The occurrence of deep venous thrombosis (DVT) following total hip arthroplasty (THA) poses a substantial risk of morbidity and mortality, highlighting the need for preoperative risk stratification and prophylaxis initiatives. However, there exists a paucity of big-data-driven predictive models for DVT risk following elective hip arthroplasty. Therefore, this study aimed to develop and assess machine learning (ML) models in predicting DVT risk following THA using a national patient cohort.

We hypothesized that machine learning models would accurately predict patient-specific DVT risk in patients undergoing elective total hip arthroplasty.

The ACS-NSQIP national database was queried to identify 70,733 THA patients from 2013 to 2020, including 317 patients (0.45%) with DVT. Artificial neural network, random forest, histogram-based gradient boosting, k-nearest neighbor, and support vector machine algorithms were trained and utilized to predict the risk of DVT following THA. Model performance was assessed using discrimination, calibration, and potential clinical utility.

Histogram-based gradient boosting demonstrated the best prediction performance with an area under the receiver operating curve of 0.93 (discrimination), a slope of 0.92 (closely aligned with actual outcomes), an intercept of 0.18 (minimal prediction bias), and a Brier score of 0.010 (high accuracy). The model also demonstrated clinical utility with greater net benefit than alternative decision criteria in the decision curve analysis. Length of stay, international normalized ratio, age, and partial thromboplastin time were the strongest predictors of DVT after primary THA.

Machine learning models demonstrated excellent predictive performance in terms of discrimination, calibration, and decision curve analysis. Further research is warranted in terms of external validation to realize the potential of these algorithms as a valuable adjunct tool for risk stratification in patients undergoing THA.

III; Retrospective study.