A Bayesian Mixture Model Approach to Examining Neighborhood Social Determinants of Health Disparities in Endometrial Cancer Care in Massachusetts

TL;DR

This study develops a Bayesian mixture model to classify neighborhoods based on social determinants of health and examines their impact on endometrial cancer care in Massachusetts, revealing disparities linked to neighborhood profiles.

Contribution

Introduces a multidimensional Bayesian clustering approach to characterize neighborhood social determinants and assess their influence on cancer care disparities.

Findings

Five distinct neighborhood SDoH profiles identified.

Profiles were not significantly associated with care quality.

Patients in non-advantaged profiles had lower odds of optimal care.

Abstract

Many studies have examined social determinants of health (SDoH) independently, overlooking their interconnected nature. Our study uses a multidimensional approach to construct a neighborhood-level measure that explores how multiple SDoH jointly impact care received for endometrial cancer (EC) patients in Massachusetts (MA). Using 2015-2019 American Community Survey data, we implemented a Bayesian multivariate Bernoulli mixture model to identify neighborhoods with similar SDoH features in MA. Five neighborhood SDoH (NSDoH) profiles were derived and characterized: (1) advantaged non-Hispanic White; (2) disadvantaged racially/ethnically diverse, more renter-occupied housing with limited English proficiency; (3) working class, lower educational attainment; (4) racially/ethnically diverse and greater economic security and educational attainment; and (5) racially/ethnically diverse, more renter-occupied housing with limited English proficiency. Based on residential information, we assigned these profiles to EC patients in the Massachusetts Cancer Registry. We used these profile assignments as the primary exposure in a Bayesian logistic regression to estimate the odds of receiving optimal EC care, adjusting for patient-level sociodemographic and clinical characteristics. NSDoH profiles were not significantly associated with receiving optimal EC care. However, compared to patients assigned to Profile 1, patients in all other profiles had lower odds of receiving optimal care. Our findings demonstrate how a flexible model-based clustering approach can account for the interconnected and multidimensional nature of NSDoH in a practical and interpretable way. Deriving and geospatially mapping NSDoH profiles may allow for identifying areas of need and inform targeted public health interventions tailored to each neighborhood's specific social determinants to improve healthcare delivery.