Machine Learning as a Diagnostic and Prognostic Tool for Predicting Thrombosis in Cancer Patients: A Systematic
Adham H El-Sherbini1, Stefania Coroneos1, Ali Zidan1
1Department of Biomedical and Molecular Sciences, School of Medicine, Queen's University, Kingston, Ontario, Canada.
Abstract:
Khorana score (KS) is an established risk assessment model for predicting cancer-associated thrombosis. However, it ignores several risk factors and has poor predictability in some cancer types. Machine learning (ML) is a novel technique used for the diagnosis and prognosis of several diseases, including cancer-associated thrombosis, when trained on specific diagnostic modalities. Consolidating the literature on the use of ML for the prediction of cancer-associated thrombosis is necessary to understand its diagnostic and prognostic abilities relative to KS. This systematic review aims to evaluate the current use and performance of ML algorithms to predict thrombosis in cancer patients. This study was conducted per Preferred Reporting Items for Systematic Reviews and Meta-Analysis guidelines. Databases Medline, EMBASE, Cochrane, and ClinicalTrials.gov, were searched from inception to September 15, 2023, for studies evaluating the use of ML models for the prediction of thrombosis in cancer patients. Search terms "machine learning," "artificial intelligence," "thrombosis," and "cancer" were used. Studies that examined adult cancer patients using any ML model were included. Two independent reviewers conducted study selection and data extraction. Three hundred citations were screened, of which 29 studies underwent a full-text review, and ultimately, 8 studies with 22,893 patients were included. Sample sizes ranged from 348 to 16,407 patients. Thrombosis was characterized as venous thromboembolism (n = 6) or peripherally inserted central catheter thrombosis (n = 2). The types of cancer included breast, gastric, colorectal, bladder, lung, esophageal, pancreatic, biliary, prostate, ovarian, genitourinary, head-neck, and sarcoma. All studies reported outcomes on the ML's predictive capacity. The extreme gradient boosting appears to be the best-performing model, and several models outperform KS in their respective datasets.
