Which modeling approach can estimate prevalence ratio in cross-sectional data when using regression?

• A. Hazard ratio
• B. Prevalence ratio
• C. Odds ratio
• D. Incidence rate ratio

Answer: (B)

In cross-sectional data the outcome is presence or absence at one point in time, so the measure you want to compare between exposure groups is the prevalence ratio. To estimate this with regression, fit a model that directly yields a ratio of probabilities. A log-binomial model does this by modeling P(Y=1|X) with a log link, so the exponentiated coefficient for exposure equals the prevalence ratio. If that model has convergence issues, a Poisson regression with a log link and robust standard errors provides a reliable approximation to the same prevalence ratio. This approach is preferred over hazard ratio (time-to-event), incidence rate ratio (incidence over person-time), or logistic regression (which gives an odds ratio, not a prevalence ratio, and can mislead when the outcome is common).