Optimizing pain management in breast cancer care: Utilizing ‘All of Us’ data and deep learning to identify patients at elevated risk for chronic pain

Abstract

Purpose: The aim of the study was to develop a prediction model using deep learning approach to identify breast cancer patients at high risk for chronic pain. Design: This study was a retrospective, observational study. Methods: We used demographic, diagnosis, and social survey data from the NIH ‘All of Us’ program and used a deep learning approach, specifically a Transformer-based time-series classifier, to develop and evaluate our prediction model. Results: The final dataset included 1131 patients. We evaluated the deep learning prediction model, which achieved an accuracy of 72.8% and an area under the receiver operating characteristic curve of 82.0%, demonstrating high performance. Conclusion: Our research represents a significant advancement in predicting chronic pain among breast cancer patients, leveraging deep learning model. Our unique approach integrates both time-series and static data for a more comprehensive understanding of patient outcomes. Clinical Relevance: Our study could enhance early identification and personalized management of chronic pain in breast cancer patients using a deep learning-based prediction model, reducing pain burden and improving outcomes.