Deep Learning-Based Risk Prediction Model for Major Adverse Cardiovascular Events in Long-Term Breast Cancer Survivors

BackgroundClinical practice guidelines recommend cardiovascular toxicity risk restratification including evaluation of new cardiovascular risk factors and cardiovascular diseases 5 years after cancer therapy in adult cancer survivors. We aimed to develop prediction models for the risk of 10-year cardiovascular complications in five-year breast cancer survivors. MethodsWe used the Korean National Health Insurance Service databases between 2005 and 2021, including 5,131 five-year female breast cancer survivors diagnosed in 2006. The study population was randomly split in a 4:1 ratio into the derivation and validation cohort. The primary outcome was the occurrence of major adverse cardiovascular events (MACEs) at any time before the final follow-up at 10 years. We developed a deep learning survival model (DeepSurv) and compared its performance with traditional Cox proportional hazards (CPH) model using the same dataset. Model performance was assessed by time-dependent concordance (Ctd) index. Shapley Additive Explanations (SHAP) was used to assess feature importance. ResultsIn the validation cohort, the DeepSurv and CPH model yielded Ctd index values of 0.738 (95% CI, 0.712-0.779) and 0.733 (95% CI, 0.649-0.777), respectively. In SHAP analysis, age, history of stroke, dyslipidemia, anthracycline use, and aromatase inhibitor therapy ranked highly in both models. ConclusionsA deep learning survival model that incorporates both conventional and breast cancer treatment-related cardiovascular risk factors outperformed traditional regression model in predicting 10-year MACEs among individual five-year breast cancer survivors.