Mechanical ventilation involves titrating the fraction of inspired oxygen to maintain arterial oxygen saturation (SpO2). The SpO2 target that results in the best outcomes for critically ill adults has historically been unknown. Randomized trials comparing use of a higher SpO2 target (96-100%) vs a lower SpO2 target (88-92%) have not found an average treatment effect among patients overall. However, the optimal SpO2 target may differ for patients with different characteristics. Recently, data from randomized trials of SpO2 targets were used to derive and validate a statistical model that predicts which SpO2 target will result in the best outcomes for an individual patient based on his or her unique characteristics (personalized SpO2 target). This statistical model has been incorporated into the electronic health record at Vanderbilt such that, for each patient receiving mechanical ventilation in the medical intensive care unit, information on which SpO2 target is predicted to result in the best outcome for the patient can be made available to clinicians. However, the use of personalized SpO2 targets for critically ill adults receiving mechanical ventilation has never been examined in a randomized trial and whether using such a personalized SpO2 target in clinical care can improve patient outcomes remains unknown. This randomized trial will examine the effect of using information on the SpO2 target that is predicted to be best for a patient based on his or her unique characteristics (personalized SpO2 target) versus usual care.

Each year, 2-3 million critically ill adults in the United States receive invasive mechanical ventilation. In-hospital mortality among critically ill adults receiving mechanical ventilation remains approximately 25-35%. Approaches to care that decrease mortality for critically ill adults receiving invasive mechanical ventilation are urgently needed.

Mechanical ventilation universally involves titrating the fraction of inspired oxygen (FiO2) to maintain arterial oxygen saturation - as assessed by pulse oximetry (SpO2) or blood gas analysis (SaO2) - or arterial oxygen tension (e.g., PaO2). Using higher SpO2 targets (96-100%) provides a margin of safety against hypoxemia, but increases exposure to excess FiO2, hyperoxemia, and tissue hyperoxia, potentially causing oxidative damage and inflammation. Using lower SpO2 targets (88-92%) minimizes these risks but may increase exposure to hypoxemia and hypoxia-induced organ injury. Historically, the effects of higher versus lower SpO2 targets on patient outcomes were unknown.

Our recent randomized trial comparing higher versus lower SpO2 targets among 2,541 critically ill adults receiving mechanical ventilation in the medical intensive care unit (ICU) found that use of a higher versus lower SpO2 target did not result in overall differences in short-term outcomes (e.g., 28-day mortality) or long-term outcomes (e.g., cognition at 12 months). Results have been similar in multiple other large, randomized trials in different settings.

Randomized trials traditionally report the average effect of treatment on outcomes for the overall population. However, the effect of treatment on outcomes may differ for patients with different characteristics. Such nonrandom variation in the magnitude or direction of treatment effect is called heterogeneity of treatment effect. To understand which treatment will produce the best outcomes for a given patient, clinicians and patients need randomized trials to move beyond reporting the average treatment effect to reporting the effect of treatment on outcomes for an individual patient based on the patient's unique characteristics, referred to as individualized treatment effect.

Recently, we used the dataset from our trial of higher versus lower SpO2 targets to develop a statistical model to predict the effect of use of a higher versus lower SpO2 target on 28-day mortality for an individual patient, considering each of the patient's baseline characteristics simultaneously. This statistical model uses 24 patient characteristics available at the time of initiation of invasive mechanical ventilation to predict which SpO2 target will result in the best outcome for that patient. The model inputs are each patient's value for each of the 24 baseline characteristics. The model output is the predicted absolute effect of using a higher or lower SpO2 target on 28-day in-hospital mortality for the patient, conditional on all of the patient's values for the baseline characteristics. To validate the accuracy of this statistical model, we applied it to the dataset from a second, geographically and temporally distinct randomized trial of higher vs lower SpO2 targets. We found that, despite no significant average treatment effect in either trial, the effect of use of a higher versus lower SpO2 target on mortality ranged widely for individual patients, with many patients appearing to benefit from either a lower or a higher SpO2 target.

Before the statistical model is widely applied in a clinical care, a randomized trial is required to determine whether using information from the model to guide oxygen therapy improves patient outcomes, compared with usual care. The EXPRESS trial will be a randomized trial comparing a personalized SpO2 target group (in which clinicians receive information on the SpO2 target predicted to result in the best outcome for each patient) vs a usual care group (in which clinicians do not receive information on the SpO2 target predicted to result in the best outcome for each patient) among adults receiving mechanical ventilation in the ICU at Vanderbilt.

Inclusion Criteria:

• Patient is located in a participating unit
• Patient is receiving invasive mechanical ventilation

Exclusion Criteria:

• Patient is known to be less than 18 years old
• Patient is known to be pregnant
• Patient is known to be a prisoner
• Patient is receiving extracorporeal membrane oxygenation
• Clinician has determined that a specific approach to oxygen therapy is required or contraindicated for the optimal care of the patient

Our approach to data management and sharing in the proposed EXPRESS trial will comply with the 2023 FINAL NIH Policy for Data Management and Sharing. We will also comply with the NHLBI policy for data sharing from clinical trials and epidemiological studies including submitting data to NHLBI BioData Catalyst (BDC) or a comparable repository active at the time of trial result publication. Procedures are planned for sharing de-identified data, as well as statistical coding and any software developed. In addition to sharing data, we will also make available the clinical protocol and statistical analysis plan. This will be done as supplemental material to the primary study manuscript so that it is shared in context with the clinical data.

Following publication. No end date