What is the statistical outcome typically used in cross-sectional studies, and how is it analyzed?

• A. Incidence rate ratio analyzed with Cox regression.
• B. Prevalence ratio (or odds ratio if logistic regression) estimated by Poisson or log-binomial models.
• C. Hazard ratio estimated with Kaplan-Meier.
• D. Risk difference estimated with linear regression.

Answer: (B)

In cross-sectional studies, you’re capturing prevalence at a single point in time, so the familiar measure to compare groups is the prevalence ratio—the ratio of the outcome’s prevalence in one group to that in another. To estimate this, you typically use models that yield a prevalence ratio directly: Poisson regression with robust standard errors or log-binomial regression. These approaches model the binary outcome with a log link to produce a PR, which is easy to interpret as “how many times more (or less) common the outcome is in one group versus the other.”

Logistic regression can also be used, but it estimates an odds ratio, which can misrepresent the strength of association when the outcome is not rare. The odds ratio tends to overstate the association compared with the prevalence ratio in common outcomes, so PR is preferred for cross-sectional prevalence comparisons when possible.

Other options don’t fit as well because they rely on time-to-event data or incidence over time (hazard ratios from Cox models or incidence rate ratios requiring person-time), which aren’t inherent to a single-point cross-sectional snapshot. A linear risk difference model is less commonly used for prevalence with binary outcomes, though it can be applied in some contexts but doesn’t provide the same straightforward interpretable ratio as PR.