Using Bayesian networks with Max-Min Hill-Climbing algorithm to detect factors related to multimorbidity

Area of Science:

• Gerontology
• Epidemiology
• Biostatistics

Background:

• Multimorbidity (MMD) is prevalent, associated with adverse health outcomes and high medical costs.
• Understanding the complex relationships between MMD and its determinants is crucial for effective health management.

Purpose of the Study:

• To construct Bayesian networks (BNs) using the Max-Min Hill-Climbing (MMHC) algorithm to explore network relationships of MMD.
• To compare the performance of BNs with traditional multivariate logistic regression for MMD analysis.

Main Methods:

• Utilized data from the CHARLS 2018 database, including demographic, health status, functioning, and lifestyle variables.
• Employed Random Forest for missing value imputation, followed by logistic regression and BNs model construction.
• BNs structural learning used MMHC algorithm; parameter learning used maximum likelihood estimation.

Main Results:

• Analysis of 19,752 individuals showed 53.8% had MMD.
• Logistic regression identified physical activity, sex, age, sleep duration, napping, smoking, and alcohol consumption associated with MMD.
• BNs revealed direct links between MMD and age, sleep duration, and physical activity, with indirect links via sleep duration for education and residence.

Conclusions:

• BNs offer a superior graphical representation of complex MMD-related factor networks compared to logistic regression.
• BNs facilitate risk reasoning for MMD through Bayesian inference, aligning with clinical practice and showing application potential.