Time-dependent associations of documented low-dose aspirin administration with 28-day mortality and early safety outcomes in adults with sepsis: a MIMIC-IV cohort study

Abstract

Associations between aspirin use and sepsis outcomes may reflect treatment timing and clinical selection. This retrospective Medical Information Mart for Intensive Care IV (MIMIC-IV) study included 31,888 adults with Sepsis-3. T0 was defined as the later of intensive care unit (ICU) admission or suspected infection; sepsis ascertainment could occur subsequently. Piecewise time-dependent Cox models evaluated actual enteral aspirin administration of 75–100 mg and 28-day restricted in-hospital and ICU mortality. Baseline sequential organ failure assessment (SOFA) measurements were restricted to T0 or earlier. Sensitivity analyses examined exposure history, eligibility timing, and a 24-h matched landmark cohort. A continuous-time marginal structural model (MSM) incorporated lagged clinical information. Modified Poisson models evaluated laboratory and transfusion outcomes. Eligible aspirin administration was recorded at or after T0 for 4,771 patients, including 3,646 within 24 h. Adjusted in-hospital mortality hazard ratios (HRs) were 0.48 (95% CI 0.31–0.75) during 0–24 h and 0.63 (0.51–0.78) during 24–72 h, without clear associations thereafter. ICU estimates were similar. Qualifying SOFA scores occurred after T0 in 24,916 patients; restriction to the 6,972 patients with qualifying SOFA scores by T0 yielded HRs of 0.54 (0.28–1.03) and 0.62 (0.44–0.88) in the first two intervals. The MSM had residual imbalance and truncation-sensitive early estimates. After multiple imputation, the first two in-hospital HRs were 0.50 (0.32–0.79) and 0.66 (0.53–0.82), close to the primary estimates. A hemoglobin decline of ≥2 g/dL during 0–72 h was associated with early aspirin administration [adjusted risk ratio (RR) 1.10, 1.03–1.17], but 606 early recipients had a qualifying decline recorded before their first dose. The temporality-restricted 24–72-h analysis showed no clear association (RR 1.05, 0.91–1.21). Documented aspirin administration was associated with lower early restricted mortality. Retrospective eligibility, residual confounding, and uncertain prior exposure prevent causal interpretation; these findings do not justify initiating aspirin solely for sepsis.