Prognostic value and clinical thresholds of the C-reactive protein–triglyceride–glucose index in critically ill patients with acute myocardial infarction: a dual-cohort machine-learning study

Despite the utility of the C-reactive protein–triglyceride–glucose index (CTI) as an integrative marker of metabolic-inflammatory dysregulation, whether the CTI is associated with mortality in patients with acute myocardial infarction (AMI) remains unknown. We conducted a retrospective cohort study using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database, version 3.0. Independent multicenter validation was performed using the eICU database. All patients were stratified into tertiles according to their CTI values. The association between CTI and mortality in AMI patients was analyzed using Cox regression models. Restricted cubic splines (RCS) regression was used to examine potential nonlinear relationships. Additionally, five machine-learning models—including random survival forest (RSF), conditional inference tree (Ctree), Elastic Net (Enet), extreme gradient boosting (XGboost), and Cox proportional hazards (CoxPH)—were applied. The SHapley Additive exPlanations (SHAP) method was used to interpret the optimal model and evaluate feature importance. Sensitivity and interaction analyses were performed to assess the stability of the findings. Among 1077 included AMI patients, Cox regression identified CTI as a significant independent risk factor for both 28-day (HR=6.63, 95% CI: 2.23–19.68) and 365-day mortality (HR=5.36, 95% CI: 3.02–9.52). RCS analysis indicated an approximately linear dose–response relationship between CTI and mortality risk. Optimal cutoffs were identified at 6.615 (28-day) and 6.475 (365-day), with patients exceeding these thresholds facing significantly higher risks (6.20-fold for 28-day and 3.27-fold for 365-day mortality, respectively). Among the five machine-learning models evaluated, the CoxPH model showed the best overall performance (internal AUC=0.865; external eICU AUC=0.670) and provided a positive net clinical benefit within a 5%–25% threshold probability range. Furthermore, sensitivity analyses on complete-case data were highly consistent with primary findings, and no significant interaction effects were observed across predefined subgroups, including diabetes, triglycerides, and CRP levels (all Interaction P > 0.05). Higher CTI values were independently associated with increased all-cause mortality in critically ill AMI patients. These associations are hypothesis-generating rather than confirmatory; the CTI-integrated modeling framework and the proposed cutoffs require external prospective validation, and comparison with established risk scores, before any clinical application.