory failure in about 14% of all cases and requires intensive care in about 5% of cases (Wu 2020). Reported mortality ranges from 1% to 10% (Phillips 2022).

The lungs are the primary target organ of the virus, as SARS‐CoV‐2 is transmitted primarily by aerosols. Primary viral replication occurs in the bronchial epithelium of the upper respiratory tract in the nasopharynx, with further replication in the lower respiratory tract and gastrointestinal mucosa (Cheung 2020; Lin 2020). COVID‐19 can progress in three phases: early infection is followed by pulmonary manifestation and subsequently a severe hyperinflammatory phase may develop (Channappanavar 2017; Siddiqi 2020). As part of this severe, primarily local inflammation, severe damage to the lung parenchyma may occur with progressive respiratory failure, hypoxemia and cardiovascular stress reactions. As a consequence of the systemic inflammatory reaction, multiorgan failure may occur as the disease progresses (Trougakos 2021). In case of insufficient development of a humoral immune response and thus insufficient inactivation and elimination of SARS‐CoV‐2, the hyperinflammatory phase occurs with the frequent occurrence of organ failure and, in particular, the possibility of further lung damage similar to classic acute respiratory distress syndrome (ARDS; Camporota 2022).  The development of ARDS is independent of the damaging agent and can occur in pneumonia as well as in the context of single or multiple organ damage outside the lungs. In all manifestations, the Berlin criteria of the definition of ARDS may be fulfilled, but it has been shown that COVID‐19 pneumonia exhibits differences from the known changes of classic ARDS, such as those seen in the context of septic shock or bacterial pneumonia (Camporota 2022).

The severity of infection, immune response, functional capacity and concomitant diseases, ventilatory response of the patient to hypoxemia (respiratory drive), and the time between the first symptoms and the start of clinical treatment determine the severity of respiratory failure (Piraino 2019). Factors increasing the risk of more severe disease progression, resulting in longer hospital stays and higher mortality rates are: being aged 65 years or older, male, a smoker, and having comorbidities such as obesity, type 2 diabetes mellitus, heart disease, chronic kidney disease, chronic obstructive pulmonary disease (COPD, Table 2), cancer, immunodeficiency, or sickle cell anemia (Huang 2020; Karagiannidis 2020; Liang 2020; Petrilli 2020; WHO 2020; Williamson 2020). Typical complications during the course of COVID‐19‐associated ARDS (CARDS) include acute renal failure (29%), liver elevations (29%), and cardiac injury (23% to 33%) (Arentz 2020). Survivors of COVID‐19, especially patients who have developed long‐lasting CARDS, are also at high risk for long‐term physical and mental impairment, so interdisciplinary therapeutic approaches are essential (Attaway 2021).

1. Glossary and abbreviations.

Phrase/word	Definition
Acute respiratory distress syndrome (ARDS)	Acute respiratory distress syndrome (ARDS) is characterized by a massive response of the respiratory system to a wide variety of external and internal noxious stimuli. There is a disturbance of oxygen uptake and an acute onset. ARDS is the common result of a wide variety of diseases leading to a severe systemic inflammatory response. The condition should be distinguished from disturbances of respiration caused by cardiac diseases.
APACHE II	The Acute Physiology and Chronic Health Evaluation II (APACHE II) is one of the ICU scoring systems that classify disease outcomes in hospitalized patients. It is calculated by a medical professional within 24 hours of admission and provides an integer score that establishes patient prognosis and mortality risk. The score ranges from 0 to 71, with higher scores indicating greater severity of the patient’s condition. A direct connection between increases in score and increases in mortality risk has been found in both surgical and non‐surgical patients.
Adverse event	An adverse event in the context of clinical trials is an unwanted medical occurrence in patients receiving a pharmacological or nonpharmacological treatment, or both. An adverse event may not necessarily be considered to be related to the treatment.
Bias	Conscious or unconscious distortion and misinterpretation of research results, especially those obtained experimentally. The most important sources for bias are as follows. Selection bias: people are more likely to be included in the study if they have a certain characteristic (age, gender, ethnicity, social class, etc.) Information bias: the data collected as part of the study is subject to error. Publication bias: studies that show statistically significant results are published preferentially. Confounding: the result of a study is distorted by interference.
Chronic obstructive pulmonary disease (COPD)	COPD is a progressive lung disease in which chronic, incompletely reversible poor airflow (airflow limitation) and inability to breathe out fully (air trapping) exist.
Continuous positive airway pressure (CPAP)	Continuous positive airway pressure (CPAP) is a form of positive airway pressure (PAP) ventilation in which a constant level of pressure greater than atmospheric pressure is continuously applied to the upper respiratory tract of the patient.
Controlled non‐randomized study	A study in which the effects of a pharmacological or non‐pharmacological measure, or both, are compared between different groups of participants. The term 'controlled' means that the measure under investigation (intervention, verum) is compared with another measure (placebo or another intervention). The group of participants receiving the intervention under study is known as the intervention group. The group of participants who do not receive the intervention is known as the control group. A controlled non‐randomised study is easier to conduct than a randomised controlled trial, but has much less power (see bias).
COVID‐19‐related acute respiratory distress syndrome (CARDS)	The coronavirus disease of 2019 (COVID‐19) mainly involves the respiratory system, with some patients rapidly progressing to acute respiratory distress syndrome (ARDS). COVID‐19‐related acute respiratory distress syndrome (CARDS) is the leading cause of death in COVID‐19 patients.
Dichotomous	Dichotomy describes a system that can have exactly two mutually exclusive states. Example: either one has a certain disease (state A), or one does not have this disease (state B). The co‐occurrence of state A and state B is impossible.
Heart rate, acidosis, consciousness, oxygenation, and respiratory rate score (HACOR)	The HACOR score is a scale based on clinical and laboratory parameters, including heart rate, respiratory rate, acidosis (assessed by pH), consciousness (evaluated by Glasgow Coma Scale), and oxygenation (assessed by PaO2/FiO2 ratio). An elevated HACOR score, measured one hour after the start of non‐invasive ventilation treatment, is associated with an increased risk of needing intubation and increased risk of death.
Heterogeneous	Heterogeneity can be translated as 'non‐uniformity'. It is the opposite of homogeneity. In the context of meta‐analyses, heterogeneity is a measure of the comparability of clinical trials. For example, studies that examine different populations (e.g. children versus adults) have limited comparability and can lead to misleading conclusions when the data from such studies are pooled in a meta‐analysis.
Intervention	The term 'intervention' in the context of clinical trials refers to a measure of which the effect (superiority, inferiority, non‐inferiority) on a specific condition is to be assessed in comparison to other measures. An intervention need not always consist of the administration of a specific drug (so‐called non‐pharmacological interventions).
Mechanical ventilation	Mechanical ventilation is the term used to describe a procedure in which oxygen is supplied to the patient with the aid of ventilators or other devices. This measure is very restrictive and not without risk, and is therefore used only if the patient can no longer take in enough oxygen through natural breathing (spontaneous respiration). In this review, the following procedures are subsumed under the term 'mechanical ventilation'. Invasive mechanical ventilation: the patient is intubated (a breathing tube is inserted into the trachea) and ventilated by a machine, or the patient is tracheotomised (a tracheal cannula is inserted into the trachea) and ventilated by a machine.
Middle East respiratory syndrome (MERS)	Middle East respiratory syndrome (MERS) is a respiratory disease caused by a coronavirus (MERS‐CoV). Most cases of the disease are asymptomatic. Diarrhoea is a common accompanying symptom. Severe cases may present with pneumonia.
Non‐invasive mechanical ventilation (NIV)	The patient is assisted in breathing by applying pressure during exhalation and/or inhalation, for example via a tight‐fitting mask or a ventilation helmet. As a rule, the patient is awake during this process. Sensitive guidance of the patient is particularly important.
Non‐randomized studies of interventions (NRSI)	Non‐randomized studies of interventions are defined here as any quantitative study estimating the effectiveness of an intervention (harm or benefit) that does not use randomization to allocate units (individuals or clusters of individuals) to intervention groups. Such studies include those in which allocation occurs in the course of usual treatment decisions or according to peoples’ choices (i.e. studies often called 'observational').
Observational study	Data collection in a specific population under a specific research question. The essential characteristic of an observational study is that no intervention/experiment is carried out.
Randomized controlled trial (RCT)	Randomized controlled trials are the best way to obtain conclusions regarding the efficacy and effectiveness of a pharmacological or non‐pharmacological intervention, or both. The term 'controlled' means that the measure under investigation (intervention) is compared with another measure (placebo or another intervention). The term 'randomized' means that the participants in the study are randomly assigned to one of two or more prespecified treatment groups. The group of participants receiving the intervention under study is known as the intervention group. The group of participants who do not receive the intervention is known as the control group.
Severe acute respiratory syndrome (SARS)	A disease caused by SARS‐CoV, which, similar to COVID‐19, results in fever and muscle pain in combination with other flu‐like symptoms. In severe cases, atypical pneumonia may occur.
Systematic review	Scientific process of critical judgement of the data available with regard to a specific question. A 'systematic' approach is taken. This includes: formulation of a research question; systematic and comprehensive search for data (studies); clearly defined criteria that the identified studies must fulfil in order to be included in the evaluation; repeatable and uniform guidelines for data analysis. A systematic review can include a meta‐analysis, but this is not required. The aim of a systematic review is to answer the defined research question or, if this is not possible, to identify gaps in the scientific coverage of the research question.

The course of COVID‐19 can sometimes last for weeks, with an often continuous deterioration of pulmonary compliance (the ability of the lungs to stretch and expand), and increasingly severe COVID‐19 pneumonia (Gattinoni 2020). The treatment of severe courses of COVID‐19 infection is undergoing constant change. However, persistent respiratory failure often requires prolonged intubation with multiple cycles of prone positioning and neuromuscular blockade (Abate 2020; Ferrando 2020).

Recent studies suggest an incidence of CARDS in hospitalized COVID‐19 patients of 17% to 29% (Chen 2020; Yang 2020). According to Yang and colleagues, deceased critically ill COVID‐19 patients retrospectively had higher rates of CARDS (81% versus 45% in survivors) and mechanical ventilation (94% versus 35% in survivors; Yang 2020). Among COVID‐19 patients who required mechanical ventilation, the median duration of ventilation was 17 days. According to some reports, up to 81% of mechanically ventilated patients died within 28 days (Chen 2020; Yang 2020).

Studies have demonstrated a median duration of invasive ventilation of eight days and a 28‐day mortality of 34.8% for classic ARDS prior to the onset of CARDS (Bellani 2016). Follow‐up in recent studies suggests that CARDS has a worse prognosis overall than classic ARDS. From these data, it can be concluded that the course of COVID‐19 can extend well beyond 10 days, mortality in CARDS is high, and aggressive therapy may be required (Mecham 2020).

Description of the intervention

Tracheostomy describes a procedure used to create an opening in the anterior tracheal wall to provide ventilation to a patient. Prolonged ventilation, acute respiratory failure and weaning failure from the ventilator are among the most important indications for the placement of a tracheostoma in critical care medicine (Abe 2018; Blot 2005). Furthermore, tracheostomy facilitates the nursing management of the patient. For example, oral hygiene is simplified, mobility is increased, and verbal communication and oral feeding are enabled (Jaeger 2002; Nieszkowska 2005). In addition, because a subglottic tracheostomy tube does not trigger a gag reflex, the need for analgesics and sedatives is reduced (Nieszkowska 2005). A less sedated patient can therefore be more active and achieve greater autonomy (Nieszkowska 2005). The design of a tracheostomy tube offers physical advantages because it has a larger diameter and is shorter than an endotracheal tube, which reduces airway resistance and thus the work of breathing (Davis 1999). In addition, modern, properly adjusted ventilators sufficiently compensate for the resistance caused by the tube or cannula, so that the clinical relevance of this point has been lost (Elsasser 2003). Other, less common indications for tracheostomy include greatly increased and unmanageable secretion production and upper airway obstruction (Clec'h 2007). Overall, an increase in the number of patients with acute respiratory failure requiring prolonged ventilation has been observed for some time (Mauri 2008). Carson 2008 found that approximately 10% of ventilated patients required prolonged ventilation, and that prolonged ventilation is also associated with certain risks. The most common complications of prolonged ventilation include ventilator‐induced lung injury, ventilator‐associated pneumonia, and an increased need for intensive care and prolonged hospitalization (El‐Khatib 2008).

Facing the pandemic caused by the novel virus and the risk to medical staff, it was necessary to re‐evaluate the previous recommendations. The previously known indications for performing tracheostomy in people with ARDS are largely transferable to CARDS, but in COVID‐19 patients, the question arises as to the expected course of disease with the presumed higher mortality rate in ventilator‐dependent patients and the associated likelihood of successful weaning from the ventilator (Martin‐Villares 2020). Although the benefits of tracheostomy are well studied, the indication for the procedure must be considered in light of this potentially poor prognosis, a limited work environment, and risk to medical personnel. Tracheostomy is a procedure that generates aerosols, which, in turn, can pose a high risk of infection to medical personnel through droplet transmission. Therefore, careful consideration was needed at the onset of the pandemic when indications were made for tracheostomy in COVID‐19 patients (Mata‐Castro 2021).

There are complications described that may be associated with the performance of tracheostomy, regardless of the technique used (open surgical and percutaneous dilative). Early complications occur peri‐interventionally, within 24 hours of starting the procedure. The most commonly occurring adverse events are bleeding, vital sign abnormalities, difficulty inserting the tracheostomy tube, and pneumothorax (Davis 1999; Massick 2001). Later complications during or after the intensive care stay include tracheal cannula dislocation, bleeding, tracheal obstruction, or stoma infection (Massick 2001). Patients with a history of tracheostomy, difficult neck anatomy, or coagulopathies are among the contraindications to a percutaneous technique (Durbin 2005).

Immediately prior to the 2019 pandemic outbreak, the network meta‐analysis by Iftikhar 2019 showed that all tracheostomy techniques were comparable in terms of the complications associated with the procedure that occurred, but dilatation techniques require significantly less time, and therefore, with consideration of contraindications, should generally be preferred.

In Bier‐Laning 2021's review of perioperative protocols and practices of tracheostomies during the COVID‐19 pandemic, many protocols did not mention contraindications to performing tracheostomy in COVID‐19 patients, but some protocols included the recommendation to postpone tracheostomy in case of a positive SARS‐CoV‐2 test, as the high risk of infection for the medical staff involved could be considered an indirect complication of the procedure (Bier‐Laning 2021).

How the intervention might work

Although there have been numerous retrospective studies, some prospective randomized trials, and meta‐analyses, there is still no consensus on the optimal timing for tracheostomy, as some pre‐pandemic studies had already led to conflicting results. In most cases, the prospective studies were not very conclusive because of small numbers of participants. There have been few methodologically robust randomized controlled trials (RCTs) on this topic, most of which were not multicenter, included few participants, and were conducted exclusively in the USA (Durbin 2005; Hazard 1991; Massick 2001; Oliver 2007; Rumbak 2004).

A previous systematic review and meta‐analysis on this topic assessed eight randomized trials of the timing of tracheostomy in critically ill patients before the COVID‐19 pandemic. It was shown that early tracheostomy (< 10 days after intubation) is associated with lower mortality (relative risk 0.83, 95% confidence interval (CI) 0.70 to 0.98; P = 0.03; 1903 participants) and shorter ICU treatment duration (relative risk 1.29, 95% CI 1.08 to 1.55; P = 0.006; 1903 participants; Andriolo 2015). Based on this evidence, recommendations for performing a tracheostomy 7 to 10 days after intubation in non‐COVID‐19 patients were generated in many institutions, and we have also prioritized these timings in our review (Andriolo 2015). However, even in current guidelines, there is neither a consensus on the optimal timing for tracheostomy, nor a universal definition of the timing of early or late tracheostomy (Geiseler 2021; Schönhofer 2019).

Bier‐Laning's comparative document analysis for the timing of tracheostomy in critically ill COVID‐19 patients also was unable to provide a consistent recommendation (Bier‐Laning 2021). Some countries (Israel, Spain, Brazil, the Netherlands) recommend performing an early tracheostomy, sometimes within three days after intubation. Other recommendations ‐ most notably in the USA ‐ recommend a longer waiting period, in some cases depending on negative COVID‐19 test results, even up to 21 days after intubation (Bier‐Laning 2021). As the pandemic has progressed, many hospitals have changed their standards and are generally moving to a more conventional ‐ earlier ‐ timing for tracheostomy (Bier‐Laning 2021).

In summary, the advantages of tracheostomy in COVID‐19 patients should outweigh the disadvantages of tracheostomy in COVID‐19 patients.

Why it is important to do this review

During the course of the COVID‐19 pandemic, the number of infections worldwide increased and the number of patients with severe infection requiring invasive mechanical ventilation also steadily increased. The treatment of COVID‐19 patients is subject to constant change, but often requires intensive respiratory therapy, with consecutive intubation and prolonged ventilation, in the setting of severe respiratory failure. Prior to the outbreak of the pandemic, evidence supported that placement of a tracheostomy has been shown to improve the process of weaning from mechanical ventilation in patients requiring prolonged mechanical ventilation. However, there was no consensus on the optimal timing for tracheostomy, neither before the pandemic (Liu 2015), nor since its onset (Bier‐Laning 2021).

Andriolo 2015 stated that the results of his systematic review, "only indicate the superiority of early versus late tracheostomy, as no high‐quality information was available for specific subtypes". Therefore, we aimed to assess whether this group of critically ill COVID‐19 patients and the high‐quality RCT provided better information to answer the question.

There is a clear and urgent need for more evidence‐based information to guide clinical decision‐making for COVID‐19 patients. This systematic review will fill current gaps by identifying, describing, evaluating, and synthesizing all evidence for early tracheostomy on clinical outcomes in COVID‐19. There is a need for a thorough understanding and an extensive review of the current body of evidence regarding early tracheostomy for COVID‐19 patients. The primary goal of this review is to provide practicing clinicians, healthcare providers, and interested laypeople with reliable and evidence‐based information that will lead to improvement in the treatment of COVID‐19.

Objectives

To assess the benefits and harms of early tracheostomy compared to late tracheostomy in critically ill COVID‐19 patients.

Methods

Criteria for considering studies for this review

Types of studies

The description of methods is based on on a template from the Cochrane Haematology working group in accordance with Cochrane Reviews investigating treatments and therapies for COVID‐19. Parts of the review's methods section are adapted from templates from Cochrane Haematology and a protocol published by Ansems 2021, Piechotta 2020, and the corresponding review (Chai 2020). However, specific adjustments were made where necessary in relation to the research question. The protocol for this review was registered with PROSPERO on 10 May 2021 (Dahms 2021).

To assess the effects of early tracheostomy compared to late tracheostomy in critically ill COVID‐19 patients, we included RCTs, as this study design provides the best evidence for interventional therapies in highly controlled therapeutic settings. We used the recommended methods in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022a). We did not find more than two RCTs, so we also included non‐randomized studies of interventions (NRSIs) according to the methods in the Cochrane Handbook for Systematic Reviews of Interventions (Reeves 2022). We did not include non‐standard RCT designs such as cluster‐randomized trials or cross‐over trials (Higgins 2022b), or quasi‐randomized trials, such as those assigned by alternation (Reeves 2022), as explained in the Differences between protocol and review section.

We included the following formats, if sufficient information was available on study design, participant characteristics, interventions, and outcomes.

• Full‐text publications

• Preprint articles

• Abstract publications

• Results published in study registries

• Personal communication with study authors

We included preprints and conference abstracts to provide a complete overview of ongoing research activity, especially to track emerging studies of tracheostomy in critically ill patients with COVID‐19. We did not apply any restriction to the length of follow‐up.

Types of participants

We included adults with a confirmed diagnosis of COVID‐19 (as described in the study) and did not exclude study participants on the basis of sex, ethnicity, severity of illness, or setting. All participants were admitted to the intensive care unit due to respiratory failure, were mechanically ventilated, and underwent surgical or percutaneous tracheostomy.

We excluded studies that examined the timing of tracheostomy in the context of treatment for other coronavirus diseases such as severe acute respiratory syndrome (SARS) or Middle East respiratory syndrome (MERS) or other viral diseases. When studies included populations with mixed viral diseases or those exposed to them, we would include them only if the study authors provided clear subgroup data for patients with a confirmed COVID‐19 diagnosis.

Types of interventions

We included the following interventions.

• Early tracheostomy, when no serious attempt has been made to wean the patient off the ventilator (on the average of the reviews and meta‐analyses of ARDS available to date, a tracheostomy is defined as early if it is performed two to 10 days after intubation, based on clinical or laboratory findings).

• Late tracheostomy when attempts to wean from the ventilator have not been successful to date (on the average of the reviews and meta‐analyses of ARDS available to date, a tracheostomy is defined as late if it is performed 10 days or later after intubation, based on clinical or laboratory findings).

Types of outcome measures

We defined outcome sets with primary and secondary outcomes for early versus late tracheostomy in critically ill COVID‐19 patients for two study types:

• RCTs

• NRSIs

The core outcomes were defined in a consensus conference with methodologists and clinicians in accordance with the Core Outcome Measures in Effectiveness Trials (COMET 2021), initiative for COVID‐19 patients (Dinglas 2020). We added other relevant outcomes prioritized by consumer representatives and the German guideline panel for hospitalized patients with SARS‐CoV‐2 infection (Malin 2021). All adjustments of the predefined outcomes are explained in the Differences between protocol and review section. Primary outcomes, critical to this review are in bold.

• Mortality:

  • overall mortality

  • in‐hospital mortality

  • at up to day 28 (± 2)

  • at day 60

  • at day 90

  • time to event

• Improvement of clinical status:

  • duration to liberation from invasive mechanical ventilation

  • need for invasive mechanical ventilation

  • liberation from invasive mechanical ventilation

  • ventilator‐free days

  • duration to decannulation

• Worsening of clinical status:

  • adverse events (any grade)

  • ventilator‐associated pneumonia

  • need for renal replacement therapy

  • postoperative bleeding

  • airway obstruction

  • tracheal stenosis

  • need for extracorporeal membrane oxygenation (ECMO)

  • ventilatory problems

  • serious adverse events

• ICU length of stay, or time to discharge from ICU

• Hospital length of stay, or time to discharge from hospital

• Quality of life

• Viral clearance

The outcome 'duration to liberation from mechanical ventilation' was a continuous outcome measuring how many days a participant needed invasive mechanical ventilation. The outcome 'liberation from invasive mechanical ventilation' was a dichotomous outcome measuring how many participants needed invasive mechanical ventilation.

Table 1 includes only primary outcomes reported in the RCT.

Search methods for identification of studies

Electronic searches

Our Information Specialist (IM) conducted six systematic searches in the following sources, from the inception of each database to 14 June 2022 (date of last search for all databases), placing no restrictions on language or population size.

• Cochrane COVID‐19 Study Register (CCSR) (inception to 14 June 2022; www.covid-19.cochrane.org) comprising:

  • Cochrane Central Register of Controlled Trials (CENTRAL), monthly updates;

  • PubMed, daily updates;

  • Embase.com, weekly updates;

  • ClinicalTrials.gov (www.ClinicalTrials.gov), daily updates;

  • World Health Organization International Clinical Trials Registry Platform (WHO ICTRP) (https://trialsearch.who.int/), weekly updates;

  • medRxiv (www.medrxiv.org), weekly updates.

• Web of Science Clarivate (inception to 14 June 2022)

  • Science Citation Index Expanded (1945 to present);

  • Emerging Sources Citation Index (2015 to present).

• World Health Organization (WHO) COVID‐19 Global literature on coronavirus disease (inception to 14 June 2022; search.bvsalud.org/global-literature-on-novel-coronavirus-2019-ncov).

For detailed search strategies, see Appendix 1

Searching other resources

We further searched for potentially eligible studies or supplementary publications by screening the reference lists of included studies, systematic reviews, and meta‐analyses. We also contacted the authors of the included studies to obtain additional information on the retrieved studies.

We performed a search for gray literature by searching trials registries such as ClinicalTrials.gov and the WHO ICTRP (https://www.who.int/clinical-trials-registry-platform), which is included in the CCSR, as well as searching preprint servers and gray literature indices in the CCSR and WHO COVID‐19 Global Literature on Coronavirus Diseases. After identifying the included studies, we searched for preprints via Europe PubMed Central to check if preprints for included studies had been published since our database search.

Data collection and analysis

Selection of studies

Three authors (AS, KD, KA) independently screened titles and abstracts from the results of the search for eligibility using Covidence software. The abstracts were coded as either 'include' or 'exclude'. If there were discrepancies in the judgments or if the abstract was not clearly sufficient for categorization, we retrieved the full‐text publication for further discussion. Two of three review authors (AS, KD, KA) then assessed the full‐text articles of the selected studies. If two review authors could not reach consensus, they consulted a third review author to make a final decision.

We documented the process of study selection in a flow chart, as recommended in the PRISMA statement showing the total number of retrieved references and the number of included and excluded studies (Liberati 2009; Moher 2009). We listed all excluded studies after full‐text assessment and the reasons for their exclusion in the Excluded studies section.

Data extraction and management

We extracted data according to Cochrane guidelines (Li 2022). Two review authors (AS, KD, KA) extracted data independently and in duplicate using a customized data extraction form developed in Microsoft Excel (Mircosoft 2018). We resolved any discrepancies through discussion or, if necessary, consultation with a third review author.

Three review authors (AS, KD, KA) independently assessed included studies for methodological quality and risk of bias. If two review authors were unable to reach consensus, we consulted a third review author. Where reported, we extracted the following information.

• General information: author, title, source, publication date, country, language, duplicate publications

• Study characteristics: trial design, setting, dates, source of participants, inclusion/exclusion criteria, comparability of groups, treatment cross‐overs, compliance with assigned treatment, length of follow‐up

• Participant characteristics: age, gender, number of participants recruited/allocated/evaluated, additional diagnoses, severity of disease, previous treatments, concurrent treatments, comorbidities (e.g. diabetes, respiratory disease, hypertension, immunosuppression, obesity, heart failure)

• Interventions: early tracheostomy, timing, technique, setting, duration of follow‐up

• Control interventions: late tracheostomy, timing, technique, setting, duration of follow‐up

• Outcomes: as specified in Types of outcome measures section

• Risk of bias assessment: randomization process, deviations from the intended interventions, missing outcome data, measurement of the outcome, selection of the reported result

• Notes: funding for trial and notable conflicts of interest of study authors

Assessment of risk of bias in included studies

We used the RoB 2 tool (beta version 7) to assess the risk of bias of the included RCT (Sterne 2019). Of interest in this review was the effect of the assignment to the intervention (the intention‐to‐treat effect) as it is considered the gold standard for RCTs and includes data from all participants initially randomized. Thus, we performed all assessments with RoB 2 on this effect. The outcomes we assessed are those specified for inclusion, as described in Types of outcome measures.

Two review authors (from AS, KD, KA, NS, TB, CB) independently assessed risk of bias for each outcome. When discrepancies arose between the assessments and the two authors could not reach consensus, they consulted a third review author to make a final decision. We assessed the following types of bias, as described in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022c).

• Bias arising from the randomization process

• Bias due to deviations from the intended interventions

• Bias due to missing outcome data

• Bias in measurement of the outcome

• Bias in selection of the reported result

To address these types of biases, we made judgments according to signaling questions recommended in RoB 2 and graded according to the following options.

• 'Yes': if there is firm evidence that the question is fulfilled in the study (i.e. the study is at low or high risk of bias given the direction of the question).

• 'Probably yes': a judgment has been made that the question is fulfilled in the study (i.e. the study is at low or high risk of bias given the direction of the question).

• 'No': if there is firm evidence that the question is unfulfilled in the study (i.e. the study is at low or high risk of bias given the direction of the question).

• 'Probably no': a judgment has been made that the question is unfulfilled in the study (i.e. the study is at low or high risk of bias given the direction of the question).

• 'No information': if the study report does not provide sufficient information to permit a judgment.

For RCTs, we used the algorithms proposed by RoB 2 to assign each domain to one of the following levels of bias.

• Low risk of bias

• Somewhat concerning

• High risk of bias

In accordance with the following suggestions, we assessed an overall risk of bias rating for each outcome we evaluated in each included study.

• 'Low risk of bias': we judged the trial to be at low risk of bias for all domains for the result.

• 'Some concerns': we judged the trial to raise some concerns in at least one domain for the result, but not to be at high risk of bias for any domain.

• 'High risk of bias': we judged the trial to be at high risk of bias in at least one domain for the result, or we judged the trial to have some concerns for multiple domains.

To implement RoB 2, we used the RoB 2 Excel tool (beta version 7, available at riskofbias.info). Details of the RoB 2 assessments are described in the Results section (Risk of bias in included studies) and presented as a supplemental table (zenodo.org/records/7895589).

For NRSIs, we assessed risk of bias according to the modified version of the Risk Of Bias In Non‐randomized Studies of Interventions (ROBINS‐I) tool as recommended for Cochrane Reviews (Schünemann 2019). The outcomes we assessed were those specified for inclusion, as described in Types of outcome measures.

Two review authors (from AS, KD, KA, NS, TB, CB) independently assessed the risk of bias for each outcome of all included studies. We discussed any discrepancies and resolved them by consensus. Otherwise, we consulted a third review author to make a final decision. We assessed the following types of bias as outlined in Chapter 25 of the Cochrane Handbook for Systematic Reviews of Interventions (Sterne 2022).

• Bias due to confounding

• Bias in selection of participants into the study

• Bias in classification of interventions

• Bias due to deviations from intended interventions

• Bias due to missing data

• Bias in measurement of outcomes

• Bias in selection of the reported result

For NRSIs, we used the algorithms proposed in Chapter 25 of the Cochrane Handbook for Systematic Reviews of Interventions according to the ROBINS‐I tool to assign each domain one of the following levels of bias (Sterne 2022). We subsequently derived an overall risk of bias rating for each prespecified outcome in each study in accordance with the suggestions for a risk of bias judgment listed in the Cochrane Handbook for Systematic Reviews of Interventions (Sterne 2022). We used the algorithms proposed to assign each domain one of the following levels of bias.

• Low risk of bias

• Moderate risk of bias

• Serious risk of bias

• Critical risk of bias

We subsequently derived an overall risk of bias rating for each prespecified outcome in each study in accordance with the following suggestions.

• 'Low risk of bias': we judged the trial to be at low risk of bias for all domains for this result and comparable to a well‐performed randomized trial.

• 'Moderate risk of bias': we judged the trial to be at low or moderate risk of bias for all domains. The study appears to provide sound evidence for a non‐randomized study but cannot be considered comparable to a well‐performed randomized trial.

• 'Serious risk of bias': we judged the trial to be at serious risk of bias in at least one domain, not at critical risk of bias in any domain, but the study has one or more important problems.

• 'Critical risk of bias': we judged the trial to be at critical risk of bias in at least one domain. The study is too problematic to provide any useful evidence and should not be included in any synthesis.

We used the ROBINS‐I tool to implement ROBINS‐I (available at riskofbias.info). ROBINS‐I assessments are described in the Results section (Risk of bias in included studies) and presented as additiontal Table 3 and Table 4.

2. Overview risk of bias ‐ ROBINS‐I (≤ 10 days vs > 10 days after intubation).

Reference	Outcome	ROBINS‐I Domains
		Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from intended interventions	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall bias
Avilés‐Jurado 2020 Prospective cohort study	Duration to liberation from IMV	Moderatea,b	Criticalc	Seriousd,e	Low	Low	Seriousf	Low	Critical
	Duration to decannulation	Moderatea,b	Criticalc	Seriousd,e	Low	Low	Seriousf	Low	Critical
	Postoperative bleeding	Moderatea,b	Criticalc	Moderated	Low	Low	Seriousf	Low	Critical
	Ventilatory problems	Moderatea,b	Criticalc	Moderated	Low	Low	Seriousf	Low	Critical
Chandran 2021 Prospective observational study	Mortality	Criticala,g	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
Evrard 2021 Retrospective cohort study	Tracheal stenosis	Criticala,g	Criticalc	Moderated	Low	Low	Low	Low	Critical
	ICU LoS	Criticala,g	Criticalc	Moderated	Low	Low	Low	Low	Critical
Kwak 2021 Retrospective cohort study	Mortality	Moderatea,b	Criticalc	Moderated	Low	Low	Low	Low	Critical
	Liberation from IMV	Moderatea,b	Criticalc	Moderated	Low	Low	Low	Low	Critical
Polok 2021 Prospective observational study	Mortality	Moderatea,b	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
	ICU LoS	Moderatea,b	Criticalc	Moderated	Low	Low	Low	Low	Critical
Prats–Uribe 2021 Prospective multicentre cohort study	Mortality	Moderatea,b	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
Volo 2021 Retrospective cohort study	Mortality	Moderatea,b	Criticalc	Moderated	Low	Low	Low	Low	Critical
	ICU LoS	Moderatea,b	Criticalc	Moderated	Low	Low	Low	Low	Critical
ECMO: extracorporeal membrane oxygenation; ICU: intensive care unit; MV: mechanical ventilation; LoS: length of stay; SAE: serious adverse event

^aDue to possible confounders.
^bRegression modeling was used to control for confounding domains.
^cSelection of participants into the study (or into the analysis) could have been based on participant characteristics observed after the start of intervention
^dNo protocol/trial registry available.
^eClassification of intervention status could have been affected by knowledge of the outcome or risk of the outcome.
^fThe outcome measure could have been influenced by knowledge of the intervention received.
^gNo appropriate analysis method was used to control for all the important confounding domains.
^h Competing risk model was used to control for confounding domains.

3. Overview risk of bias ‐ ROBINS‐I (≤ 14 days vs > 14 days after intubation).

Reference	Outcome	Risk of bias domains
		Bias due to confounding	Bias in selection of participants into the study	Bias in classification of interventions	Bias due to deviations from intended interventions	Bias due to missing data	Bias in measurement of outcomes	Bias in selection of the reported result	Overall bias
Angel 2021 Prospective cohort study	Mortality	Moderatea,b	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
	Liberation from IMV	Moderatea,b	Criticalc	Seriousd	Low	Low	Seriousf	Low	Critical
Arnold 2022 Retrospective observational study	Mortality	Moderatea,g	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
	Liberation from IMV	Moderatea,g	Criticalc	Moderated	Low	Low	Low	Low	Critical
	Postoperative bleeding	Moderatea,g	Criticalc	Moderated	Low	Low	Low	Low	Critical
	Need for renal replacement	Moderatea,g	Criticalc	Moderated	Low	Low	Low	Low	Critical
	Need for ECMO	Moderatea,g	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
	Length of stay	Moderatea,g	Criticalc	Moderated	Low	Low	Low	Low	Critical
Battaglini 2021 Retrospective multicentre study	Mortality	Moderatea,g	Criticalc	Low	Low	Low	Low	Low	Critical
	Ventilator‐associated pneumonia	Moderatea,g	Criticalc	Moderated	Low	Low	Low	Low	Critical
Breik 2020 Prospective observational study	Mortality	Criticala,h	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
	Duration to liberation from IMV	Criticala,h	Criticalc	Seriousd,e	Low	Criticali	Seriousf	Low	Critical
Dal 2022 Retrospective observational study	Mortality	Moderatea,g	Criticalc	Low	Low	Low	Low	Low	Critical
Glibbery 2020 Prospective institutional study	Duration to liberation from IMV	Criticala,h	Criticalc	Seriousd,e	Low	Criticali	Seriousf	Low	Critical
Hernandez 2022 Retrospective propensity‐matched cohort study	Ventilator‐associated pneumonia	Moderatea,b	Criticalc	Moderate	Low	Low	Low	Low	Critical
Kuno 2021 Retrospective study	Mortality	Moderatea	Criticalc	Moderate	Low	Low	Low	Low	Critical
Mahmood 2021 Retrospective multicenter study	Mortality	Criticala,h	Criticalc	Moderate	Low	Low	Low	Low	Critical
	Liberation from IMV	Criticala,h	Criticalc	Moderate	Low	Low	Low	Low	Critical
	Ventilator‐associated pneumonia	Criticala,h	Criticalc	Moderate	Low	Low	Low	Low	Critical
Takhar 2020 Prospective single‐center observational study	Mortality	Criticala,h	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
	Duration to liberation from IMV	Criticala,h	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
	Liberation from IMV	Criticala,h	Criticalc	Seriousd,e	Low	Low	Low	Low	Critical
Tang 2020 Retrospective multicenter observational study	Mortality	Moderatea,g	Criticalc	Moderate	Low	Low	Low	Low	Critical
	Postoperative bleeding	Moderatea,g	Criticalc	Moderate	Low	Low	Low	Low	Critical
	Need for ECMO	Moderatea,g	Criticalc	Moderate	Low	Low	Low	Low	Critical
ECMO: extracorporeal membrane oxygenation; ICU: intensive care unit; IMV: invasive mechanical ventilation

^aDue to possible confounders.
^bPropensity score matching approach was used to control for confounding domains.
^cSelection of participants into the study (or into the analysis) could have been based on participant characteristics observed after the start of intervention.
^dNo protocol/trial registry available.
^eClassification of intervention status could have been affected by knowledge of the outcome or risk of the outcome.
^fThe outcome measure could have been influenced by knowledge of the intervention received.
^gRegression modeling was used to control for confounding domains.
^hNo appropriate analysis method was used to control for all the important confounding domains.
ⁱDue to missing outcome data.

For the domain 'Bias due to missing outcome data', we considered death as a competing risk factor, particularly for dichotomous clinical progression outcomes. Improvement may have a high risk of bias due to missing data because death during follow‐up may have influenced or impeded relief from ventilatory support and therefore missing data on improvement depend on its true value.

Measures of treatment effect

For continuous outcomes, we recorded the total number of participants in the treatment and control groups, the mean and standard deviation (SD). If continuous outcomes were measured on the same scale, we performed analyses using the mean difference (MD) with 95% confidence intervals (CIs). For continuous outcomes measured with different scales, we used standardized mean difference (SMD). In interpreting SMDs, we re‐expressed SMDs in the original units of a particular scale with the most clinical relevance and impact (e.g. clinical symptoms using the WHO Clinical Progression Scale (WHO 2020)).

For dichotomous outcomes, we recorded the number of events and the total number of participants in both the treatment and control groups. We reported the pooled risk ratio (RR) with its associated 95% CI, and risk difference (RD) with its associated 95% CI (Deeks 2022).

If adequate information was available, we extracted hazard ratios (HRs) and reported them for time to event (e.g. time to hospital discharge; Skoetz 2020). If HRs were not available, we made every effort to estimate HRs as accurately as possible from available data using methods suggested by Parmar and Tierney (Parmar 1998; Tierney 2007). If a sufficient number of studies provided HRs, we used HRs instead of RRs or MDs in a meta‐analysis as they provide more information.

Unit of analysis issues

The aim of this review was to summarize studies that analyzed data at the individual level. We based the unit of analysis on the individual participant (unit randomly assigned to the interventions being compared) (Li 2022). We summarized multiple reports of a given study so that each study and not each report was the unit of analysis.

Dealing with missing data

Chapter 10 of the Cochrane Handbook for Systematic Reviews of Interventions discusses a variety of potential sources of missing data that we have considered: at study level, at outcome level, and at summary data level (Deeks 2022). At all levels, it is important to distinguish between 'incidental missing' data, which are often unbiased, and 'nonincidental missing' data, which may bias the study and thus the results of the review.

In the case of missing data, we requested this information from the study authors; details are provided in the Included studies section.

Assessment of heterogeneity

We used the Chi² test to quantify the heterogeneity of treatment effects between studies with a significance level of P < 0.1. We used the I² statistic (Higgins 2002), and visualization of the forest plots to assess potential heterogeneity (I² > 30% indicates moderate heterogeneity, I² > 75% indicates substantial heterogeneity; Deeks 2022).

If the I² statistic value was above 80%, we planned to analyze possible causes of heterogeneity through sensitivity analyses. If we could not find a reason for heterogeneity, we decided not to pool studies in a meta‐analysis, but would comment on the results of all studies and present them in tables.

Assessment of reporting biases

We searched trials registries to identify completed studies that had not yet been published elsewhere to minimize publication bias or to determine whether publication bias was present.

We planned to assess publication bias or a systematic difference between smaller and larger studies (small‐study effects) by creating a funnel plot (study effect versus study size) and statistically testing it with a linear regression test for meta‐analyses, if a sufficient number of studies and results for the same outcomes were available (at least 10 trials; Copas 2000; Sterne 2022).

Data synthesis

If the clinical and methodological characteristics of individual studies were sufficiently homogeneous, we pooled the data in a meta‐analysis. We planned to conduct analyses for all eligible studies, however conclusions would only be based on the results of studies with a low risk of bias and some concerns. We performed the analyses according to the recommendations in the Cochrane Handbook for Systematic Reviews of Interventions (Deeks 2022).

We used Review Manager Web software (RevMan Web 2021) for the analyses. One review author entered the data into the software, and a second review author checked them for accuracy (AS, KD, KA). We planned to use the random‐effects model to meta‐analyze quantitative data, because of substantial clinical and methodological heterogeneity among studies, which in itself could generate substantial statistical heterogeneity. If data from primary studies were nonparametric (e.g. effects were reported as medians, quartiles, etc.) or reported without sufficient statistical information (e.g. standard deviations, number of participants, etc.) and meta‐analysis was not possible, we planned to describe and comment on them narratively with the results of all studies. In addition, we presented any clinically relevant estimate of effect in the overview of included studies table (Table 5).

4. Overview of included non‐randomized studies of interventions ‐ Studies A ‐ G.

	Angel 2021	Arnold 2022	Avilés‐Jurado 2020	Battaglini 2021	Breik 2020	Chandran 2021	Dal 2022	Evrard 2021	Glibbery 2020
Setting	Inpatient New York, USA	Inpatient Illinois, USA	Inpatient Spain	Inpatient Italy	Inpatient UK	Inpatient India	Inpatient Turkey	Inpatient France	Inpatient UK
Design	Mulitcenter, prospective cohort study	Prospective observational cohort study	Prospective cohort study	Multicenter retrospective observational study	Prospective observational cohort study	Prospective observational cohort study	Retrospective observational study	Multicenter retrospective cohort study	Prospective institutional review
Study protocol	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
Statistical analysis plan	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
Allocated participants (n)	394	72	50	153	85	51	33	48	28
Number of participants per trial arm (allocated/evaluated)	Early: 116 Late: 89	Early: 14 Late: 58	Early: 32 Late: 18	Early: 76 Late: 77	Early: 56 Late: 29	Early: 32 Late: 19	Early: 18 Late: 15	Early: 10 Late: 38	Early: 9 Late: 19
Recruitment dates	11 March to 29 April 2020	March 2020 to April 2021	March 16 to April 10, 2020	20 April to 30 June 2020	9 March 2020 to 21 April 2020	1 April 2020 to 31 January 2021	1 November 2020 and February 1, 2021	January 27 to May 18, 2020	15 March to 20 May 2020
Time of follow‐up	During hospitalization + 1‐2 months after discharge	Median length of follow‐up 45 (IQR 16 – 135) days	From the time of ICU admission until death/ withdrawal of IMV/ ICU discharge/ end of the study (8 May 2020)	Not reported	Not reported	Follow‐up ranged from 15 days to 12 months with a median follow‐up period of 7.5 months	Follow‐up as intubated in the ICU	Mean follow‐up time 277 (range 42–532) days	Mean follow‐up time: 57.7 days (11.7), total follow‐up time: 74.7 days (11.6).
Intervention
Intervention (early tracheostomy)(days)	Early tracheostomy (≤ 13 days)	Early tracheostomy (< 14 days)	Early tracheostomy (< 10 days)	Early tracheostomy (< 15 days)	Early tracheostomy (< 14 days)	Early tracheostomy (≤ 10 days)	Early tracheostomy (< 14 days)	Early tracheostomy (≤ 10 days)	Early tracheostoy (≤ 14 days)
Control (late tracheostomy) (days)	Late tracheostomy (> 13 days)	Late tracheostomy (> 14 days)	Late tracheostomy (> 10 days)	Late tracheostomy (≥ 15 days)	Late tracheostomy (> 14 days)	Late tracheostomy (> 10 days)	Late tracheostomy (> 14 days)	Late tracheostomy (> 10 days)	Late tracheostomy (> 14 days)
Technique	Percutaneous dilational tracheostomy	Percutaneous bedside tracheostomy	Bedside open tracheostomy	Percutaneous or surgical tracheostomy	Percutaneous or surgical tracheostomy	Open surgical tracheostomy	Bedside percutaneous dilatation tracheostomy	Percutaneous or surgical tracheostomy	Percutaneous or surgical tracheostomy
Subgroups	NA	NA	NA	NA	NA	NA	NA	NA	NA
Demographics
Age (years)	Median (IQR) Early: 59 (46–67) Late: 64 (55–70)	Median (IQR) 66 (61‐71)	Mean (SD) Early: 62.2 (11.6) Late: 64.53 (8.2)	Mean (SD) Early: 63.8 (9.24) Late: 62.9 (9.48)	NA	Median (range) 52 (23‐83)	Mean (SD) Early: 62.28 (12.84) Late: 68.80 (14.920)	Median (IQR) Early: 52 (48‐68) Late: 57 (46‐64)	Mean 60.5 (range 25‐82)
Gender (male (n(%)))	Early: 93 (80) Late: (71)	Total: 51 (71)	Early: 22 (68.7) Late: 11 (61.1)	Early: 60 (78.9) Late: 58 (75.3)	NA	Total: 32 (62.74)	Early: 16 (88.9) Late: 9 (60.0)	Early: 8 (80) Late: 28 (74)	Total: 20 (71.4)
Comorbidities at baseline (n (%))
Diabetes	Early: 33 (28) Late: 31 (35)	Total: 32 (54%)	Early: 8 (25) Late: 1 (5.6)	Early: 20 (26.3) Late: 14 (18.2)	NA	Total: 15 (29.41)	Early: 3 (16.7) Late: 4 (26.7)	Early: 3 (30) Late: 11 (29)	Total: 8 (28.6)
Hypertension	NA	Total: 36 (61%)	Early: 19 (59.4) Late: 8 (44.4)	Early: 42 (55.3) Late: 40 (51.9)	NA	Total: 17 (33.33)	Early: 9 (50) Late: 9 (60)	Early: 2 (20) Late: 21 (55)	Total: 12 (42.9)
Cardiac disease	Hyperlipidemia, hypertension, coronary disease, heart failure Early: 61 (53) Late: 49 (55)	Coronary artery disease Total: 14 (19) Heart failure Total: 10 (14)	NA	Early: 12 (15.8) Late: 11 (14.3)	NA	Cardiovascular disease Total: 4 (7.8)	Cardiovascular disease Early: 2 (11.1) Late: 4 (26.7)	Early: 2 (20) Late: 4 (11)	Total: 4 (14.3)
Respiratory disease	Asthma or COPD Early: 5 (4) Late: 0	Total: 5 (7%)	NA	Early: 11 (14.7) Late: 5 (6.5)	NA	Total: 4 (7.8)	NA	Early: 0 Late: 6 (16)	Total: 1 (3.6)
Asthma	NA	NA	NA	NA	NA	NA	Early: 1 (5.6) Late: 0 (0.0)	NA	NA
COPD	NA	NA	Early: 5 (15.6) Late: 4 (22.2)	NA	NA	NA	Early: 1 (5.6) Late: 1 (6.7)	NA	total: 2 (7.1)
Obesity (BMI ≥ 30 kg/m2)	Early: 38 (33) Late: 28 (31)	Total: 24 (41%)	NA	NA	NA	NA	NA	Early: 3 (30) Late: 18 (47)	Total: 12 (42.9)
Immunosuppression therapy	NA	NA	Early: 2 (6.3) Late: 2 (11.1)	NA	NA	NA	NA	Early: 4 (40) Late: 2 (5)	Total: 1 (3.6)
SOFA score, mean (SD)	NA	NA	Early: 6.3 (2.1) Late: 6 (2.5)	NA	NA	NA	Median Early: 5.5 (2.0–7.0) Late: 4.0 (2.0–6.0)	Median (IQR) Early (n=10): 3 (2–7) Late (n=32): 4 (2–4)	NA
Outcomes
Primary outcomes	Discontinuation from MV Length of hospitalization Overall survival	Overall mortality Decannulation rates	Infections among the surgeons Short‐term complications Weaning Total required days of invasive ventilation	Median time to tracheostomy	30‐day survival	30‐day mortality rate following tracheostomy Association with various prognostic risk factors	Mortality ICU Length of stay Duration of MV	Patient outcome: Date of first COVID‐19 symptoms and PCR results Level of respiratory support (oxygen therapy, MV support) ICU and hospital outcomes (length of MV, length of stay at hospital, vital status) MV and tracheostomy complications post‐ICU (after 6 months) Procedure outcome: Timing of tracheostomy Length of procedure Complications	Post‐tracheostomy outcomes: Weaning from intravenous sedation Weaning from mechanical ventilation Successful decannulation Intensive care unit discharge to a general ward Hospital discharge Complications (return to the operating theatre, failed decannulation, intensive care unit re‐admission and death)
Secondary outcomes	NA	Time to weaning from MV	NA	Survival rate Length of ICU stay Post‐tracheostomy complications	Time to waking after ceasing sedation Duration of sedation and mechanical ventilation Discharge from ICU Tracheostomy decannulation Complications	Length of ICU stay Days from confirmed COVID‐19 to tracheostomy Days from intubation to tracheostomy Days from tracheostomy to ICU and hospital discharge Perioperative complications Risk of infection to the operating team Decannulation rate	NA	NA	NA
Notes		Authors were contacted and provided requested data.			As the data reported were inconclusive and we did not receive a response from the author, we excluded their data from our analyses.

COPD: chronic obstructive pulmonary disease; ECMO: extracorporeal membrane oxygenation; ICU: intensive care unit; IQR: interquartile range; IMV: invasive mechanical ventilation; MV: mechanical ventilation; NA: not applicable; NIV: non‐invasive mechanical ventilation; PCR: polymerase chain reaction; SOFA score: sequential organ failure assessment score

Review outcomes

• Mortality:
  • overall mortality;
  • in‐hospital mortality;
  • at up to day 28 (± 2);
  • at day 60;
  • at day 90;
  • time‐to‐event.
• Improvement of clinical status:
  • duration to liberation from invasive mechanical ventilation;
  • need for invasive mechanical ventilation;
  • liberation from invasive mechanical ventilation;
  • ventilator‐free days;
  • duration to decannulation.
• Worsening of clinical status:
  • adverse events (any grade);
  • ventilator associated pneumonia;
  • need for renal replacement therapy;
  • postoperative bleeding;
  • airway obstruction;
  • tracheal stenosis;
  • need for ECMO;
  • ventilatory problems.
• ICU length of stay, or time to discharge from ICU.
• Hospital length of stay, or time to discharge from hospital.

If meta‐analysis was possible, we assessed the effects of potential biases in sensitivity analyses (see Sensitivity analysis). For binary outcomes, we estimated between‐study variance using the Mantel‐Haenszel method. For continuous outcomes or outcomes for which HRs were available, we used the inverse variance method. For NRSIs we used the adjusted HR, when available.

We analyzed RCTs and NRSIs separately.

Subgroup analysis and investigation of heterogeneity

If data were available, we planned to conduct subgroup analyses on participant age and pre‐existing conditions to evaluate heterogeneity in accordance with our research question and use the tests for interaction to test for differences between subgroup results.

Sensitivity analysis

We performed sensitivity analyses on the following study characteristics for our primary outcomes, as described in the Types of outcome measures section.

• Comparison of intention‐to‐treat‐analysis with per protocol analysis

• Components assessing risk of bias (studies with low risk of bias or some concerns versus studies with high risk of bias)

• Comparison of preprints with peer‐reviewed articles

• Comparison of prematurely terminated studies with completed studies

Summary of findings and assessment of the certainty of the evidence

We created Table 1 and evaluated the certainty of the evidence using the GRADE approach for interventions evaluated in RCTs and NRSIs.

Summary of findings

We used GRADEpro GDT software to establish a summary of findings table and to assess the certainty of evidence.

Chapter 14 of the updated Cochrane Handbook for Systematic Reviews of Interventions states that the most critical or important health outcomes, desirable and undesirable, limited to seven or fewer outcomes, should be included in the summary of findings table(s) (Schünemann 2022). We included outcomes prioritized according to the Core Outcome Set for Intervention Trials (COMET 2021) and patient relevance; these are listed below.

• Mortality: overall mortality is preferred; if not reported, we will include in‐hospital mortality, followed by mortality at day 28, day 60, day 90 or time‐to‐event estimate in the summary of findings table.

• Improvement of clinical status: duration to liberation from invasive mechanical ventilation is preferred; if not reported, we will include need for invasive mechanical ventilation, liberation from invasive mechanical ventilation, ventilator‐free days or duration to decannulation.

• Worsening of clinical status:

  • adverse events (any grade);

  • ventilator‐associated pneumonia;

  • need for renal replacement therapy;

  • if none of these worsening outcomes were reported, we included postoperative bleeding, airway obstruction, tracheal stenosis, need for ECMO or ventilatory problems.

• ICU length of stay; if not reported, we will include time to discharge from ICU.

Assessment of the certainty of the evidence

We used the GRADE approach to assess the certainty of the evidence for the outcomes listed above. In this approach, five domains are considered (risk of bias, consistency of effect, imprecision, indirectness, and publication bias) to assess the certainty of the evidence for each prioritized outcome.

We downgraded the certainty of the evidence for:

• Serious (‐1) or very serious (‐2) risk of bias

• Serious (‐1) or very serious (‐2) inconsistency

• Serious (‐1) or very serious (‐2) uncertainty regarding directness

• Serious (‐1) or very serious (‐2) inaccurate or sparse data

• Serious (‐1) or very serious (‐2) likelihood of reporting bias

The GRADE system uses the following criteria for grading the certainty of the evidence.

• 'High': we are very confident that the true effect is close to the estimated effect.

• 'Moderate': we are moderately confident in the effect estimate: the true effect is likely to be close to the effect estimate, but there is a possibility that it will be substantially different.

• 'Low': our confidence in the effect estimate is limited: the true effect may differ substantially from the effect estimate.

• 'Very low': we have very low confidence in the effect estimate: the actual effect is likely to differ substantially from the effect estimate.

We used the current GRADE guidelines, as recommended in Chapter 14 of the Cochrane Handbook for Systematic Reviews of Interventions (Schünemann 2022), in these assessments.

The overall risk of bias assessment was derived from the RoB 2 Excel tool to make a decision on downgrading the certainty of the evidence for risk of bias. We phrased the outcomes and certainty of evidence as suggested in the informative statements' guidance (Santesso 2020).

For NRSIs, we assessed risk of bias according to the modified version of the Risk Of Bias In Non‐randomized Studies of Interventions (ROBINS‐I) tool as recommended for Cochrane Reviews (Schünemann 2019).

Results

Description of studies

See Characteristics of included studies and Characteristics of excluded studies.

The characteristics of included NRSIs are also compared in Table 5; Table 6; Table 7: 'Overview of included non‐randomized studies of interventions'.

5. Overview of included non‐randomized studies of interventions ‐ Studies H ‐ P.

	Hansson 2022	Hernandez 2022	Karna 2022	Kwak 2021	Kuno 2021	Livneh 2021	Mahmood 2021	Navaratnam 2022	Polok 2021	Prats–Uribe 2021
Setting	Inpatient Sweden	Inpatient Spain	Inpatient India	Inpatient New York, USA	Inpatient New York, USA	Inpatient Israel	Inpatient USA	Inpatient England	Inpatient Multinational	Inpatient Spain
Design	Multicenter retrospective observational study	Multicenter propensity‐matched cohort study	Retrospective observational cohort study	Retrospective cohort study	Retrospective	Retrospective cohort study	Multicenter, retrospective study	Muliticenter, retrospective observational study	Multicenter prospective observational study	Multicenter prospective cohort study
Study protocol	Not reported	Reported	Not reported	Not reported	Not reported	Not reported	Reported	Not reported	Reported	Not reported
Statistical analysis plan	Not reported	Reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported	Not reported
Allocated participants (n)	117	682	65	148	133	38	118	2200	450	696
Number of participants per trial arm (allocated/evaluated)	Early: 56 Late: 61	≤ 7 (n = 65) 8‐10 (n = 126) 11‐14 (n = 191) 15‐20 (n = 197) ≥ 21 (n = 103)	Early: 41 Late: 24	Early: 52 Late: 96	Early: 86 Late: 47	Early: 19 Late: 19	Early: 9 Late: 109	Available data: 1777 Early: 851 Late: 926	Early: 135 Late: 315	Early: 142 Late: 554
Recruitment dates	14 March 2020 to 13 March 2021	15 February 15 to 15 May 2020	1 May 2020 to 30 April 2021	1 March to 7 May 2020	1 March 2020 to 7 May 2020	March 2020 to January 2021	1 February 2020 to 4 September 2020	1 March to 31 October 2020	12 February to 31 December 2020	11 March 2020 to 20 July 2020
Time of follow‐up	Not reported	60 days	Postdischarge follow‐up in pulmonary OPD till 60 days of ICU admission	From their admission until the defined end point of the study	> 60 days	Patients were followed until death or decannulation	Data reviewed last on 15 January 2021	Not reported	Not reported	Follow‐up from the day of the tracheostomy (T0) until death, weaning, or the end of 20 July 2020
Intervention
Intervention (early tracheostomy)(days)	Early tracheostomy (< 7 days)	Early tracheostomy (≤ 7 days, 8‐10 days, 11‐14 days, 15‐20 days)	Early tracheostomy (< 7 days)	Early tracheostomy (< 10 days)	Early tracheostomy (≤ 14 days)	Early tracheostomy (≤ 7 days)	Early tracheostomy (≤ 14 days)	Early tracheostomy (≤ 14 days)	Early tracheostomy (≤ 10 days)	Early tracheostomy (7‐10 days postintubation)
Control (late tracheostomy) (days)	Late tracheostomy (≥ 7 days)	Late tracheostomy (≥ 21 days)	Late tracheostomy (> 7 days)	Late tracheostomy (≥ 10 days)	Late tracheostomy (> 14 days)	Late tracheostomy (> 8 days)	Late tracheostomy (> 14 days)	Late tracheostomy (> 14 days)	Late tracheostomy (> 10 days)	Late tracheostomy (> 10 days)
Technique	Percutaneous or surgical tracheostomy	NA	Percutaneous or surgical tracheostomy	Percutaneous or surgical tracheostomy	Percutaneous or surgical tracheostomy	Open surgical tracheostomy	Percutaneous or surgical tracheostomy	NA	NA	NA
Subgroups	NA	NA	NA	NA	NA	NA	NA	6 subgroups: 18‐39, 40‐49, 50‐59, 60‐69, 70‐79, ≥ 80	NA	NA
Demographics
Age (years)	Median (range) Early: 67(22‐87) Late: 65(18‐83)	Median (IQR) ≤ 7: 62 (55‐70) 8‐10: 65 (56‐69) 11‐14: 64 (57‐71) 15‐20: 64 (57‐69) ≥ 21: 65 (56‐72)	Median (IQR) Early: 55 (43, 65) Late: 45 (38, 54)	Mean (SD) 58.1 (15.8)	Early: 61.2 (13.3) Late: 61.2 (11.4)	Median Early: 60 (54–67) Late: 68 (59–74)	Median 54 (42.5–65.0)	Continous data	Mean (SD) Early: 74 (71, 76) Late: 74 (72, 77)	Early: 63.2 (9.2) Late: 63.0 (10.4)
Gender (male (n(%)))	Early: 46 (82) Late: 44 (72)	≤ 7: 42 (64.6) 8‐10: 88 (69.8) 11‐14: 136 (71.2) 15‐20: 149 (73.8) ≥ 21: 74 (75.5)	Early: 29 (71) Late: 13 (54)	Total: 120 (81)	Early: 53 (61.6) Late: 27 (57.4)	Early: 16 (84) Late: 17 (89)	Total: 75 (63.6)	Total: 1528 (70.7)	Early: 108 (80) Late: 240 (76.19)	Early: 393 (70.6) Late: 89 (63.4)
Comorbidities at baseline (n (%))
Diabetes	Diabetes Type 1 Early: 2 (4) Late: 2 (3) Diabetes Type 2 Early: 13 (23) Late: 20 (33)	NA	Early: 19 (46) Late: 7 (29)	47 (32)	Early: 22 (25.6) Late: 12 (25.5)	NA	Total: 49 (41.5)	Total: 642 (29.2)	Early: 35 (25.9) Late: 97 (30.8)	Early: 24.6% Late: 20.8%
Hypertension	Early: 27 (48) Late: 37 (61)	NA	Early: 19 (46) Late: 11 (46)	80 (54)	Early: 27 (31.4) Late: 13 (27.7)	NA	Total: 53 (44.9)	NA	Early: 99 (73.3) Late: 198 (63.1)	Early: 54.2% Late: 44.6%
Cardiac disease	Cardiovascular disease Early: 9 (16) Late: 15 (25)	≤ 7: 6 (9.2) 8‐10: 10 (7.9) 11‐14: 16 (8.4) 15‐20: 15 (7.4) ≥ 21: 20 (20.4)	Chronic atery disease Early: 5 (7.8%) Late: 4 (10%)	NA	Atrial fibrillation Early: 6 (7.1) Late: 4 (9.3) Heart failure Early: 2 (3.1) Late: 0 (0.0)	NA	Cardiovascular disease Total: 13 (11)	Congestive heart failure Total: 149 (6.8) Acute myocardial infarction Total: 99 (4.5)	Ischemic heart disease Early: 20 (15.0) Late:63 (20.3) Congestive heart failure Early: 12 (8.9) Late: 33 (10.6)	Ischemic cardiopathy Early: 11.3% Late: 8.8%
Respiratory disease	Early: 7 (13) Late: 9 (15)	≤ 7: 6 (9.2) 8‐10: 6 (4.8) 11‐14: 24 (12.6) 15‐20: 31 (15.3) ≥ 21: 15 (15.3)	NA	NA	NA	NA	NA	Total: 480 (21.8)	Early: 26 (19.3) Late: 64 (20.4)	NA
Asthma	Early: 6 (11) Late: 8 (13)	NA	NA	NA	NA	NA	Total: 16 (13.6)	NA	NA	NA
COPD	Early: 6 (11) Late: 5 (8)	≤ 7: 2 (3.1) 8‐10: 2 (2.4) 11‐14: 11 (5.8) 15‐20: 8 (4) ≥ 21: 4 (4.1)	Early: 0 (0) Late: 1 (4.2)	NA	Early: 2 (2.3) Late: 1 (2.1)	NA	Total: 2 (1.7)	NA	NA	Early: 7.7% Late: 7.0%
Obesity (BMI ≥ 30 kg/m2)	Early: 21 (38) Late: 32 (53)	≤ 7: 28 (43.1) 8‐10: 52 (41.3) 11‐14: 88 (46.1) 15‐20: 74 (36.6) ≥ 21: 41 (41.8)	NA	NA	NA	NA	NA	Total: 397 (18.0)	NA	NA
Immunosuppression therapy	Early: 3 (6) Late: 12 (20)	NA	Early: 0 (0) Late: 0 (0)	NA	NA	NA	NA	NA	NA	Late: 4.9% Early: 7.6%
SOFA score, mean (SD)	NA	NA	At admission Early: 3.0 (3.0, 5.0), Late: 4.0 (3.0, 4.0) At intubation Early: 4.0 (4.0, 4.0), Late: 4.0 (4.0, 4.0) At tracheostomy Early: 3.0 (3.0, 4.0), Late: 3.0 (3.0, 5.0)	NA	NA	NA	NA	NA	Early: 7.00 (4.00, 8.50) Late: 6.00 (4.00, 8.50)	Early: 6.7 (4.4) 19.0% missing Late: 6.0 (3.4) 22.6% missing
Outcomes
Primary outcomes	Duration of IMV	Ventilator‐free days at 28 days	Tracheostomy complications Weaning outcomes Sedation wean Days on pressure support mode Tracheostomy to ventilator liberation Decannulation success Tracheostomy to decannulation duration Duration of mechanical ventilation Duration of ICU stay Duration of hospital stay	Time from symptom onset to endotracheal intubation or tracheostomy Time from endotracheal intubation to tracheostomy Time from tracheostomy to tracheostomy tube downsizing or decannulation Total time on mechanical ventilation Total length of stay	Overall mortality	Mortality rate	Intra‐ and postprocedural data Hospital course including ventilator weaning and length of stay Adverse events Survival	In‐hospital mortality Length of hospital stay Length of critical care stay Tracheostomy malfunction	All‐cause mortality within 3 months after admission to ICU	Time to weaning
Secondary outcomes	Number of days on IMV before tracheostomy Time of decannulation Number of hours on mechanical ventilation in the prone position ICU length of stay All‐cause mortality (within 30 days of ICU admission) Complications associated with tracheostomy	Ventilator‐free days at 60 days Modified ICU or hospital bed‐free days at 28 days or 60 days	NA	NA	NA	Time to weaning from IMV Time to decannulation Time to discharge	NA	NA	ICU length of stay Duration of mechanical ventilation	Death Rates of intraoperative bleeding Postoperative bleeding Ventilatory complications (air leak)
Notes		Authors were contacted and provided requested data.					Authors were contacted and provided requested data.

COPD: chronic obstructive pulmonary disease; ECMO: extracorporeal membrane oxygenation; ICU: intensive care unit; IQR: interquartile range; IMV: invasive mechanical ventilation; MV: mechanical ventilation; NA: not applicable; NIV: non‐invasive mechanical ventilation; PCR: polymerase chain reaction; SOFA score: sequential organ failure assessment score

Review outcomes

• Mortality:
  • overall mortality;
  • in‐hospital mortality;
  • at up to day 28 (± 2);
  • at day 60;
  • at day 90;
  • time‐to‐event.
• Improvement of clinical status:
  • duration to liberation from invasive mechanical ventilation;
  • need for invasive mechanical ventilation;
  • liberation from invasive mechanical ventilation;
  • ventilator‐free days;
  • duration to decannulation.
• Worsening of clinical status:
  • adverse events (any grade);
  • ventilator associated pneumonia;
  • need for renal replacement therapy;
  • postoperative bleeding;
  • airway obstruction;
  • tracheal stenosis;
  • need for ECMO;
  • ventilatory problems.
• ICU length of stay, or time to discharge from ICU.
• Hospital length of stay, or time to discharge from hospital.

6. Overview of included non‐randomized studies of interventions ‐ Studies T ‐ V.

	Takhar 2020	Tang 2020	Tetaj 2021	Tsonas 2022	Volo 2021
Setting	Inpatient UK	Inpatient China	Inpatient Italy	Inpatient Netherlands	Inpatient Italy
Design	Prospective single‐center observational study	Multi‐center, retrospective, observational stuy	Retrospective observational study	Multicenter observational study	Retrospective cohort study
Study protocol	Not reported	Not reported	Not reported	Reported	Not reported
Statistical analysis plan	Not reported	Not reported	Not reported	Reported	Not reported
Allocated participants (n)	81	80	451	189	23
Number of participants per trial arm (allocated/evaluated)	Early: 24 Late: 57	Early: 30 Late: 50	Early: 61 Late: 59	Early: 96 Late: 93	Early: 9 Late: 14
Recruitment dates	21 March to 20 May 2020	8 January 2020 to 25 March 2020	1 April 2020 to 31 March 2021	1 March 2020 to 1 June 2020	22 February 2020 to 26 April 2020
Time of follow‐up	Followed up until discharge from hospital or death	Treatment was recorded for the duration of hospitalization. The living status at 60 days after intubation was also recorded	NA	NA	Median follow‐up of 50 days (IQR 30.0–71.0 days)
Intervention
Intervention (early tracheostomy)(days)	Early tracheostomy (< 14 days)	Early tracheostomy (< 14 days)	Early tracheostomy (≤ 12 days)	Early tracheostomy (≤ 21 days)	Early tracheostomy (≤ 10 days)
Control (late tracheostomy) (days)	Late tracheostomy (≥ 14 days)	Late tracheostomy (> 14 days)	Late tracheostomy (> 12 days)	Late tracheostomy (> 21 days)	Late tracheostomy (> 10 days)
Technique	Percutaneous or surgical tracheostomy	Percutaneous or surgical tracheostomy	Percutaneous tracheostomy	Surgical or percutaneous technique (not recorded)	Mostly surgical tracheostomy
Subgroups	NA	NA	NA	NA	NA
Demographics
Age (years)	Mean (SD) Early 58.4 (11.8) Late 50.6 (11.8)	Mean (SD) Early: 66.5 (15.1) Late: 62.3 (13.2)	Median (SD) 68.5 (61–75) Early: 70 (64–77) Late: 65 (69–73)	Median (IQR) Early: 65.0 (59.0‐72.0) Late: 65.0 (60.0‐71.0)	Median 69 (42‐84)
Gender (male (n(%)))	Early: 15 (62.5) Late: 40 (70.1)	Early: 21 (70.0) Late: 34 (68.0)	Early: 42 (68.9) Late: 38 (64.4)	Early: 75 (78.1) Late: 77 (82.8)	Total: 21 (91)
Comorbidities at baseline (n (%))
Diabetes	NA	Early: 6 (20) Late: 8 (16)	Early: 15 (24.6) Late: 13 (22.0)	Early: 22 (22.9) Late: 21 (22.6)	Total: 7 (30)
Hypertension	NA	Early: 12 (40.0) Late: 20 (40.0)	Early: 45 (73.7) Late: 35 (59.3)	Early: 29 (30.2) Late: 33 (35.5)	Total: 11 (47)
Cardiac disease	NA	Coronary heart disease Early: 3 (10.0) Late: 14 (28.0)	Early: 15 (24.6) Late: 9 (15.2)	Heart failure Early: 4 (4.2) Late: 6 (6.5)	Cardiovascular disease Total: 4 (17%)
Respiratory disease	NA	NA	COPD/bronchial asthma Early: 12 (19.7) Late: 9 (15.2)	NA	NA
Asthma	NA	NA	NA	Early: 10 (10.4) Late: 9 (9.7)	NA
COPD	NA	Early: 0 Late: 2 (4%)	NA	Early: 6 (6.2) Late: 8 (8.6)	Total: 88 (34)
Obesity (BMI ≥ 30 kg/m2)	Early: 12 (50) Late: 27 (47.36)	NA	Early: 25 (41) Late: 29 (49.1)	NA	Total: 3 (13)
Immunosuppression therapy	NA	NA	NA	Early: 4 (4.2) Late: 1 (1.1)	NA
SOFA score, mean (SD)	NA	Early: 6 (4, 9) Late: 5 (4, 7)	Early: 5 (4–7) Late: 5 (3–8)	NA	SOFA score intubation > 6 n = 9 (39)
Outcomes
Primary outcomes	Duration of ventilation post‐tracheostomy	Treatments Details of the tracheostomy procedure Successful weaning after tracheostomy Living status	ICU length of stay Mechanical ventilation	Duration of ventilation ICU and hospital length of stay ICU and hospital 28‐day and 90–day mortality Factors associated with timing	Mortality Sedation Date to sub‐intensive unit Date of weaning from MV Date of decannulation SOFA score at the day of intubation and of the day before tracheostomy D‐dimer level
Secondary outcomes	Days until Sedation was stopped The patient was discharged from the ICU Decannulation of tracheostomy occurred The patient was discharged from hospital Death occurred	NA	NA	NA	NA
Notes			Authors were contacted and provided requested data.

COPD: chronic obstructive pulmonary disease; ECMO: extracorporeal membrane oxygenation; ICU: intensive care unit; IQR: interquartile range; IMV: invasive mechanical ventilation; MV: mechanical ventilation; NA: not applicable; NIV: non‐invasive mechanical ventilation; PCR: polymerase chain reaction; SOFA score: sequential organ failure assessment score

Review outcomes

• Mortality:
  • overall mortality;
  • in‐hospital mortality;
  • at up to day 28 (± 2);
  • at day 60;
  • at day 90;
  • time‐to‐event.
• Improvement of clinical status:
  • duration to liberation from invasive mechanical ventilation;
  • need for invasive mechanical ventilation;
  • liberation from invasive mechanical ventilation;
  • ventilator‐free days;
  • duration to decannulation.
• Worsening of clinical status:
  • adverse events (any grade);
  • ventilator associated pneumonia;
  • need for renal replacement therapy;
  • postoperative bleeding;
  • airway obstruction;
  • tracheal stenosis;
  • need for ECMO;
  • ventilatory problems.
• ICU length of stay, or time to discharge from ICU.
• Hospital length of stay, or time to discharge from hospital.

Results of the search

We searched the databases in June 2022 and identified 1203 records. After removing 103 duplicates, we performed title and abstract screening of 1091 records and excluded 1046 records that did not meet our predefined inclusion criteria. We checked the full texts or, if these were not available, the registry entries of the remaining 45 references. This resulted in the exclusion of 19 entries that did not meet our inclusion criteria. Reasons for exclusion are listed in the Characteristics of excluded studies section. Overall, we included 26 records (26 studies) in our narrative analysis and 24 records (24 studies) in our meta‐analyses. We depicted the selection process in a PRISMA flow diagram (see Figure 1).

1.

Included studies

We included one RCT with 150 COVID‐19 patients. Study participants' mean age was 65 years and 79% were men (Eeg‑Olofsson 2022).

Further, we included 24 NRSIs with 6372 COVID‐19 patients, whose mean age was 61.8 years; 72% were men (Angel 2021; Arnold 2022; Avilés‐Jurado 2020; Battaglini 2021; Breik 2020; Chandran 2021; Dal 2022; Evrard 2021; Glibbery 2020; Hansson 2022; Hernandez 2022; Karna 2022; Kuno 2021; Kwak 2021; Livneh 2021; Mahmood 2021; Navaratnam 2022; Polok 2021; Prats–Uribe 2021; Takhar 2020; Tang 2020; Tetaj 2021; Tsonas 2022; Volo 2021).

When our prioritized outcomes were not reported, we attempted to contact the authors of the included studies. We contacted 16 study authors (Angel 2021; Arnold 2022; Avilés‐Jurado 2020; Battaglini 2021; Breik 2020; Chandran 2021; Eeg‑Olofsson 2022; Glibbery 2020; Hansson 2022; Hernandez 2022; Kwak 2021; Mahmood 2021; Takhar 2020; Tang 2020; Tetaj 2021; Tsonas 2022). We received responses from six study authors, who provided us with requested data that had not been published and were missing for our analyses (Arnold 2022; Eeg‑Olofsson 2022; Hernandez 2022; Mahmood 2021; Tetaj 2021). If data were still missing after this step, we had to make explicit assumptions about the methods used in the included studies.

As the data reported by Breik 2020 were inconclusive, and we did not receive a response from the study author, we decided to exclude their data from our analyses.

During the detailed full‐text search, we found that the authors of two articles belonged to the same research group and were involved in both publications (Angel 2021; Kwak 2021). Both studies included patients admitted to the same hospital, and the recruitment period of the two studies overlapped, raising the suspicion that the included cohorts might overlap. We sent an inquiry to the study authors, but we received no response. The participants in the study by Angel 2021 received early tracheostomy, defined as 14 days or fewer after intubation, and Kwak 2021 defined an early tracheostomy as ten days or fewer after intubation. Therefore, we were able to ensure that the cohorts were not compared with each other in our analyses.

In total, 6522 participants (mean age of 62 years. 73% men) diagnosed with COVID‐19 from 25 studies were included in our review. The majority of included studies were conducted in high‐ and middle‐income countries; the only lower‐middle‐income country was India. For a better overview, we summarized the characteristics of the included NRSIs in Table 5; Table 6; Table 7 and in the Characteristics of included studies section.

Study design and setting

Randomized controlled trial

Eeg‑Olofsson 2022 was a multicenter, parallel, single‐blinded RCT, conducted in Sweden from 6 June 2020 to 20 April 2021. It was approved by the Swedish Ethics Review Authority, and was conducted in accordance with the Declaration of Helsinki. An interim analysis was conducted using both the ITT and the per protocol cohorts.

Non‐randomized studies of interventions (NRSIs)

The other 24 included studies were non‐randomized; nine of these NRSIs had a prospective design (Angel 2021; Arnold 2022; Avilés‐Jurado 2020; Breik 2020; Chandran 2021; Glibbery 2020; Polok 2021; Prats–Uribe 2021; Takhar 2020), and 15 had a retrospective design (Battaglini 2021;Dal 2022; Evrard 2021; Hansson 2022; Hernandez 2022; Karna 2022; Kuno 2021; Kwak 2021; Livneh 2021; Mahmood 2021; Navaratnam 2022; Tang 2020; Tetaj 2021; Tsonas 2022; Volo 2021).

Eleven NRSIs had a multicenter design and 13 were single‐center studies (see Table 5). Angel 2021 included 394 patients from two New York University Health hospitals. Battaglini 2021 included 153 patients admitted to 11 Italian ICUs. Evrard 2021 performed a retrospective analysis of prospectively collected data on 48 patients at two university hospitals in the Paris region. Hansson 2022 examined 117 patients in the ICUs of three nonacademic, rural hospitals in the southern Swedish county of Jönköping. Hernandez 2022 included 682 patients in 15 Spanish ICUs. Mahmood 2021 included 118 patients from seven hospitals in five academic tertiary centers in the USA. Navaratnam 2022 performed a retrospective analysis of 2200 completed episodes of hospitalization in England (UK). Polok 2021 collected data from 152 centers in 16 European countries and enrolled 2078 patients, where 461 (26.5%) patients had a tracheostomy. Polok 2021 was part of the Very Old Intensive Care (VIP) Project (www.vipstudy.org), funded by the European Society of Intensive Care Medicine (ESICM). Prats–Uribe 2021 included 696 patients from 36 hospitals in Spain. Tang 2020 evaluated 80 patients in ICUs at 23 hospitals in Hubei Province, China. Tsonas 2022 analyzed data from 22 Dutch centers and included 189 patients.

The data collection periods of 14 studies were two to approximately three months, from March to June 2020, representing the first pandemic wave (Angel 2021; Avilés‐Jurado 2020; Battaglini 2021; Breik 2020; Dal 2022; Glibbery 2020; Karna 2022; Kuno 2021; Hernandez 2022; Kwak 2021; Takhar 2020; Tang 2020; Tsonas 2022; Volo 2021). Polok 2021 included patients from February to December 2020. Dal 2022 included patients from November 2020 to January 2021, the second pandemic wave. Karna 2022 and Livneh 2021 chose a period from March 2020 to April 2021, representing the third pandemic wave. Three studies were stopped early after approximately one month (Arnold 2022; Avilés‐Jurado 2020; Breik 2020), and five studies after three to four months (Dal 2022; Kuno 2021; Kwak 2021; Prats–Uribe 2021; Volo 2021). The data collection periods of three studies were six months (Evrard 2021; Mahmood 2021; Navaratnam 2022), and of four studies approximately one year (Arnold 2022; Chandran 2021; Hansson 2022; Tetaj 2021).

Data synthesis

The studies by Angel 2021 and Hernandez 2022 used a propensity score‐matching approach to compare outcomes. Battaglini 2021 followed the recommendations of Strengthening the Reporting of Observational studies in Epidemiology (STROBE). Navaratnam 2022 performed a retrospective analysis using the Hospital Episode Statistics administrative dataset, also according to STROBE guidelines. The associations between demographic factors, comorbidity, and tracheostomy and the relationship between tracheostomy, tracheostomy timing, and outcomes were investigated using multilevel modelling.

Polok 2021 and Prats–Uribe 2021 sent a data collection form at the beginning of the studies, and study investigators at each hospital recorded data from ICU admission through weaning, death, or the end of the study. Tsonas 2022 included patients in a secondary analysis of the PRoVENT trial (Botta 2021).

Participants

All studies enrolled adult, critically ill COVID‐19 patients who had a positive nasopharyngeal swab for SARS‐CoV‐2 with real‐time polymerase chain reaction (RT‐PCR) and were admitted to the ICU due to respiratory failure, required mechanical ventilation, and underwent surgical or percutaneous tracheostomy.

The information on the presence of comorbidities in the included cohort at baseline varied widely. Most studies provided information on the presence of hypertension, cardiac disease, respiratory disease (e.g. asthma, COPD), obesity (body mass index (BMI) ≥ 30 kg/m²), immunosuppressive therapy, or sequential organ failure assessment score (SOFA score). Detailed information on reported comorbidities is listed in Table 5, Table 6 and Table 7.

Notably, the substudy of the COVIP (www.vipstudy.org), trial, Polok 2021, specifically included elderly critically ill COVID‐19 patients, over 70 years of age, who were admitted to the ICU and required invasive mechanical ventilation followed by tracheostomy.

Intervention

Randomized controlled trial

A total of 72 participants from Eeg‑Olofsson 2022 were randomized to receive early tracheostomy, defined as seven days or fewer after intubation, whereas late tracheostomy was performed not earlier than 10 days or fewer after intubation.

Non‐randomized studies of interventions

Three studies defined early tracheostomy as seven days or fewer after intubation (Hansson 2022; Karna 2022; Livneh 2021), or in seven studies as 10 days or fewer after intubation (Avilés‐Jurado 2020; Chandran 2021; Evrard 2021; Kwak 2021; Polok 2021; Prats–Uribe 2021; Volo 2021), see Table 8. Tetaj 2021 defined early tracheostomy as 12 days or fewer after the start of invasive mechanical ventilation. The majority of participants (n = 820) from the NRSIs received early tracheostomy, defined as 14 days or fewer (±1 day) after intubation or start of invasive mechanical ventilation (Angel 2021; Arnold 2022; Battaglini 2021; Breik 2020; Dal 2022; Glibbery 2020; Hernandez 2022; Kuno 2021; Mahmood 2021; Takhar 2020; Tang 2020). Tsonas 2022 defined early tracheostomy as 21 days or fewer after start of invasive mechanical ventilation. Navaratnam 2022 defined early tracheostomy as 14 days or fewer after admission to critical care.

7. Intervention definitions.

Study ID	Early tracheostomy (days)	Time to tracheostomy defined by timing of:	Techniques
Angel 2021	≤ 13 days	Intubation	Bedside PDT with modified visualization and ventilation
Arnold 2022	≤ 14 days	Intubation	Bedside PDT
Avilés‐Jurado 2020	≤ 10 days	Intubation	Surgical tracheostomy was performed for all patients following recommended criteria for use of PPE
Battaglini 2021	< 15 days	Intubation	Comparison PDT versus open surgical technique
Breik 2020	≤ 14 days	Intubation	PDT without bronchoscopy to minimise aerosol generation
Chandran 2021	≤ 10 days	Intubation	Open surgical tracheostomy
Dal 2022	≤ 14 days	Intubation	Bedside PDT
Eeg‑Olofsson 2022	≤ 7 ‐ 10 days	Intubation	PDT and open surgical technique
Evrard 2021	≤ 10 days	Intubation	PDT and open surgical technique
Glibbery 2020	< 14 days	Intubation	Surgical or percutaneous tracheostomy
Hansson 2022	≤ 7 days	Mechanical ventilation	All but one were performed with an open surgical technique
Hernandez 2022	≤ 7 days, 8‐10 days and 11‐14 days (in analyse ≤ 14 days)	Intubation	No standardization of the procedure due to the international multicenter design
Karna 2022	≤ 7 days	Intubation	69% PDT and 31% open surgical technique
Kwak 2021	< 10 days	Intubation	Bedside PDT with modified visualization and ventilation
Kuno 2021	≤ 14 days	Intubation	No standardization of the procedure
Livneh 2021	≤ 7 days	Intubation	Only open surgical tracheostomy procedures
Mahmood 2021	≤ 14 days	Intubation	78% percutaneous technique 22% open surgical technique ‐ due to anatomic concerns that precluded percutaneous access. (Standard percutaneous and surgical techniques, in addition with PPE and measures to reduce aerosol formation.)
Navaratnam 2022	≤ 14 days	Admission to critical care	No standardization of the procedure due to the national multicenter design
Polok 2021	< 10 days	Intubation	No standardization of the procedure due to the international multicenter design
Prats–Uribe 2021	≤ 7 ‐ 10 days	Intubation	No standardization of the procedure due to the multicenter design
Takhar 2020	≤ 14 days	Intubation	93.8% PDT, 6.2% via a hybrid or open surgical technique, 96.3% at the ICU bedside
Tang 2020	≤ 14 days	Intubation	78.8% percutaneous techniques, 95.0% at the ICU bedside
Tetaj 2021	≤ 12 days	Intubation	PDT under bronchoscopic visualization (Frova PercuTwist or Ciaglia Blue Rhino)
Tsonas 2022	≤ 21 days	Invasive ventilation	No standardization of the procedure due to the multicenter design
Volo 2021	≤ 10 days	Intubation	Open surgical tracheostomy procedures at the ICU
ICU: intensive care unit; PDT: percutaneous dilatational tracheostomy; PPE: personal protective equipment

All analyses are presented as early compared to late tracheostomy in the Data and analyses section. However, because of the heterogeneous timing of the interventions, it is not ideal to perform meta‐analyses from the data of all included studies. As described above, we divided the studies into two groups to achieve comparability of the data:

• early tracheostomy (≤ 10 days after intubation) compared to late tracheostomy (> 10 days);

• early tracheostomy (≤ 14 (± 1) days after intubation) compared to late tracheostomy (> 14 (± 1) days).

The authors of the included studies often refer to existing recommendations on the timing of tracheostomy procedure in COVID‐19, some of which recommended waiting as long as possible, but at least two weeks, before performing a tracheostomy (Mahmood 2021; Tsonas 2022). The prospective studies had established indications for tracheostomy, partly based on international recommendations, but finally always based on the decision of the treating physicians or a defined tracheostomy team, and in some cases with the presence of a standard protocol (Avilés‐Jurado 2020; Takhar 2020).

The majority of the multicenter studies could not establish a standardization on the indication, as well as on the applied technique of the procedure (Hernandez 2022; Navaratnam 2022; Polok 2021; Prats–Uribe 2021; Tsonas 2022). The final decision to perform a tracheostomy, as well as the choice of the timing of the procedure, was always left to the intensivist treating the patient in all studies and was primarily influenced by the patient's overall clinical condition, prognosis, and expected weaning tolerance.

Tracheostomy was performed in 11 of the studies as a bedside dilatative approach (Angel 2021; Arnold 2022; Battaglini 2021; Breik 2020; Dal 2022; Karna 2022; Kwak 2021; Mahmood 2021; Takhar 2020; Tang 2020; Tetaj 2021). Otherwise, an open surgical approach was chosen. Most studies did not compare the two approaches, that is, a dilated or surgical approach, but focused on the timing of the procedure. Only one study compared the two approaches (Battaglini 2021). Eeg‑Olofsson 2022, Evrard 2021 and Glibbery 2020 used and reported these two techniques equally. Four studies reported only surgically performed tracheostomies (Avilés‐Jurado 2020; Chandran 2021; Livneh 2021; Volo 2021).

All definitions of the interventions in the included studies are summarized in Table 8.

Outcomes

Sixteen studies reported mortality as a primary outcome (Angel 2021; Arnold 2022; Breik 2020; Chandran 2021; Dal 2022; Eeg‑Olofsson 2022; Hansson 2022; Kuno 2021; Livneh 2021; Mahmood 2021; Navaratnam 2022; Polok 2021; Prats–Uribe 2021; Takhar 2020; Tsonas 2022; Volo 2021). However, the outcome of mortality was not standardized in these studies. We extracted outcomes related to mortality, such as prioritized overall mortality or ‐ secondarily ‐ mortality up to day 28, day 30, day 60, or day 90 and time to event. We asked the study authors for the exact definition of the mortality they reported. The "death in the ICU" reported in Eeg‑Olofsson 2022 meets our definition of overall mortality. In response to our inquiry, the study authors also reported three additional participants who died within the follow‐up period of 90 days after intubation but after ICU discharge (two who were randomized to the early tracheostomy group in the ITT group and one who was randomized to the late tracheostomy group in the per protocol group), which we could thus include in our analyses.

The primary endpoints in all studies included parameters that described weaning from mechanical ventilation. Eight studies reported duration of mechanical ventilation as a primary outcome, that is, how many patients could be liberated from mechanical ventilation (Breik 2020; Dal 2022; Hansson 2022; Karna 2022; Kwak 2021; Polok 2021; Tetaj 2021; Tsonas 2022). One study also reported the time interval to liberation from mechanical ventilation, that is, how many days were required until liberation from mechanical ventilation (Karna 2022), and one study reported ventilator‐free days (Hernandez 2022). Other outcomes related to duration of mechanical ventilation included the initiation of weaning from mechanical ventilation in nine studies (Arnold 2022; Avilés‐Jurado 2020; Glibbery 2020; Karna 2022; Livneh 2021; Mahmood 2021; Prats–Uribe 2021; Tang 2020; Volo 2021), and duration to decannulation in 10 studies (Arnold 2022; Breik 2020; Chandran 2021; Glibbery 2020; Hansson 2022; Karna 2022; Kwak 2021; Livneh 2021; Takhar 2020; Volo 2021).

Parameters representing clinical worsening, for instance, safety outcomes such as the number of participants with adverse events and serious adverse events, were also reported as primary or secondary outcomes, and extracted from the available data of the included studies. One study reported this outcome as adverse events (Eeg‑Olofsson 2022). Six studies reported bleeding (Arnold 2022; Avilés‐Jurado 2020; Eeg‑Olofsson 2022; Hernandez 2022; Karna 2022; Tang 2020). Two studies reported tracheal stenosis (Evrard 2021; Karna 2022). One study reported airway obstruction (Eeg‑Olofsson 2022). Three studies reported ventilator problems (Avilés‐Jurado 2020; Hernandez 2022; Tetaj 2021). Five studies reported ventilator‐associated pneumonia (Battaglini 2021; Eeg‑Olofsson 2022; Hernandez 2022; Mahmood 2021; Tsonas 2022). One study reported need for renal replacement therapy (Arnold 2022). Three studies reported need for ECMO therapy (Arnold 2022; Tang 2020; Tsonas 2022). Fourteen studies reported length of ICU stay (Arnold 2022; Breik 2020; Eeg‑Olofsson 2022; Evrard 2021; Glibbery 2020; Hansson 2022; Hernandez 2022; Karna 2022; Navaratnam 2022; Polok 2021; Takhar 2020; Tetaj 2021; Tsonas 2022; Volo 2021), and eight studies reported length of hospital stay (Arnold 2022; Evrard 2021; Hernandez 2022; Karna 2022; Kwak 2021; Navaratnam 2022; Takhar 2020; Tsonas 2022). A detailed summary of all reported outcome measures for each of the included studies can be found in the overview of included studies table (Table 5; Table 6; Table 7).

Because of the different definitions, we searched the studies for exact descriptions of each parameter before including them in our analyses and comparing them with each other. In the case of missing data, we requested this information from the study authors.

Ongoing studies

We did not identify any ongoing studies.

Excluded studies

We excluded 19 references that did not meet our inclusion criteria (see Characteristics of excluded studies).

• Two studies did not define early and late tracheostomy groups.

• The full text of four studies was not retrievable.

• Four studies did not compare early tracheostomy to late tracheostomy.

• Nine references were comments without results.

Risk of bias in included studies

We assessed the risk of bias in the included RCT, Eeg‑Olofsson 2022, using the RoB 2 tool as recommended in Chapter 8 of the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2022c; Sterne 2019). Results are reported according to our review protocol (Dahms 2021).

We assessed the risk of bias in the included NRSIs using the modified version of the ROBINS‐I tool recommended for assessment on interventions in Cochrane Reviews (Schünemann 2019).

We could not assess reporting bias for any outcome due to an insufficient number of studies.

The completed RoB 2 tool and ROBINS‐I tool with responses to all assessed signalling questions are available online at https://doi.org/10.5281/zenodo.7895589.

Mortality

Randomized controlled trial

Eeg‑Olofsson 2022 reported this outcome as "death in the ICU", which, after requesting further information from the study authors, met our definition of overall mortality (see Table 27). We assessed this outcome at the study level for the ITT analysis. We have some concerns regarding bias due to the randomization process (domain 1), since no information regarding allocation concealment is provided, which could suggest potential issues with the randomization process. Therefore, overall, we rated the risk of bias for mortality as 'some concerns'.

Risk of bias for analysis 1.1 Overall mortality.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Non‐randomized studies of interventions

Thirteen NRSIs reported this outcome as mortality (Angel 2021; Arnold 2022; Battaglini 2021; Chandran 2021; Dal 2022; Kuno 2021; Kwak 2021; Mahmood 2021; Polok 2021; Prats–Uribe 2021; Takhar 2020; Tang 2020; Volo 2021). We rated confounding bias as moderate because of possible confounders that may have influenced the decision to assign to the early or late tracheostomy group and thus may have affected the outcome (Angel 2021; Arnold 2022; Battaglini 2021; Dal 2022; Kuno 2021; Kwak 2021; Polok 2021; Prats–Uribe 2021; Tang 2020; Volo 2021). We rated confounding bias as critical if no appropriate analysis method was provided for controlling important confounding areas (Chandran 2021; Mahmood 2021; Takhar 2020). We rated study participant selection bias in all 13 studies as critical because selection of participants into the study could have been based on participant characteristics observed after the intervention began or because characteristics could have led to selection of the patient. We rated intervention classification bias as moderate, when no protocol or study registry was available (Kwak 2021; Volo 2021), or as serious, if intervention classification could have been influenced by knowledge of the outcome or risk of the outcome (Angel 2021; Arnold 2022; Chandran 2021; Polok 2021; Prats–Uribe 2021; Takhar 2020). In the other domains, we did not identify bias. We rated the overall risk of bias for this outcome as critical in all studies (see Table 3 ≤ 10 days; Table 4 ≤ 14 days).

Improvement of clinical status: duration to liberation from invasive mechanical ventilation

Randomized controlled trial

Eeg‑Olofsson 2022 reported this outcome as duration of mechanical ventilation (see Analysis 1.2). We assessed this outcome at the study level for the ITT analysis. There were some concerns regarding bias due to the randomization process (domain 1), as described above. Therefore, overall, we rated the risk of bias for this outcome with 'some concerns'.

1.2. Analysis.

Comparison 1: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 2: Duration to liberation from invasive mechanical ventilation

Non‐randomized studies of interventions

Three NRSIs reported this outcome as duration to liberation from invasive mechanical ventilation (Avilés‐Jurado 2020; Glibbery 2020; Takhar 2020). Five studies reported this outcome as liberation from mechanical ventilation (Angel 2021; Arnold 2022; Livneh 2021; Mahmood 2021; Takhar 2020). We rated confounding bias in these studies as moderate because of possible confounders that may have influenced the decision to assign to the early or late tracheostomy group and thus may have affected the outcome, but the studies reported the method they used to control for confounding domains (Angel 2021; Arnold 2022; Avilés‐Jurado 2020; Kwak 2021). We rated confounding bias as critical if studies did not provide an appropriate analysis method for controlling important confounding areas (Glibbery 2020; Mahmood 2021; Takhar 2020). We rated study participant selection bias as critical in all seven studies because selection of participants into the study (or into the analysis) could have been based on participant characteristics observed after the intervention began or because characteristics could have led to selection of the patient. We rated intervention classification bias as moderate, because in most cases no protocol or study registry was available (Arnold 2022; Kwak 2021), or serious, if intervention classification could have been influenced by knowledge of the outcome or risk of the outcome (Angel 2021; Avilés‐Jurado 2020; Glibbery 2020Takhar 2020). We rated outcome measurement bias as serious for this outcome, when outcome measurement could have been influenced by knowledge of the intervention received (Angel 2021; Avilés‐Jurado 2020; Glibbery 2020). In the other domains, we did not identify potential problems that could bias reported outcomes. We rated the overall risk of bias for this outcome as critical in all studies (Table 3 ≤ 10 days; Table 4 ≤ 14 days).

Worsening of clinical status: adverse events (any grade)

Randomized controlled trial

Eeg‑Olofsson 2022 reported this outcome as any complications, matching our definition of adverse events (any grade); see Analysis 1.3. We assessed this outcome at the study level for the ITT analysis. There were some concerns regarding bias due to the randomization process (as described above) and in the measurement of the outcome (domains 1 and 4). Assessors were aware of the intervention status when measuring adverse events and not all details of the measurements were disclosed. Therefore, overall, we rated the risk of bias for this outcome as 'some concerns'.

1.3. Analysis.

Comparison 1: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 3: Adverse events

Non‐randomized studies of interventions

No NRSIs reported this outcome as adverse events (any grade).

Worsening of clinical status: ventilator‐associated pneumonia

Randomized controlled trial

Eeg‑Olofsson 2022 reported this outcome as ventilator‐associated pneumonia (see Analysis 1.4). We assessed this outcome at the study level for the ITT analysis. There were some concerns regarding bias due to the randomization process (as described above) and in the measurement of the outcome (domains 1 and 4). Assessors were aware of the intervention status when measuring adverse events and not all details of the measurements were disclosed. Therefore, overall, we rated the risk of bias for this outcome as 'some concerns'.

1.4. Analysis.

Comparison 1: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 4: Ventilator‐associated pneumonia

Non‐randomized studies of interventions

Three NRSIs reported this outcome as ventilator‐associated pneumonia (Battaglini 2021; Hernandez 2022; Mahmood 2021). Overall, we rated the risk of bias for the incidence of ventilator‐associated pneumonia in these studies as critical (Table 3 ≤ 10 days; Table 4 ≤ 14 days). In particular, we rated confounding bias in these studies as moderate because of possible confounders that may have influenced the decision to assign to the early or late tracheostomy group and thus may have affected the outcome (Battaglini 2021; Hernandez 2022). We rated confounding bias as critical if studies did not provide an appropriate analysis method for controlling important confounding areas (Mahmood 2021). We rated study participant selection bias in all three studies as critical because selection of participants into the study (or into the analysis) could have been based on participant characteristics observed after the intervention began or because characteristics could have led to the patient being selected. In the other domains, we did not identify any potential problems that could bias the reported results. We rated the overall risk of bias for this outcome as critical in all studies.

Worsening of clinical status: need for renal replacement therapy

Randomized controlled trial

The RCT did not report data for this outcome.

Non‐randomized studies of interventions

One NRSI reported this outcome as need for renal replacement therapy (Arnold 2022). Overall, we rated the risk of bias for the need for renal replacement therapy in this study as critical (Table 4 ≤ 14 days). In particular, we rated confounding bias in this study as moderate because of possible confounders that may have influenced the decision to assign to the early or late tracheostomy group and thus may have affected the outcome. We rated study participant selection bias as critical because selection of participants into the study (or into the analysis) could have been based on participant characteristics observed after the intervention began or because characteristics could have led to the patient being selected. We judged the bias from intervention classification to be moderate because a protocol or study registry was not available. In the other domains, we did not identify potential problems that could bias the reported results. We rated the overall risk of bias for this outcome as critical in this study.

Worsening of clinical status: postoperative bleeding

Randomized controlled trial

Eeg‑Olofsson 2022 reported this outcome as tracheal bleeding, matching our definition of postoperative bleeding (see Analysis 1.5). We assessed this outcome at the study level for the ITT analysis. There were some concerns regarding bias due to the randomization process (as described above) and in the measurement of the outcome (domains 1 and 4). Assessors were aware of the intervention status when measuring adverse events and not all details of the measurements were disclosed. Therefore, overall, we rated the risk of bias for this outcome with 'some concern'.

1.5. Analysis.

Comparison 1: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 5: Postoperative bleeding

Non‐randomized studies of interventions

Three NRSIs reported this outcome as postoperative bleeding (Arnold 2022; Avilés‐Jurado 2020; Tang 2020). Overall, we rated the risk of bias for the incidence of postoperative bleeding in these studies as critical (Table 3 ≤ 10 days; Table 4 ≤ 14 days). In these three studies, we rated confounding bias as moderate because of possible confounders that may have influenced the decision to assign to the early or late tracheostomy group and thus may have affected the outcome, but the studies reported the method they used to control for all important confounding areas. We rated study participant selection bias in these three studies as critical because selection of participants into the study (or into the analysis) could have been based on participant characteristics observed after the intervention began or because characteristics could have led to selection of the patient. We rated the bias from intervention classification to be moderate when a protocol or study registry was not available (Arnold 2022), or serious, if intervention classification could have been influenced by knowledge of the outcome or risk of the outcome (Avilés‐Jurado 2020). In the other domains, we did not identify potential problems that could bias reported outcomes. We rated the overall risk of bias for this outcome as critical in all studies.

Worsening of clinical status: airway obstruction

Randomized controlled trial

Eeg‑Olofsson 2022 reported this outcome as airway obstruction, matching our definition of airway obstruction (see Analysis 1.6). We assessed this outcome at the study level for the ITT analysis. There were some concerns regarding bias due to the randomization process (as described above) and in the measurement of the outcome (domains 1 and 4). Assessors were aware of the intervention status when measuring adverse events and not all details of the measurements were disclosed. Therefore, overall, we rated the risk of bias for this outcome as 'some concerns'.

1.6. Analysis.

Comparison 1: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 6: Airway obstruction

Non‐randomized studies of interventions

We did not find data on this outcome in the included NRSIs.

ICU length of stay or time to discharge from ICU

Randomized controlled trial

Eeg‑Olofsson 2022 reported this outcome as ICU length of stay in days (see Analysis 1.7). We assessed this outcome at the study level for the ITT analysis. There were some concerns regarding bias due to the randomization process (domain 1), as described above. Therefore, overall, we rated the risk of bias for this outcome as 'some concerns'.

1.7. Analysis.

Comparison 1: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 7: ICU length of stay

Non‐randomized studies of interventions

One NRSI reported outcomes that met our definition of ICU length of stay (Arnold 2022). Overall, we rated the risk of bias for ICU length of stay in this study as critical (Table 3 ≤ 10 days; Table 4 ≤ 14 days). In particular, we rated confounding bias in this study as moderate because of possible confounders that may have influenced the decision to assign to the early or late tracheostomy group and thus may have affected the outcome. We rated study participant selection bias as critical because selection of participants into the study (or into the analysis) could have been based on participant characteristics observed after the intervention began or because characteristics could have led to the patient being selected. We judged the bias from intervention classification to be moderate because a protocol or study registry was not available. In the other domains, we did not identify potential problems that could bias the reported results. We rated the overall risk of bias for this outcome as critical in this study.

Effects of interventions

See: Table 1

We presented the summary of findings and the certainty of evidence for adult, hospitalized, critically ill COVID‐19 patients with confirmed SARS‐CoV‐2 infection and respiratory failure resulting in ICU admission for intubation and mechanical ventilation where performing an early tracheostomy was compared with a late tracheostomy in Table 1. Because we identified only one RCT, we have concentrated mainly on these results in the summary of findings table, as they represent the highest level of evidence currently available. We report our prioritized outcomes for the NRSIs narratively, and we have included them in the analyses when possible. All analyses are presented as early versus late tracheostomy in the Data and analyses section.

Because of the difference in timing of the interventions, we pooled the results of the NRSIs into groups, to ensure a better comparison of the outcomes of the interventions, and then pooled them in the analyses: early (≤ 10 days) compared to late (> 10 days) tracheostomy; and early (≤ 14 ± 1 days) compared to late (> 14 ± 1 days) tracheostomy. We selected the groups in which an early tracheostomy was defined as fewer than 10 days after intubation because it was most similar to the recommendations for early tracheostomy for the treatment of comparable conditions before the onset of the pandemic, and the groups in which an early tracheostomy was defined as fewer than 14 days after intubation because it was most similar to the recommendations during the pandemic; the majority of participants from the observational studies received an early tracheostomy defined as 14 days or fewer (± 1 day) after intubation. Therefore, we describe only the results of early tracheostomy as 10 days or fewer and 14 days or fewer after intubation. Results to all tracheostomy time points are reported in Analysis 4.1 to Analysis 4.18.

4.1. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 1: Overall mortality and in‐hospital mortality

4.18. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 18: Hospital length of stay

Adaptations made in our outcome set compared to the protocol are outlined in Differences between protocol and review.

Mortality

Randomized controlled trial

One study reported overall mortality for 150 participants in the ITT analysis (see Analysis 1.1). Considering the reported event rates, we found that early tracheostomy (≤ 10 days) may result in little to no difference in overall mortality compared with late tracheostomy (RR 0.82, 95% CI 0.52 to 1.29; RD 67 fewer per 1000, 95% CI 178 fewer to 108 more; 1 study, 150 participants; low‐certainty evidence). Our main reasons for downgrading were serious imprecision due to the small number of participants in only one study and wide confidence intervals.

1.1. Analysis.

Comparison 1: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 1: Overall mortality

Non‐randomized studies of interventions

Two studies reported this outcome as overall mortality for 719 participants for early tracheostomy (≤ 10 days), and six studies reported in‐hospital mortality for 615 participants for early tracheostomy (≤ 14 days; see Analysis 2.1; Analysis 3.1). Additionally, we found data in the early tracheostomy (≤ 10 days) studies for 28‐day mortality, 90‐day mortality and mortality as time to event (see Analysis 2.2; Analysis 2.3; Analysis 2.4). In the early tracheostomy (≤ 14 days) studies we found data for 28‐, 60‐ and 90‐day mortality (see Analysis 3.2; Analysis 3.3; Analysis 3.4). For all these mortality outcomes, we are uncertain whether early tracheostomy (≤ 10 days or ≤ 14 days) increases or decreases mortality because of very low‐certainty evidence. Our main reasons for downgrading were serious risk of bias due to confounding, serious inconsistency and indirectness due to statistical heterogeneity, point estimates varying widely, inconsistent direction, extremely serious imprecision due to point estimates varying widely, and wide confidence intervals.

2.1. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 1: Overall mortality

3.1. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 1: In‐hospital mortality

2.2. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 2: 28‐day (± 2) mortality

2.3. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 3: 90‐day mortality

2.4. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 4: Mortality (time to event)

3.2. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 2: 28‐day (± 2) mortality

3.3. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 3: 60‐day mortality

3.4. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 4: 90‐day mortality

For an overview of all available data, including NRSIs comparing early tracheostomy (≤ 7 days, ≤ 12 days, ≤ 21 days) with late tracheostomy, see Analysis 4.1; Analysis 4.2; Analysis 4.3; Analysis 4.4 and Analysis 4.5.

4.2. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 2: 28‐day (± 2) mortality

4.3. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 3: 60‐day (± 3) mortality

4.4. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 4: 90‐day (± 1) mortality

4.5. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 5: Mortality (time to event)

Improvement of clinical status: duration to liberation from invasive mechanical ventilation

Randomized controlled trial

One study reported this outcome for 150 participants in the ITT analysis (see Analysis 1.2). Considering the reported event rates, we found that early tracheostomy (≤ 10 days) may result in little to no difference in duration to liberation from mechanical ventilation compared with late tracheostomy (MD 1.50 days fewer, 95% CI 5.74 days fewer to 2.74 days more; 1 study, 150 participants; low‐certainty evidence). Our main reasons for downgrading were serious imprecision due to the small number of participants in only one study and wide confidence intervals.

We did not find any data for the remaining outcomes.

Non‐randomized studies of interventions

One study reported this outcome for 50 participants for early tracheostomy (≤ 10 days), and two studies reported this outcome for 109 participants for early tracheostomy (≤ 14 days); (see Analysis 2.5; Analysis 3.6).

2.5. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 5: Duration to liberation from invasive mechanical ventilation

3.6. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 6: Duration to liberation from invasive mechanical ventilation

Improvement of clinical status: secondary outcomes

Additionally, we found data for liberation from invasive mechanical ventilation and time to decannulation in the early tracheostomy (≤ 14 days) studies (see Analysis 3.7; Analysis 3.8). In the early tracheostomy (≤ 10 days) studies we found data for time to decannulation (see Analysis 2.6). We did not find any data for need for invasive mechanical ventilation and ventilator‐free days.

3.7. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 7: Liberation from mechanical ventilation

3.8. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 8: Duration to decannulation

2.6. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 6: Duration to decannulation

For all these improvement outcomes, we are uncertain whether early tracheostomy (≤ 10 days or ≤ 14 days) increases or decreases improvement of clinical status because of very low‐certainty evidence. Our main reasons for downgrading were very serious risk of bias due to confounding, very serious inconsistency due to point estimates varying widely, inconsistent direction, and extremely serious imprecision due to the wide confidence intervals and the few participants included in the studies.

For an overview of all available data, including NRSIs comparing early tracheostomy (≤ 7 days, ≤ 12 days, ≤ 21 days) with late tracheostomy, see Analysis 4.6; Analysis 4.7 and Analysis 4.8.

4.6. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 6: Duration to liberation from invasive mechanical ventilation

4.7. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 7: Liberation from invasive mechanical ventilation

4.8. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 8: Duration to decannulation

Worsening of clinical status: adverse events (any grade)

Randomized controlled trial

One study reported this result for 150 participants in the ITT analysis (see Analysis 1.3). Considering the reported event rates, we found that early tracheostomy (≤ 10 days) may result in little to no difference in the incidence of adverse events compared with late tracheostomy (RR 0.94, 95% CI 0.79 to 1.13; RD 47 fewer per 1000, 95% CI 164 fewer to 102 more; 1 study, 150 participants; low‐certainty evidence). Our main reasons for downgrading were serious imprecision due to the small number of participants in only one study and the 95% confidence interval including both benefits and harms.

Non‐randomized studies of interventions

We did not find data for adverse events (any grade) in NRSIs.

For an overview of all available data, see Analysis 4.9.

4.9. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 9: Adverse events

Worsening of clinical status: ventilator‐associated pneumonia

Randomized controlled trial

One study reported this outcome for 150 participants in the ITT analysis (see Analysis 1.4). Considering the reported event rates, we found that early tracheostomy (≤ 10 days) may result in little to no difference in ventilator‐associated pneumonia compared with late tracheostomy (RR 1.08, 95% CI 0.23 to 5.20; RD 3 more per 1000, 95% CI 30 fewer to 162 more; 1 study, 150 participants; low‐certainty evidence). Our main reasons for downgrading were serious imprecision due to the small number of participants in only one study, and the 95% confidence interval including both benefit and harms.

Non‐randomized studies of interventions

Three studies reported this outcome as ventilator‐associated pneumonia for early tracheostomy (≤ 14 days) for 953 participants (see Analysis 3.9). Considering the event rates reported in the studies, we are uncertain whether early tracheostomy (≤ 14 days) increases or decreases the incidence of ventilator‐associated pneumonia, because we graded the certainty of evidence as very low (RR 0.78, 95% CI 0.66 to 0.91; I² = 0%; RD 111 fewer per 1000, 95% CI 171 fewer to 45 fewer; 3 studies, 953 participants; very low‐certainty evidence). Our main reasons for downgrading were serious risk of bias due to confounding and serious imprecision due to the 95% confidence interval including both benefits and harms.

3.9. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 9: Ventilator‐associated pneumonia

For an overview of all available data, see Analysis 4.10.

4.10. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 10: Ventilator‐associated pneumonia

Worsening of clinical status: need for renal replacement therapy

Randomized controlled trial

We did not find data for this outcome.

Non‐randomized studies of interventions

One study reported this outcome as need for renal replacement therapy for early tracheostomy (≤ 14 days) for 72 participants (see Analysis 3.10). Considering the event rates reported in the study, we are uncertain whether early tracheostomy (≤ 14 days) increases or decreases the need for renal replacement therapy because we graded the certainty of evidence as very low (RR 0.08, 95% CI 0.01 to 1.30; I² = 0%; RD 365 fewer per 1000, 95% CI 393 fewer to 119 more; 1 study, 72 participants; very low‐certainty evidence). Our main reasons for downgrading were serious risk of bias due to confounding, serious imprecision due to the small number of participants in only one study, wide confidence intervals, and 95% confidence interval including both benefit and harms.

3.10. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 10: Need for renal replacement therapy

For an overview of all available data, see Analysis 4.11.

4.11. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 11: Need for renal replacement therapy

Worsening of clinical status: secondary outcomes

Randomized controlled trial

One study reported postoperative bleeding for 150 participants in the ITT analysis (see Analysis 1.5). Considering the reported event rates, we found that early tracheostomy (≤ 10 days) may result in little to no difference regarding tracheal bleeding compared with late tracheostomy (RR 0.99, 95% CI 0.47 to 2.11; RD 2 fewer per 1000, 95% CI 82 fewer to 171 more; 1 study, 150 participants; low‐certainty evidence). Our main reasons for downgrading were serious imprecision due to the small number of participants in only one study, wide confidence intervals, and the 95% confidence interval including both benefits and harms.

One study reported airway obstruction for 150 participants in the ITT analysis (see Analysis 1.6). Considering the reported event rates, we found that performing an early tracheostomy (≤ 10 days) may result in a slight increase in the incidence of airway obstruction compared with a late tracheostomy (RR 2.44, 95% CI 0.78 to 7.57; RD 74 more per 1000, 95% CI 11 fewer to 337 more; 1 study, 150 participants; low‐certainty evidence). Our main reasons for downgrading were serious imprecision due to the small number of participants in only one study, wide confidence intervals, and the 95% confidence interval including both benefits and harms.

We did not find any data for tracheal stenosis, need for ECMO, ventilatory problems and serious adverse events.

Non‐randomized studies of interventions

One study reported postoperative bleeding for 50 participants for early tracheostomy (≤ 10 days; Analysis 2.7), and two studies reported this outcome for 152 participants for early tracheostomy (≤ 14 days; Analysis 3.11). For both analyses we are uncertain whether early tracheostomy (≤ 10 days or ≤ 14 days) increases or decreases the risk of postoperative bleeding because we graded the certainty of evidence as very low(≤ 10 days: RR 2.81, 95% CI 0.36 to 22.24; RD 101 more per 1000, 95% CI 36 fewer to 1180 more; 1 study, 50 participants; very low‐certainty evidence; and ≤ 14 days: RR 1.22, 95% CI 0.33 to 4.45; I² = 58%; RD 31 more per 1000, 95% CI 93 fewer to 471 more; 2 studies, 152 participants; very low‐certainty evidence). Our main reasons for downgrading in the analysis for 10 days or fewer were serious risk of bias due to confounding, bias in classification of interventions, and serious imprecision because of only one study. We downgraded both analyses due to serious imprecision due to the small number of participants and wide confidence intervals in the study with the 95% confidence interval including both benefits and harms.

2.7. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 7: Postoperative bleeding

3.11. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 11: Postoperative bleeding

One study reported this outcome as tracheal stenosis for early tracheostomy (≤ 10 days) for 50 participants (see Analysis 2.8). Considering the event rates reported in the study, we are uncertain whether early tracheostomy (≤ 10 days) increases or decreases the incidence of tracheal stenosis, because we graded the certainty of evidence as very low (RR 17.73, 95% CI 0.92 to 342.69; RD 200 fewer per 1000, 95% CI 200 fewer to 200 fewer; 1 study, 48 participants; very low‐certainty evidence). Our main reasons for downgrading were serious risk of bias due to confounding, serious imprecision due to the small number of participants in only one study, wide confidence intervals, and the 95% confidence interval including both benefit and harms.

2.8. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 8: Tracheal stenosis

Two studies reported this outcome as need for ECMO for early tracheostomy (≤ 14 days) for 152 participants (see Analysis 3.12). Considering the event rates reported in the study, we are uncertain whether early tracheostomy (≤ 14 days) increases or decreases the need for ECMO because we graded the certainty of evidence as very low (RR 0.69, 95% CI 0.03 to 15.21; I² = 0%; RD 57 fewer per 1000, 95% CI 180 fewer to 2631 more; 2 studies, 152 participants; very low‐certainty evidence). Our main reasons for downgrading were serious risk of bias due to confounding, bias in selection of participants and classification of interventions, serious imprecision due to the small number of participants, wide confidence intervals, and the 95% confidence interval including both benefit and harms.

3.12. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 12: Need for ECMO

One study reported this outcome as ventilatory problems for early tracheostomy (≤ 10 days) for 50 participants (see Analysis 2.9). Considering the event rates reported in the study, we are uncertain whether early tracheostomy (≤ 10 days) increases or decreases the incidence of ventilatory problems because we graded the certainty of evidence as very low (RR 2.81, 95% CI 0.36 to 22.24; RD 101 fewer per 1000, 95% CI 36 fewer to 1180 more; 1 study, 50 participants; very low‐certainty evidence). Our main reasons for downgrading were serious risk of bias due to confounding, serious imprecision due to the small number of participants in only one study, wide confidence intervals, and the 95% confidence interval including both benefit and harms.

2.9. Analysis.

Comparison 2: NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy, Outcome 9: Ventilatory problems

We did not find data for airway obstruction and serious adverse events in the included NRSIs (≤ 10 days versus > 10 days; ≤ 14 days versus > 14 days).

For an overview of all available data, including NRSIs that compared early tracheostomy (≤ 7 days, ≤ 12 days, ≤ 21 days) with late tracheostomy, see Analysis 4.10; Analysis 4.12; Analysis 4.13; Analysis 4.14 and Analysis 4.15.

4.12. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 12: Postoperative bleeding

4.13. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 13: Airway obstruction

4.14. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 14: Tracheal stenosis

4.15. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 15: Need for ECMO

For an additional narrative description, see Table 5; Table 6; Table 7.

ICU length of stay or time to discharge from ICU

Randomized controlled trial

One study reported this outcome for 150 participants in the ITT analysis (see Analysis 1.7). Considering the reported event rates, we found that early tracheostomy (≤ 10 days) may result in little to no difference in terms of the ICU length of stay compared with late tracheostomy (MD 0.5 days fewer, 95% CI 5.34 days fewer to 4.34 days more; 1 study, 150 participants; low‐certainty evidence). Our main reasons for downgrading were serious imprecision due to the small number of participants in only one study and wide confidence intervals.

Non‐randomized studies of interventions

Three NRSIs reported this outcome for 519 participants with early tracheostomy (≤ 10 days) as medians, which could not be included in the meta‐analyses. Evrard 2021 reported a median of 14 days in the early tracheostomy group (n = 10) and 38 days in the control group (n = 38). Polok 2021 reported a median of 23.3 days in the early tracheostomy group (n = 135) and 33.17 days in the control group (n = 315). Volo 2021 reported a median of 20 days in the early tracheostomy group (n = 8) and 31.4 days in the control group (n = 13). Considering reported event rates, we are uncertain whether early tracheostomy (≤ 10 days) increases or decreases the ICU length of stay compared with late tracheostomy because we graded the certainty of evidence as very low. Our main reasons for downgrading were serious risk of bias due to confounding, serious imprecision due to the small number of participants, wide confidence intervals and missing data.

One NRSI reported this outcome for 72 participants with early tracheostomy (≤ 14 days (Arnold 2022); see Analysis 3.14). Considering reported event rates, we are uncertain whether early tracheostomy (< 14 days) increases or decreases the ICU length of stay compared with late tracheostomy because we graded the certainty of evidence as very low (MD 6.70 days fewer, 95% CI 16.46 days fewer to 3.06 days more; 1 study, 72 participants; very low‐certainty evidence). Our main reasons for downgrading were serious risk of bias due to confounding, and serious imprecision due to the wide confidence intervals with the 95% confidence interval including both benefits and harms.

3.14. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 14: ICU length of stay

All results for these outcomes are presented in Analysis 4.17.

4.17. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 17: ICU length of stay

Hospital length of stay

We did not find any data for this outcome.

Quality of life

We did not find any data for this outcome.

Viral clearance

We did not find any data for this outcome.

Subgroup analyses

We did not find data for our predefined subgroup analyses.

Polok 2021 specifically included elderly critically ill patients over 70 years of age, but did not report any of our prioritized outcomes, so we could not perform a subgroup analysis for comparison with younger patients from any of the other included studies.

Sensitivity analyses

Comparison of intention‐to‐treat‐analysis with per protocol analysis

Eeg‑Olofsson 2022 conducted all analyses for the two randomization groups in the ITT population and the per protocol population. We performed sensitivity analyses comparing ITT and per protocol for our primary endpoints.

For duration to liberation from invasive mechanical ventilation, the ITT analysis showed no statistically significant difference between early and late tracheostomy (MD 1.50 days fewer, 95% CI 5.74 days fewer to 2.74 days more; 1 study, 150 participants; low‐certainty evidence). The per protocol analysis suggests that early tracheostomy may reduce the duration of invasive mechanical ventilation (MD 8.00 days fewer, 95% CI 13.73 days fewer to 2.27 days fewer; 1 study, 61 participants; low‐certainty evidence); see Analysis 5.2.

5.2. Analysis.

Comparison 5: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)), Outcome 2: Time to liberation from invasive mechanical ventilation

For the ICU length of stay, ITT analysis showed no statistically significant difference between early and late tracheostomy (MD 0.5 days fewer, 95% CI 5.34 days fewer to 4.34 days more; 1 study, 150 participants; low‐certainty evidence). The per protocol analysis suggests that early tracheostomy may shorten ICU length of stay (MD 6.6 days fewer, 95% CI 13.67 days fewer to 0.47 days more; 1 study, 61 participants; low‐certainty evidence); see Analysis 5.7.

5.7. Analysis.

Comparison 5: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)), Outcome 7: ICU length of stay

After sensitivity analysis, the results in effect did not differ for the following outcomes: overall mortality, worsening of clinical status ‐ adverse events (any grade), ventilator‐associated pneumonia, postoperative bleeding, and airway obstruction; see Analysis 5.1; Analysis 5.3; Analysis 5.4; Analysis 5.5; Analysis 5.6.

5.1. Analysis.

Comparison 5: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)), Outcome 1: Overall mortality

5.3. Analysis.

Comparison 5: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)), Outcome 3: Adverse events

5.4. Analysis.

Comparison 5: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)), Outcome 4: Ventilator‐associated pneumonia

5.5. Analysis.

Comparison 5: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)), Outcome 5: Postoperative bleeding

5.6. Analysis.

Comparison 5: RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)), Outcome 6: Airway obstruction

Risk of bias assessment components (studies with low risk of bias or some concerns versus studies with a high risk of bias)

The overall risk of bias judgment from all NRSIs was critical. Therefore, sensitivity analysis was not necessary.

Comparison of preprints versus peer‐reviewed articles

We did not include any preprints.

Comparison of prematurely terminated studies with completed studies

The interim analysis by an independent Data and Safety Monitoring Board (DSMB) showed that, after 50% of patients were enrolled in the Eeg‑Olofsson 2022 RCT, there were no differences between the two study arms in terms of ICU death or serious tracheostomy complications. However, per protocol analysis showed a statistically significant difference (P = 0.014) for time to relief from invasive mechanical ventilation in favor of the early tracheostomy group. The DSMB judged these results to be stable even when 180 participants were included. The DSMB recommended stopping enrollment of patients after analyzing safety and futility parameters.

None of the other included studies were terminated prematurely, so we did not perform a sensitivity analysis for comparison of prematurely terminated studies with completed studies for the outcomes.

Discussion

Summary of main results

The aim of this systematic review is to assess the safety and effects of early tracheostomy compared to late tracheostomy in critically ill COVID‐19 patients. This is the first version of this systematic review. We included one RCT and 24 NRSIs with 6522 participants. The studies included only participants with severe disease progression of COVID‐19 who had to be admitted to the ICU with respiratory failure during the further course of the disease and had to be mechanically ventilated. The time to liberation from mechanical ventilation largely determines the course of hospitalization. Thus, this duration has a strong impact on acute health‐related quality of life, functional independence, and autonomy of the individual patient. Therefore, we analyzed the duration of mechanical ventilation after tracheostomy as a surrogate for the severity of COVID‐19 disease progression and a shortening as an improvement in clinical condition. Complications associated with the intervention may negatively affect the course and, thus, are a measure of worsening of the clinical condition. In summary, this is represented by hospital length of stay and overall mortality.

Early (≤ 10 days after intubation) versus late tracheostomy (> 10 days after intubation)

We have focused first on the results reported in the RCT, as this represents the highest‐quality evidence at present in a review of the available studies (see Table 1).

Early tracheostomy (≤ 10 days) may result in little to no difference in overall mortality compared with late tracheostomy.

We are uncertain whether early tracheostomy (≤ 10 days) increases or decreases the chance of clinical improvement in terms of duration to liberation from mechanical ventilation, because early tracheostomy may result in little to no difference in duration to liberation from mechanical ventilation compared with late tracheostomy.

We are uncertain whether early tracheostomy (≤ 10 days) increases or decreases the risk of clinical worsening, because early tracheostomy may result in little to no difference in the incidence of adverse events (any grade) compared with late tracheostomy and the incidence of ventilator‐associated pneumonia.

Early tracheostomy (≤ 10 days) may result in little to no difference in ICU length of stay compared with late tracheostomy.

After reviewing all results of NRSIs for mortality, we are uncertain whether early tracheostomy (≤ 10 days) increases or decreases overall mortality. Based on the analyses of NRSIs, we are uncertain whether early tracheostomy (≤ 10 days) increases or decreases duration to liberation from mechanical ventilation. We did not find data for adverse events (any grade) or the incidence of ventilator‐associated pneumonia in NRSIs investigating early tracheostomy (≤ 10 days). We are uncertain whether early tracheostomy (≤ 10 days) increases or decreases ICU length of stay. We rated the certainty of evidence from the majority of NRSIs as very low, because of critical risk of bias, few participants, or wide CIs. A finding that there is a difference may reflect the insufficiency of the studies rather than there being an actual benefit to performing an early tracheostomy.

Overall completeness and applicability of evidence

We identified one RCT and 24 NRSIs, with a total number of 6522 participants. Of these, 2356 participants were assigned to an early tracheostomy group. All included studies were conducted primarily in hospitals in high‐ and middle‐income countries and investigated the performance of early or late tracheostomy in hospitalized adults with COVID‐19 and its impact on clinical outcomes.

A detailed review of the design of the NRSIs and their analysis revealed substantial heterogeneity. The definition of early and late tracheostomy groups varied widely, because the timing of tracheostomy has not yet been standardized and, in all studies, the decision to perform tracheostomy lies within the attending physician, which may bias the results. Definitions of early tracheostomy ranged from 7 to 21 days after intubation or start of invasive mechanical ventilation (116 participants for early tracheostomy ≤ 7 days; 412 participants for early tracheostomy ≤ 10 days; 61 participants for early tracheostomy ≤ 12 days; 820 participants for early tracheostomy ≤ 14 days and 96 participants for early tracheostomy ≤ 21 days). The definition of what constitutes the start of the timing also differed between the studies. The majority of the included studies (n = 22) used 'intubation' as the starting time point; in two studies, the start of mechanical ventilation was considered day 0 before the tracheostomy, and in one study, admission to critical care was chosen as the starting point before tracheostomy.

Limitations of the evidence

Our confidence in the evidence is limited because we found only one RCT of good quality with a small number of participants.

In particular, the differences in definitions of the timing in the early and late tracheostomy groups, and the differences in reported outcomes, made comparisons of outcomes and pooling in the analyses almost impossible, making it difficult to draw conclusions from the results. Although we contacted almost all authors of the NRSIs, particularly with regard to a detailed description of our defined end points, we received few responses supplying missing data. The differences in the reported data and the incomplete data sets have limited the conducting of the planned analyses. In particular, because it is thus not possible to identify the effects of the intervention on the incidence of serious or rare adverse events. Patients' pre‐existing conditions and condition at the time of the procedure were reported very differently and therefore could not be included in the analyses. It would have been important to define subgroups based on baseline characteristics to see which patients might benefit from an early intervention. The studies also sometimes reported other COVID‐19 treatment options performed or combinations of these treatments. It would also have been important to define subgroups based on concomitant treatment options to determine which patients might benefit from this combination with an early intervention.

Therefore, uncertainties remain regarding the validity of the results of the NRSIs due to the partly poor study quality and the incompleteness of the data (see Quality of the evidence and Potential biases in the review process).

Quality of the evidence

We included data from one RCT to assess the safety and impact of early tracheostomy compared to late tracheostomy in critically ill COVID‐19 patients. We report data of NRSIs only because they provide additional insight into the existing evidence. We assessed the certainty of the evidence using the GRADE approach and explained the rationale for each downgrade (see Table 1). The evidence for effect and safety outcomes was of low certainty.

Overall mortality

We downgraded by two levels to low certainty because of imprecision in the analysis due to the 95% CI including both benefits and harms, and the small number of participants studied in only one RCT.

For NRSIs, the main reasons for downgrading were serious risk of bias due to confounding, serious inconsistency and indirectness due to statistical heterogeneity, widely varying point estimates and inconsistent direction, as well as extremely serious imprecision due to widely varying point estimates and wide CIs across studies, resulting in very low‐certainty evidence overall for this endpoint.

Improvement of clinical status: duration to liberation from invasive mechanical ventilation

We downgraded by two levels to low certainty because of imprecision in the analysis due to the 95% CI including both benefits and harms, and the small number of participants studied in only one RCT.

For NRSIs (≤ 10 days and ≤ 14 days), our main reasons for downgrading were very high risk of bias due to confounding, very high inconsistency due to widely varying point estimates and inconsistent direction, and extremely high imprecision due to wide CIs and small numbers of participants included in the studies, resulting in very low‐certainty evidence overall for this end point.

Worsening of clinical status: adverse events (any grade)

We downgraded by two levels to low certainty because of imprecision in the analysis due to the 95% CI including both benefits and harms, and the small number of participants studied in only one RCT.

None of the included NRSIs reported this outcome in this way.

Worsening of clinical status: ventilator‐associated pneumonia

We downgraded by two levels to low certainty because of imprecision in the analysis due to the 95% CI including both benefits and harms, and the small number of participants studied in only one RCT.

For NRSIs (≤ 14 days), our main reasons for downgrading were serious risk of bias due to confounding, serious imprecision due to the 95% CI including both benefit and harm, resulting in very low‐certainty evidence overall for this end point.

Worsening of clinical status: need for renal replacement therapy

The RCT did not report data on this outcome.

For one NRSI (≤ 14 days), the main reasons for downgrading were serious risk of bias due to confounding, serious imprecision due to the small number of participants in only one study, wide CIs in the study, and the 95% CI including both benefits and harms, resulting in very low‐certainty evidence overall for this outcome.

ICU length of stay

We downgraded by two levels to low certainty because of imprecision in the analysis due to the 95% CI including both benefits and harms, and the small number of participants studied in only one RCT.

For NRSIs (≤ 10 days and ≤ 14 days), our main reasons for downgrading were serious risk of bias due to confounding and serious imprecision due to missing data, wide CIs in the studies, and the 95% CI including both benefit and harm, resulting in very low‐certainty evidence overall for this end point.

Potential biases in the review process

Experienced medical information specialists from the CEOsys consortium developed the multiple comprehensive and sensitive search strategies to identify all completed and ongoing studies for inclusion in this systematic review to answer our research question. The search included relevant electronic databases and clinical trials registries. Reference lists of included studies were also consulted to supplement the search. Primarily, peer‐reviewed, full‐text articles were considered, but also preprints. Because preprints are of lower quality and changes may occur when peer‐reviewed journal publications become available, searches were repeated several times, and several articles originally available as preprints were also included as the most recently published articles. Because of the large amount of missing data, we attempted to contact the authors of the studies (see Included studies). An overview of the included studies is provided in Table 5; Table 6; Table 7.

We are confident that we identified all relevant studies that were available at the time of the searches.

Agreements and disagreements with other studies or reviews

The timing of tracheostomy in critically ill patients was a subject of debate before the COVID‐19 pandemic (Andriolo 2015; Siempos 2015). This COVID‐19 Cochrane Review is based on the existing evidence from a Cochrane Review by Andriolo 2015. The authors stated that additional, high‐quality RCTs are needed to better assess benefits and harms between early and late tracheostomy in critically ill patients, and that the evidence should be improved by standardizing the results so that they can be included in meta‐analyses. But even after this review and several decades of intense debate, there is still no clear evidence on the optimal timing of tracheostomy in ventilated patients in the ICU because of conflicting study results (Littlewood 2001; McWhorter 2003).

Recommendations on the optimal time point to perform tracheostomy in critically ill COVID‐19 patients vary widely in published guidelines and studies (Bier‐Laning 2021; Faris 2020; Tay 2020). Early in the pandemic, there were reports of mortality rates ranging from 52% to 86% in critically ill COVID‐19 patients who were mechanically ventilated (Yang 2020; Wu 2020), so the presumed futility of tracheostomy in this patient population has been debated and has also led to a 'wait‐and‐see' approach. However, as the pandemic progressed, higher‐quality studies showed mortality rates of 24% to 26% in critically ill COVID‐19 patients requiring mechanical ventilation (Grasselli 2020; Richardson 2020), which are more consistent with mortality rates known for the population of ventilated critically ill non‐COVID‐19 patients. Thus, the pathophysiology and prognosis of CARDS were more frequently compared to other forms of ARDS, and recommendations on ventilatory therapy were derived from the guidelines on invasive ventilation in acute respiratory failure (Pfeifer 2020). Furthermore, especially in the earlier publications on COVID‐19, there was a remarkable heterogeneity in definitions of specific disease states, different but unnamed severity levels, different timing of disease and different populations, and different reported concomitant symptoms of COVID‐19, which made comparisons extremely difficult. In addition, studies were published very rapidly, especially at the beginning of the pandemic, with very small cohorts. Many of the studies published and subsequently cited require further review because many had not undergone a standard review process (Pfeifer 2020). For these reasons, previously available and evidence‐based guidance for the treatment of similar diseases was used to provide recommendations for COVID‐19 therapy.

Due to this unclear evidence base, the intention of our analysis was to provide an overview of the studies from the beginning of the pandemic to the present time and to derive evidence‐based recommendations. Notably, no previous review has been able to bring together this many studies on the timing of tracheostomy in patients with COVID‐19 in meta‐analyses.

During our work on this review, we also identified two other systematic reviews and meta‐analyses on the comparison between early and late timing of tracheostomy in critically ill COVID‐19 patients and compared them with our results (Chong 2022; Ji 2022). Both publications included only NRSIs, which are all also included in our analysis but are not comparable with each other after close examination of their designs, cohorts, outcomes, and definitions of the two groups. The comparison of the outcomes as presented by Chong 2022 and Ji 2022 in their systematic reviews does not seem to be appropriate when the early and late groups partially overlap with such heterogeneous groups in the cohorts.

Hence, we subdivided the groups according to the timing of early tracheostomy to create more comparable groups. This way, we were able to compare cohorts that received early tracheostomy up to day 10 after intubation and groups in which early tracheostomy was performed up to day 14 after intubation. The groups in which an early tracheostomy (< 10 days) was performed are most comparable to the definitions of early tracheostomy that were investigated for the treatment of comparable diseases before the onset of the pandemic and have now been used to guide the treatment of CARDS. We have only presented the group with tracheostomies performed fewer than 14 days after intubation to represent a group that most studies have formed because they followed the recommendations prevalent at the onset of the pandemic.

Another difference from the two previously published systematic reviews is that our most recent search in June 2022 identified one RCT (Eeg‑Olofsson 2022). We considered this recently published RCT to provide the best currently available evidence, and we therefore included only this study in the summary of findings. Due to the quality of the RCT and its informative value, we have presented the highest level of evidence currently available on the question at hand, when compared to previous publications, as NRSIs alone, due to their substantial weaknesses, cannot present such a high level of evidence.

Similar limitations have been reported by the existing evidence from studies and reviews on other indications for long‐term ventilation. The results of our analysis are consistent with most previous systematic reviews on other indications for long‐term ventilation, with similar limitations mentioned earlier (Andriolo 2015). The reviews by Dunham 2006, Griffiths 2005, Liu 2015 and Meng 2016 found no benefit for either early or late tracheostomy in reducing mortality based on the data considered by the authors. The meta‐analysis by Siempos 2015 reported that early tracheostomy may be associated with lower ICU mortality compared with late tracheostomy. The systematic review by Shan 2013 also reported that early tracheostomy reduced ICU length of stay, duration of mechanical ventilation, and mortality, but only six NRSIs were included, which may have overemphasized the results.

The lack of high‐quality studies prior to the COVID‐19 pandemic is also striking. The majority of the evidence consists of non‐RCTs, some of which have been systematically pooled in reviews. Most of the studies have shown that early tracheostomy can reduce the duration of ventilatory support (Arabi 2004; Arabi 2009; Blot 1995; Dunham 2006; Gandía‐Martínez 2010; Lesnik 1992; Zagli 2010), can shorten the length of stay in the ICU (Arabi 2004; Arabi 2009; Gandía‐Martínez 2010; Griffiths 2005; Lesnik 1992; Zagli 2010), and may also shorten the length of hospital stay (Arabi 2004; Arabi 2009; Blot 1995) compared with late tracheostomy. However, all results have limitations because of large clinical, methodological, and timing differences between studies.

Meta‐analyses such as the present one often face heterogeneity of data. Heterogeneity may result from both methodological differences in study conduct and clinical differences in study populations. In previous meta‐analyses also investigating early compared to late tracheostomy, the main sources of heterogeneity, as also reported here, are the different inclusion and exclusion criteria for the intervention, the different patient populations, different tracheostomy techniques, and the different definitions of early and late tracheostomy (Liu 2015) in particular. These differences may lead to differences in intervention effects, and a pooled result could be confounded because it would not capture a true effect in any of the populations. For this reason, Liu 2015 pooled data only when there was moderate or low heterogeneity (I² statistic ≤ 60%), and showed a benefit for the early tracheostomy group only for length of ICU stay. Thus, no conclusions were derived from the included NRSIs, because there was considerable heterogeneity within these studies.

An important limitation of studies addressing the optimal timing of tracheostomy is the decision on the timing of tracheostomy, which in most studies is left to the attending physician and for which protocols are rarely available. Here, the accuracy with which the duration of mechanical ventilation can be predicted presents a problem. In all studies included in our meta‐analysis and in most included in previous meta‐analyses, supervising physicians with clinical expertise predicted prolonged mechanical ventilation and provided the indication for tracheostomy. No consistent predictive criteria for presumed long‐term ventilation can be found at this time, and thus the indication remains subjective.

Authors' conclusions

Implications for practice.

Based on the single included randomized controlled trial (RCT), we found low‐certainty evidence that early tracheostomy (≤ 10 days after intubation) may result in little to no difference in overall mortality compared with late tracheostomy in critically ill COVID‐19 patients requiring prolonged mechanical ventilation. In terms of clinical improvement, early tracheostomy (≤ 10 days) may result in little to no difference in duration to liberation from mechanical ventilation compared with late tracheostomy. Considering the low‐certainty evidence, we are not sure whether early tracheostomy influences clinical worsening in terms of the incidence of adverse events, need for renal replacement therapy, ventilator‐associated pneumonia, or the length of stay in the intensive care unit.

The available evidence should be viewed with extreme caution because many studies were published rapidly and did not undergo a standard review process. Reporting is highly variable and characterized by missing data, therefore we do not have sufficient information to draw conclusions about subgroups. Clinical heterogeneity is a typical characteristic of studies addressing this question and can significantly influence the results by leading to confounding variables that may confound the true effects of the investigated intervention. This may affect the applicability of the results to real clinical situations and may make it difficult to draw conclusions about the overall effectiveness of the treatment in different patient groups.

Implications for research.

It is remarkable that even before the pandemic, there was no consensus on the timing of tracheostomy. Thus, a new clinical entity was met without clear recommendations. From this, we can also derive the importance of targeted questions and disclosure of research findings for guideline development. Due to the stabilization of the pandemic, we do not expect further studies to address our question regarding critically ill COVID‐19 patients. Nevertheless, we should seek consensus on whether there is a potential benefit of early tracheostomy compared to late tracheostomy in critically ill patients with similar disease entities to ensure a more precise indication for this intervention and thus greater patient safety.

The RCT by Eeg‑Olofsson 2022 was terminated prematurely because a survival benefit for the group of early tracheostomized patients became apparent even before the planned number of participants was reached, and the ethics committee determined that this effect would persist even if additional patients were included. Therefore, it did not seem ethically justifiable to include additional patients and presumably expose them to an increased risk of death, and it does not seem justifiable today to conduct similar RCTs to assess the potential benefit of early tracheostomy compared to late tracheostomy in critically ill patients (with or without COVID‐19).

The NRSIs failed to provide greater insight into this question, due to the qualitative inferiority manifested in the described methodological weaknesses and high risk of confounding, as we found very low‐certainty evidence in all NRSIs. Because it is very difficult for the treating intensivist to accurately predict the need for mechanical ventilation when early tracheostomy is indicated, it is unavoidable that a number of patients will receive this intervention unnecessarily. Therefore, future research should include the development of validated scoring systems to predict prolonged mechanical ventilation more accurately. These could then be used in new RCTs to ensure a more precise indication and thus greater patient safety. NRSIs could also provide valuable answers to clinical questions. Here, it would be desirable to conduct high‐quality NRSIs that implement measures against bias due to confounding, bias in selection of participants, and bias in classification of interventions, such as random sequence generation and allocation concealments. Blinding of participants and involved personnel would be difficult to perform for this question, but at least partial blinding of outcome assessment could be performed. Selective reporting should also be avoided; the time point of intervention and the complete outcome data sets should be disclosed.

To improve the evidence on the potential benefit of interventions in severe COVID‐19 or similar clinical conditions and establish comparability, the following endpoints should be considered in a standardized manner: mortality rates at longer follow‐up, up to 12 months; duration of mechanical ventilation; as well as length of hospital and ICU stay. Other potentially relevant outcomes such as adverse events and serious adverse events, related to the procedure, need for ECMO, and in‐hospital infections should also be considered. Quality of life after disease survival and successful weaning should also be considered during longer follow‐up periods. It would be of great benefit to standardize the results and make the raw data of studies available so that they can be summarized in meta‐analyses. Pooling multiple studies for the same outcome in meta‐analyses increases the certainty of the evidence by increasing the number of participants. This would provide more accurate data after re‐analysis to identify patients who might benefit from early tracheostomy. In addition, it would be helpful to generate data on subgroups and reduce clinical heterogeneity to draw conclusions about which groups of patients might particularly benefit from an early intervention.

Risk of bias

Risk of bias for analysis 1.2 Duration to liberation from invasive mechanical ventilation.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. But assessment of the outcome was unlikely to be influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Risk of bias for analysis 1.3 Adverse events.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 1.4 Ventilator‐associated pneumonia.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 1.5 Postoperative bleeding.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 1.6 Airway obstruction.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 1.7 ICU length of stay.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. But assessment of the outcome is unlikely to be influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Risk of bias for analysis 5.1 Overall mortality.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 5.1.1 ITT

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Subgroup 5.1.2 PP

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Risk of bias for analysis 5.2 Time to liberation from invasive mechanical ventilation.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 5.2.1 ITT

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. But assessment of the outcome was unlikely to be influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Subgroup 5.2.2 PP

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. But assessment of the outcome was unlikely to be influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Risk of bias for analysis 5.3 Adverse events.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 5.3.1 ITT

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Subgroup 5.3.2 PP

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 5.4 Ventilator‐associated pneumonia.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 5.4.1 ITT

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Subgroup 5.4.2 PP

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 5.5 Postoperative bleeding.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 5.5.1 ITT

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Subgroup 5.5.2 PP

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 5.6 Airway obstruction.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 5.6.1 ITT

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Subgroup 5.6.2 PP

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Some concerns

Outcome assessors were not blinded. Thus, assessment of the outcome could have been influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available and outcome assessors were not blinded.

Risk of bias for analysis 5.7 ICU length of stay.

Study

Bias

Randomisation process

Deviations from intended interventions

Missing outcome data

Measurement of the outcome

Selection of the reported results

Overall

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Authors' judgement

Support for judgement

Subgroup 5.7.1 ITT

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. But assessment of the outcome is unlikely to be influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Subgroup 5.7.2 PP

Eeg‑Olofsson 2022

Some concerns

No information regarding allocation concealment

Low risk of bias

There were no deviations from intended interventions, patients were blinded and an appropriate analysis was used to estimate the effect of assignment to intervention.

Low risk of bias

Outcome data was available for all patients

Low risk of bias

Outcome assessors were not blinded. But assessment of the outcome is unlikely to be influenced by knowledge of intervention received.

Low risk of bias

Selection of reported results was done appropriately, a research protocol was prepared

Some concerns

No information regarding allocation concealment was available

Appendices

Appendix 1. Search strategies

Cochrane COVID‐19 Study Register

tracheost* OR tracheot*

Intervention assignment: 
Randomised & Unclear
Study design:
Parallel/Crossover & Single Arm/Controlled Before After & Time Series & Unclear

Web of Science – Science Citation Index and Emerging Sources Citation Index

# 1 TI=((COVID OR COVID19) OR ("SARS‐CoV‐2" OR "SARS‐CoV2" OR SARSCoV2 OR"SARSCoV‐2" OR "SARS coronavirus 2") OR ("2019 nCoV" OR "2019nCoV" OR "2019‐novel CoV" OR "nCov 2019" OR "nCov 19") OR ("severe acute respiratory syndrome coronavirus 2" OR "novel coronavirus disease" OR "novel corona virus disease" OR "corona virus disease 2019" OR "coronavirus disease 2019" OR "novel coronavirus pneumonia" OR "novel corona virus pneumonia") OR ("severe acute respiratory syndrome coronavirus 2") )

# 2 AB=((COVID OR COVID19) OR ("SARS‐CoV‐2" OR "SARS‐CoV2" OR SARSCoV2 OR "SARSCoV‐2" OR "SARS coronavirus 2") OR ("2019 nCoV" OR "2019nCoV" OR "2019‐novel CoV" OR "nCov 2019" OR "nCov 19") OR ("severe acute respiratory syndrome coronavirus 2" OR "novel coronavirus disease" OR "novel corona virus disease" OR "corona virus disease 2019" OR "coronavirus disease 2019" OR "novel coronavirus pneumonia" OR "novel corona virus pneumonia") OR ("severe acute respiratory syndrome coronavirus 2") )

# 3 #1 OR #2

# 4 TI=(tracheost* OR tracheot*)

# 5 AB=(tracheost* OR tracheot*)

# 6 #4 OR #5

# 7 #3 AND #6

WHO COVID‐19 Global literature on coronavirus disease

Advanced search in search fields: title, abstract, subject

tracheost* OR tracheot*

Data and analyses

Comparison 1. RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
1.1 Overall mortality	1	150	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.52, 1.29]
1.2 Duration to liberation from invasive mechanical ventilation	1	150	Mean Difference (IV, Random, 95% CI)	‐1.50 [‐5.74, 2.74]
1.3 Adverse events	1	150	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.79, 1.13]
1.4 Ventilator‐associated pneumonia	1	150	Risk Ratio (M‐H, Random, 95% CI)	1.08 [0.23, 5.20]
1.5 Postoperative bleeding	1	150	Risk Ratio (M‐H, Random, 95% CI)	0.99 [0.47, 2.11]
1.6 Airway obstruction	1	150	Risk Ratio (M‐H, Random, 95% CI)	2.44 [0.78, 7.57]
1.7 ICU length of stay	1	150	Mean Difference (IV, Random, 95% CI)	‐0.50 [‐5.34, 4.34]

Comparison 2. NRSI: early (≤ 10 days) vs late (> 10 days) tracheostomy.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
2.1 Overall mortality	2	719	Risk Ratio (M‐H, Random, 95% CI)	1.47 [0.43, 5.00]
2.2 28‐day (± 2) mortality	1	51	Risk Ratio (M‐H, Random, 95% CI)	1.24 [0.80, 1.93]
2.3 90‐day mortality	1	450	Risk Ratio (M‐H, Random, 95% CI)	1.06 [0.87, 1.29]
2.4 Mortality (time to event)	2	844	Hazard Ratio (IV, Random, 95% CI)	0.94 [0.62, 1.42]
2.5 Duration to liberation from invasive mechanical ventilation	1	50	Mean Difference (IV, Random, 95% CI)	1.98 [‐0.16, 4.12]
2.6 Duration to decannulation	1	50	Mean Difference (IV, Random, 95% CI)	‐4.20 [‐10.33, 1.93]
2.7 Postoperative bleeding	1	50	Risk Ratio (M‐H, Random, 95% CI)	2.81 [0.36, 22.24]
2.8 Tracheal stenosis	1	48	Risk Ratio (M‐H, Random, 95% CI)	17.73 [0.92, 342.69]
2.9 Ventilatory problems	1	50	Risk Ratio (M‐H, Random, 95% CI)	2.81 [0.36, 22.24]

Comparison 3. NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
3.1 In‐hospital mortality	6	615	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.79, 1.37]
3.2 28‐day (± 2) mortality	2	225	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.76, 1.46]
3.3 60‐day mortality	3	305	Risk Ratio (M‐H, Random, 95% CI)	1.23 [0.81, 1.89]
3.4 90‐day mortality	2	225	Risk Ratio (M‐H, Random, 95% CI)	1.01 [0.96, 1.07]
3.5 Mortality (time to event)	2	213	Hazard Ratio (IV, Random, 95% CI)	0.65 [0.18, 2.30]
3.6 Duration to liberation from invasive mechanical ventilation	2	109	Mean Difference (IV, Random, 95% CI)	‐4.69 [‐17.21, 7.84]
3.7 Liberation from mechanical ventilation	4	449	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.88, 1.20]
3.8 Duration to decannulation	2	109	Mean Difference (IV, Random, 95% CI)	‐4.66 [‐16.80, 7.49]
3.9 Ventilator‐associated pneumonia	3	953	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.66, 0.91]
3.10 Need for renal replacement therapy	1	72	Risk Ratio (M‐H, Random, 95% CI)	0.08 [0.01, 1.30]
3.11 Postoperative bleeding	2	152	Risk Ratio (M‐H, Random, 95% CI)	1.22 [0.33, 4.45]
3.12 Need for ECMO	2	152	Risk Ratio (M‐H, Random, 95% CI)	0.69 [0.03, 15.21]
3.13 Ventilatory problems	1	682	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.76, 1.66]
3.14 ICU length of stay	1	72	Mean Difference (IV, Random, 95% CI)	‐6.70 [‐16.46, 3.06]
3.15 Hospital length of stay	1	72	Mean Difference (IV, Random, 95% CI)	‐5.40 [‐15.36, 4.56]

3.5. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 5: Mortality (time to event)

3.13. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 13: Ventilatory problems

3.15. Analysis.

Comparison 3: NRSI: early (≤ 14 ± 1 day) vs late (> 14 ± 1 day) tracheostomy, Outcome 15: Hospital length of stay

Comparison 4. Early vs late tracheostomy.

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
4.1 Overall mortality and in‐hospital mortality	11		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.1.1 RCT: early ≤ 10 days vs late > 10 days	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.34, 1.17]
4.1.2 Early ≤ 7 days vs late > 7 days	1	38	Risk Ratio (M‐H, Random, 95% CI)	0.57 [0.32, 1.03]
4.1.3 Early ≤ 10 days vs late > 10 days	2	719	Risk Ratio (M‐H, Random, 95% CI)	1.29 [0.28, 6.02]
4.1.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.1.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	6	615	Risk Ratio (M‐H, Random, 95% CI)	1.04 [0.79, 1.37]
4.1.6 Early ≤ 21 days vs late > 21 days	1	176	Risk Ratio (M‐H, Random, 95% CI)	2.19 [1.11, 4.31]
4.2 28‐day (± 2) mortality	8		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.2.1 Early ≤ 7 days vs late > 7 days	2	182	Risk Ratio (M‐H, Random, 95% CI)	1.29 [0.77, 2.18]
4.2.2 Early ≤ 10 days vs late > 10 days	2	747	Risk Ratio (M‐H, Random, 95% CI)	1.00 [0.66, 1.50]
4.2.3 Early ≤ 12 days vs late > 12 days	1	120	Risk Ratio (M‐H, Random, 95% CI)	1.83 [0.89, 3.77]
4.2.4 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	2	225	Risk Ratio (M‐H, Random, 95% CI)	1.05 [0.76, 1.46]
4.2.5 Early ≤ 21 days vs late > 21 days	1	189	Risk Ratio (M‐H, Random, 95% CI)	22.29 [1.33, 372.88]
4.3 60‐day (± 3) mortality	4		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.3.1 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.3.2 Early ≤ 10 days vs late > 10 days	1	696	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.60, 1.12]
4.3.3 Early ≤ 12 days vs late > 12 days	1	120	Risk Ratio (M‐H, Random, 95% CI)	1.17 [0.72, 1.91]
4.3.4 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	2	233	Risk Ratio (M‐H, Random, 95% CI)	1.34 [0.78, 2.30]
4.3.5 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.4 90‐day (± 1) mortality	5		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.4.1 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.4.2 Early ≤ 10 days vs late > 10 days	2	1146	Risk Ratio (M‐H, Random, 95% CI)	0.84 [0.51, 1.38]
4.4.3 Early ≤ 12 days vs late > 12 days	1	120	Risk Ratio (M‐H, Random, 95% CI)	1.17 [0.72, 1.91]
4.4.4 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	2	273	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.85, 1.26]
4.4.5 Early ≤ 21 days vs late > 21 days	1	169	Risk Ratio (M‐H, Random, 95% CI)	1.89 [1.01, 3.52]
4.5 Mortality (time to event)	4		Hazard Ratio (IV, Random, 95% CI)	Subtotals only
4.5.1 Early ≤ 7 days vs late > 7 days	0		Hazard Ratio (IV, Random, 95% CI)	Not estimable
4.5.2 Early ≤ 10 days vs late > 10 days	2		Hazard Ratio (IV, Random, 95% CI)	0.94 [0.62, 1.42]
4.5.3 Early ≤ 12 days vs late > 12 days	0		Hazard Ratio (IV, Random, 95% CI)	Not estimable
4.5.4 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	2		Hazard Ratio (IV, Random, 95% CI)	0.65 [0.18, 2.30]
4.5.5 Early ≤ 21 days vs late > 21 days	0		Hazard Ratio (IV, Random, 95% CI)	Not estimable
4.6 Duration to liberation from invasive mechanical ventilation	5		Mean Difference (IV, Random, 95% CI)	Subtotals only
4.6.1 RCT: early ≤ 10 days vs late > 10 days	1	61	Mean Difference (IV, Random, 95% CI)	‐8.00 [‐13.73, ‐2.27]
4.6.2 Early ≤ 7 days vs late > 7 days	1	38	Mean Difference (IV, Random, 95% CI)	3.00 [‐9.78, 15.78]
4.6.3 Early ≤ 10 days vs late > 10 days	1	50	Mean Difference (IV, Random, 95% CI)	1.98 [‐0.16, 4.12]
4.6.4 Early ≤ 12 days vs late > 12 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.6.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	2	109	Mean Difference (IV, Random, 95% CI)	‐4.69 [‐17.21, 7.84]
4.6.6 Early ≤ 21 days vs late > 21 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.7 Liberation from invasive mechanical ventilation	5		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.7.1 Early ≤ 7 days vs late > 7 days	1	38	Risk Ratio (M‐H, Random, 95% CI)	2.20 [0.95, 5.12]
4.7.2 Early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.7.3 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.7.4 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	4	449	Risk Ratio (M‐H, Random, 95% CI)	1.03 [0.88, 1.20]
4.7.5 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.8 Duration to decannulation	3		Mean Difference (IV, Random, 95% CI)	Subtotals only
4.8.1 RCT: early ≤ 10 days vs late > 10 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.8.2 Early ≤ 7 days vs late > 7 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.8.3 Early ≤ 10 days vs late > 10 days	1	50	Mean Difference (IV, Random, 95% CI)	‐4.20 [‐10.33, 1.93]
4.8.4 Early ≤ 12 days vs late > 12 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.8.5 Early ≤ 14 (±1) days vs late > 14 (±1) days	2	109	Mean Difference (IV, Random, 95% CI)	‐4.66 [‐16.80, 7.49]
4.8.6 Early ≤ 21 days vs late > 21 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.9 Adverse events	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.9.1 RCT: early ≤ 10 days vs late > 10 days	1	150	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.79, 1.13]
4.9.2 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.9.3 Early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.9.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.9.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.9.6 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.10 Ventilator‐associated pneumonia	5		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.10.1 RCT: early ≤ 10 days vs late > 10 days	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.06, 6.58]
4.10.2 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.10.3 Early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.10.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.10.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	3	953	Risk Ratio (M‐H, Random, 95% CI)	0.78 [0.66, 0.91]
4.10.6 Early ≤ 21 days vs late > 21 days	1	188	Risk Ratio (M‐H, Random, 95% CI)	2.94 [0.12, 71.20]
4.11 Need for renal replacement therapy	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.11.1 RCT: early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.11.2 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.11.3 Early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.11.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.11.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	1	72	Risk Ratio (M‐H, Random, 95% CI)	0.08 [0.01, 1.30]
4.11.6 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.12 Postoperative bleeding	6		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.12.1 RCT: early ≤ 10 days vs late > 10 days	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.37, 2.39]
4.12.2 Early ≤ 7 days vs late > 7 days	1	65	Risk Ratio (M‐H, Random, 95% CI)	1.27 [0.56, 2.90]
4.12.3 Early ≤ 10 days vs late > 10 days	1	50	Risk Ratio (M‐H, Random, 95% CI)	2.81 [0.36, 22.24]
4.12.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.12.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	3	834	Risk Ratio (M‐H, Random, 95% CI)	1.64 [0.75, 3.59]
4.12.6 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.13 Airway obstruction	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.13.1 RCT: early ≤ 10 days vs late > 10 days	1	150	Risk Ratio (M‐H, Random, 95% CI)	2.44 [0.78, 7.57]
4.13.2 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.13.3 Early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.13.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.13.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.13.6 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.14 Tracheal stenosis	2		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.14.1 RCT: early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.14.2 Early ≤ 7 days vs late > 7 days	1	65	Risk Ratio (M‐H, Random, 95% CI)	2.98 [0.15, 59.52]
4.14.3 Early ≤ 10 days vs late > 10 days	1	48	Risk Ratio (M‐H, Random, 95% CI)	17.73 [0.92, 342.69]
4.14.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.14.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.14.6 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.15 Need for ECMO	3		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.15.1 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.15.2 Early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.15.3 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.15.4 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	2	152	Risk Ratio (M‐H, Random, 95% CI)	0.69 [0.03, 15.21]
4.15.5 Early ≤ 21 days vs late > 21 days	1	188	Risk Ratio (M‐H, Random, 95% CI)	2.94 [0.12, 71.20]
4.16 Ventilatory problems	2		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
4.16.1 RCT: early ≤ 10 days vs late > 10 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.16.2 Early ≤ 7 days vs late > 7 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.16.3 Early ≤ 10 days vs late > 10 days	1	50	Risk Ratio (M‐H, Random, 95% CI)	2.81 [0.36, 22.24]
4.16.4 Early ≤ 12 days vs late > 12 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.16.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	1	682	Risk Ratio (M‐H, Random, 95% CI)	1.12 [0.76, 1.66]
4.16.6 Early ≤ 21 days vs late > 21 days	0	0	Risk Ratio (M‐H, Random, 95% CI)	Not estimable
4.17 ICU length of stay	5		Mean Difference (IV, Random, 95% CI)	Subtotals only
4.17.1 RCT: early ≤ 10 days vs late > 10 days	1	148	Mean Difference (IV, Random, 95% CI)	‐0.50 [‐5.37, 4.37]
4.17.2 Early ≤ 7 days vs late > 7 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.17.3 Early ≤ 10 days vs late > 10 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.17.4 Early ≤ 12 days vs late > 12 days	1	120	Mean Difference (IV, Random, 95% CI)	‐11.60 [‐18.15, ‐5.05]
4.17.5 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	3	181	Mean Difference (IV, Random, 95% CI)	‐5.97 [‐13.29, 1.36]
4.17.6 Early ≤ 21 days vs late > 21 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.18 Hospital length of stay	1		Mean Difference (IV, Random, 95% CI)	Subtotals only
4.18.1 Early ≤ 7 days vs late > 7 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.18.2 Early ≤ 10 days vs late > 10 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.18.3 Early ≤ 12 days vs late > 12 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable
4.18.4 Early ≤ 14 (± 1) days vs late > 14 (± 1) days	1	81	Mean Difference (IV, Random, 95% CI)	8.10 [1.75, 14.45]
4.18.5 Early ≤ 21 days vs late > 21 days	0	0	Mean Difference (IV, Random, 95% CI)	Not estimable

4.16. Analysis.

Comparison 4: Early vs late tracheostomy, Outcome 16: Ventilatory problems

Comparison 5. RCT: early (≤ 10 days) vs late (> 10 days) tracheostomy (intention‐to‐treat (ITT) vs per protocol (PP)).

Outcome or subgroup title	No. of studies	No. of participants	Statistical method	Effect size
5.1 Overall mortality	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
5.1.1 ITT	1	150	Risk Ratio (M‐H, Random, 95% CI)	0.82 [0.52, 1.29]
5.1.2 PP	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.34, 1.17]
5.2 Time to liberation from invasive mechanical ventilation	1		Mean Difference (IV, Random, 95% CI)	Subtotals only
5.2.1 ITT	1	150	Mean Difference (IV, Random, 95% CI)	‐1.50 [‐5.74, 2.74]
5.2.2 PP	1	61	Mean Difference (IV, Random, 95% CI)	‐8.00 [‐13.73, ‐2.27]
5.3 Adverse events	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
5.3.1 ITT	1	150	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.79, 1.13]
5.3.2 PP	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.91 [0.76, 1.08]
5.4 Ventilator‐associated pneumonia	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
5.4.1 ITT	1	150	Risk Ratio (M‐H, Random, 95% CI)	1.08 [0.23, 5.20]
5.4.2 PP	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.63 [0.06, 6.58]
5.5 Postoperative bleeding	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
5.5.1 ITT	1	151	Risk Ratio (M‐H, Random, 95% CI)	0.98 [0.46, 2.08]
5.5.2 PP	1	61	Risk Ratio (M‐H, Random, 95% CI)	0.94 [0.37, 2.39]
5.6 Airway obstruction	1		Risk Ratio (M‐H, Random, 95% CI)	Subtotals only
5.6.1 ITT	1	150	Risk Ratio (M‐H, Random, 95% CI)	2.44 [0.78, 7.57]
5.6.2 PP	1	61	Risk Ratio (M‐H, Random, 95% CI)	2.52 [0.69, 9.15]
5.7 ICU length of stay	1		Mean Difference (IV, Random, 95% CI)	Subtotals only
5.7.1 ITT	1	150	Mean Difference (IV, Random, 95% CI)	‐0.50 [‐5.34, 4.34]
5.7.2 PP	1	61	Mean Difference (IV, Random, 95% CI)	‐6.60 [‐13.67, 0.47]

Characteristics of studies

Characteristics of included studies [ordered by study ID]

Angel 2021.

Study characteristics
Methods	Trial design: multicenter prospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 11 March 2020‐30 July 2020, follow‐up until death or the end of 30 July 2020 Country: USA Language: English Number of centers: ICUs at 2 large metropolitan hospitals in New York City, USA
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 541/394, intervention ‐ ET: 116, control ‐ LT: 89, no tracheostomy: 189 Age (years, median): intervention group 59, control group 64 Gender (male n (%)): intervention group 93 (80), control group 60 (71) Comorbidities: not reported Inclusion criteria Adult patients, ≥ 18 years, admitted to the ICU at 2 New York University (NYU) Langone Health Hospitals, suffering from respiratory failure caused by SARS CoV‐2 infection, confirmed by PCR, requiring IMV were included The selected patients met the following requirements on MV: PEEP < 12 cm H2O, FiO2 0.6 respiratory rate < 30 breaths/min PaCO2 < 60 mm Hg Exclusion criteria Significant extrapulmonary organ dysfunction (except for acute renal failure on dialysis Required high dose of vasopressors (> 0.05 µg/kg/min of norepinephrine or equivalent) Showing active bleeding secondary to severe coagulopathy
Interventions	The main exposure variable was ET versus LT. The definition of ET vs late PDT was based on consensus statement recommendations of avoiding ETs (defined as < 14 d) in patients with COVID‐19 disease. Procedure: bedside PDT with modified visualization and ventilation Treatment details of intervention group: 'early' on median time of 9 days after initiation of mechanical ventilation (IQR 7–12 d) Treatment details of control group: 'late' on median time of 19 days after initiation of mechanical ventilation (IQR 16–24 d)
Outcomes	Primary study outcomes Discontinuation from MV by median duration of MV and probability of discontinuation of MV Length of hospitalization Mortality by overall survival All complications during the PDTs and the follow‐up period were reported (e.g. dialysis, ECMO) Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (R) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Arnold 2022.

Study characteristics
Methods	Trial design: single‐center prospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: March 2020‐April 2021, final date for data cut‐off was April 2021, median length of follow‐up of 45 (IQR: 16 ‐135) days Country: USA Language: English Number of centers: 1 tertiary care, teaching hospital in Chicago, Illinois (ICU)
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 79/79, intervention ‐ ET: 14/14, control ‐ LT: 58/58 Age (total(IQR)): 66 (58‐71) Gender (female, n (%)): 51 (71) Comorbidities (N (%)): hypertension 44 (61) diabetes mellitus 36 (50) obesity 24 (35) coronary artery disease 14 (19) chronic kidney disease 13 (18) solid organ malignancy 10 (14) heart failure 10 (14) atrial fibrillation 9 (13) obstructive sleep apnea 8 (11) cerebrovascular disease 6 (8) chronic lung disease 5 (7) solid organ transplantation 5 (7) end‐stage renal disease 5 (7) deep vein thrombosis/pulmonary embolism 4 (6) cirrhosis 2 (3) hematologic malignancy 2 (3) Inclusion criteria Patients with SARS‐CoV‐2 who had percutaneous bedside tracheostomy performed by the Interventional Pulmonary team for prolonged respiratory failure from March 2020 to April 2021 Exclusion criteria Not defined
Interventions	The main exposure variable was ET versus LT Procedure: percutaneous tracheostomies were performed at the bedside by providers wearing powered air purifying respirators, gowns, and gloves. The procedures followed standard practices for percutaneous tracheostomy, with additional steps taken to minimize aerosolization including packing the oropharynx with gauze to minimize aerosolization when the cuff on the endotracheal tube was deflated Treatment details of intervention group: 'early' ‐ patients who were on MV for ≤ 14 days prior to surgery Treatment details of control group: 'late' was defined as occurring >14 days after orotracheal intubation
Outcomes	Primary study outcome Overall mortality and decannulation rates Secondary study outcome Time to weaning from MV Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (R) at day 60 (R) at day 90 (R) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (R) ventilator‐free days (NR) time to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (R) postoperative bleeding (R) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (R) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (R) Quality of life (NR) Viral clearance (NR)
Notes	We contacted study authors, who provided requested data that had not yet been published and could be included in our analyses for the outcomes mortality at day 60, mortality at day 90, need for renal replacement therapy, postoperative bleeding, need for ECMO, ICU length of stay, hospital length of stay

Avilés‐Jurado 2020.

Study characteristics
Methods	Trial design: single‐center prospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 16 March‐10 April 2020, final date for data cut‐off was 14 April 2021 Country: Spain Language: English Number of centers: 1 tertiary referral center (ICU)
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 50/50, intervention ‐ ET: 32/32, control ‐ LT: 18/18 Age (mean(SD)): intervention group 62.6 (11.6), control group 64.53 (8.2) Gender (male, n (%)): intervention group 22 (68.7), control group 11 (61.1) Comorbidities (intervention group vs control group (n/N (%))): cardiovascular disease 9/56 (16) vs 15/61 (25) hypertension 19/32 (59.4) vs 8/18 (44.4) diabetes mellitus 8/32 (25) vs 1/18 (5.6) immunosuppression 2/32 (6.3) vs 2/18 (11.1) autoimmune disease 2/32 (6.3) vs 1 (5.6) > 2 comorbidities 18/32 (56.2) vs 7/18 (38.8) SOFA score, mean (SD) 6.3/32 (2.1) vs 6/18 (2.5) APACHE II score, mean (SD) 14.16/32 (4.3) vs 11.9/18 (3.9) Inclusion criteria Adult patients with confirmed COVID‐19 who were admitted to the ICU and required tracheostomy between 16 March and 10 April 2020 Exclusion criteria Not defined
Interventions	The main exposure variable was ET versus LT Procedure: a surgical tracheostomy was performed for all patients at the bedside (in the ICU) following recommended criteria for use of PPE Treatment details of intervention group: 'early' ‐ patients who were on MV for ≤ 10 days prior to surgery Treatment details of control group: 'late' was defined as occurring later than 10 days after orotracheal intubation
Outcomes	Primary study outcome Number of subthyroid operations Tracheal entrance protocol Ise of PPE Secondary study outcome Infections among the surgeons were monitored weekly by RT–PCR of nasopharyngeal swab samples, short‐term complications, weaning, and the association of timing of tracheostomy (early [≤ 10 days] vs late [> 10 days]) with total required days of IMV were assessed Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (R) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) time to decannulation (R) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (R) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (R) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Battaglini 2021.

Study characteristics
Methods	Trial design: multicenter, retrospective, observational study Type of publication: journal publication Setting: inpatient Recruitment dates: from 20 April 2020‐30 June 2020, follow‐up to 120 days Country: Italy Language: English Number of centers: 11 hospitals (ICUs)
Participants	Baseline characteristics Number randomised (in total and per arm) (included): 153, intervention ‐ ET: 76, control ‐ LT: 77 Age (years, mean (SD)): intervention group 63.8 (9.24), control group 62.9 (9.48) Gender (male n (%)): total 118 (77.1), intervention group 60 (78.9), control group 58 (75.3) Comorbidities (intervention group vs control group (n/N (%))): hypertension 42/76 (55.3) vs 40/77 (51.9) diabetes mellitus 20/76 (26.3)vs 20/77 (18.2) chronic respiratory disease 11/76 (14.7) vs 5/77 (6.5) chronic cardiac disease 12/76(15.8) vs 11/77 (14.3) malignancy 5/76 (6.6) vs 7/77 (9.1) chronic liver disease 3/76 (3.9) vs 3/77 (3.9) chronic neurologic disease 9/76 (11.8) vs 4/77 (5.2) chronic kidney disease 5/76 (6.6) vs 5/77 (5.2) smoke 5/76 (8.5) vs 5/77 (7.1) Inclusion criteria Critically ill patients with a positive RT‐PCR nasopharyngeal swab for SARS‐CoV‐2, admitted to the study ICUs, requiring IMV and subsequent surgical or percutaneous tracheostomy Exclusion criteria Patients who did not receive tracheostomy and aged < 18 years with a missing tracheostomy
Interventions	The main exposure variable was ET versus LT Stratified by tracheostomy timing (early versus late) and technique (surgical versus percutaneous) Procedure: PDTs, primarily performed at the bedside in the ICUs by intensivists, specialist head and neck surgeons performed surgical tracheostomies in patients with difficult anatomy or other presumed obstacles to percutaneous technique, mostly performed in the operating theater, and occasionally at the bedside in the ICU, as necessary Treatment details of intervention group: 'early' was defined as a tracheostomy performed within the first 15 days after endotracheal intubation Treatment details of control group: 'late' was defined as a tracheostomy performed ≥ 15 days after endotracheal intubation The reason for selecting 15 days as the cut‐off for early and LT was based on the median time to tracheostomy performance in the cohort. Reasons for tracheostomy were categorized as follows: prolonged weaning expected, neurological impairment (inability to maintain patient airways because of neurological condition), extubation failure (failure of SBT and/or extubation needing re‐intubation), and airway failure (inability to maintain patient airways because of upper airway causes).
Outcomes	Primary study outcome Median time to tracheostomy in critically ill COVID‐19 patients (defined as the time elapsed from endotracheal intubation to tracheostomy) Secondary study outcomes Post‐tracheostomy complications, stratified by tracheostomy timing (early vs late) and technique (surgical vs percutaneous) Days to weaning by tracheostomy timing, since orotracheal intubation and since tracheostomy (time to weaning was defined as the time from endotracheal intubation to the first SBT) SBT (defined as the first attempt to reduce respiratory support before extubation (removal of endotracheal tube and respiratory support), the execution of a SBT did not automatically lead to extubation) Time to extubation (defined as the time between the insertion and the removal of an artificial airway such as an endotracheal tube or tracheostomy tube) Extubation (defined as the removal of an artificial airway, this term was used either for the removal of an endotracheal tube or tracheostomy tube) Time to ICU admission (defined as the time from hospital admission to ICU admission) Time to endotracheal intubation in ICU (defined as the time from ICU admission to endotracheal intubation) ICU discharge (defined as the last day of ICU stay, irrespective of death or discharge to another non‐ICU ward) Length of ICU stay (defined as the time between ICU admission and ICU discharge, irrespective of death or discharge to another non‐ICU ward) Post‐tracheostomy ICU length of stay (the time between tracheostomy and ICU discharge) Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (R); at day 60 (R); at day 90 (R); time‐to‐event (NR). Improvement of clinical status: duration to liberation from invasive mechanical ventilation (NR); need for invasive mechanical ventilation (NR); liberation from invasive mechanical ventilation (NR); ventilator‐free days (NR); duration to decannulation (NR). Worsening of clinical status: adverse events (any grade) (NR); ventilator associated pneumonia (R); need for renal replacement therapy (NR); postoperative bleeding (NR); airway obstruction (NR); tracheal stenosis (NR); need for ECMO (NR); ventilatory problems (NR); SAEs (NR). ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Breik 2020.

Study characteristics
Methods	Trial design: single‐center prospective observational cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 9 March 2020‐21 April 2020, final date for data cut‐off was 21 April 2021 Country: UK Language: English Number of centers: 1 center (ICU)
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 100/100, intervention ‐ ET: 64/64, control ‐ LT: 36/36 Age (total (IQR)): 55.2 (21–78) Gender (male (%)): 71 Comorbidities (N (%))): BMI ((kg m−2) [SD]) : 32.0 [7.0] APACHE II score, mean (SD): 14 (4) Inclusion criteria All patients admitted to the ICU with severe respiratory failure requiring mechanical ventilation at the Queen Elizabeth Hospital Birmingham, UK, from 9 March 2020‐21 April 2020 were included. SARS‐CoV‐2 positivity was confirmed by RT‐PCR testing of nasopharyngeal swabs or non‐directed bronchial lavage/aspirate. Exclusion criteria Not defined
Interventions	The main exposure variable was ET vs LT Procedure: through multidisciplinary agreement, parameters to guide selection for tracheostomy were defined before the study period. Patients with physiology outside of these parameters were still considered for tracheostomy on a case‐by‐case basis Percutaneous and surgical techniques: the choice of performing a surgical or percutaneous tracheostomy depended only on patient body habitus, adequate neck extension, and grade of direct laryngoscopy Treatment details of intervention group: 'early' ‐ patients who were on MV for ≤ 14 days prior to surgery Treatment details of control group: 'late' was defined as occurring later than 14 days after orotracheal intubation
Outcomes	Primary study outcome 30‐day survival (from date of ICU admission), compared between tracheotomised patients and those who had no tracheostomy (primarily extubated) Secondary study outcomes Time to waking after ceasing sedation, duration of sedation and MV, discharge from ICU, tracheostomy decannulation rate, and complications The endpoint for ventilatory support was defined as when the patient wore a tracheostomy mask for at least 24 h 30‐day survival was compared for patients with APACHE scores of < 17 and > 17 Subgroup analyses performed on the timing of tracheostomy affected survival, time on ventilator, and length of ICU stay Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (R) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (R) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR). ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes	As the data reported were inconclusive and we did not receive a response from the study authors, we decided to exclude their data from our analyses.

Chandran 2021.

Study characteristics
Methods	Trial design: single‐center prospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: March 2020‐April 2021, final date for data cut‐off was April 2021, median length of follow‐up of 45 (IQR: 16‐135) days Country: USA Language: English Number of centers: 1 tertiary care, teaching hospital in Chicago, Illinois (ICU)
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 79/79, intervention ‐ ET: 14/14, control ‐ LT: 58/58 Age (total(IQR)): 66 (58‐71) Gender (male, n (%)): 32 (62.74) Comorbidities (N (%))): hypertension 44 (61) diabetes mellitus 36 (50) obesity 24 (35) coronary artery disease 14 (19) chronic kidney disease 13 (18) solid organ malignancy 10 (14) heart failure 10 (14) atrial fibrillation 9 (13) obstructive sleep apnea 8 (11) cerebrovascular disease 6 (8) chronic lung disease 5 (7) solid organ transplantation 5 (7) end‐stage renal disease 5 (7) deep vein thrombosis/pulmonary embolism 4 (6) cirrhosis 2 (3) hematologic malignancy 2 (3) Inclusion criteria Patients with SARS‐CoV‐2 who had percutaneous bedside tracheostomy performed by the interventional pulmonary team for prolonged respiratory failure from March 2020‐April 2021 Exclusion criteria Not defined
Interventions	The main exposure variable was ET vs LT Procedure: PDTs were performed at the bedside by providers wearing powered air purifying respirators, gowns, and gloves. The procedures followed a standard practices for PDT, with additional steps taken to minimize aerosolization including packing the oropharynx with gauze to minimize aerosolization when the cuff on the endotracheal tube was deflated Treatment details of intervention group: 'early' ‐ patients who were on mechanical ventilation for ≤ 10 days prior to surgery Treatment details of control group: 'late' was defined as occurring > 10 days after orotracheal intubation
Outcomes	Primary study outcome Overall mortality and decannulation rates Secondary study outcome Time to weaning from mechanical ventilation Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (R) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: adverse events (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Dal 2022.

Study characteristics
Methods	Trial design: single‐center retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 1 November 2020‐1 February 2021, follow‐up until hospital discharge Country: Turkey Language: English Number of centers: 1 pandemic hospital in Ankara
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 33/33, intervention ‐ ET: 18, control ‐ LT: 15, no tracheostomy: 0 early Age (years, median): total 54 (42.5–65.0) Gender (male %): total: 75 (63.6) Comorbidities (intervention group vs control group (n/N (%))): SAPS II score on admission to ICU, median (Q1–Q3): 35 (29–45) APACHE II 20.72±6.51/18.33±5.09 SOFA 5.5 (2.0–7.0)/4.0 (2.0–6.0) GCS 14.0 (10.5–15.0)/14.0 (14.0–15.0) hypertension 9 (50.0)/9 (60.0) diabetes mellitus 3 (16.7)/4 (26.7) COPD 1 (5.6)/1 (6.7) asthma 1 (5.6)/0 (0.0) cardiovascular disease 2 (11.1)/4 (26.7) malignancy 2 (11.1)/0 (0.0) chronic renal disease 0 (0.0)/1 (6.7) cerebrovascular disease 2 (11.1)/4 (26.7) at least 1 comorbidity 11 (61.1)/10 (66.7) Inclusion criteria Patients intubated in the ICU and underwent a bedside percutaneous tracheostomy due to a prolonged intubation period and whose data could be accessed Exclusion criteria Not defined
Interventions	The main exposure variable was ET vs LT Procedure: bedside PDT Treatment details of intervention group: ‘early’ (≤ 14 days) Treatment details of control group: ‘late’ (> 21 days)
Outcomes	Primary study outcomes Mortality ICU Length of stay Duration of MV Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Eeg‑Olofsson 2022.

Study characteristics
Methods	Trial design: RCT Allocation: randomized Intervention model: parallel assignment Intervention model description: randomized, single‐blinded, controlled trial Masking: single (participant) Description of masking: patients were not informed about the result of the randomization Primary purpose: treatment Type of publication: journal article Setting: inpatient Recruitment dates: from 6 June 2020‐20 April 2021 Country: Sweden Language: English Number of centers: 3
Participants	Baseline characteristics Ages eligible for study: ≥ 18 years (adult, older adult) Gender eligible for study: all Mean age was 65 years 79% were men Inclusion criteria Adult patients (≥ 18 years) Patients who were intubated due to RT–PCR‐verified, SARS‐CoV‐2 infection with ARDS according to the Berlin definition Patients who were hospitalized at the Sahlgrenska university hospital in Gothenburg or at 2 other county hospitals within the region Västra Götaland of Sweden (Södra Älvsborg Hospital, Borås and NU Hospital group, Trollhättan) Patients in whom a need for MV for > 14 days after intubation could not be ruled out (as assessed by the team of anaesthesiologists at the ICU in agreement with the study co‐ordinators) Exclusion criteria Patients where a tracheostomy performed within 7 days after intubation could be life‐threatening due to a poor medical condition (as assessed by the team of anaesthesiologists at the ICU in agreement with the study co‐ordinators) Patients with an anatomical abnormality of the neck impeding the tracheostomy procedure (as assessed by the anaesthesiologist or otolaryngologist) Patients with no informed consent
Interventions	The main exposure variable was ET versus LT Procedure: tracheostomy ‐ surgical procedure Treatment details of intervention group: 'early' tracheostomy was defined as occurring within 7 days after intubation. Treatment details of control group: 'late' tracheostomy was defined as occurring after at least 10 days after intubation The definition of early and LT was decided based on the 'Swedish national recommendations for tracheostomy' regarding the time span for which tracheotomies are generally recommended The 3‐day span between early and LT was regarded as not too short to hide differences and not too long to induce an increased risk of protocol deviations
Outcomes	Primary study outcomes Explore the optimal timing of tracheostomy in relation to the need for MV, specifically to compare ET (≤ 7 days after intubation) and LT (≥ 10 days after intubation) regarding the total number of days needed for MV (primary endpoint) number of days with MV Time frame: through the individual ICU stay assessed up to 60 days Secondary study outcomes Total number of days from intubation to tracheostomy Type of tracheostomy ICU stay: total number of days in the ICU number of days at ICU Sedation: total number of days that patients were in need of sedation total number of days that patients were in need of sedation AEs: need for reintubation, and complications various AEs associated with the tracheostomy/tracheostomy Mortality: mortality in the ICU, and mortality within 90 days of intubation (time frame: through the individual ICU stay assessed up to 60 days) Review outcomes Mortality: overall mortality (R) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (R) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: adverse events (any grade) (R) ventilator‐associated pneumonia (R) need for renal replacement therapy (NR) postoperative bleeding (R) airway obstruction (R) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes	Interim analysis according to the ITT and per protocol (PP) principles After inclusion of 90 patients (90 days after inclusion), the Data and Safety Monitoring Board (DSMB) recommended, after analysis of safety and futility parameters, to terminate the inclusion of patients, due to a statistically significant difference of PP analysis in favor of the early ET group We contacted study authors, who provided requested data that had not yet been published and could be included in our analyses.

Evrard 2021.

Study characteristics
Methods	Trial design: multicenter retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: 27 January 2020 (date of first COVID‐19 admission) and 18 May 2020 (last tracheostomy performed) Country: France Language: English Number of centers: 2 university hospitals in the Paris region
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 48/48, intervention ‐ ET: 10, control ‐ LT: 38, no tracheostomy: 0 early Age (years, median): total 56 (47–65) Gender (male %): total 36 (75) Comorbidities (intervention group vs control group (n/N (%))): BMI > 30 kg/m2 3 (30)/18 (47) chronic heart disease 2 (20)/4 (11) chronic kidney disease 0/4 (11) obstructive lung diseasea 0/6 (16) obstructive sleep apnea 0/4 (11) immunosuppression 4 (40)/2 (5) pregnancy 0 (0)/1 (3) hypertension 2 (20)/21 (55) diabetes mellitus 3 (30)/11 (29) current smoker 3 (30)/8 (21) Inclusion criteria All patients tracheostomized for a COVID‐19‐related ARDS Exclusion criteria Not defined
Interventions	The main exposure variable was ET versus LT Procedure: surgical and PDT Treatment details of intervention group: ‘early’ (≤ 10 days) Treatment details of control group: ‘late’ (> 10 days)
Outcomes	Primary study outcomes Patient outcome Date of first COVID‐19 symptoms and PCR results Level of respiratory support (oxygen therapy, MV support) ICU and hospital outcomes (length of MV, length of stay at hospital, vital status) MV and tracheostomy complications post‐ICU (after 6 months) Procedure outcome Timing of tracheostomy Length of procedure Complications Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (R) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Glibbery 2020.

Study characteristics
Methods	Trial design: prospective institutional review Type of publication: journal publication Setting: inpatient Recruitment dates: 15 March‐20 May 2020 Country: UK Language: English Number of centers: 1 UK tertiary referral centre (Addenbrooke’s Hospital, Cambridge)
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 28/28, intervention ‐ ET: 9, control ‐ LT: 19, no tracheostomy: 0 early Age (years, median): total 60.5 (25‐82) Gender (male %): total 20 (71.4) Comorbidities (n(%)): BMI (in kg/m2) underweight (BMI < 18.5) 0 (0.0) normal weight (BMI = 18.5–24.9) 3 (10.7) overweight (BMI = 25–29.9) 9 (32.1) moderately obese (BMI = 30–34.9) 7 (25.0) severely obese (BMI = 35–39.9) 5 (17.9) very severely obese (BMI ≥ 40) 4 (14.3) smoking status ex‐smoker 10 (35.7) current smoker 1 (3.6) asthma 5 (17.9) COPD 2 (7.1) other respiratory 1 (3.6) hypertension 12 (42.9) other cardiac 4 (14.3) immunosuppression 1 (3.6) renal disease 3 (10.7) diabetes 8 (28.6) other 9 (32.1) APACHE II score mean (SD) 13.5 (3.5) median (IQR) 14 (5) Inclusion criteria All COVID‐19 patients who underwent a surgical or percutaneous tracheostomy for weaning off mechanical ventilation Exclusion criteria Not defined
Interventions	The main exposure variable was ET versus LT Procedure: surgical and PDT Treatment details of intervention group: ‘early’ (≤ 14 days) Treatment details of control group: ‘late’ (> 14 days)
Outcomes	Primary study outcomes Post‐tracheostomy outcomes Weaning from intravenous sedation Weaning from MV Successful decannulation ICU discharge to a general ward Hospital discharge Complications (return to the operating theatre, failed decannulation, ICU re‐admission and death) Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (R) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (R) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Hansson 2022.

Study characteristics
Methods	Trial design: multicentre retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 14 March 2020‐14 March 2021, final date for data cut‐off was 14 April 2021 Country: Sweden Language: English Number of centers: 3 hospitals (ICUs)
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 117/117, intervention ‐ ET: 56/56, control ‐ LT: 61/61 Age (median/range): intervention group 67/ 22‐87, control group 66/18‐87 Gender (male, n (%)): intervention group 46 (82), control group 44 (72) Comorbidities (intervention group vs control group (n/N (%))): cardiovascular disease 9/56 (16) vs 15/61 (25) hypertension 27/56 (48) vs 37/56 (61) diabetes type 2 13/56 (23) vs 20/61 (33) liver cirrhosis 0/56 (0) vs 1/61 (2) neuromuscular disease 3/56 (6) vs 5/61 (3) immunosuppression 3/56 (6) vs 12/61 (20) asthma 6/56 (11) vs 12/61 (13) COPD 6/56 (11) vs 5/61 (8) other pulmonary disease 7/56 (13) vs 9/61 (15) active or previous smoking 19/59 (34) vs 21/61 (34) BMI > 30 kg/m2 21/59(38) vs 32/61 (53) high‐dose prophylactic LMWH 48/59 (86) vs 38/61 (62) SAPS 3, median [IQR] 60[49‐63] vs 58 [52‐63] Inclusion criteria Adult patients (> 18 years old) who were admitted to the ICU in Jönköping County between 14 March 2020 and 13 March, 2021. All included patients were diagnosed with severe ARDS caused by SARS CoV‐2 infection, as confirmed by positive PCR results using nasopharyngeal swabs, requiring IMV and subsequent tracheostomy. Exclusion criteria Patients with a missing tracheostomy, orotracheal intubation, or outcome date or missing age or sex were excluded. Patients who were treated in the ICU in Jönköping County but underwent tracheostomy elsewhere were excluded.
Interventions	The main exposure variable was ET vs LT Procedure: no standardization of the procedure due to the multicenter design Treatment details of intervention group: 'early' ‐ those who were on MV for < 7 days prior to surgery Treatment details of control group: 'late' was defined as occurring on ≥ day 7 after orotracheal intubation
Outcomes	Primary study outcome Duration of IMV Secondary study outcomes Included number of days on IMV prior to tracheostomy Time of decannulation Number of hours on mechanical ventilation in the prone position ICU length of stay All‐cause mortality (within 30 days of ICU admission) Complications associated with tracheostomy (hemorrhage, aspiration, displaced tracheal cannula, tracheal injury, failed surgery, accidental decannulation, stoma infection, perioperative hypoxemia, pneumothorax/pneumomediastinum/subcutaneous emphysema, fistulas, and airway obstruction related to tracheostomy, and procedure‐related death) Days to weaning by tracheostomy timing, since orotracheal intubation and since tracheostomy Mortality by cumulative incidence of death by tracheostomy timing Associations of tracheostomy with intraoperative and postoperative complications incidence (bleeding, ventilatory problems) Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (R) at day 60 (NR) at day 90 (NR) time‐to‐event (NR). Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Hernandez 2022.

Study characteristics
Methods	Trial design: multicentre propensity‐matched cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: between 15 February and 15 May 2020, final date for data cut‐off was 60 days after intubation Country: Spain Language: English Number of centers: 15 Spanish ICUs
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 682/682, intervention ‐ ET ≤ 14 days (≤ 7 days; 8‐10 days; 11‐14days): 382/382, control ‐ LT > 14 days (15‐20 days; ≥ 21 days): 300/300 Age (mean (IQR)): intervention groups (≤ 7 days; 8‐10 days; 11‐14 days) 62 (55‐70); 65 (56‐69); 64 (57‐71), control groups (15‐20 days; ≥ 21 days) 64 (57‐69); 65 (56‐72) Gender (male, n (%)): intervention groups (≤ 7 days; 8‐10 days; 11‐14 days) 42 (64.6); 88 (69.8); 136 (71.2), control groups (15‐20 days; ≥ 21 days) 149 (73.8); 74 (75.5) Comorbidities (intervention groups (≤ 7 days; 8‐10 days; 11‐14 days) vs control groups (15‐20 days; ≥ 21 days) (n/N (%))): BMI > 30 kg/m2 28 (43.1); 52 (41.3); 88 (46.1) vs 74 (36.6); 41 (41.8) heart disease 6 (9.2); 10 (7.9); 16 (8.4) vs 15 (7.4); 20 (20.4) COPD 2 (3.1); 2 (2.4); 11 (5.8) vs 8 (4); 4 (4.1) other respiratory disease 6 (9.2); 6 (4.8); 24 (12.6) vs 31 (15.3); 15 (15.3) Inclusion criteria All patients in 15 spanish ICUs diagnosed with hypoxemic respiratory failure secondary to RT‐PCR‐confirmed COVID‐19 pneumonia who underwent tracheostomy between 15 February and 15 May 2021 Exclusion criteria To prevent competing‐risk bias, patients with factors associated with tracheostomy: admission to the ICU with positive PCR results for COVID‐19, but without indications for MV for COVID‐19 pneumonia Admission after otorhinolaryngology surgery Low level of consciousness Swallowing dysfunction Neuromuscular disease other than ICU‐acquired weakness Tracheostomy Advanced directives to withhold life‐sustaining interventions or being expected to die before hospital discharge Post‐tracheostomy factors that could lead to immortal time bias were excluded, except the use of high‐flow oxygen therapy during weaning. For matched comparisons, patients in the LT cohort were selected according to the propensity score from among the remaining patients (≥ 8 days, ≥ 11 days, and ≥ 15 days, respectively).
Interventions	The main exposure variable was ET versus LT Procedure: no standardization of the procedure due to the multicentre design Treatment details of intervention group: ‘early’ ≤ 7 days, 8‐10 days, 11‐14 days Treatment details of control group: ‘late’ > 14 days, 15‐20 days, ≥ 21 days
Outcomes	Primary study outcome The primary outcomes were ventilator‐free days at 28 days (VFD28) , calculated as VFD28 = 28 – x, where x represents the number of days from intubation to liberation from ventilation or death Secondary study outcome Ventilator‐free days at 60 days (VFD60)(VFD60 = 60 – x, where x represents the number of days from intubation to liberation from ventilation or death) and modified ICU or hospital bed‐free days (BFD) at 28 days (BFD28 = 28 – y, where y represents the number of days from ICU or hospital admission to discharge to the ward or home or death) and at 60 days (BFD60 = 60 – y, where y represents the number of days from ICU or hospital admission to discharge to the ward or home or death) Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (R) need for renal replacement therapy (NR) postoperative bleeding (R) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (R) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes	The STROBE guidelines for reporting observational studies were followed. To prevent residual selection bias resulting from the lack of randomization of the timing of tracheostomy, cohorts were matched based on propensity scores. We contacted the study authors, who provided requested data that had not yet been published and could be included in our analyses.

Karna 2022.

Study characteristics
Methods	Trial design: single‐center retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: 1 May 2020‐30 April 2021 Country: India Language: English Number of centers: tertiary care institute in Central India
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 65/65, intervention ‐ ET: 41, control ‐ LT: 24, no tracheostomy: 0 early Age (years, median (IQR)): total 53 (40, 62) Gender (male %): total 42 (65%) Comorbidities (intervention group vs control group (n/N (%))): chronic kidney disease 4 (9.8)/0 (0) obstructive lung diseasea 0 (0)/1 (4.2) hypertension 19 (46)/11 (46) diabetes mellitus 19 (46)/7 (29) coronary artery disease 4 (10)/1 (4.2) chronic liver disease 0 (0)/0 (0) hypothyroidism 4 (9.8)/4 (17) autoimmune 0 (0)/0 (0) malignancy 2 (4.9)/0 (0) cerebrovascular 3 (7.3)/1 (4.2) others 3 (7.3)/3 (12) Inclusion criteria All COVID‐19 patients electively tracheostomized during intensive care from 1 May 2020‐30 April 2021 Exclusion criteria Not defined
Interventions	The main exposure variable was ET versus LT Procedure: surgical and PDT Treatment details of intervention group: ‘early’ (< 7 days) Treatment details of control group: ‘late’ (> 7 days)
Outcomes	Primary study outcomes Tracheostomy complications Weaning outcomes: Sedation wean Days on pressure support mode Time to ventilator liberation Decannulation success Tracheostomy to decannulation duration Duration of MV Duration of ICU stay Duration of hospital stay Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (R) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (R) airway obstruction (NR) tracheal stenosis (R) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (R) Quality of life (NR) Viral clearance (NR)
Notes

Kuno 2021.

Study characteristics
Methods	Trial design: single‐center retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 1 March 2020‐17 March 2020, follow‐up from their admission dates till 15 August 2020 or the date they expired Country: USA Language: English Number of centers: 1 hospital (ICU)
Participants	Baseline characteristics Number of participants (recruited/evaluated): 178/133, intervention ‐ ET: 86, control ‐ LT: 47 Age (years, mean (SD)): intervention group 61.2 (13,3), control group 61.2 (11.4) Gender (male, n (%)): intervention group 53 (61.6) control group 27 (57.4) Comorbidities (intervention group vs control group (n/N %)): hypertension 31.4/27.7 COPD 2.3/2.1 diabetes mellitus 25.6/25.5 chronic kidney disease 14.0/6.4 cancer 3.5/6.4 atrial fibrillation 7.1/9.3 heart failure 3.1/0.0 Inclusion criteria Patients suffering from respiratory failure caused by SARS CoV‐2 infection, requiring IMV and subsequent tracheostomy, performed before 17 March 2020 Exclusion criteria Patients with a missing timing of tracheostomy
Interventions	The main exposure variable was ET versus LT Procedure: no standardization of the procedure Treatment details of intervention group: 'Early' ‐ patients who underwent tracheostomy ≤ 14 days after intubation Treatment details of control group: 'late' ‐ patients who underwent tracheostomy > 14 days after intubation
Outcomes	Primary study outcome Overall mortality Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (R) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR); ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Kwak 2021.

Study characteristics
Methods	Trial design: single‐center retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 1 March 2020‐7 May 2020, follow‐up until death or the end of July 2020 Country: USA Language: English Number of centers: ICU at a single tertiary care medical center in New York City
Participants	Baseline characteristics Number of participants (in total and per arm) (included): 148, intervention ‐ ET: 52, control ‐ LT: 96 Age (years, overall mean (SD)) 58.1 (15.8) Gender (male n (%)): 120 (81) Comorbidities: BMI, mean (SD) 30.57 (6.48) Charlson Comorbidity Index score n (%) 0: 29 (19.6) 1: 27 (18.2) 2: 39 (26.4) 3: 29 (20.0) 4: 16 (10.8) 5: 4 (2.7) 6: 3 (2.0) 7‐8: 0 9: 1 (0.7) ≥10: 0 Inclusion criteria All adult patients tested positive for SARS‐CoV‐2 on RT‐PCR assay testing, admitted to the ICU at 1 New York University (NYU) Langone Health Hospital, intubated and subsequently underwent percutaneous or open tracheostomy NYULH indications and guidelines used in the selection of patients for ET during the SARS‐CoV‐2 pandemic: COVID‐19‐infected patients on MV for at least 5 days with anticipated prolonged MV no significant extra‐pulmonary organ dysfunction (with the exception of acute renal failure on dialysis) no anticipated continuation of prone positioning low dose of vasopressors requirements (< 0.05 µg/kg/min of norepinephrine or equivalent) no major coagulopathy the selected patients met the following requirements on MV: positive end‐expiratory pressure < 12 cm H2O fraction of inspired oxygen < 60%, respiratory rate < 25 breaths per minute, and PaCO2 < 60 mmHg or any patient within 24–36 h of being placed on ECMO Exclusion criteria Not reported
Interventions	The main exposure variable was ET vs LT Procedure: bedside PDT with modified visualization and ventilation Treatment details of intervention group: 'early' < 10 days of intubation Treatment details of control group: 'late' ≥ 10 days of intubation
Outcomes	Primary study outcomes: Time from symptom onset to (1) endotracheal intubation, (2) tracheostomy Time from endotracheal intubation to tracheostomy Time from tracheostomy to (1) tracheostomy tube downsizing, (2) decannulation Total time on MV Total length of stay Major in‐hospital events were noted, including the need for ECMO, cardiac arrest, and cerebrovascular accident Secondary study outcomes Time from symptom onset, as documented by patient report in the admission history and physical, to (a) intubation, (b) tracheostomy Time from intubation to (a) tracheostomy, (b) discontinuation of MV Time from tracheostomy to (a) downsizing of tracheostomy tube, (b) decannulation Total length of stay was calculated Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (R) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Livneh 2021.

Study characteristics
Methods	Trial design: single‐center retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: between March 2020 and January 2021, follow‐up until death or decannulation Country: Israel Language: English Number of centers: 1 Israeli tertiary referral center (ICU)
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 38/38, intervention ‐ ET: 19/19, control ‐ LT: 19/19 Age: years, median (IQR): intervention group 60 (54–67), control group 68 (59–74) Gender (male, n (%)): intervention group 16 (84), control group 17 (89) Comorbidities (intervention group vs control group (n/N (%))): chronic disease (includes obesity (BMI ≥ 30), diabetes, hypertension, and chronic lung disease) 16/19 (84) vs 17/19 (89) smoking, currently 2/19 (11) vs 1/19 (5) smoking, past 4/19 (21) vs 2/19 (11) Inclusion criteria Adult patients with confirmed COVID‐19 who were admitted to the ICU and required tracheostomy between between March 2020 and January 2021. ET group included all patients who were operated on within 7 days of intubation. LT group included all patients who were operated on in ≥ 8 days of intubation. Exclusion criteria Not defined
Interventions	The main exposure variable was ET versus LT Procedure: a surgical tracheostomy was performed for all patients at the bedside (in the designated COVID‐19 ICU Treatment details of intervention group: 'Early' ≤ 7 days of intubation Treatment details of control group: 'late' ≥ 8 days of intubation
Outcomes	Primary study outcome Mortality rate Secondary study outcomes Time to weaning from IMV Time to decannulation Time to discharge The effect of the tracheostomy on the dosage of vasopressor agents and sedatives (these were recorded and measured 3 days prior and 3 days after the procedure) Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (R) need for IMV (NR) liberation from IMV (R) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Mahmood 2021.

Study characteristics
Methods	Trial design: multicenter retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 1 February 2020‐4 September 2020, follow‐up until hospital discharge Country: USA Language: English Number of centers: 5 hospitals in USA
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 118/118, intervention ‐ ET: 9, control ‐ LT: 109, no tracheostomy: 0 Age (years, median): total 54 (42.5–65.0) Gender (male %): total: 75 (63.6) Comorbidities N (%): diabetes 49 (41.5) hypertension 53 (44.9) cardiovascular disease 13 (11) cerebrovascular disease 5 (4.2) COPD 2 (1.7) asthma 16 (13.6) liver disease 3 (2.5) renal disorder 26 (22) malignancy 2 (1.7) ARDS: any 118 (100) mild 10 (8.5) moderate 33 (28.0) severe 75 (63.6) SAPS II Score on admission to ICU, median (Q1–Q3): 35 (29–45) Inclusion criteria All adult patients with COVID‐19 respiratory failure who required tracheostomy between 1 February 2020 and 4 September 2020 Exclusion criteria Not defined
Interventions	The main exposure variable was ET versus LT Procedure: PDT was used for 78.0% of procedures, the remainder were performed with open surgical technique—primarily because of anatomical concerns that precluded percutaneous approach Treatment details of intervention group: 'early' ≤ 14 days Treatment details of control group: ’middle’ 15–21 days, and 'late' > 21 days
Outcomes	Primary study outcomes: Intra‐ and postprocedural data Hospital course including ventilator weaning and length of stay AEs Survival Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (R) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (R) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes	We contacted the study authors, who provided requested data that had not yet been published and could be included in our analyses.

Navaratnam 2022.

Study characteristics
Methods	Trial design: multicentre retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: 1 March‐31 October 2020 Country: England Language: English Number of centers: all 500 National Health Service hospitals in England
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 2200/1777, intervention ‐ ET: 851, control ‐ LT: 926, no tracheostomy: 0 early Age: continous data Gender (male %): total 1528 (70.7) Comorbidities (ORs (95% CIs)): Charlson comorbidity index items: peripheral vascular disease 0.723 (0.524 to 0.997) congestive heart failure 0.498 (0.393 to 0.631) acute myocardial infarction 0.581 (0.444 to 0.759) cerebrovascular disease 1.946 (1.539 to 2.462) dementia 0.657 (0.294 to 1.471) chronic pulmonary disease 1.031 (0.894 to 1.189) connective tissue disease/rheumatic disease 0.636 (0.413 to 0.978) peptic ulcer 1.214 (0.651 to 2.262) mild liver disease 0.861 (0.648 to 1.144) moderate or severe liver disease 0.606 (0.374 to 0.980) diabetes without chronic complications 0.876 (0.762 to 1.005) diabetes with chronic complications 0.795 (0.550 to 1.150) paraplegia and hemiplegia 0.973 (0.778 to 1.217) renal disease 0.839 (0.755 to 0.933) primary cancer 0.495 (0.348 to 0.705) metastatic carcinoma 0.141 (0.055 to 0.363) obesity 0.923 (0.788 to 1.081) Inclusion criteria All completed episodes of hospital care in England with a discharge date from 1 March to 31 October 2020 that involved a diagnosis of COVID‐19 Exclusion criteria Patients aged < 18 years were excluded.
Interventions	The main exposure variable was ET vs LT Procedure: no standardization of the procedure due to the multicenter design Treatment details of intervention group: ‘early’ (≤ 14 days) Treatment details of control group: ‘late’ (> 14 days)
Outcomes	Primary study outcomes In‐hospital mortality Length of hospital stay Length of critical care stay Tracheostomy malfunction Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (R) Quality of life (NR) Viral clearance (NR)
Notes

Polok 2021.

Study characteristics
Methods	Trial design: prospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 12 February‐31 December 2020, final date for data cut‐off was 14 April 2021 Country: 16 European countries Language: English Number of centers: 152 centres
Participants	Baseline characteristics Number randomised (in total and per arm): (recruited/evaluated): 2078/1740, intervention ‐ ET: 461/135, control ‐ LT: 461/315 Age (median/SD range): intervention group 74.0 (71.0‐76.0), control group 74.0 (72.0‐77.0) Gender (male, n (%)): intervention group 27 (20.0), control group 75 (23.8) Comorbidities (intervention group vs control group (n/N (%))): hypertension 99/135 (73.3) vs 198/315 (63.1) ischemic heart disease 20/135 (15.0) vs 63/315 (20.3) congestive heart failure diabetes type II 35/135 (25.9) vs 97/315 (30.8) chronic renal failure 21/135 (15.6) vs 40/315 (12.8) pulmonary disease 26/135 (19.3) vs 64/315 (20.4) SOFA score on admission, median [IQR] 7 [4 ‐ 8] vs 6 [4 ‐ 8] Inclusion criteria Patients aged ≥ 70 years with confirmed SARS‐CoV‐2 infection admitted to the ICU who required IMV between 12 February‐31 December 2020 Exclusion criteria Patients previously enrolled in the COVIP study were excluded.
Interventions	The main exposure variable was ET vs LT Procedure: no standardization of the procedure due to the international multicenter design Treatment details of intervention group: 'early' ≤ 10 days from intubation Treatment details of control group: 'late' > 10 days after orotracheal intubation
Outcomes	Primary study outcome All‐cause mortality within 3 months after admission to ICU Secondary study outcomes ICU length of stay Duration of MV Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (R) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes	Substudy of the COVIP study, included patients aged ≥ 70 years with confirmed SARS‐CoV‐2 infection admitted to the ICU who required IMV.

Prats–Uribe 2021.

Study characteristics
Methods	Trial design: multicenter prospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: from 11 March 2020‐20 July 2020, follow‐up until death, weaning or the end of 20 July 2020 Country: Spain Language: English Number of centers: 36 hospitals (ICUs)
Participants	Baseline characteristics Number randomized (in total and per arm): (recruited/evaluated): 794/696, intervention ‐ ET: 142, control ‐ LT: 554 Age (years, mean (SD)): intervention group 63.0 (10.4), control group 63.2 (9.2) Gender (male %): Total 69.1, intervention group 36.6, control group 24.4 Comorbidities (intervention group vs control group ( %)): active or previous smoking 19.0 vs 15.3 hypertension 54.2 vs 44.6 immunosuppression 4.9 vs 76 heart failure 3.5 vs 3.4 autoimmune disease 4.2 vs 6.1 COPD 7.7 vs 7.0 pregnancy 0.0 vs 0.5 diabetes mellitus 24.6 vs 20.8 neuromuscular disease 1.4 vs 1.4 ischemic cardiopathy 11.3 vs 8.8 APACHE II (intervention group vs control group (mean (SD))) 11.2(6.1)/15.3(6.7) SOFA (intervention group vs control group (mean (SD))) 6.7(4.4)/6.0(3.4) Inclusion criteria Patients with respiratory failure caused by SARS CoV‐2 infection, confirmed by PCR, requiring IMV and subsequent tracheostomy, performed before 20 July 2020 Exclusion criteria Patients with a missing tracheostomy, orotracheal intubation, or outcome date or missing age or sex Following the target trial framework, patients with a tracheostomy performed in the first 7 days after orotracheal intubation were excluded.
Interventions	The main exposure variable was ET vs LT Procedure: no standardization of the procedure due to the multicenter design Treatment details of intervention group: 'early' on Day 7–10 Treatment details of control group: 'late' was defined as occurring ≥ 11 days after orotracheal intubation
Outcomes	Primary study outcomes Days to weaning by tracheostomy timing, since orotracheal intubation and since tracheostomy Mortality by cumulative incidence of death by tracheostomy timing Associations of tracheostomy with intraoperative and postoperative complications incidence (bleeding, ventilatory problems) Review outcomes Mortality: overall mortality (R) in‐hospital mortality (NR) at up to day 28 (± 2) (R) at day 60 (R) at day 90 (R) time‐to‐event (R) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes	Due to differences in the reported figures, we used the data reported in the text for the analysis.

Takhar 2020.

Study characteristics
Methods	Trial design: single‐centre prospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: 21 March‐20 May 2020 Country: UK Language: English Number of centers: 1 centre in the UK
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 263/81, intervention ‐ ET: 24, control ‐ LT: 57, no tracheostomy: 0 Age (mean/SD): total 52.9 (12.2), intervention group 58.4 (11.8), control group 50.6 (11.8) Gender (male %): total 55 (67.9), intervention group 15 (62.5), control group 40 (70.1) Comorbidities (intervention group vs control group (n/N (%)): BMI (kg/m2) (mean (SD)) 30.0 (6.6) vs 31.2 (7.0) Very severe co‐morbidities ≥ 1 co‐morbidities (n (%))5 (20.8) vs 6 (10.5) Mean (SD) 0.3 (0.5) vs 0.1 (0.4) APACHE II score Mean (SD) 14.9 (3.4) vs 13.5 (3.8) Inclusion criteria All patients diagnosed with laboratory‐confirmed COVID‐19, critically ill with acute hypoxemic respiratory failure and receiving IMV The decision to perform tracheostomy for anticipated prolonged respiratory weaning was made jointly by 2 critical care consultants after evaluation of each participant’s clinical course and prognosis, considering the factors defined in the local guideline. Exclusion criteria Not defined
Interventions	The main exposure variable was ET vs LT Procedure: 93.8 % PDT and the remainder via a hybrid or open technique Treatment details of intervention group: ‘early’ (< 14 days) Treatment details of control group: ‘late’ (≥ 14 days)
Outcomes	Primary study outcome Duration of ventilation post‐tracheostomy Secondary study outcome Days until: sedation was stopped the patient's discharge from the ICU decannulation of tracheostomy the patient's discharge from hospital death Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (R) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (R) need for IMV (NR) liberation from IMV (R) ventilator‐free days (NR) duration to decannulation (R) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (R) Quality of life (NR) Viral clearance (NR)
Notes

Tang 2020.

Study characteristics
Methods	Trial design: multicenter, retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: 8 January 2020‐25 March 2020 Country: China Language: English Number of centers:. ICUs of 23 hospitals in Hubei Province, China
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 80/80, intervention ‐ ET: 30, control ‐ LT: 50, no tracheostomy: 0 Age (mean/SD): total 63.9 (14.0), intervention group 66.5 (15.1), control group 62.3 (13.2) Gender (male %): total 55 (68.8), intervention group 21 (70.0), control group 34 (68.0) Comorbidities (intervention group vs control group (n/N (%)): hypertension 12 (40.0%) vs 20 (40.0%) coronary heart disease 3 (10.0%) vs 14 (28.0%) diabetes 6 (20.0%) vs 8 (16.0%) cerebrovascular disease 4 (13.3%) vs 4 (8.0%) dementia 3 (10.0%) vs 2 (4.0%) chronic renal disease 2 (6.7%) vs 2 (4.0%) chronic hepatic disease 0 (0.0%) vs 4 (8.0%) cancer 2 (6.7%) vs 1 (2.0%) COPD 0 (0.0%) vs 2 (4.0%) SOFA score [IQR] 6 [4, 9] vs 5 [4, 7] APACHE II score [IQR] 15 [11, 21] vs 11 [9, 17] Inclusion criteria All patients diagnosed with COVID‐19 and receiving elective tracheostomies The decision of tracheostomy was made by treating clinicians Exclusion criteria Not defined
Interventions	The main exposure variable was ET vs LT Procedure: most tracheostomies were performed by ICU physicians (77.5%) and using percutaneous techniques (78.8%) at the ICU bedside (95.0%) Treatment details of intervention group: ‘early’ (< 14 days) Treatment details of control group: ‘late’ (≥ 14 days)
Outcomes	Primary study outcomes Treatments Details of the tracheostomy procedure Successful weaning after tracheostomy Living status Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (R) at day 90 (NR) time‐to‐event (R) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (R) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (R) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (NR) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

Tetaj 2021.

Study characteristics
Methods	Trial design: single‐center retrospective observational study Type of publication: journal publication Setting: inpatient Recruitment dates: from 1 April‐31 March 2021 Country: Italy Language: English Number of centers: 1 hospital (National Institute for Infectious Disease “Lazzaro Spallanzani” in Rome ‐ 200‐bed hospital for infectious disease with a 55‐bed ICU)
Participants	Baseline characteristics Number of participants (in total and per arm): (recruited/evaluated): 133/120, intervention ‐ ET: 61/61, control ‐ LT: 59/59 Age (years, median (IQR)): intervention group 70 (64‐77), control group 65 (69‐73) Gender (male n (%): Total 80 (66.7%), intervention group 42 (68.9), control group 38 (64.4) Comorbidities (intervention group vs control group (n/N %)): hypertension 45 (73.7) vs 35 (59.3) other cardiopathies 15/61 (24.6) vs 9/59 (15.2) diabetes 15/61 (24.6) vs 13/59 (22.0) obesity 25/61 (41) vs 29/59 (49.1) kidney disease (stage 3–5 of CKD) 5 (8.2) vs 3/59 (5.1) moderate to severe chronic liver disease 1/61 (1.6) vs 0/59(0) COPD/bronchial asthma 12/61 (19.7) vs 9/59 (15.2) previous neoplasia (solid neoplasia or hematological malignancy in the last 5 years) 7/61 (11.4) vs 7/59 (11.8) previous surgery in last month 2/61 (3.2) vs 2/59 (3.4) previous hospitalisation last 6 months 3/61 (4.9) vs 5/59 (8.5) chronic neurological disorders 10/61 (16.4) vs 10/59 (16.9) autoimmune diseases 6/61 (9.8) vs 9/59(15.2) other chronical diseases 23/61(37.7) vs 9/59 (15.2) APACHE II (intervention group vs control group (median (IQR)) 13(610‐17) vs 12(9‐17) SOFA (intervention group vs control group (median (IQR))) 5(4‐7) vs 5(3‐8) BMI (intervention group vs control group (median IQR))) 27,6(25‐31) vs 28,4(26‐31) Inclusion criteria Patients with virologically confirmed COVID‐19 by nasal pharyngeal swab for RT‐PCR, hospitalized, admitted to the ICU, underwent orotracheal intubation with subsequent percutaneous dilatation tracheostomies at bedside The inclusion criteria for tracheostomy were acute respiratory failure and the need for prolonged mechanical ventilation Exclusion criteria The exclusion criteria for tracheostomy were infection at the site of the tracheostomy, uncontrolled coagulopathy, altered neck anatomy, marked obesity, and multiorgan failure
Interventions	The main exposure variable was ET versus LT Procedure: bedside, using the Frova PercuTwist technique with rotational dilatation of the tracheal stoma through the use of hydrophilic screws or Ciaglia Blue Rhino using a single‐beveled curved hydrophilic dilator. Both techniques were conducted under video‐assisted fibro‐bronchoscopy. Patients who underwent bedside PDT in the ICU were stratified into two groups: treatment details of intervention group: the early group, which included patients who underwent PDT within the first 12 days of orotracheal intubation treatment details of control group: the late group, which included patients in whom the procedure was performed > 12 days after orotracheal intubation
Outcomes	Primary study outcome Date of the endotracheal intubation, tracheostomy, weaning and decannulation, and ICU outcome (discharge or death) Secondary study outcomes Ventilation parameter data were recorded on the day of orotracheal intubation, PDT, and 7 days after the PDT; these data included the respiratory exchange ratio (PaO2/FiO2 ratio), the fraction of inspired oxygen (FiO2), and positive end‐expiratory pressure (PEEP) Data collected on medications administered by intravenous continuous infusion included doses of sedatives, opioids as analgesics, and inotropic agents The complications of tracheostomy were assessed during the procedure and throughout the hospital stay Serological SARS‐CoV‐2 tests and nasopharyngeal swab tests for the molecular detection of SARS‐CoV‐2 were performed for the surveillance of all involved medical staff 14 days after performing a PDT and thereafter at least once every 3 weeks, unless they had symptoms or had been in direct contact with confirmed SARS‐CoV‐2 patients Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes	We contacted study authors, who provided requested data that had not yet been published and could be included in our analyses.

Tsonas 2022.

Study characteristics
Methods	Trial design: multicentre, retrospective, observational study Type of publication: journal publication Setting: inpatient Recruitment dates: from 1 March‐1 June 2020, follow up to 120 days Country: Netherlands Language: English Number of centers: 22 Dutch ICUs
Participants	Baseline characteristics Age (years, median): intervention group 65, control group 65 Gender (male n (%)): intervention group 75 (78.1), control group 77 (82.8) Number of participants (included): 189, intervention ‐ ET: 96, control ‐ LT: 93 Comorbidities (intervention group vs control group (n/N (%)): severity of ARDS mild 18.8/23.3 moderate77.1/66.7 severe 4.2/10.0 hypertension 29/96 (30.2) vs 33/96 (35.5) heart failure 4/96 (4.2) vs 6/96 (6.5) diabetes 22/96 (22.9) vs 21/96 (22.6) chronic kidney disease 6/96 (6.2) vs 5/96 (5.4) liver cirrhosis 1/96 (1.0) vs 0/96 (0.0) COPD 6/96 (6.2) vs 8/96 (8.6) active hematological neoplasia 1/96 (1.0) vs 3/96 (3.2) active solid neoplasia 3/96 (3.1) vs 2/96 (2.2) neuromuscular disease 0/96 (0.0) vs 1/96 (1.1) immunosuppression 4/96 (4.2) vs 1/96 (1.1) asthma 10/96 (10.4) vs 9/96 (9.7) obstructive sleep apnea syndrome 5/96 (5.2) vs 5/96 (5.4) Inclusion criteria Patients were enrolled if: aged ≥ 18 years admitted to 1 of the participating ICUs had received MV for respiratory failure related to COVID–19. Exclusion criteria Patients with unknown tracheostomy status due to transfer to a non–participating hospital, were excluded from this analysis.
Interventions	The main exposure variable was ET vs LT Procedure: no standardization of the procedure due to the multicenter design Treatment details of intervention group: 'Early' ≤ 21 (17–28) days after endotracheal intubation Treatment details of control group: 'late' ≥ 21 (17–26) days after endotracheal intubation. The reason for selecting 21 days as the cut‐off for ET and LT was based on the median time to tracheostomy performance in the cohort.
Outcomes	Primary study outcome The primary endpoint was incidence of tracheostomy Secondary study outcomes Timing of tracheostomy (counted as the number of days between start of MV and tracheostomy procedure) Outcomes including duration of ventilation ICU and hospital length of stay ICU, hospital, 28–day and 90–day mortality Factors associated with timing: post‐tracheostomy complications, stratified by tracheostomy timing (early vs late) and technique (surgical versus percutaneous) Review outcomes Mortality: overall mortality (NR) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (R) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (R) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (R) Quality of life (NR) Viral clearance (NR)
Notes	Secondary analysis of the PRoVENT trial

Volo 2021.

Study characteristics
Methods	Trial design: single‐center retrospective cohort study Type of publication: journal publication Setting: inpatient Recruitment dates: 22 February‐26 April 2020 Country: Italy Language: English Number of centers: Angel Hospital and Saints Giovanni e Paolo Hospital (Venice, Italy)
Participants	Baseline characteristics Number of participants (in total and per arm) (recruited/evaluated): 23/23, intervention ‐ ET: 9, control ‐ LT: 14, no tracheostomy: 0 Age (median/IQR): total 69 (42‐84) Gender (male %): total 21 (91) Comorbidities (total (n (%)): current smoking 4 (17) diabetes 7 (30) hypertension 11 (47) cardiovascular disease 4 (17) COPD 88 (34) malignancy 0 obesity 3 (13) chronic liver disease 0
Interventions	The main exposure variable was ET vs LT Procedure: open surgical tracheostomy procedures at the ICU Treatment details of intervention group: ‘early’ (< 10 days) Treatment details of control group: ‘late’ (≥ 10 days)
Outcomes	Primary study outcomes Mortality Sedation Date to subintensive unit Date of weaning Date of decannulation SOFA score at the day of intubation and of the day before tracheostomy D‐dimer level Review outcomes Mortality: overall mortality (R) in‐hospital mortality (NR) at up to day 28 (± 2) (NR) at day 60 (NR) at day 90 (NR) time‐to‐event (NR) Improvement of clinical status: duration to liberation from IMV (NR) need for IMV (NR) liberation from IMV (NR) ventilator‐free days (NR) duration to decannulation (NR) Worsening of clinical status: AEs (any grade) (NR) ventilator‐associated pneumonia (NR) need for renal replacement therapy (NR) postoperative bleeding (NR) airway obstruction (NR) tracheal stenosis (NR) need for ECMO (NR) ventilatory problems (NR) SAEs (NR) ICU length of stay, or time to discharge from ICU (R) Hospital length of stay, or time to discharge from hospital (NR) Quality of life (NR) Viral clearance (NR)
Notes

AE: adverse event; APACHE: Acute Physiology and Chronic Health Evaluation; ARDS: acute respiratory syndrome; CI: confidence interval; COPD: chronic obstructive pulmonary disease; ECMO: extracorporeal membrane oxygenation; ET: early tracheostomy; GCS: Glasgow Coma Scale; ICU: intensive care unit; IMV: invasive mechanical ventilation; IQR: interquartile range; ITT: intention‐to‐treat; LT: late tracheostomy; MV: mechanical ventilation; NR: not reported; OR: odds ratio; PDT: percutaneous dilational tracheostomy; PEEP: positive end‐expiratory pressure; PP: per protocol; PPE: personal protective equipment; R: reported; RCT: randomized controlled trial; RT–PCR: real‐time, reverse transcription polymerase chain reaction; SAE: serious adverse event; SAPS: Simplified Acute Physiology Score; SARS‐CoV‐2: severe acute respiratory syndrome coronavirus‐2; SBT: spontaneous breathing trial; SD: standard deviation; SOFA: sequential organ failure assessment; STROBE: Strengthening the Reporting of Observational Studies in Epidemiology

Characteristics of excluded studies [ordered by study ID]

Study	Reason for exclusion
Ahn 2021	Did not compare ET to LT
Andrews 2021	Full‐text not retrievable
Brenner 2021	Comment without results
Cazzador 2020	Comment without results
Chao 2020	Did not compare ET to LT
Chao 2021	Comment without results
Exall 2021	No definition of ET and LT
Farlow 2021	No definition of ET and LT groups
Fusina 2021a	Comment without result
Fusina 2021b	Comment without results
Garas 2021	Comment without results
Geriani 2022	Full‐text not retrievable
ICTRP 2833	Full‐text not retrievable
Mesolella 2021	Did not compare ET to LT
Okazaki 2021	Comment without results
Olanipekun 2021	Comment without results
Oxman 2021	Comment without results
Pandian 2021	Full‐text not retrievable
Pauli 2021	Did not compare ET to LT

ET: early tracheostomy; LT: late tracheostomy

Differences between protocol and review

The prespecified protocol is available in the International Prospective Register for Systematic Reviews (PROSPERO). The protocol for this review was registered with PROSPERO on 10 May 2021 (Dahms 2021).

Objective

For the sake of readability, we changed our objective to:

To assess the benefits and harms of early tracheostomy compared to late tracheostomy in critically ill COVID‐19 patients.

Methods

We focused our search on standard RCT designs and decided not to include non‐standard RCT designs, as these non‐standard study designs were not applicable to the research question under the severe constraints of the ongoing pandemic.

Type of outcome measures

We specified outcomes regarding benefits and harms of early tracheostomy for individuals with COVID‐19 after a guideline consortium (CEOsys) that took place after protocol registration. This approach was implemented in all reviews of CEOsys. We created outcome categories and added/specified the following outcomes for hospitalized participants with COVID‐19, as follows.

• Mortality:

  • overall mortality;

  • in‐hospital mortality;

  • at up to day 28 (± 2);

  • at day 60;

  • at day 90;

  • time‐to‐event.

• Improvement of clinical status:

  • duration to liberation from invasive mechanical ventilation;

  • need for invasive mechanical ventilation;

  • liberation from invasive mechanical ventilation;

  • ventilator‐free days;

  • duration to decannulation.

• Worsening of clinical status:

  • adverse events (any grade);

  • ventilator associated pneumonia;

  • need for renal replacement therapy;

  • postoperative bleeding;

  • airway obstruction;

  • tracheal stenosis;

  • need for extracorporeal membrane oxygenation (ECMO);

  • need for dialysis;

  • ventilatory problems;

  • serious adverse events.

• ICU length of stay, or time to discharge from ICU.

• Hospital length of stay, or time to discharge from hospital.

• Quality of life.

• Viral clearance.

Mortality

The outcome 'mortality' is often described at different follow‐up periods (up to day 28 (± 2), day 60, day 90, time‐to‐event or in‐hospital). Therefore, we have listed the different definitions of this outcome used in the studies. 'Overall mortality' is reported as the primary outcome, while the different timings of mortality are reported as secondary outcomes.

Worsening of clinical status

For worsening of clinical status, we focused on the impact of the intervention on the incidence of adverse events (any grade), instead of grade 3 to 4 adverse events.

Due to the imprecise definition of the outcome 'hospital‐acquired infections' in which the direct correlation to the intervention was not obvious, we specified the outcome in the data analysis and therewith considered ventilator‐associated pneumonia occurring after tracheostomy.

Further, we added 'postoperative bleeding', 'tracheal stenosis', 'airway obstruction', and 'ventilatory problems' directly related to the intervention that were reported as direct complications in the studies and also evaluated these as secondary outcomes.

We removed the outcomes 'admission to the ICU' and 'need for respiratory support' because these apply to the whole patient population of the included studies since they match the inclusion criteria of the patients of interest.

For these outcomes, such as adverse events, the different follow‐up periods were not explicitly reported because the follow‐up period is less important than the focus on the hospital stay.

Effect measures

We added the hazard ratio to the effect measures where this information was available. Any change in methodology was made before analysis. We found no other potential sources of bias in our review process.

Sensitivity analysis

We expanded sensitivity analyses of the following study characteristics for our prioritized outcomes, with regard to the following.

• Comparison of intention‐to‐treat‐analysis with per‐protocol‐analysis

• Components assessing risk of bias (studies with low risk of bias or some concerns versus studies with high risk of bias)

• Comparison of preprints with peer‐reviewed articles

• Comparison of prematurely terminated studies with completed studies

Contributions of authors

AS: clinical expertise, study selection, data extraction and assessment, conception and writing of the manuscript

KD: methodological expertise, study selection, data extraction and assessment

KA: methodological expertise, study selection, data extraction and assessment, conception and writing of the manuscript

NS: methodological expertise and advice, data extraction and assessment, conception and writing of the manuscript

IM: Information Specialist, development of the search strategy

TB: clinical expertise and advice, data extraction and assessment, conception, writing and proofreading of the manuscript

CB: methodological expertise and advice, data extraction and assessment, conception, writing and proofreading of the manuscript

Sources of support

Internal sources

• University Hospital RWTH Aachen, Germany

  Department of Intensive Care Medicine

• University Hospital Cologne, Germany

  Cochrane Haematology, Department of Internal Medicine, University Hospital Cologne, Germany

External sources

• Federal Ministry of Education and Research, Germany

  This review is part of the CEOsys project funded by the Network of University Medicine (Nationales Forschungsnetzwerk der Universitätsmedizin (NUM)) by the Federal Ministry of Education and Research of Germany (Bundesministerium für Bildung und Forschung (BMBF)), grant number 01KX2021.

Declarations of interest

AS: none known

KD: is member of the CEOsys project funded by the Network of University Medicine (Nationales Forschungsnetzwerk der Universitätsmedizin (NUM)) by the Federal Ministry of Education and Research of Germany (Bundesministerium für Bildung und Forschung (BMBF)), grant number 01KX2021, paid to the institution.

KA: is member of the CEOsys project funded by the Network of University Medicine (Nationales Forschungsnetzwerk der Universitätsmedizin (NUM)) by the Federal Ministry of Education and Research of Germany (Bundesministerium für Bildung und Forschung (BMBF)), grant number 01KX2021, paid to the institution.

NS: is Co‐ordinating Editor of Cochrane Haematology but was not involved in the editorial process of this review.

IM: is an information Specialist and Cochrane editor but was not involved in the editorial process of this review.

TB: received grants by the German Research Foundation (Deutsche Forschungsgemeinschaft (DFG), grant number: BR 5308/3‐1) and by the Federal Institute for Drugs and Medical Devices (Bundesinstitut für Arzneimittel und Medizinprodukte (BfArM), funding number: V‐2020.4 / 1503_68403 / 2020).

CB: none known

These authors contributed equally to this work

New