Mechanistic Modeling of Social Conditions in Disease-Prediction Simulations via Copula-Informed Probabilistic Graphical Models: HIV Case Study

Amir Khosheghbal, Peter Haas, Chaitra Gopalappa

Keywords: Graphical models, Machine learning, Copula, Social determinants of health, Simulation, Infectious diseases

TL;DR: Copula-informed probabilistic graphical modeling to estimate multivariate joint distributions from univariate marginals.

Abstract:

This work has already been published; here, we provide a brief overview [1]. Epidemic models typically simulate the spread of diseases as functions of behaviors, e.g., sexual and care behaviors for sexually transmitted diseases. However, multilevel factors, including poverty, housing or food insecurity, mental health, substance use disorder, etc., which are called social determinants of health (SDH), are drivers of those behaviors. There is increasing awareness of the need to incorporate SDH into epidemic simulation models to evaluate structural interventions alongside behavioral interventions. However, the multivariate joint associations between SDH and behaviors needed for modeling are not available. Data for SDH are mostly available as county-level marginal distributions, and associations between SDH and behaviors are mostly bivariate. To address this problem, we combine copula theory and probabilistic graphical models to estimate multivariate joint distributions. We estimate bivariate associations between SDH using a novel copula approach that transitions from continuous to discrete copulas. We then use these bivariate associations—together with bivariate associations between SDH and behaviors from the literature—as links in an undirected graphical model to calculate the multivariate joint distributions. As a case study, we used the joint distributions to model HIV-risk related behaviors as function of SDH in a national-level HIV/AIDS (PATH 4.0) model and studied the impact of hypothetical 100% efficacious SDH interventions on HIV prevention. We found that this intervention could lead to a cumulative 10-year reduction of 29% in HIV incidence.

[1] Khosheghbal, A., Haas, P.J. & Gopalappa, C. Mechanistic modeling of social conditions in disease-prediction simulations via copulas and probabilistic graphical models: HIV case study. Health Care Manag Sci 28, 28–49 (2025). https://doi.org/10.1007/s10729-024-09694-3

Submission Number: 133