Science, Medicine, and the Anesthesiologist

Based on data from randomized trials, no oxygenation targets are known to affect patient outcome. Also, oxygenation targets specific to a patient’s characteristics have not been reported. This study was performed using machine learning in two randomized trials, determining whether an individual’s characteristics have an impact on mortality when a lower versus a higher peripheral oxygenation saturation (Spo₂) is targeted. The first study, the Pragmatic Investigation of Optimal Oxygen Targets (PILOT) trial, performed from July 2018 to August 2021 in the United States including 1,682 patients with invasive mechanical ventilation, predicted 28-day mortality (the primary outcome) based on treatment aiming at a lower versus higher Spo₂. Data from the PILOT trial were then validated in the Intensive Care Unit Randomized Trial Comparing Two Approaches to Oxygen Therapy (ICU-ROX) trial with 965 patients from 21 intensive care units in Australia and New Zealand (September 2015 to May 2018). In the ICU-ROX validation cohort, the predicted effect of an individual patient target with lower versus higher Spo₂ was found to vary from 27% absolute reduction to 34% absolute increase in 28-day mortality. On an individual basis, target predicted Spo₂ would have been superior to the randomized Spo₂ in reducing overall mortality by 6.4% (95% CI, 1.9 to 10.9%).

Take home message: This study reports that individualized oxygenation targets for critically ill patients, defined through machine learning analyses of two randomized trials, could reduce mortality. A prospective trial should be conducted to confirm the positive effect on reducing 28-day mortality using individualized instead of randomized oxygenation targets.