Residual VTE risk after early ICU thromboprophylaxis: risk stratification in MIMIC-IV and an imaging-follow-up cohort.

BACKGROUND/OBJECTIVES: Pharmacological thromboprophylaxis reduces but does not eliminate venous thromboembolism (VTE) risk in critically ill patients. We assessed whether early routine ICU data stratify clinically recognized VTE risk among patients receiving early pharmacological thromboprophylaxis. PATIENTS/METHODS: We performed a retrospective 24-hour landmark prediction-model study using MIMIC-IV v3.1. Adult ICU stays with pharmacological thromboprophylaxis within 24 h were included. The primary endpoint was clinically recognized VTE before hospital exit, defined by the VTE diagnosis phenotype and temporally anchored to the first positive radiology evidence; death was not included in the endpoint. Six models were compared using fold-contained preprocessing, five-fold outer cross-validation and three-fold inner tuning. Conventional multivariable logistic regression, cause-specific Cox regression and Fine-Gray competing-risk regression were used to identify independent determinants; VTE-free hospital exit was treated as the competing event in time-to-event analyses. A prospective observational hospital imaging-follow-up cohort provided descriptive imaging results, exploratory local classification and a landmark-aligned independent validation subset for the frozen MIMIC model. RESULTS: The MIMIC-IV cohort included 5702 ICU stays and 174 VTE events (3.1%). Elastic net achieved AUROC 0.709, average precision 0.083 and Brier score 0.029. The highest-risk 20% captured 48.3% of events (event rate 7.36%; sensitivity 48.3%; specificity 80.9%; positive predictive value 7.4%). Among patients without preceding VTE, 717 deaths and 4811 discharges were treated as competing hospital exits. Estimated cumulative VTE incidence was 1.58% at day 7 and 2.24% at day 14. Fine-Gray regression identified invasive ventilation (adjusted subdistribution hazard ratio [sHR] 1.82, 95% CI 1.28-2.60), vasopressor/inotrope use (adjusted sHR 2.02, 95% CI 1.40-2.93) and renal replacement therapy (adjusted sHR 1.92, 95% CI 1.03-3.60) as independent determinants. In the hospital cohort, 25 of 138 patients had imaging-confirmed PE (18.1%). The landmark-aligned validation subset included 96 patients and 36 post-landmark imaging-confirmed VTE events; the frozen common-variable MIMIC model yielded AUROC 0.581 and average precision 0.461. CONCLUSIONS: Early routine ICU data stratified clinically recognized VTE risk after early prophylaxis exposure. At the 4.33% threshold, the highest-risk 20% contained nearly half of recognized VTE events. This fixed threshold provides a candidate entry rule for prospective evaluation of structured reassessment and imaging-priority strategies.