Association between albumin-bilirubin score and mortality risk in patients with acute myocardial infarction: a retrospective study based on MIMIC-IV

Abstract

The albumin-bilirubin (ALBI) score, a liver function indicator, has a not fully established prognostic association with acute myocardial infarction (AMI) mortality. Our study sought to investigate the adjusted association of the ALBI score with all-cause mortality (ACM) in this population. We analyzed the first eligible ICU stay of adults with AMI in MIMIC-IV. ALBI was modeled continuously and using three prespecified grades (Grade 1 ≤ − 2.60, Grade 2 > − 2.60 to ≤ − 1.39, and Grade 3 > − 1.39). Kaplan–Meier curves, Cox models, restricted cubic splines (RCS), receiver-operating-characteristic (ROC) analyses, paired DeLong tests, and incremental discrimination metrics were used for 30-, 90-, 180-, and 360-day ACM. Five random-forest multiple imputations were analyzed and pooled with Rubin’s rules. Global and term-specific proportional-hazards assumptions were tested using scaled Schoenfeld residuals in each completed dataset. Repeated ALBI-specific violations were handled using ALBI × log(time/30 days) interactions, with separate Grade 2 and Grade 3 time interactions in categorical models. Binary RCS-derived high-versus-low ALBI propensity-score matching (PSM) was conducted as a sensitivity analysis. Among 3,153 critically ill AMI patients, 605, 2,059, and 489 were classified as Grade 1, Grade 2, and Grade 3, respectively. In fully adjusted Model 3 including SOFA, continuous ALBI had fixed-effect HRs of 1.43, 1.42, and 1.47 for 30-, 90-, and 180-day ACM. For 360-day ACM, exposure-specific time-varying models yielded day-360 h of 1.18 (0.98–1.43) for continuous ALBI and 1.21 (0.81–1.83) for Grade 3 versus Grade 1. ALBI AUCs were 0.677, 0.674, 0.677, and 0.668, while SOFA+ALBI AUCs were 0.746, 0.736, 0.730, and 0.715. The corresponding SOFA+ALBI versus SOFA ΔAUCs were 0.015, 0.018, 0.021, and 0.020. Continuous NRI and IDI indicated incremental discrimination, but the magnitude was modest. In matched Model 1 Cox analyses, binary RCS-derived high-versus-low ALBI PSM yielded HRs of 1.22, 1.25, 1.31, and 1.30 across the four endpoints. In this single-center retrospective cohort, higher ALBI was associated with higher 30-, 90-, and 180-day ACM, whereas 360-day associations were reported as time-specific estimates because the ALBI-specific proportional-hazards assumption was not met. ALBI showed limited incremental discrimination beyond SOFA. These findings do not establish causality or clinical utility and require external validation.