Individualized oxygen targets may lower mortality in mechanically ventilated patients

Key takeaways:

• A machine learning model determined if an ICU patient benefited more from a low or high oxygen target.
• When patients received the SpO₂ target they were predicted to benefit from, mortality went down.

Among ventilated adults, individualized oxygenation targets collected from a machine learning model that considers baseline characteristics may lower 28-day mortality, according to results published in JAMA.

“In two temporally and geographically distinct randomized trials, patients’ individual characteristics modified the effect of lower vs. higher SpO₂ targets on 28-day mortality,” Kevin G. Buell, MBBS, pulmonary and critical care fellow at University of Chicago Medical Center, and colleagues wrote.

Using a machine learning model derived from the Pragmatic Investigation of Optimal Oxygen Targets (PILOT) trial and validated in the Intensive Care Unit Randomized Trial Comparing Two Approaches to Oxygen Therapy (ICU-ROX) trial, Buell and colleagues assessed if ICU patients’ baseline characteristics change the predicted impact of lower vs. higher peripheral oxygenation-saturation (SpO2) targets on mortality at 28 days.

Researchers noted a large range (27.2% absolute reduction to 34.4% absolute increase) when evaluating the impact of treatment on 28-day mortality in a cohort of 965 critically ill adults receiving mechanical ventilation, of whom 484 received lower SpO₂ treatment and 481 received higher SpO₂ treatment.

When dividing the cohort based on whether they are predicted to benefit from a lower or higher SpO₂ target using multiple baseline characteristics, researchers found some common characteristics within the groups. Notably, some adults had similar predicted benefits when comparing lower vs. higher SpO₂ targets, so they made up a third group of patients.

Frequently reported characteristics of those predicted to benefit with a lower SpO₂ target included older age, male sex, presence of hypoxic or nonhypoxic brain injury and cardiovascular disease. In contrast, common characteristics among those predicted to benefit with a higher SpO₂ target included younger age, abnormally high vital signs (heart rate, mean arterial pressure, respiratory rate and temperature) and the presence of sepsis and respiratory disease.

Researchers observed reduced mortality among patients when they were given the SpO₂ target that they were predicted to benefit from (likelihood ratio test for effect modification P = .02).

Patients assigned a lower vs. higher SpO₂ target in the group predicted to have better outcomes with a lower target had reduced 28-day mortality (–6.1%; 95% CI, –4.3% to 16.5%). On the other hand, patients assigned to a lower vs. higher target in the group predicted to have better outcomes with a higher target had increased 28-day mortality (13%; 95% CI, 3.5%-22.6%).

According to researchers, if every patient received the best SpO₂ target predicted for them as opposed to a randomized target, absolute overall mortality would have dropped by 6.4% (95% CI, 1.9-10.9%).

“Further prospective validation is required before results can be generalized to other patient populations and predicted individualized treatment effects are used to inform clinical care,” Buell and colleagues wrote.