<p>Sepsis remains a major cause of mortality in intensive care units. Hypothermia is frequently observed in septic patients and is associated with greater disease severity and poorer outcomes. This study aimed to develop and validate a mortality prediction model for hypothermic sepsis and explore the clinical and biological relevance of prothrombin time (PT). A retrospective cohort study was conducted using the MIMIC-IV database, including 3,618 adult patients with sepsis classified according to ICU temperature status. Kaplan–Meier analysis compared survival across temperature groups. Five machine learning models were developed to predict in-hospital mortality among patients with hypothermic sepsis. Model performance was externally validated using the eICU database and compared with SOFA, qSOFA, and PT alone. Sensitivity analyses were performed using alternative temperature definitions and missing-data strategies. Complementary cecal ligation and puncture mouse and LPS-stimulated HepG2 cell models were used to explore biological relevance. Hypothermic patients showed poorer survival than normothermic or hyperthermic patients. The random forest model showed the best performance in the MIMIC-IV cohort (AUC = 0.813) and maintained acceptable discrimination in the eICU cohort (AUC = 0.742). It outperformed SOFA, qSOFA, and PT alone in clinical utility analyses. PT was consistently identified as the most important predictor and showed the strongest standalone predictive performance. Sensitivity analyses supported the robustness of these findings. Experimental results suggested that hypothermia may aggravate sepsis-associated coagulation dysfunction and liver injury, with reduced hepatic F2 and F7 expression and suppressed coagulation factor synthesis in HepG2 cells. Hypothermia is associated with poorer survival in sepsis, and PT is a key predictor of in-hospital mortality in hypothermic sepsis. Integrated clinical and experimental findings support PT-based risk stratification in this high-risk population.</p>