Prostate cancer and prostatic diseases · 2026
Interpretable machine learning identifies health system levers for survival outcomes of patients with prostate cancer.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Patel Milit S, Janopaul-Naylor James, Ting Frederic Ivan L, Wu James Fan, McBride Sean M, Rathkopf Dana E, Feliciano Erin Jay G, Truong Hong, Iyengar Puneeth, Willmann Jonas, Gorovets Daniel E, Nguyen Paul L, Murthy Vedang, Mahal Brandon A, Lee Nancy Y, Nagar Himanshu, Dee Edward Christopher
Аннотация
BACKGROUND: Advances in the care of patients with prostate cancer across the disease spectrum have resulted in substantial survival gains over recent decades. However, global disparities in prostate cancer survival, stage at diagnosis, and access to timely diagnosis and treatment remain among the greatest across all malignancies. We aimed to quantify country-level health system correlates of the prostate cancer mortality-to-incidence ratio (MIR).
METHODS: We applied interpretable machine learning using CatBoost with SHAP (SHapley Additive exPlanations) to all 185 countries with prostate cancer MIR estimates in GLOBOCAN 2022. National age-standardized MIR estimates were linked to 11 health system indicators from the World Health Organization Global Health Observatory, World Bank World Development Indicators, United Nations agencies, and the Directory of Radiotherapy Centres. Model performance was evaluated using repeated leave-one-country-out cross-validation with bootstrap-based uncertainty estimation. CatBoost handled missing predictor values natively, allowing inclusion of all countries without imputation.
RESULTS: The model demonstrated strong predictive performance for prostate cancer MIR (R² = 0.80, RMSE = 0.078, MAE = 0.060, Pearson correlation = 0.89). The health system features most strongly associated with lower MIR were gross domestic product per capita, the universal health coverage (UHC) service coverage index, and radiotherapy infrastructure. Country-level SHAP decompositions demonstrated heterogeneous feature contributions across settings, identifying context-specific priorities. Greater radiotherapy capacity and more comprehensive UHC were consistently associated with lower MIR, whereas higher aggregate health expenditure alone showed weaker associations with improved outcomes.
CONCLUSIONS: Interpretable machine learning identified radiotherapy infrastructure, universal health coverage, and national wealth as the health system features most consistently associated with lower prostate cancer mortality-to-incidence ratios across countries. These findings suggest that targeted investment in treatment capacity and service coverage may have greater relevance to prostate cancer outcomes than increases in aggregate health spending alone. Although ecological and hypothesis-generating, this framework may inform future health systems research, prospective evaluation, and global prostate cancer control.
METHODS: We applied interpretable machine learning using CatBoost with SHAP (SHapley Additive exPlanations) to all 185 countries with prostate cancer MIR estimates in GLOBOCAN 2022. National age-standardized MIR estimates were linked to 11 health system indicators from the World Health Organization Global Health Observatory, World Bank World Development Indicators, United Nations agencies, and the Directory of Radiotherapy Centres. Model performance was evaluated using repeated leave-one-country-out cross-validation with bootstrap-based uncertainty estimation. CatBoost handled missing predictor values natively, allowing inclusion of all countries without imputation.
RESULTS: The model demonstrated strong predictive performance for prostate cancer MIR (R² = 0.80, RMSE = 0.078, MAE = 0.060, Pearson correlation = 0.89). The health system features most strongly associated with lower MIR were gross domestic product per capita, the universal health coverage (UHC) service coverage index, and radiotherapy infrastructure. Country-level SHAP decompositions demonstrated heterogeneous feature contributions across settings, identifying context-specific priorities. Greater radiotherapy capacity and more comprehensive UHC were consistently associated with lower MIR, whereas higher aggregate health expenditure alone showed weaker associations with improved outcomes.
CONCLUSIONS: Interpretable machine learning identified radiotherapy infrastructure, universal health coverage, and national wealth as the health system features most consistently associated with lower prostate cancer mortality-to-incidence ratios across countries. These findings suggest that targeted investment in treatment capacity and service coverage may have greater relevance to prostate cancer outcomes than increases in aggregate health spending alone. Although ecological and hypothesis-generating, this framework may inform future health systems research, prospective evaluation, and global prostate cancer control.