Deep Learning Approaches for Cardiovascular Risk Prediction Incorporating Peripheral Arterial Disease Score in Cancer Survivors

Abstract

Cancer survivors face elevated risks of cardiovascular disease (CVD), often exacerbated by underdiagnosed peripheral arterial disease (PAD). This study presents a deep learning–based framework for cardiovascular risk prediction among cancer survivors, incorporating a composite five-factor PAD score derived from the NHANES 2021–2023 dataset. After cohort selection and feature engineering, eleven distinct machine learning and deep learning models were implemented, including feedforward, LSTM, CNN-LSTM, Wide & Deep, and Tab Transformer architectures. Extensive preprocessing, chi-square association testing, and ROC AUC evaluation ensured methodological robustness. While traditional models showed limitations in recall and F1-score, attention-based and convolutional architectures demonstrated improved predictive capacity. The chi-square test confirmed a significant association between PAD scores and CVD comorbidities (χ² = 112.4, p < 0.0001). These findings highlight the potential of PAD-aware deep learning models for improved risk stratification in cardio-oncology care.