Machine learning prediction of venous thromboembolism after surgeries of major sellar region tumors

Abstract

PurposeTo describe and predict the risk of venous thromboembolism (VTE) after surgical resection of major sellar region tumors. MethodPatients with sellar region tumors were identified from a database. The outcome was VTE, including deep vein thrombosis (DVT) and pulmonary embolism (PE) within 60 days after surgery. We trained regression and machine learning models to predict the outcome using baseline characteristics, surgical findings and postoperative laboratory tests. ResultsAmong 3818 patients included, 124 patients developed VTE after surgery. The total 60-day VTE incidence was 3.2 %, with incidence peak within ten days after the surgery. The risk increased in patients >65 years old (OR 2.96, p < 0.001), in patients with chordoma (OR 3.40, p = 0.006) or craniopharyngioma (OR 1.86, p = 0.036), in patients underwent craniotomy approach (OR 2.78, p = 0.017), in patients with high volume CSF leakage (OR 4.24, p < 0.001), and in patients with longer surgical duration (OR 1.78, p = 0.029). The linear discriminant analysis algorithm had the highest AUC (0.869, 95%CI, 0.840–0.898) in predicting the outcome. The specificity, accuracy, and sensitivity of the best model were 61.8 %, 93.6 %, and 92.8 %, respectively. Risk stratification using our best model suggested that 1.3 % and 24.5 % of the patient developed VTE in the low-risk group and in the high-risk group, respectively. We developed an online decision-support tool available on https://deepvep.shinyapps.io/VTEpred/. ConclusionThe overall incidence of VTE after surgical resection of major sellar region tumors was clinically significant, especially in older patients with chordoma or craniopharyngioma.