Abstract Rationale Previous research has identified phenotypes of acute respiratory distress syndrome (ARDS) and have suggested that they may respond differently to systemic corticosteroids, but whether early steroid use is associated with mortality varies by phenotype remains uncertain. Objectives To evaluate the association between early systemic corticosteroid administration and in-hospital mortality overall and within ARDS phenotypes, and to test for phenotype-specific effect modification. Methods A retrospective, multicenter cohort of adults with ARDS admitted to intensive care units was analyzed and subsequently phenotyped as hyperinflammatory or hypoinflammatory using a validated 9-variable model (n≈112k). The exposure of interest was systemic corticosteroids within 24 hours of ARDS identification. The primary outcome was in-hospital mortality. Analyses included multivariable logistic regression with a steroid-by-phenotype interaction, propensity score matching within phenotype, and causal sensitivity analyses using a marginal structural model with inverse probability of treatment weighting (IPTW; stabilized, truncated at the 1st-99th percentiles) and a doubly robust augmented inverse probability weighting (AIPW) estimator. Measurements and Main Results Unadjusted analyses showed higher mortality with steroids in hyperinflammatory ARDS (OR 1.45, 95% CI 1.30-1.62) but not hypoinflammatory (OR 1.00, 95% CI 0.89-1.11). The interaction persisted after adjustment (interaction OR 1.31; p = 0.00088). Propensity-matched comparisons were attenuated and not significant. Causal sensitivity analyses were consistent with the previous unadjusted analyses: IPTW MSM showed a significant interaction (interaction OR 1.45, 95% CI 1.23-1.71) with a null overall steroid effect (OR 0.93, 95% CI 0.84-1.04). AIPW indicated a near-zero overall risk difference (+0.13 percentage points pp, 95% CI − 0.50 to + 0.77), near-null effect in hypoinflammatory (−0.05 pp), and higher mortality in hyperinflammatory (+2.9 pp, 95% CI + 0.8 to + 5.0). Weight distributions indicated adequate overlap. Conclusions Across the ARDS cohort as a whole, there was no demonstrable mortality benefit to steroid administration in the first 24 hours of ARDS identification. However, a harmful effect was seen in the hyperinflammatory group that was not observed in the hypoinflammatory group. Despite robust statistical methods, these findings warrant cautious interpretation given the potential for residual confounding and time-fixed exposure definition. This abstract is funded by: None
Fay et al. (Fri,) studied this question.