A Real-time Risk-Prediction Model for Pediatric Venous Thromboembolic Events.

Abstract

Hospital-associated venous thromboembolism (HA-VTE) is an increasing cause of morbidity in pediatric populations, yet identification of high-risk patients remains challenging. General pediatric models have been derived from case-control studies, but few have been validated. We developed and validated a predictive model for pediatric HA-VTE using a large, retrospective cohort. The derivation cohort included 111 352 admissions to Monroe Carell Jr. Children's Hospital at Vanderbilt. Potential variables were identified a priori, and corresponding data were extracted. Logistic regression was used to estimate the association of potential risk factors with development of HA-VTE. Variable inclusion in the model was based on univariate analysis, availability in routine medical records, and clinician expertise. The model was validated by using a separate cohort with 44 138 admissions. A total of 815 encounters were identified with HA-VTE in the derivation cohort. Variables strongly associated with HA-VTE include history of thrombosis (odds ratio [OR] 8.7; 95% confidence interval [CI] 6.6-11.3; P < .01), presence of a central line (OR 4.9; 95% CI 4.0-5.8; P < .01), and patients with cardiology conditions (OR 4.0; 95% CI 3.3-4.8; P < .01). Eleven variables were included, which yielded excellent discriminatory ability in both the derivation cohort (concordance statistic = 0.908) and the validation cohort (concordance statistic = 0.904). We created and validated a risk-prediction model that identifies pediatric patients at risk for HA-VTE development. We anticipate early identification of high-risk patients will increase prophylactic interventions and decrease the incidence of pediatric HA-VTE.