Body temperature as a dynamic host-response biomarker in sepsis: trajectories and time-varying mortality risk.

Background: Body temperature is a readily available physiological signal that may reflect dynamic host responses in sepsis, yet most studies rely on single or early measurements and do not capture the dynamic nature of thermoregulation during critical illness. Whether the association between body temperature and mortality varies over the course of sepsis remains unclear. Methods: In this multicenter retrospective cohort study, patients with sepsis were identified from the MIMIC-IV database as the derivation cohort and externally validated using the eICU Collaborative Research Database and a single-center intensive care unit (ICU) cohort from Nanjing Drum Tower Hospital. Body temperature was summarized into 6-hour intervals during the ICU stay. The primary outcome was 28-day in-hospital mortality. Joint latent class mixed models were used to identify temperature trajectory classes and generate landmark-based dynamic survival predictions. Piece-wise exponential additive mixed models were applied to estimate the nonlinear and time-varying association between infection-related temperature responses and in-hospital mortality. Results: A total of 8681 patients were included in the derivation cohort and 13717 in the validation cohort. Five distinct temperature trajectory classes were identified, demonstrating heterogeneous thermoregulatory patterns during the ICU stay. Mortality risk differed significantly across trajectory classes. The low-increase-moderate hyperthermia trajectory was associated with the highest mortality risk in both early and later periods of hospitalization. Dynamic prediction models incorporating longitudinal temperature trajectories showed better overall discrimination than models based on single temperature measurements. Time-varying analysis revealed a nonlinear U-shaped relationship between temperature and mortality, with the lowest risk near the upper range of normothermia. Both hypothermia and hyperthermia were associated with increased mortality, and the prognostic impact of infection-related temperature responses varied over the ICU course, peaking at approximately 4 days after ICU admission. Conclusion: Body temperature may serve as a dynamic host-response marker in sepsis, with prognostic significance determined by both temporal trajectory patterns and disease stage. Longitudinal temperature monitoring could provide a clinically accessible strategy for dynamic risk stratification and future temperature-guided precision management in sepsis.