Unveiling Clinical Heterogeneity in Statin-Treated Sepsis Patients: A Machine Learning-Based Subphenotyping Study Leveraging the MIMIC-IV Database

Background: The therapeutic efficacy of statins in sepsis has long been a subject of debate, largely because the sepsis population is profoundly heterogeneous. To address this controversy, this study analyzed data from the MIMIC-IV database to evaluate the overall survival benefit of statins and identify data-driven, hidden subphenotypes among statin users. Methods: Data from 28,838 adult sepsis patients were extracted. Initially, we utilized 1:1 propensity score matching (PSM) to establish the baseline impact of statin therapy on the generalized cohort. Subsequently, to uncover variations in treatment response, we focused exclusively on the statin users (N=12118) by deploying an unsupervised machine learning pipeline—combining LASSO regression for feature selection and K-prototypes for clustering. Comprehensive multivariable Cox proportional hazards models were then constructed to evaluate the prognostic independence of the identified subphenotypes. Results: PSM successfully balanced the population (N=12,912) and demonstrated that statin administration is unequivocally linked to improved 28-day all-cause mortality ( P <0.0001). However, within the statin-user cohort, our data-driven approach delineated three distinct clinical subphenotypes. The most pivotal finding was the identification of Cluster 3, a subphenotype heavily burdened by advanced age, chronic kidney disease (CKD), and heart failure (HF), which experienced a precipitous decline in survival. Multivariable Cox regression revealed that, compared to the low-risk Cluster 2, belonging to Cluster 3 is a profound, independent predictor of 28-day mortality (Hazard Ratio = 2.121, 95% Confidence Interval: 1.937-2.322, P <0.001). Conclusions: The efficacy of statins in sepsis is profoundly modulated by a patient’s chronic disease architecture. While low-risk and metabolic phenotypes derive substantial benefit, the therapeutic window narrows in patients with complex multimorbidities like LC and HF. These findings advocate for a subphenotype-informed, precision-medicine approach in sepsis management.