JMIR Publications

ABSTRACT

Background:

With the risk prediction models based on machine learning(ML) for venous thromboembolism(VTE) in patients increasing, the quality and applicability of these models in practice and future research remain unknown. How ML is predicting and how many factors are selected have been research hotspots in VTE prediction.

Objective:

To systematically review relevant literatures on the predictive value of machine learning for venous thromboembolism.

Methods:

A comprehensive literature search was conducted across multiple databases, including PubMed, Web of Science, MEDLINE, Embase, CINAHL and the Cochrane Library for relevant studies on predictive models for venous thromboembolism. The novel Prediction Model Risk of Bias Assessment Tool (PROBAST) was used to assess the risk of bias of the ML models. Sensitivity, specificity and area under the curve (AUC) were used to evaluate the performance of the prediction models to investigate the predictive value of machine learning for venous thromboembolism.

Results:

27 studies were included in the systematic review. The pooled sensitivity, specificity, and area under the curve were 0.79 (95% CI 0.78-0.80), 0.82 (95% CI 0.81-0.82) and 0.8774, respectively. The studies were found to exhibit a considerable degree of bias, primarily due to shortcomings in the handling of missing data and the reporting of the study design. Age was used more often in prediction models. Random Forest (RF) was the superior ML model in predicting venous thromboembolism.

Conclusions:

It was effectively to predict venous thromboembolism in patients with machine learning, and may provide a reference for the development or updating of subsequent scoring systems. Clinical Trial: This systematic review and meta-analysis were conducted in accordance with the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) guidelines and registered with PROSPERO (CRD420251041604).