Machine learning to individualize oxygenation targets during mechanical ventilation
Multiple randomized trials in heterogeneous patients requiring mechanical ventilation have not shown any benefit from any particular oxygenation target (higher vs. lower).
But what if those trials’ negative results hid a more complex reality, in which patients do often respond differently to oxygen—with many surviving because their oxygen target was selected “correctly” (albeit randomly), with others perishing from an O2 target unmatched to their real needs? In a well-randomized trial, equal numbers of clinically important differences in outcomes in opposite directions could average out to a null result, hiding the variability between patients.
Heterogeneity of treatment effects (HTE) poses a research conundrum: it could exist at problematic levels in any given trial, but without access to parallel universes, there’s no way to reliably detect or measure it. Subgroup analysis (e.g., “there was a benefit in people over 65 with diabetes” in an overall negative trial) has been the most-oft…
Keep reading with a 7-day free trial
Subscribe to PulmCCM to keep reading this post and get 7 days of free access to the full post archives.