JMIR Publications

ABSTRACT

Background:

There is increasing research on machine learning in predicting venous thromboembolism after joint arthroplasty, but the quality and clinical applicability of these models are unclear.

Objective:

This systematic review and meta-analysis aims to evaluate the predictive performance and methodological quality of machine learning models for venous thromboembolism risk after joint replacement surgery, and to provide insights for further clinical application.

Methods:

Web of Science, Embase, Scopus, CNKI, Wanfang, Vipro, and PubMed were searched until December 15, 2024. The risk of bias and applicability were evaluated using the prediction model Bias Risk Assessment Tool (PROBAST) checklist. Quantitative synthesis and meta-analysis included models reporting AUC value with 95% confidence intervals.

Results:

There were 34 prediction models from 9 studies, and the most used machine learning models were extreme gradient boosting and logistic regression. 24 models with reported confidence intervals were incorporated into the meta-analysis, and the total area under the curve was 0.826 (95% CI 0.775-0.876). All studies indicated a high risk of bias and considerable heterogeneity. Age, gender, diabetes, and hypertension were the most frequently used predictive factors.

Conclusions:

The predictive performance of machine learning models varies greatly, and the AUC value of the report indicates that most of the models have good discriminative ability. These models have a high risk bias, and it is necessary to take this into account when they are used in clinical practice. Future studies should adopt a prospective study design, ensure appropriate data handling, and use external validation to improve model robustness and applicability. Clinical Trial: The protocol for this study is registered with PROSPERO (registration number: CRD42024625842).