Validation of risk assessment models for venous thromboembolism and bleeding in critically ill adolescents

Abstract

Objective: To determine the performance of risk assessment models that were developed for adults, in predicting venous thromboembolism (VTE) and bleeding in critically ill adolescents. Study design: We conducted a retrospective cohort study of adolescents 12 to 17 years old admitted to the pediatric intensive care unit who received cardiopulmonary support but did not have VTE on admission nor received anticoagulation. Discrimination, using areas under the receiver operating characteristic (AUROC) and precision-recall (AUPRC) curves, and calibration, using Hosmer-Lemeshow test, of the Geneva, Padua, IMPROVE VTE and IMPROVE Bleed models were calculated. Results: Of 536 adolescents analyzed, 7 (1.3%) developed VTE and 13 (2.4%) bled. AUROCs of the Geneva, Padua and IMPROVE VTE models ranged from 0.46 to 0.59, with 95% confidence intervals (CI) including 0.5. AUPRCs ranged from 0.011 to 0.017, with 95% CIs including 0.013. Only IMPROVE VTE model had non-statistically significant Hosmer-Lemeshow test. IMPROVE Bleed model had AUROC and AUPRC of 0.75 and 0.062, with 95% CIs excluding 0.5 and 0.024, respectively. Hosmer-Lemeshow test was not statistically significant. Conclusion: Despite similarities in coagulation between adolescents and adults, risk assessment models for VTE in adults should not be used for critically ill adolescents. The model for bleeding may be useful.