Machine learning-based prediction of cardiovascular adverse events in patients with cancer: a systematic review
Year: 2026
Type: article
Study design: Systematic Review
Abstract: Cardiovascular (CV) adverse events are increasingly recognized in patients with cancer. Previous reviews of AI/ML have focused on single cancer types, imaging-based data, and lacked evaluation of methodological rigor. This systematic review synthesized AI/ML models developed to predict CV adverse events from patient-level clinical data across diverse cancer populations. This review followed the PRISMA 2020 guidelines. PubMed and Web of Science were searched through 26 October 2025. Study characteristics, model development, and handling of features and missing data were extracted. Study quality was assessed using the IJMEDI checklist. Of 32 included studies, 18 compared multiple algorithms and 14 used a single algorithm. Random forest and XGBoost were the most common methods (n = 17, respectively), and XGBoost was most often the best-performing model in multi-algorithm studies, although substantial study heterogeneity precludes concluding general algorithmic superiority. Common limitations were unreported missing data handling (n = 17), limited external validation (n = 8), and rare calibration assessment (n = 4). Most studies were rated medium quality (n = 28). AI/ML models show promise for predicting CV adverse events in patients with cancer; however, clinical applicability is constrained by insufficient preprocessing transparency, limited external validation, and inadequate calibration reporting. Machine learning is increasingly used to predict cardiovascular complications in patients with cancer, but how to build, evaluate, and report these models is still debated. This review included 32 studies. Random forest and XGBoost were the most common methods, and XGBoost most often performed best, although differences across studies make it hard to say any one method is truly superior overall. However, several issues were noted, including poor reporting of missing data handling, limited external validation, and infrequent assessment of calibration. Future studies should use clearer reporting standards, test models in diverse populations, and apply explainable AI to support clinical use in cardio-oncology.
Source: Figshare
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7203533332
