Predicting regional ischemic heart disease mortality in Estonia with data science tools

Abstract Ischemic heart disease (IHD) remains one of the leading causes of death in Estonia, especially among older adults and disadvantaged socioeconomic groups. While descriptive statistics inform retrospective trends, health authorities increasingly require predictive tools for more proactive planning. This project proposes a data science approach to forecasting IHD mortality, aiming to identify high-risk populations and support targeted public health interventions. We propose to develop a machine learning model trained on national health data to predict IHD-related mortality by region, sex, and age group over a 3-5-year horizon. Data will be sourced from the Estonian Health Insurance Fund, Statistics Estonia, the National Institute for Health Development, and international sources such as WHO and Eurostat. The modeling pipeline will include: data preprocessing (handling ICD-10/ICD-11 inconsistencies, missing values, and ill-defined deaths), exploratory analysis, and training using logistic regression, random forests, and gradient boosting methods. Fairness audits and explainability tools (e.g., SHAP) will ensure interpretability and equity. Ethical safeguards aligned with GDPR will be integrated throughout. Although results are forthcoming, we anticipate that this approach will enable dynamic risk stratification and scenario testing across Estonian regions. The model will be designed to update with new data and disaggregate outputs by subgroup characteristics, supporting equity-focused health policy decisions. This project introduces a scalable, data-driven method for predicting IHD mortality that bridges epidemiology and data science. By forecasting where and in whom mortality risks are highest, the model aims to help regional health authorities in Estonia allocate resources more effectively and intervene earlier. The framework may serve as a template for similar predictive analytics in other disease areas or countries. Key messages • Predictive modeling can support more timely and targeted public health responses to ischemic heart disease in Estonia. • Transparency, fairness audits, and stakeholder engagement are key to ensuring trust and utility in health-related machine learning applications.