Heart Failure sub-phenotyping and in-hospital and 28-day mortality prediction based on mean arterial pressure trajectory modeling.

Background: Acute decompensated heart failure patients follow a broad range of clinical pathways during hospitalization. Efficient patient sub-phenotyping based on early in-hospital trajectories, particularly when paired with clinical measurements, may inform optimal treatment strategies using digital twins. Methods: Using the MIMIC-IV database, we selected patients admitted to the ICU from 2008 to 2022. Patients were categorized into three distinct classes based on changes in mean arterial pressure (MAP) during the first 24 h of hospitalization, using Group-Based Trajectory Modeling (GBTM). Additional measures (including vital signs, serum biomarkers, hemodynamic data, and respiratory metrics) were then used as inputs to logistic regression and random forest analyses to identify important covariates associated with in-hospital and 28-day mortality. Results: A total of 4817 patients had sufficient data for analysis and were categorized into three trajectory classes: 2737 with relatively steady MAP (Class 1), 721 with increasing MAP (Class 2), and 1361 with decreasing MAP (Class 3). Overall mortality did not differ significantly among groups; however, variables associated with in-hospital and 28-day mortality varied across classes. Conclusion: This data-driven, trajectory-based sub-phenotyping approach of changes in MAP during the first 24 h of a hospital stay can characterize the heterogeneity of mortality risk among heart failure patients and can inform digital-twin-based scenario modeling to evaluate targeted management strategies for critically ill heart failure patients.