Data-driven subtypes of postoperative delirium in surgical ICU patients: a secondary analysis of a prospective cohort study.

BACKGROUND: Postoperative delirium (POD) is a heterogeneous syndrome associated with adverse outcomes. Traditional classifications may obscure underlying pathophysiological differences. Data-driven subtyping may reveal this heterogeneity and guide more precise management. METHODS: We performed a secondary exploratory analysis of a prospective cohort of adult surgical ICU patients who developed POD. Variables were obtained from prospectively collected cohort data, electronic health records, and bedside monitoring systems. Latent class analysis was applied to derive risk-profile-based subtypes. Delirium features and clinical outcomes were compared across subtypes. RESULTS: A total of 298 patients with POD were included (mean age, 67.3 years [SD, 13.6]; 60.4% male). Based on model fit indices and clinical interpretability, a three-class solution was selected as the optimal POD subtype structure. Subtype 1 included 120 patients (40.3%) and was characterized by greater intraoperative blood loss (median, 500 mL) and a higher rate of perioperative blood transfusion (74.2%). Subtype 2 included 104 patients (34.9%) and comprised the oldest patients (mean age, 70.0 years) with the most stable postoperative profile. Subtype 3 included 74 patients (24.8%) and had the highest proportion of urgent surgery (71.6%), the highest procalcitonin level (median, 28.1 ng/mL), and the lowest oxygenation index (median, 235). Delirium duration was longest in subtype 3 (median, 3 days), which also had the highest proportions of delirium recurrence (13.5%) and mechanical ventilation at first delirium identification (67.6%). Clinical outcomes showed descriptive differences across subtypes. Subtype 3 had the highest in-hospital mortality (18.2%), subtype 2 showed the most favorable short-term outcomes, and subtype 1 had the highest hospitalization costs, although not all between-subtype differences remained statistically significant in adjusted sensitivity analyses. CONCLUSION: This study identified three data-driven risk-profile-based subtypes of POD that differed in delirium features and clinical outcomes. These findings highlight POD heterogeneity and may inform subtype-based risk stratification and management. REGISTRATION: The PREDICt study was registered at ClinicalTrials.gov (NCT03704324).