Abstract
BACKGROUND: Absolute time-dependent risk of hospital-acquired venous thromboembolism (HA-VTE) is essential to evaluate the net clinical benefit of pharmacologic thromboprophylaxis. The multicenter Children's Hospital Acquired Thrombosis (CHAT) risk assessment model (RAM) identified risk factors for pediatric HA-VTE but was not designed to predict this risk. We aimed to predict the absolute time-dependent risk of HA-VTE in hospitalized children. METHODS: We conducted a single center case-cohort study of children ≤18 years old admitted from 2013 to 2022. Children with radiologically confirmed, symptomatic HA-VTE were considered cases. A 5 % random sample of eligible admissions formed the subcohort. Using Kaplan-Meier estimates and Cox regression with risk factors from the CHAT RAM and time to HA-VTE as outcome, we calculated the absolute time-dependent risk of HA-VTE of each child. Performance was assessed using area under the receiver operating characteristic curve (AUROC), calibration plot, and area under the precision-recall curve (AUPRC). RESULTS: We identified 81 children with HA-VTE from 23,287 admissions. Adolescent age, slightly limited mobility, and central venous catheter were associated with time to HA-VTE. Median absolute time-dependent risk of HA-VTE was 1.7 % (IQR: 1.0 %, 2.1 %) by day 12 of admission when the revised RAM performed best. AUROC by day 12 was 0.81 (95 % confidence interval, CI: 0.70, 0.93) with calibration slope of 1.13 (95 % CI: 0.38, 1.87), calibration intercept of 0.004 (95 % CI: -0.64, 0.65), and AUPRC of 0.04 (95 % CI: 0.004, 0.07). CONCLUSIONS: We accurately predicted the absolute time-dependent risk of HA-VTE. Similar predictions should be developed for bleeding.