Machine Learning Models for Risk Prediction of Cancer Associated Thrombosis: A Systematic Review and Meta-Analysis.

Abstract

Background: While the number of models for predicting the risk of cancer-associated thrombosis has been rising, there is still a lack of comprehensive assessment for machine learning prediction models.

Objective: To critically appraise and quantify the performance studies using machine learning to predict cancer-associated thrombosis.

Methods: We conducted searches on PubMed, Embase, The Cochrane Library, Cumulative Index to Nursing and Allied Health Literature, and other related databases for the related publications (from inception to December 1, 2023). The Prediction Model Risk of Bias Assessment Tool checklist was employed to evaluate the risk of bias and applicability. The Grading of Recommendations Assessment, Development and Evaluation system was used to evaluate the quality of evidence in systematic reviews. Meta-analyses were conducted using R (version 4.3.2).

Results: A total of thirty-two studies were included. Mostly included literature exhibits a high risk of bias, and the applicability of the prediction models is deemed acceptable. The twenty-one included studies in the meta-analysis demonstrated the high predictive capacity of the machine learning models for cancer-associated thrombosis.

Conclusion: Most of the prediction models included in the study showed good applicability and excellent prediction performance, but there was a high risk of bias.