Time Series Modeling and Forecasting of Breast Cancer Mortality among U.S. Women

Abstract

This study examines temporal mortality patterns among U.S. female breast cancer patients, addressing persistent health inequities across racial and ethnic groups through time series forecasting models. Using follow-up data from 4,024 cases via the U-BRITE platform, it systematically compared ARIMA, SARIMAX, ETS, and TBATS models. Results demonstrate that ARIMA and ETS achieved superior predictive accuracy with satisfactory residual diagnostics. SARIMAX, despite incorporating seasonal components and exogenous variables, failed to improve forecasts, while TBATS exhibited substantial prediction uncertainty. Analysis reveals elevated mortality risk during early post-diagnosis periods, suggesting systemic barriers in timely diagnosis and treatment. Based on identified high-risk windows, it proposes four intervention strategies: expanding screening coverage, enhancing medical support during peak-risk periods, establishing targeted programs for vulnerable populations, and reforming healthcare coverage systems. Methodologically, parsimonious models outperformed complex seasonal frameworks for monthly mortality data. Future research should integrate patient-level covariates and external factors to enhance predictive precision and inform evidence-based public health policy.