Modeling Area-Level Health Rankings

Area of Science:

• Epidemiology
• Biostatistics
• Health Services Research

Background:

• Existing county health rankings (CHRs) rely on assigned weights for mortality and morbidity variables.
• There is a need to quantify uncertainty in CHRs, considering factors like population size, spatial correlation, and missing data.

Purpose of the Study:

• To develop and apply a Bayesian factor analysis model for ranking county health.
• To improve upon existing CHRs by using data-derived weights and quantifying uncertainty.

Main Methods:

• Utilized secondary county data from the National Center for Health Statistics and Behavioral Risk Factor Surveillance System.
• Employed a Bayesian factor analysis model incorporating data-derived weights and accounting for population, spatial correlation, and missing data.
• Applied the model to Wisconsin (comprehensive data) and Texas (substantial missing data) for comparative analysis.

Main Results:

• Rankings generated by the Bayesian model showed higher correlation with existing CHRs in Wisconsin (0.89) than in Texas (0.65).
• Data-derived factor weights aligned better with assigned weights in Wisconsin compared to Texas.
• Significant uncertainty in rankings was observed for Texas due to extensive missing data.

Conclusions:

• The reliability of CHRs is state-dependent, influenced by data completeness and weighting methods.
• It is recommended to focus on counties consistently ranked as least healthy across different weighting methods and uncertainty analyses.
• The study underscores the critical need for improved geographic coverage and completeness of health data.