[Using log-binomial model for estimating the prevalence ratio].

Area of Science:

• Biostatistics
• Epidemiology
• Statistical Modeling

Context:

• Assessing the association between smoking status and attitudes toward smoking-ban legislation is crucial for public health policy.
• Traditional logistic regression models estimate odds ratios, which can overestimate prevalence when the outcome is common.
• Log-binomial models provide direct estimation of prevalence ratios, offering a more accurate measure of association in such scenarios.

Purpose:

• To compare prevalence ratios estimated by log-binomial models with odds ratios from logistic regression for analyzing attitudes towards smoking-ban legislation.
• To evaluate the performance of log-binomial models, including methods for handling convergence issues with continuous covariates.
• To provide SAS programs for calculating prevalence ratios in regression analyses.

Summary:

• The study utilized log-binomial models to estimate prevalence ratios and compared them with odds ratios from logistic regression for individuals' attitudes towards smoking-ban legislation.
• Results indicated that prevalence ratios provided a more accurate measure of association than odds ratios, especially in men where smoking prevalence was higher.
• The log-binomial model required the COPY method for convergence when age was included as a continuous covariate.

Impact:

• Demonstrates that prevalence ratios are a more appropriate measure than odds ratios for high-prevalence outcomes in epidemiological studies.
• Highlights the utility of log-binomial regression for analyzing public health issues like smoking-ban legislation.
• Provides practical guidance and tools (SAS programs) for researchers to implement log-binomial models effectively.