Machine learning prediction of mortality in venous thromboembolism patients: the Birmingham Black Country Venous Thromboembolism (BBC-VTE) cohort

Abstract Introduction Venous thromboembolism (VTE), including deep vein thrombosis (DVT) and pulmonary embolism (PE), is one of the main causes preventable death in hospitals UK. Current clinical risk scores to predict mortality patients with VTE are severity index (PESI) simplified PESI (sPESI) which have similar predictive power. Purpose To evaluate ability machine learning algorithms admitted compare their capability sPESI score for 30-day mortality. Methods The BBC-VTE was a retrospective multicentre patient cohort established determine features novel aspects prediction (and VTE-related complications) contemporary cohort. We include 1554 (mean age 65.6 years; 53% female) who represent all consecutive admissions final diagnosis 3 regional West Midlands, UK during years 2012–2014. dataset split into training (70%) validation (30%) cohorts. trained two tree-based models, Random Forests (RF) XGBoost (XG), using 5-fold cross-validation on This validated held-out compared simple logistic regression model. provide comparison score, we extracted sub-group (n=652) had values oxygen saturation, systolic blood pressure, heart rate, history cancer, cardiopulmonary disease, age. used RF using: i) only variables listed and; ii) available us. then against standard this C-indices (AUC) were comparison. Results c-indices XG full 0.85 [95% CI: 0.80 – 0.90] (Fig. 1a) 0.82 0.77 - 0.87], c-index being 0.83 0.78 0.88]. reported significantly smaller (p<0.05) than (0.75 0.69–0.80]). most important indicated by algorithm age, admission levels, discharge oral anticoagulation, previous malignancy 2). subgroup found be 0.72. In comparison, same gives larger 0.73 0.83]. When increased 1b). Conclusion Application hospital settings can improve post-VTE event above-and-beyond current score. Prospective study warranted validate external datasets construct individualised predictions. FUNDunding Acknowledgement Type funding sources: None. Figure 1. ROC curve comparisons 2. Feature Importances