Association between early body temperature dynamic trajectory and mortality and infection rates for patients with cardiac arrest in ICU: a retrospective study.

Background: In-hospital cardiac arrest (IHCA) is a common critical event with poor prognosis. Targeted temperature management (TTM) is the only approved therapy for cerebral ischemia post-cardiac arrest (CA). Body temperature (BT) is a key physiological indicator, but the dynamic evolution of BT is usually neglected, which may be closely associated with patient outcomes. This study aimed to investigate the dynamic BT trajectories within the first 72 hours after ICU admission in CA patients, identify distinct BT trajectory subphenotypes, and evaluate their associations with mortality and infection risk. Methods: A retrospective study was conducted using data from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database and the eICU Collaborative Research Database (eICU-CRD), including 3,052 intensive care unit (ICU) CA patients. The time-series k-means algorithm combined with dynamic time warping (DTW) was used to cluster dynamic BT trajectories within the first 72 hours of ICU admission, identifying subphenotypes. To evaluate the association between BT trajectory subphenotypes and mortality, this study employed Kaplan-Meier analysis, Cox regression and logistic regression models, and sensitivity and subgroup analyses. The tabular prior-data fitted network (TabPFN) was used to construct prediction models for mortality and underwent external validation. Infection status was also compared across subphenotypes. Results: Four BT trajectory subphenotypes were identified, including "hyperthermia", "normothermia", "hypothermic, slow rise" (HSR), and "normothermic, rapid decline" (NRD). The NRD subphenotype had the highest in-hospital, 7-day, 28-day, and 180-day mortality rates, which remained robust across multiple models. There was no significant difference in infection incidence among subphenotypes, but pathogen distribution varied slightly. The TabPFN models showed excellent performance in predicting mortality, with the area under the receiver operating characteristic curve (AUC) values ranging from 0.840 [95% confidence interval (CI): 0.688-0.979] to 0.991 (95% CI: 0.982-1.000). The above results remained consistent in external validation. Conclusions: The NRD subphenotype is significantly associated with higher mortality in ICU CA patients. Dynamic BT trajectories can provide insights for personalized clinical management and timely intervention in ICU patients with CA.