Rationale and design of a randomized controlled clinical trial; Titration of Oxygen Levels (TOOL) during mechanical ventilation

Abstract

Background:

Both hyperoxemia and hypoxemia are deleterious in critically ill patients. Targeted oxygenation is recommended to prevent both of these extremes, however this has not translated to the bedside. Hyperoxemia likely persists more than hypoxemia due to absence of immediate discernible adverse effects, cognitive biases and delay in prioritization of titration.

Methods:

We present the methodology for the Titration Of Oxygen Levels (TOOL) trial, an open label, randomized controlled trial of an algorithm-based FiO₂ titration with electronic medical record-based automated alerts. We hypothesize that the study intervention will achieve targeted oxygenation by curbing episodes of hyperoxemia while preventing hypoxemia. In the intervention arm, electronic alerts will be used to titrate FiO₂ if SpO₂ is ≥94% with FiO₂ levels ≥0.4 over 45 min. FiO₂ will be titrated per standard practice in the control arm. This study is being carried out with deferred consent. The sample size to determine efficacy is 316 subjects, randomized in a 1:1 ratio to the intervention vs. control arm. The primary outcome is proportion of time during mechanical ventilation spent with FiO₂ ≥ 0.4 and SpO₂ ≥ 94%. We will also assess proportion of time during mechanical ventilation spent with SpO₂ < 88%, duration of mechanical ventilation, length of ICU and hospital stay, hospital mortality, and adherence to electronic alerts as secondary outcomes.

Conclusion:

This study is designed to evaluate the efficacy of a high fidelity, bioinformatics-based, electronic medical record derived electronic alert system to improve targeted oxygenation in mechanically ventilated patients by reducing excessive FiO₂ exposure.

1. Introduction

Therapeutic oxygen delivery during mechanical ventilation is a life sustaining intervention used for approximately 2–3 million critically ill patients annually [1–3]. This need has escalated several-fold during the COVID-19 global pandemic [4]. Clinical practice guidelines continue to recommend targeted fractional inspired oxygen (FiO₂) titration to avoid excessive or insufficient oxygenation in acutely ill patients [5]. While bedside providers are very cognizant of preventing acute hypoxemia due to insufficient oxygen, excessive oxygen delivery in the ICU persists, contributing to increased morbidity and mortality in critically ill patients [6–9].

Toxicity from high concentrations of oxygen is well-defined; yet regulating deleterious adverse effects from hyperoxia is critical and complex. High oxygen delivery is necessary to address hypoxemia in acute respiratory failure. However, the prolonged FiO₂ exposure then promotes acute lung injury, manifesting clinically as diffuse alveolar damage, the pathologic correlate of Acute Respiratory Distress Syndrome (ARDS) [10,11]. In addition to directly promoting ARDS, hyperoxia augments ventilator induced lung injury and predisposes to ventilator associated pneumonias, delaying liberation from the ventilator while increasing length of stay and hospital mortality [12–14]. Even with numerous animal models and human studies we can only estimate, but do not precisely know, of the point at which oxygen exposure becomes toxic. Therefore, in acutely ill patients it may be best to deescalate FiO₂ supplementation to the lowest level necessary to maintain physiologically safe oxygenation. However, even in ICU’s with established oxygenation protocols, staff do not rigorously implement FiO₂ titration in an attempt to shield patients from episodic hypoxemia, thus resulting in persistent liberal oxygenation [15,16]. Addressing education and implementation measures to optimize rigorous oxygen titration can be effective, as shown by our group and others [15,17,18]. Electronic medical record (EMR)-based automation measures have been effective in the ICU for implementation of low tidal volumes, diagnosis of acute lung injury, and early detection and management of sepsis [19–21]. To that end, leveraging electronic health records to increase the efficiency of interventions in critical care trials is identified as a research priority by NHLBI [22]. Our group conducted a pilot study to demonstrate the feasibility, safety and preliminary efficacy of a high fidelity, bioinformatics-based, EMR-derived electronic alert (e-alert) system for hyperoxia notifications in the ICU. We present here the methodology for the Titration of Oxygen Levels (TOOL) trial, a randomized controlled trial of an algorithm-based FiO₂ titration with EMR based automated e-alerts. We hypothesize that the study intervention will achieve targeted oxygenation by preventing episodes of hyperoxemia or hypoxemia.

1.1. Goal of this trial

The goal of this pragmatic, multi-disciplinary open label clinical trial is to implement targeted FiO₂ titration by engaging respiratory therapists through EMR based alerts to prioritize FiO₂ titration in mechanically ventilated patients. We will use this e-alert strategy to inform respiratory therapists of excessive oxygen delivery real time and evaluate the efficacy of the targeted oxygenation protocol.

2. Materials and methods

2.1. Trial design and oversight

TOOLs is a single center, open label, 1:1 randomized control trial conducted in the Medical ICU at The Ohio State University Medical Center and James Cancer Hospital. Based on the planned intervention and implementation of the protocol, the study cannot be blinded. The trial is funded by the National Center for Advancing Translational Sciences (NCATS) and the Ohio State University Center for Clinical and Translational Science (OSU CCTS). The protocol was approved by the Ohio State University (OSU) IRB (Protocol #2014H0236) and is registered at ClinicalTrials.gov (NCT04481581). Progress and safety of the trial is monitored by an independent Data and Safety Monitoring Board (DSMB). The trial protocol. as developed according to the SPIRIT guidelines (Appendix A in Supplementary Materials) and was launched in January 2021.

2.2. Trial participants

Inclusion criteria for the study are, 1) patient age 18 years or greater and 2) requirement of mechanical ventilation when admitted to the medical ICU. Those in whom mechanical ventilation was initiated for more than 6 h or initiated at another hospital; pneumothorax; carbon monoxide poisoning; requiring hyperbaric oxygen therapy, COVID 19 infection; Myocardial infarction with acute ST elevation; prisoner status; pregnant patients and those intubated only for procedural intent, such as bronchoscopy, esophageoduodenoscopy or colonoscopy will be excluded. See inclusion and exclusion criteria in Fig. 1. Because oxygen use is ubiquitous in critically ill patients and the potential for hyperoxic injury is relevant in most patient populations, we chose to include a broad population of patients requiring mechanical ventilation in which there was not a defined indication for hyperoxia or reason to believe that increased oxygen delivery may be beneficial. These inclusion criteria are concordant with those used for other clinical trials of oxygen titration [23–25].

Fig. 1.

TOOL study design.

2.3. Justification of exclusion criteria

Patients with pneumothorax, carbon monoxide poisoning, hyperbaric oxygen therapy and acute ST elevation myocardial infarction may need liberal oxygenation for therapeutic indications, therefore frequent titrations in these patients may not be justified. Pregnant women, as defined by the presence of positive urine or serum pregnancy test, or the patient’s personal history of current pregnancy, are excluded primarily because effects of varying supply of oxygen in the fetus is not known. Prisoners are excluded being considered a vulnerable population. To standardize FiO₂ exposure that may occur prior to enrollment, we decided to arbitrarily limit the maximal pre-enrollment FiO₂ exposure to 6 h or less, therefore excluding patients who are mechanically ventilated for greater than 6 h. For similar reasons, we cannot account for the FiO₂ exposure in mechanically ventilated patients transferred from other hospitals and therefore needed to exclude them. Patients with COVID-19 pneumonia were excluded to reduce both the exposure risk for study personnel and Respiratory therapists as well as to conserve protective personal equipment which may be required with repetitive access to the patients’ room for titration.

2.4. Deferred consent

This study is being carried out with deferred consent. Consent is deferred when a patient is incapable of providing informed consent and their legal authorized representative (LAR) is incapable of providing consent or is not available. Consent is then obtained at the earliest possible time that a LAR is available; otherwise, the patient is approached for consent after extubation. In the event that the LAR or patient denies consent, then the patient is withdrawn from the study. No additional data is collected for research purposes.

2.5. Randomization and blinding

Eligible patients are randomly assigned to the intervention and control arms in 1:1 fashion (Fig. 1) using block randomization with blocks of size six. Respiratory therapists and study coordinators are unblinded. There is no possible way to carry out the study by blinding respiratory therapists. Study investigators, medical monitor and statisticians are blinded. The statistician may be unblinded if needed by the DSMB.

Fig. 1 ¹A member of the study team audits each alert and records 1) its appropriateness, 2) whether FiO2 is titrated, and 3) FiO₂ before and after the alert.

2.6. Study intervention

In the interventional arm, e-alerts are sent to respiratory therapists for FiO₂ titration. This intervention is based on our previous work showing the efficacy of electronic alerts reducing excessive oxygen supplementation and improving compliance to oxygenation protocols as well as supported by unpublished pilot data for this trial [26].

2.6.1. E-Alerts

Levels of both FiO₂ and SpO₂ are transmitted from the bedside monitor and ventilator to an interim server connected to the EMR. The biomedical research informatics team at the Ohio State University has developed a real-time algorithm which continuously analyzes data at one-minute intervals to generate e-alerts if the clinical criteria are met. Please see details of alert development in Appendix B.

2.6.2. E-Alert criteria

Patients must meet the following criteria for an e-alert to be sent:

1. Presence of mechanical ventilation for ≥45 min and

2. SpO₂ target ≥94% when the supplied FiO₂ levels were ≥ 0.4.

On meeting the above criteria, e-alerts are fired every 45 min. No more than 4 alerts are sent in a 6-h period. The bedside nurse monitors the patients after titration. Flow of events in the interventional arm, including the oxygen titration protocol and decision support tool are noted in Fig. 2 (a, b).

Fig. 2.

a and b Notification algorithm (2a) and Decision support tool (2b).

a. Notification Algorithm - Flow Diagram: This figure shows the process of oxygen titration in the intervention arm. Alert notification starts only after 45 min of intubation. When the criteria for SpO₂ and FiO₂ are met as noted above, then a pager/text alert is sent on the CISCO phone to the Respiratory Therapist. SpO₂ and FiO₂ criteria for e-alert vs no e-alert are noted above. The bedside monitor will alarm if SpO₂ drops below 88% for 5 min.

b. Decision Support Tool for Intervention Arm: The figure above shows the decision support tool used for oxygen titration in the intervention arm by the Respiratory Therapist after receiving an alert. The first column indicates the respective conditions for which an alert will be sent, the second column defines the criteria. The third column indicates directions to be followed for each respective condition.

In the control arm, FiO₂ titration is done per standard practice, in which the clinical team makes assessments to decrease FiO₂ in response to FiO2 without prompting from an alert system. This is executed by respiratory therapists or occasionally physicians themselves. The ICU ventilator protocol is followed for guidance. Respiratory Therapists are encouraged to review oxygen needs at least once every 4 h. (Appendix C).

2.7. Outcomes

The primary outcome is the proportion of time during mechanical ventilation spent with FiO₂ ≥ 0.4 and SpO₂ ≥ 94% during mechanical ventilation. Proportion of time spent with SpO₂ < 88%, duration of mechanical ventilation, length of ICU and hospital stay, hospital mortality, accuracy and adherence to e-alerts will be calculated as secondary outcomes. In addition, data for self-reported race along with arterial saturation (SaO2); arterial oxygen tension (PaO2) and SpO₂ will be collected to further corroborate correlation between these variables. We will collect primary comorbid conditions which have been shown to be valuable for developing precision and personalization in oxygen titration to identify their trajectories and sub stratify their outcomes (e.g. Coronary Artery Disease; Chronic Obstructive Pulmonary Disease; Acute ischemic Stroke; post-cardiac arrest; hypoxic ischemic encephalopathy [23,27]). Any patient who is enrolled in the study and is re-intubated within 48 h of extubation will continue in the same arm as before [28]. Reintubation within such a short time can be clinically considered as a part of the same disease process. However, if re-intubated after 48 h, the same patient cannot be enrolled in the study again. Time duration of hyperoxemia in both arms is counted only when the patient is under study protocol.

2.8. Missing data

The data variables (FiO₂ and SpO₂) are collected every 1 min. We will use the following strategy to address missing FiO₂ and SpO₂ values. For periods of contiguous time ≤ 15 min that are missing FiO₂ and/or SpO₂ values, the last observation carried forward approach will impute the missing FiO₂ and/or SpO₂ values during the time period. For contiguous time periods >15 min with missing FiO₂ and/or SpO₂ values, these values will not be imputed and this period of time will be omitted from analysis of the primary endpoint. Additional secondary outcome data (date of admission, date of discharge, death, time of extubation) are collected as a standard part of clinical care during a patient’s hospital stay, and we have not encountered missing data for these outcomes in previous studies.

2.9. Statistical analysis

Normality of the primary outcome variable will be evaluated using the Shapiro-Wilk test. Differences in the primary outcome between the two study arms will be tested using either a two-sample t-test if the outcome is normally distributed (p-value from Shapiro-Wilk test >0.05), or the Wilcoxon rank-sum test (p-value from Shapiro-Wilk test <0.05). An α = 0.05 level for statistical significance will be used for testing differences in the primary outcome between groups, based on intention to treat analysis. Covariate balance between arms will be evaluated using standardized differences. A similar analysis will be conducted for the proportion of time spent with SpO₂ < 88%.

Length of hospital and ICU stay, ventilator free days (time to extubation), and in-hospital mortality will be analyzed using competing risks regression, as recommended by several recent articles [28,29]. In-hospital mortality will be treated as a competing event for length of hospital stay, length of ICU stay, and time to extubation. Patient discharge will be a competing event for in-hospital mortality. Per recommendations, day 0 for all time-to-event analyses will be day of randomization, with length of mechanical ventilation, ICU and overall hospital stay prior to randomization reported and considered as covariates. Cumulative incidence functions will summarize the time-to-event outcomes and estimate the percentage of patients with each outcome (discharged, removed from MV, deceased) by a given day. Differences in cumulative incidence functions between study arms will be tested using Gray’s test [30]. Subdistribution proportional hazards regression will estimate subdistribution hazard ratios between study arms for each outcome [31]. All time-to-event outcomes except in-hospital mortality will be administratively censored at 28 days, since the majority of relevant outcomes occur within this timeframe and durations longer than this can skew results and adversely impact power. Please see the statistical analysis plan in appendix E.

2.10. Fidelity monitoring

With the ability to extract covariate real time from the EMR (values every minute), fidelity monitoring is conducted by auditing the EMR for 1) alert accuracy, 2) percent response to total alerts per patient and 3) percent adherence to the protocol. Data related to inaccurate alerts, or alert aberrancy is included in the intention to treat analysis.

2.11. Practice diffusion

To account for practice diffusion we will analyze the treatment effect in our cohort over sub-sequential times through study duration. This will be done by including time of enrollment in the regression model for the primary and secondary outcomes and testing for potential interaction between treatment and time of enrollment.

2.12. Sample size and power

The planned sample size for the trial is 316 subjects. The sample size calculation is based on published data from similar interventions demonstrating a median reduction in the percentage of excess exposure time ranging between 5% and 10% with standard deviations (SD) between 13.5% and 15% [32,33]. With 300 total subjects (150 subjects per arm), there is 82% power to detect a 5% difference with a SD of 15% at α = 0.05, based on the two-sample t-test assuming equal variance. This is more conservative than the observed difference in the aforementioned studies (difference = 8%, SD = 13.5%). An additional 16 subjects are included to account for potential 5% loss after recruitment. Since data are collected during a patient’s hospital stay, missing data are expected to be minimal.

2.13. Education, adherence, and stakeholder engagement

Approximately 100 respiratory therapists were trained for this trial prior to its initiation via three modalities:

1. A protocol training session was held on four occasions with the respiratory leadership to accommodate any scheduling conflicts.

2. Protocol details and updates were included in the ICU newsletter twice prior to the initiation of the study and monthly thereafter.

3. Respiratory break rooms were supplied with protocol packets, including all required and supplemental study information.

Respiratory therapists were engaged in protocol development, study design, and actively in implementation. Being primarily responsible for performing FiO₂ titration, they are encouraged to discuss any concerns regarding e-alerts and oxygen titration with treating physicians and/or study investigator. Respiratory therapists must document and provide reasoning for protocol deviations in the EMR flowsheet. A medical monitor is established for the study. The study coordinator conducts routine review of study progress and screens for protocol adherence or deviation. The respiratory leadership may additionally perform spot checks of therapist performance for protocol compliance.

Adverse events and patient safety: The bedside nurse monitors a patient after titration. Will inform the physician in charge immediately if patient develops hypoxemia (SpO₂ less than 88% for 5 min). They will then inform the study team in case of adverse events. The study is monitored by a data and safety monitoring board and an independent medical monitor has also been appointed. Attribution of adverse events is detailed in Appendix D.

3. Discussion

TOOLs is a unique, pragmatic and multidisciplinary effort to make headway in targeted oxygenation in the ICU with the goal of preventing both hypoxemia and hyperoxia. TOOL’s vision is based on leveraging EMRs in developing algorithmic feedback for FiO₂ titration based upon 24 h real-time screening. Below we discuss the rationale for key aspects of the TOOL study design.

3.1. Rationale for current oxygen targets used in this study

The optimal target range in the interventional arm of this trial is to use supplemental FIO₂ less than 0.4 to achieve a SpO₂ range of 88–94%. Patients may spontaneously achieve a SpO₂ range higher than the target and that is acceptable as long as FiO₂ levels are at or lower than 0.4. The goal of this protocol is to prevent direct or systemic FiO₂ toxicity in lungs primed with systemic or pulmonary disease [34–36]. The target SpO₂ range for this trial (88–94%) was modified from our conservative feasibility trial goal of 88–92%. The upper target was increased from 92% to 94% to mitigate hypoxemic episodes in patients with acute hypoxemic respiratory failure/ARDS who were noted to have recurrent hypoxemia with our previous tighter SpO₂ goal. This decision is also supported by the French LOCO2 trial which noted higher incidence of mesenteric ischemia and 90 day mortality in an ARDS cohort in which hypoxemic events were documented [24]. We posit that this extended SpO2 range maintains arterial oxygen tension within a physiological range while avoiding hyperoxemia, as recommended in recent literature and clinical practice guidelines. The extended range also allows providers flexibility for opting a preferred approach to oxygenation management. For example, physicians might want to maintain a conservative goal (closer to 88–92%) for patients with COPD [5] or those with concomitant hypoxic encephalopathy [23], while they may want to stay towards the higher end of the SpO₂ range (90–94%) when needed for patients with hemorrhagic shock and critical anemia. Also, the extended liberal range of targets can be used to adjust for SpO₂ variation in patients with various racial and ethnic backgrounds with varied skin tones, given that higher SpO₂ goals are required to prevent hypoxemia in black patients [37,38]. Data generated by our study regarding racial variation in pulse oximetry and correlation with arterial cooximetry and arterial oxygen tension will define any existing differences to further improve accuracy of oxygenation in people of color. While equipoise for liberal and conservative oxygen targets in critically ill patients remains, we emphasize that achieving that goal with the least required FiO₂ is essential to prevent direct pulmonary toxicity due to oxidative stress and its associated systemic adverse effects. Our trial is unique in that provides a roadmap for implementing early FiO₂ titration.

3.2. Deferred consent to support time-sensitive enrollment

The highest potential for hyperoxia is noted during the initial hours of mechanical ventilation [39]. To make maximal impact in reducing hyperoxic exposure, patients need to be enrolled as soon as possible after intubation. We therefore aimed to enroll patients as early as possible following initiation of mechanical ventilation, accordingly deferred consent was approved for this study.

Deferred consent has been in use since 1996 for emergency research and is used increasingly in trauma, cardiac arrest, in neonates, emergency room research and other oxygenation studies [40–42]. Procedures for deferred consent are also discussed as Exception from Informed Consent are detailed by the FDA39 (EFIC; 21 CFR 50.24) [43]. Several factors influence emergent consent in critically ill patients prior to onset of a time sensitive study. Some commonly noted factors include lack of decisional capacity, absence of a surrogate decision maker or Legal authorized representative (LAR) and poor receptiveness of research due to emotional burden of a sick member [44]. EFIC is of paramount importance in conducting emergency research since exclusion of such patients leads to lack of internal validity and generates bias in critical care studies. Previously, enrollment with the deferred consenting process has been successful in our pilot study.

3.3. Electronic medical records based alerts for implementation of automated FiO₂ delivery during mechanical ventilation

The algorithm was developed by the biomedical informatics research team at the Ohio State University the EMR EPIC health systems (version 2013). See Appendix B. Electronic “assistance” through EMR or computerized decision support in the ICU has widely helped improve compliance to guidelines, e.g. low tidal volumes, detection and management of sepsis, etc [20,45] These e-alerts for FiO₂ titration will serve as real time “reminders” without increasing provider workload (as it often occurs with outdated, out of context EMR reminders). Cochrane Effective Practice and Organization of Care (EPOC) Review Group list “reminders” as a specific professional intervention for implementing a change in practice or in provider behavior [46]. Eighteen reviews looked at “Reminders” alone as an intervention, both paper and computer-based clinical decision support systems and have found them about 73–78% effective at creating an effective change in provider behavior [47]. Our approach of using reminders and decision support is supported by the normalization process theory for implementation of interventions [48,49]. The normalization process theory has been previously applied as a theoretical framework for implementation of electronic health systems and guidelines [50,51].

One of our aims with EMR based alerts is to progressively refine feedback and improve our algorithmic accuracy for automation. Automated systems have been shown to improve the precision of closely monitored variables (such as SpO₂) and reduce the workload in clinical practice without increasing the risks to patients [52]. Such closed loop systems are being studied for anesthesia, continuous drug delivery and ventilator management [53–56]. Employed in the ICU, ventilator modes to aid weaning such as Proportional Assist Ventilation, Neurally-Adjusted Ventilator Assist work on similar principles of algorithmic feedback and have shown to reduce time on the ventilator [57,58]. Closed-loop systems responding to SpO2 levels can be integrated in ventilators or even as portable adjuncts in a resource limited setting [59,60]. In addition, automated closed loop systems for FiO₂ titration through methods other than SpO₂ feedback can also be generated for situation where pulse oximetry use is limited (racial and ethnic variation of skin tones, low perfusion state, dyshemoglobinemias and nail deformities [37,38,61]). Further research and development of automated titrating devices is considered as a future research priority to prevent oxygen toxicity in critically ill patients [27].

3.4. Limitation of study design

Responsiveness of respiratory therapists to the e-alerts is a critical component of the intervention. However, studying efficacy by randomizing providers (in this case, the respiratory therapists) is not feasible given practical constraints of staffing patterns and the one to one randomization used in the study design. Currently, the same respiratory therapist could be providing care for intervention and control patients. Given this pattern, diffusion of study intervention between groups by providers is also a potential limitation. As stated in the statistical analysis this will be studied by analyzing the treatment effect over sub-sequential times through the study and studying the interaction between treatment effect and enrollment time.

4. Conclusions

The strengths of our proposed study include its pragmatic nature with active engagement of relevant stakeholders for study design and implementation. We believe this strategy will lead to success of intervention efficacy, compliance and enrollment [62]. Feedback from this stakeholder group in our pilot trial has shaped the protocol to improve alert fatigue. Our approach in this study of leveraging EMR to develop a real-time and a continuous monitoring approach with automated alerts is both practical and powerful in ensuring compliance with recommended treatments to avoid adverse outcomes. Embedding the large amount of data generated by EMR for clinical research to conduct pragmatic clinical trials by creating a “learning health care” environment is encouraged by NIH through the NIH collaboratory for pragmatic clinical trials [63]. The high fidelity of our electronic algorithm employing minute by minute values of FiO₂ and SpO₂ over 24 h is novel and to our knowledge will be studied in this context for the first time. EPIC is a widely used EMR [64]. This real time screening algorithm can be employed in other targeted oxygen studies and can be exported to other institutions with EPIC health systems for implementation. Even in institutions with other EMRs, an algorithm with similar principles can be instituted with varying fidelity. Lastly, our study sets the stage for developing artificial intelligent mechanisms for precise oxygenation which in addition to reducing work load can be channeled towards delivery personalized oxygenation goals for patients.

Abbreviations list:

ARDS

Acute Respiratory Distress Syndrome

FiO₂

Fraction of Inspired Oxygen

SpO₂

Arterial Oxygen Saturation

ICU

Intensive Care Unit

MV

Mechanical Ventilation

PEEP

Positive End Expiratory Pressure