Interpretable AI for Understanding Social Determinants of Health in Diseases of Despair

Research Poster Engineering 2025 Graduate Exhibition

Presentation by Haya Alshayji

Exhibition Number 190

Abstract

Social Determinants of Health (SDoH) are environmental and socioeconomic factors that are increasingly associated with diseases of despair (DoD), such as substance use disorders, alcohol-related ailments, and suicide. Traditional risk models, however, frequently lack transparency, which reduces their use in clinical and political decision-making. This research employs interpretable machine learning techniques to enhance the understanding of how SDoH influences DoD outcomes, bridging the gap between complex predictive models and actionable insights. Our study integrates electronic health records and publicly available socioeconomic data to develop a risk stratification framework for DoD. We employ machine learning models, including logistic regression, XGBoost, and random forests, to predict high-risk individuals while incorporating SHAP values to explain model predictions. By highlighting key risk factors, including SHAP-based feature importance analysis, we give policymakers and physicians understandable information about DoD vulnerabilities. To ensure equity in AI-driven decision-making, we also assess potential predictions' biases across demographic groupings. Our framework improves predictive modeling transparency by displaying individual risk factors and quantifying their contributions. This provides a better understanding of the intricate relationships between DoD risk and SDoH. This approach helps build trust in AI-driven analytics by making model outputs interpretable and actionable for public health and clinical practice stakeholders. This research contributes to the growing field of ethical AI in healthcare by demonstrating how interpretable models can improve trust and applicability in predictive analytics. Future work will refine model performance and integrate real-time clinical decision support systems.

Importance

Understanding how social and environmental factors influence health is essential for tackling public health issues like Diseases of Despair (DoD). This research enhances transparency in AI-driven risk prediction by using interpretable machine learning to identify key socioeconomic drivers of DoD. Our approach breaks down complex models into simpler terms, making it easier for policymakers and healthcare professionals to make informed decisions that promote fairness in health interventions. This work is very important in assuring responsible use of artificial intelligence in healthcare by making sure predictive analytics enhance risk assessment and assist ethical, fair, and pragmatic public health projects.

DEI Statement

This research advances health equity by using interpretable AI to identify and address disparities in Diseases of Despair (DoD), which disproportionately affect marginalized communities. By analyzing Social Determinants of Health (SDoH), we highlight systemic risk factors contributing to substance use disorders, suicide, and alcohol-related conditions. Our method guarantees openness in predictions powered by artificial intelligence, therefore lowering prejudices and enhancing fair public health decision-making. This work promotes fairness in healthcare analytics, empowering policymakers and clinicians with actionable insights to design targeted, data-driven interventions that address health disparities and improve outcomes for underserved populations.