Temporal Disease Trajectories Derived from Electronic Health Record Data in Critical Care Patients

Electronic health records contain detailed longitudinal diagnostic information, yet most prognostic models rely on static representations of comorbidity. Here we analyzed disease trajectories in a large critical care cohort of 364,627 patients and over six million diagnoses from the MIMIC-IV database. We identified 1,258 statistically significant disease transitions corresponding closely to known clinical progressions. We used graph-based clustering to group, these diagnoses into 74 multimorbidity clusters that, captured coherent clinical entities such as cardiorenal syndrome. Certain disease sequences were strongly predictive of mortality, notably progression from sepsis to respiratory failure and shock, associated with elevated death risk. These findings show that temporal diagnosis trajectories provide complementary risk signals beyond traditional static representations of comorbidity.