International Normalized Ratio as an Effect Modifier of the Heparin-Mortality Association in Sepsis: A Retrospective Cohort Study

Abstract Heparin is widely used in sepsis, but whether baseline coagulation status modifies its association with mortality is unknown. In 38,431 septic ICU patients from MIMIC-IV, we modeled the international normalized ratio (INR) as a continuous effect modifier of the heparin-mortality association using restricted cubic splines in Cox models, with inverse probability of treatment weighting, propensity score matching, and competing-risk models for robustness, and negative-control exposures to assess severity confounding. INR significantly modified the heparin-mortality association (p<0.001), with a crossover from harm to benefit at INR ≈ 2.8: associations were adverse at INR ≤ 2.4 (adjusted absolute risk increase ≈ 5.7%) and favorable at INR ≥ 3.4 (absolute risk reduction ≈ 5.0%). Negative-control exposures without antithrombotic mechanisms showed INR-dependent interactions (renal replacement therapy: p=0.035; corticosteroids: p=0.078), formally demonstrating severity-of-illness confounding, and therapeutic heparin was almost never administered to high-INR patients in external cohorts. Observational data therefore cannot resolve whether heparin benefits or harms coagulopathic patients with sepsis. We propose a three-tier INR-based prognostic enrichment hypothesis as a design blueprint for an INR-enriched randomized trial rather than for clinical decisions.