Most randomized trials in critical care report no mortality benefit; this may reflect competing pathogenic mechanisms, patient heterogeneity, or true ineffectiveness of interventions. We hypothesize that in acute respiratory distress syndrome (ARDS), randomizing only those patients who show a favorable physiological response to an intervention would help ensure that only those likely to benefit would be entered into the study. If true, this would decrease study "noise" and reduce required sample size, thereby increasing the chances of finding true-positive outcomes. It would also lessen the chances of exposing patients to treatments that are unlikely to help or that could cause harm. We present a reanalysis of randomized clinical trials of positive end-expiratory pressure in ARDS that support this hypothesis.