Obtaining adjusted prevalence ratios from logistic regression models in cross-sectional studies

Area of Science:

• Epidemiology
• Biostatistics
• Statistical Modeling

Background:

• The prevalence ratio (PR) is increasingly preferred over the odds ratio for measuring association in cross-sectional studies.
• Calculating PR using traditional statistical models presents challenges, including convergence issues, tool availability, and assumption violations.

Purpose of the Study:

• To address the difficulties in estimating the prevalence ratio (PR) using statistical models.
• To implement and evaluate a direct approach for estimating PR from binary regression models.

Main Methods:

• A direct approach was implemented to estimate PR from binary regression models, utilizing methods by Wilcosky & Chambless.
• The implemented approach was compared against log-binomial regression, Poisson regression, and the prevalence odds ratio (POR) using three examples.
• Crude and adjusted PR estimates were compared across methods.

Main Results:

• The direct approach yielded PR estimates comparable to those from log-binomial and Poisson regression models.
• The prevalence odds ratio (POR) consistently overestimated the prevalence ratio (PR).
• The implemented direct approach demonstrated no numerical instability and assumed adequate probability distributions.

Conclusions:

• The direct approach offers a robust and numerically stable method for estimating the prevalence ratio (PR) in cross-sectional studies.
• This method is readily available through the R statistical package, enhancing its accessibility for researchers.
• The direct approach provides a reliable alternative to the prevalence odds ratio (POR), which tends to overestimate associations.