Practices in Triage and Transfer of Critically Ill Patients: A Qualitative Systematic Review of Selection Criteria

Supplemental Digital Content is available in the text.

Objectives:

To identify and appraise articles describing criteria used to prioritize or withhold a critical care admission.

Data Sources:

PubMed, Embase, Medline, EBM Reviews, and CINAHL Complete databases. Gray literature searches and a manual review of references were also performed. Preferred Reporting Items for Systematic reviews and Meta-Analyses guidelines were followed.

Study Selection:

We sought all articles and abstracts of original research as well as local, provincial, or national policies on the topic of ICU resource allocation. We excluded studies whose population of interest was neonatal, pediatric, trauma, or noncritically ill. Screening of 6,633 citations was conducted.

Data Extraction:

Triage and/or transport criteria were extracted, based on type of article, methodology, publication year, and country. An appraisal scale was developed to assess the quality of identified articles. We also developed a robustness score to further appraise the robustness of the evidence supporting each criterion. Finally, all criteria were extracted, evaluated, and grouped by theme.

Data Synthesis:

One-hundred twenty-nine articles were included. These were mainly original research (34%), guidelines (26%), and reviews (21%). Among them, we identified 200 unique triage and transport criteria. Most articles highlighted an exclusion (71%) rather than a prioritization mechanism (17%). Very few articles pertained to transport of critically ill patients (4%). Criteria were classified in one of four emerging themes: patient, condition, physician, and context. The majority of criteria used were nonspecific. No study prospectively evaluated the implementation of its cited criteria.

Conclusions:

This systematic review identified 200 criteria classified within four themes that may be included when devising triage programs including the coronavirus disease 2019 pandemic. We identified significant knowledge gaps where research would assist in improving existing triage criteria and guidelines, aiming to decrease arbitrary decisions and variability.

The decision to admit a patient to the ICU is the result of a complex process in which several human and logistic factors intertwine. Resource availability, patient characteristics, and physician assessment’s variability all contribute to patients’ outcome uncertainty (1). A lack of a standardized process to evaluate patient referrals and optimize resource allocation accentuates those differences leading to several critically ill patients being prevented access to ICU resources, which directly impacts their prognosis. The current coronavirus disease 2019 (COVID-19) pandemic shines a light on the ethical and practical issues clinicians face when triaging patients.

Triage centers, or the process of regionalization, can potentially help solve these issues. It involves access to a larger pool of resources provided by members of a network and allocation, as appropriate, according to the needs of each individual patient triaged (2). Drawing from the trauma literature, where implementation of organized and coordinated triage systems was shown to decrease mortality (3) and reduce time-consuming inter-hospital transfers (4), several groups have proposed that similar structures could be beneficial to critical care patients (5, 6). However, despite two decades of advocacy (7), only few initiatives were developed.

In the Province of Quebec (Canada), such a proposition has garnered the interest of policymakers and clinicians alike (8). More recently, the COVID-19 pandemic has shined a light on strained critical medical resources and the need for a triage process (9–11). However, there is currently a lack of universally agreed specific recommendations as to the triage criteria that should be used outside of the trauma population for ICU transport and admission. Namely, recommendations are especially vague when it comes to describing circumstances where the triage should result in the withholding of critical care resources including mass casualties such as COVID-19. Therefore, the purpose of this review was to systematically identify and appraise the published literature that defined criteria under which ICU admissions (triage and transport) should be withheld or not prioritized. We then propose a general framework to help clinicians use the results of this review to develop local triage guidelines adapted to the COVID-19 pandemic context.

MATERIALS AND METHODS

Search Strategy

In order to capture the broadest scope of articles, we aimed to include articles pertaining to critical care triage for admission to the unit or for interventions that mandatorily require critical care admission (e.g., extracorporeal membrane oxygenation [ECMO], heart-lung transplant) We searched PubMed, Medline (Ovid), Embase (Ovid), EBM Reviews (Ovid), and CINAHL Complete (EBSCO) using relevant keywords from inception to November 8, 2016. For each database, we used terms from controlled vocabulary (MESH, EMTREE, and CINAHL headings) and also performed a free text searching in title, abstract, and author keywords fields. A gray literature search was also executed in the following sources: Health Development Agency; National Guideline Clearing House; National Institute for Health and Clinical Excellence; National Institutes of Health; Research Service Delivery and Organisation Programme; Research Register for Social Care; and Google Scholar and OpenGrey. We limited our search to English and French languages. We also manually searched the reference lists of all articles remaining at the full-text review step for any potentially relevant article missed by our electronic searches. Our complete search strategy can be found in Supplemental Digital Content 1 (http://links.lww.com/CCM/F760).

Inclusion and Exclusion Criteria

We sought all articles and abstracts of original research, such as trials and observational studies, guidelines, reviews, editorials, and commentaries published in peer-reviewed journals, which listed criteria for ICU admission. Furthermore, we sought to include local, provincial, and/or national policies on the topic of ICU resource allocation. We excluded studies whose population of interest was neonatal, pediatric, trauma, or noncritically ill. Furthermore, we excluded studies if the ICU selection criteria did not permit to discriminate between ICU candidates or if they did not reflect systematic practice.

Study Identification and Selection

After removal of duplicates, one reviewer (J.D.) systematically reviewed the results and performed a title screen. All potentially relevant records, as well as those that did not contain enough information to determine eligibility, were retained for abstract screening. We went on to perform an abstract screen and retained all records that met inclusion criteria for full-text review. If no abstract was available, the citation was automatically selected for full-text review.

Data Extraction and Analysis

Data extraction was completed by one team member (J.D.). Data were collected on an electronically prepared Excel-based data collection tool (Microsoft Corp, Seattle, WA). Information extracted were the type of article and study design when applicable, year and country of publication, mechanism and setting of the triage and/or transport process, including whether it was condition-specific, and whether the criteria were proposed or currently in use. Given the heterogeneity of the studies, we developed a three-level appraisal scale to allow quality grading: level 1: randomized controlled trials, society guidelines, and national policies; level 2: rigorous reviews and multicenter observational studies; and level 3: single-center studies, editorials/commentaries where criteria are not backed up by evidence, and other articles with methodological flaws. Of note, certain exceptions, where papers that became highly cited references with time, were assigned a higher level (Supplemental Digital Content 2, http://links.lww.com/CCM/F761). To appraise the strength of the triage and transport criteria, we developed a robustness score (RS) which factors the number of studies listing each criteria as well as their quality (RS = [n_L1 × F_L1] + [n_L2 × F_L2] + [n_L3 × F_L3], where n = number of studies; F = factor; L₁ = level 1 quality [highest]; L₂ = level 2 quality; and L₃ = level 3 quality [lowest]). Level 1, 2, and 3 quality studies were given a factor of 2 points, 1 point, and 0.25 point, respectively. From this score, we classified the robustness of the evidence supporting all criteria within one of three categories: strongly robust evidence (defined as > two sds from the average RS), averagely robust evidence (above average RS but < 2 sds), and weakly robust evidence (below average RS). Additionally, we assessed whether extracted criteria met the following criteria: specific, scientifically sound, measurable, feasible to implement, and usable (12, 13). (Supplemental Digital Content 3, http://links.lww.com/CCM/F762; Supplemental Digital Content 4, http://links.lww.com/CCM/F763; Supplemental Digital Content 5, http://links.lww.com/CCM/F764; and Supplemental Digital Content 6, http://links.lww.com/CCM/F765). Finally, all articles were evaluated by identifying key themes. Criteria were grouped by theme and classified as patient-, condition-, physician-, or context-related. Context-related criteria are those criteria that are to be triggered only under specific circumstances. For example, they may pertain to time, location, or resource shortages, such as during pandemics or mass disasters.

Reporting Guidelines

We published our review protocol on International prospective register of systematic reviews (CRD42016047239). We also followed the Preferred Reporting Items for Systematic reviews and Meta-Analyses guideline as applicable to this study’s design (14) (Supplemental Digital Content 7, http://links.lww.com/CCM/F766).

RESULTS

Eligible Studies

The literature search identified 5,818 unique articles. Review of titles and abstracts resulted in the retrieval of 416 potentially full-text articles. Manual search of the reference lists of these articles led to 29 further full-text reviews. Of these, we identified 129 articles that met inclusion criteria and were included in this review (Fig. 1)

Figure 1.

Flow diagram of studies identified and included in the systematic review.

Characteristics of Studies

Table 1 summarizes the characteristics of the articles. All articles were in English except for five written in French (3.8%). Most articles were original research studies (n = 44, 34%) of which the vast majority were cohort studies (n = 31, 70%). Of note, no randomized controlled trials were identified. Pertaining to triage criteria, 62 articles (48%) were specific to certain patient populations or situations. Most articles described a triage process which relied on complete exclusion from ICU admission (n = 91, 70.5%) rather than a prioritization process (n = 22, 17%), whereas a few more articles described a combination of both. About a third of identified articles listed criteria that addressed more than one theme (n = 48, 37%). The most common theme for triage criteria was condition-related (n = 63, 48.8%) followed by patient-related criteria (n = 48, 37%).

Table 1.

Characteristics of Studies Included in Review (n = 129 Articles)

Characteristics	Articles, n (%)
Type of article
Original research	44 (34)
Cohort study	31 (24)
Systematic review	1 (1)
Nonrandomized control trial with contemporaneous controls	1 (1)
Series of consecutive cases	2 (2)
Policy analysis	1 (1)
Qualitative study	1 (1)
Other original research	7 (5)
Policy/guideline	33 (26)
Editorial/commentary	19 (15)
Review	27 (21)
Other	6 (5)
Country of origin
United States	63 (48)
France	21 (16)
United Kingdom	13 (10)
Canada	7 (5)
Language of publication
English	124 (96)
Year of publication
1970–1989	6 (5)
1990–1999	23 (18)
2000–2009	36 (28)
2010–2016	64 (50)
Type of process
Triage	124 (96)
Transport	3 (2)
Both	2 (2)
Process type
Complete exclusion	91 (71)
Prioritization/waiting lists	22 (17)
Both	12 (9)
Other	4 (3)
Themes
Patient-related	48 (37)
Condition-related	63 (49)
Physician-related	45 (35)
Context-related	37 (29)

Triage and Transport Criteria

A total of 200 unique triage and transport criteria were extracted from the 129 articles (1, 6, 15–141). Only five articles (4%) discussed transport criteria. A complete list of all triage and transport criteria are listed in Supplemental Digital Content 7 (http://links.lww.com/CCM/F766). As previously mentioned, we grouped identified criteria under four themes. Where possible, synonymous criteria were grouped unless it was felt that the differences in terminology evocated significant clinically relevant nuances. Condition-related criteria consisted in the most diverse and populous theme.

Analysis for patient-related criteria yielded eight categories and 18 unique criteria stemming from 48 citations. Patient preference was the most common reason cited to exclude patients from ICU admission (n = 29, 60.4%). Articles citing functional status (n = 13, 27%) and age to exclude patients were also common (n = 7, 14.6%), but only two citations used a specific age cutoff (i.e., 65 and 70 yr old) (33, 97). Social support and technical considerations were found but related to specific interventions (transplant or ECMO).

We found 63 articles that contained at least one condition-related criteria. Analysis yielded 11 categories and 87 unique criteria of which most related to comorbidities or diagnosis (n = 69). The most common criteria cited for ICU refusal was when no further oncological treatment options were available (n = 12) followed by persistent vegetative state and terminal diagnosis (n = 8). The majority of criteria had only one citation supporting it (n = 67). Thirteen articles pertained to a prioritization process, but only one article described a prioritization that was exclusively condition-related (42).

A total of 45 articles addressed physician-related criteria. Only two categories were extracted from this theme: prognosis and physician evaluation. All articles contained a criterion attributable to prognosis, although one article also listed incomplete physician evaluation as a reason to deny ICU admission. Within the prognosis category, 14 unique criteria were found. Most citations were encompassed by two criteria: patients who were judged by the physician to either be too well (n = 20) or too sick (n = 15). Fifteen citations used a prioritization scheme rather than exclusion.

Finally, analysis for context-related criteria yielded six categories and 81 unique criteria stemming from 37 articles. Citations citing metastatic malignant disease (n = 14), advanced and irreversible neurologic disease (n = 13), and Sequential Organ Failure Assessment (SOFA) score greater than 11 (n = 11) led to these three criteria being the most common within this theme. The most populous category is that of epidemics and pandemics (including influenza) or other mass disasters (24 citations, 65.9%). Within this category, we further divided the data into 19 themes and 70 criteria. Several subthemes pertained to clinical conditions (n = 13 criteria, 68.4%), while the balance pertained to age (n = 1, 5.3%) or other forms of prognostication (n = 4, 21.1%). Within the context-related criteria, only seven studies were published prior to 2006, and none of these addressed pandemics or other large-scale disasters. Other relevant categories pertained to bed availability (n = 7 citations, 18.9%) or other lack of resources (n = 5, 13.5%). Furthermore, only one study specifically pertained to transport criteria and related to unsuitable flying conditions.

Developing and ranking criteria per our robustness scale provided further insight. We extracted the top 10% most robust criteria for a total of 20 criteria (Table 2). While the most robust criteria related to patient preference (i.e., avoiding admission of patients or families who decline intensive care), 75% of those triage criteria pertained to epidemics, which falls under the context theme.

Table 2.

Criteria Achieving the Highest Robustness Score (Top 10%)

Rank	Theme/Category	Specific Criteria	Robustness Score
1	Patient/patient preference	Patients or families who decline intensive care or some of its components (e.g., mechanical ventilation, do not resuscitate)	26.25
2	Context/epidemics	Metastatic malignant disease	22.25
3	Context/epidemics	Advanced and irreversible neurologic event or condition	21.25
4	Context/epidemics	If cardiac arrest: Any of: unwitnessed cardiac arrest, witnessed cardiac arrest not responsive to electrical therapy, recurrent cardiac arrest	18.25
5	Context/epidemics	End-stage lung failure: Primary pulmonary hypertension with NYHA class III or IV heart failure, right atrial pressure > 10 mm Hg, or mean pulmonary artery pressure > 50 mm Hg	18.25
6	Context/epidemics	If burn injury: Any two of: > 60 yr old, > 40% total body area, inhalational injury	17.25
7	Context/epidemics	End-stage lung failure: Chronic obstructive pulmonary disease < 25% or Pao2 < 55 mm Hg or secondary pulmonary hypertension or on home o2 (10)	17.25
8	Context/epidemics	Advanced untreatable neuromuscular disease	16.25
9	Context/epidemics	End-stage heart failure: NYHA III or IV	16.25
10	Context/epidemics	End-stage lung failure: Cystic fibrosis with post-bronchodilator forced expiratory volume in 1 s < 30% or baseline Pao2 < 55 mm Hg	16.25
11	Context/epidemics	End-stage lung failure: Pulmonary fibrosis with vital capacity or total lung capacity < 60% predicted or Pao2 < 55 mm Hg or secondary pulmonary hypertension	16.25
12	Context/epidemics	End-stage liver failure: Child-Pugh score ≥ 7	16.25
13	Context/epidemics	Elective palliative surgery	15.25
14	Context/epidemics	Sequential Organ Failure Assessment score > 11	14.5
15	Physician/prognosis	Too well	12.5
16	Physician/prognosis	Too sick	12
17	Condition/diagnosis	No further oncological treatment options	11.25
18	Context/epidemics	Severe baseline cognitive impairment	11.25
19	Context/epidemics	Age > 85 yr old	11.25
20	Condition/diagnosis	Persistent vegetative state	9.5

NYHA = New-York Heart Association.

The overlap between context and the three other themes was further analyzed. All but five of the 81 context criteria (Supplemental Digital Content 8, http://links.lww.com/CCM/F767, in red) were not akin to being classified under patient-, condition-, or physician-related criteria. Furthermore, the terminology used for context-related criteria tended to be more specific as 66% of criteria were measurable as opposed to patient (51.6%) or condition (44.8%) related. None of the physician-related criteria were measurable (Supplemental Digital Content 5, http://links.lww.com/CCM/F764). Furthermore, we analyzed the data across geographic regions using the three most commonly cited criteria for each theme (Table 3). This analysis generally showed international consistency.

Table 3.

Thematic Top Three Most Commonly Cited Criteria by Geographic Region

Specific Criteria	Citations	All Articles, n	North America, n (%)	Europe, n (%)	Asia, n (%)	Other, n (%)
Patient
Patients or families who decline intensive care or some of its components	(36, 37, 44, 50, 55, 59, 73, 94, 105, 127, 133) (1, 19, 30, 47, 53, 56, 66–68, 93, 102, 103, 117–119, 125, 136, 140)	29	10 (16)	14 (30)	4 (33)	1 (14)
Bedridden	(47, 53, 79, 109, 129)	5	0 (0)	5 (11)	0 (0)	0 (0)
Advanced age	(23, 62, 69, 111)	4	1 (2)	3 (9)	0 (0)	0 (0)
Condition
No further oncological treatment options	(20, 47, 48, 53, 56, 66, 77, 79, 113, 118, 129, 133)	12	2 (3)	8 (17)	2 (17)	0 (0)
Persistent vegetative state	(39, 66, 82, 103, 115, 117, 118, 133)	8	6 (9)	1 (2)	1 (8)	0 (0)
Terminal diagnosis	(20, 35, 87, 100, 109, 118, 119, 133)	8	4 (6)	2 (4)	2 (17)	0 (0)
Physician
Too well	(15, 16, 26, 32, 45, 50, 62, 63, 66–69, 87, 95, 98, 110, 112, 125, 136, 138, 141)	21	8 (13)	9 (20)	1 (8)	3 (43)
Too sick	(15, 16, 26, 32, 50, 62, 63, 67–69, 95, 98, 101, 112, 138, 141)	16	6 (10)	7 (15)	1 (8)	2 (29)
Likelihood of benefit	(25, 31, 66, 103, 133, 135, 139, 141)	8	3 (5)	3 (7)	2 (17)	0 (0)
Context
Metastatic	(17, 27, 40, 41, 65, 70, 74, 75, 89–91, 95, 104, 131, 137)	15	9 (14)	4 (9)	1 (8)	1 (14)
Advanced and irreversible neurologic event or condition	(17, 27, 40, 41, 65, 70, 74, 75, 90, 91, 95, 124, 131, 137)	14	8 (13)	4 (9)	1 (8)	1 (14)
Sequential Organ Failure Assessment score > 11	(17, 27, 40, 46, 70, 75, 90, 91, 95, 104, 124, 131)	12	6 (9)	5 (9)	1 (8)	0 (0)

DISCUSSION

The objective of this systematic review was to identify published criteria about recommended or used criteria for the triage or transport of patients to a critical care facility. We identified 200 unique criteria, stemming from 129 articles and 23 countries depicting a wide variety of ICU organizational models. Yet, despite an ICU exclusion mechanism identified in 71% of the articles, several cited exclusion criteria do not meet minimal standards of a good criterion as was previously defined (Supplemental Digital Content 3, http://links.lww.com/CCM/F762; Supplemental Digital Content 4, http://links.lww.com/CCM/F763; Supplemental Digital Content 5, http://links.lww.com/CCM/F764; and Supplemental Digital Content 6, http://links.lww.com/CCM/F765). For example, none of the physician-related criteria are quantitative, while only 45% to 66% of the patient-, condition-, and context-related criteria are measurable. Furthermore, while we assessed 16% of the articles to be of high quality (i.e., national policies, society guidelines), none of the criteria stemmed from randomized controlled trials. More importantly, not a single study evaluated the impact of the implementation of its triage criteria.

Vague wording of ICU triage guidelines such as “likelihood of benefit,” “futility,” or “advanced age” provides practical issues, notably not allowing for measurement of compliance rates (142, 143). It also results in unintended externalities such as lack of transparency (144) or, at least, that of significant physician decision-making variation (145–153). Hence, despite the Society of Critical Care Medicine recommending each unit develops its own admission policy as early as in 1999 (112) and again in 2016 (36), there is no universally accepted set of specific and measurable admission criteria. This probably relates to the importance that physician autonomy holds in modern medical practice and its consequent freedom of treatment options based on best clinical judgment (154).

Nonetheless, our results show that there may be an emerging trend in the past decade, where exclusion criteria are becoming more specific. The temporal trend seems to correlate with the publication in 2006 of the article by Christian et al describing a triage protocol for an Influenza pandemic which described 12 exclusion criteria (91). The interest generated by this highly cited study largely accounts for the findings of our study, where criteria were ranked by robustness. Indeed, out of the 20 most robust criteria according to the score we developed, 15 pertain to epidemics situations and, when not identical, draw inspiration from the propositions by Christian et al (91). This is a testimony to the acceptability within the triage community of such specific criteria.

Our study adds to existing reviews documenting the process of ICU triage and transport (15, 95, 136, 155). Namely, our study is the first systematic review documenting criteria, either currently in use or proposed, adopted homogeneously within a team or organization. Indeed, previous studies aimed to document individual practices and usually used survey methodology. Such studies identified different themes than ours, mainly religion (156), country of practice (157), socioeconomic status (158), or even nursing morale (18).

Moving Forward

Several frameworks have been proposed to help triage patients and allocate scarce resources during the current COVID-19 (severe acute respiratory syndrome coronavirus 2) pandemic (9–11, 159). Guidance to help front-line clinicians make critical rationing decisions have been recently proposed (160). Our review provides guidance on the type of criteria to include in triage programs and systems. An earlier version of this work was submitted to the provincial critical care pandemic preparedness working group in Quebec (Canada). The work informed the design of a provincial triage protocol. Robust criteria were validated against what was then known about COVID-19. Hence, patients’ preferences to decline intensive care as well as comorbidities which decreased short- and long-term prognosis were chosen as exclusion criteria (Table 2). Such exclusion criteria become more stringent as surge increases (159). We recommend using criteria that are specific, scientifically sound, measurable, feasible, and usable in order to favor reproducibility of the decisions and accountability (Supplemental Digital Content 3, http://links.lww.com/CCM/F762; Supplemental Digital Content 4, http://links.lww.com/CCM/F763; Supplemental Digital Content 5, http://links.lww.com/CCM/F764; and Supplemental Digital Content 6, http://links.lww.com/CCM/F765). Criteria which do not discriminate in the context of COVID-19, such as the SOFA score (161), should be avoided despite their robustness. Finally, as many COVID-19 infected patients present with few comorbidities, clinicians should familiarize themselves with nonclinical triage principles such as the multiplier effect (i.e., can saving this patient’s life help save many others?), the life-cycle principle (i.e., how many further life stages is the patient expected to live through?), and randomization. Such principles may help discriminate between seemingly similar patients during mass triage. Others have integrated such concepts into a point system (123). Our approach is summarized in Figure 2.

Figure 2.

Proposed ICU triage algorithm for access to critical care resources during the coronavirus disease 2019 pandemic. COPD = chronic obstructive pulmonary disease, MELD = model for end-stage liver disease, NYHA = New-York Heart Association.

Limitations

Despite the strengths of our study, its conclusions are bound by a few limitations. First, we did not include articles published in languages other than English or French. We also may have missed local or national policies that are not cross-referenced in any of our search engines or that are only available offline. However, such an endeavor would have been beyond the scope of this work. Finally, only a single author (J.D.) performed the screens and full-text reviews which may have impacted the sensitivity of the review.

Future Research

Optimization of triage and transport practices for critically ill patients will require further research to help address the process deficiencies identified by our study. Namely, environmental scans should be undertaken to identify organizations where triage criteria were objectively evaluated after successful implementation to help benchmark expected process and outcome changes. These environmental scans should not be limited to the realm of intensive care as other patient populations with similar challenges may offer valuable insight (e.g., trauma, PICU, organ transplant) Then, acceptability of identified triage and transport criteria should be determined using Delphi methodology with knowledge users. Finally, formal triage and transport criteria performance should be rigorously tested. A stepped wedge cluster randomized trial may provide the ideal design if the new triage and transfer criteria are implemented within the context of a change in regional health policies (162).

CONCLUSIONS

A systematic review aimed at identifying triage and transport criteria used to prioritize or exclude certain patient populations under different settings helped to generate a list of 200 criteria classified within four themes (patient-, condition-, physician-, and context-related). These criteria may help clinicians and decision-makers devise local, regional, or national ICU triage criteria. A practical example is proposed using the COVID-19 pandemic. However, further high-quality studies or policies yielding specific and measurable criteria tailored to clearly defined patient populations are needed to promote wider clinical adoption in an effort to decrease practice variability and improve transparency.

ACKNOWLEDGMENTS

We acknowledge the work of Nicole Leduc who reviewed an early version of this work.