Abstract
OBJECTIVES: The study aimed to investigate the association between features extracted from time-series of vital signs and hemodynamic instabilities in critically ill patients with sepsis. METHODS: The present study included a total of 9,248 sepsis patients from the eICU Collaborative Research Database (Version 2.0). The patients were admitted to the intensive care unit (ICU) for their initial hospitalization, had a minimum duration of ICU stay exceeding 6 h, and were aged between 15 and 80 years. Respiration permutation entropy (RPE), a time series feature extracted by tsfresh from 6-hour continuous vital signs monitoring records (heart rate, respiratory rate, and peripheral oxygen saturation) preceding hemodynamic instability events was analysed. The primary outcomes were the occurrence of the first hemodynamic instability event, ICU mortality, and hospital mortality. The prognostic value of respiration permutation entropy was evaluated by multivariable logistic regression models. Model performance was assessed by calculating the area under the receiver operating characteristic (ROC) curve. RESULTS: A total of 2200 patients (23.8%) experienced an episode of hemodynamic instability. Multivariable logistic models demonstrated that a significant association between RPE and an increased incidence of hemodynamic instability events in critically ill patients with sepsis with duration of ICU stay exceeding 6 h (odds ratio [OR] = 1.224, 95% confidence interval [CI] = 1.137–1.318). Furthermore, RPE was found to be associated with higher ICU mortality (OR = 1.311, 95% CI = 1.182–1.453) and hospital mortality (OR = 1.234, 95% CI = 1.129–1.349), particularly in male patients, with longer length of stay in ICU, higher sequential organ failure assessment scores, and cirrhosis, respectively. Importantly, this model achieved area under the curve (AUC) of 0.811 (95%CI = 0.801–0.822) for hemodynamic instability events, 0.761 (95%CI = 0.749–0.773) for hospital mortality and 0.779 (95%CI = 0.767–0.7990) for ICU mortality, respectively, in the ROC curve analyses. CONCLUSIONS: Elevated respiratory permutation entropy is associated with an increased incidence of hemodynamic instability events, ICU mortality, and in-hospital mortality, potentially guiding timely and effective resuscitation in sepsis patients. SUPPLEMENTARY INFORMATION: The online version contains supplementary material available at 10.1186/s12879-025-12436-7.