Methods for Estimating Adjusted Risk Ratios

Abstract

The risk ratio can be a useful statistic for summarizing the results of cross-sectional, cohort, and randomized trial studies. I discuss several methods for estimating adjusted risk ratios and show how they can be executed in Stata, including 1) Mantel–Haenszel and inverse-variance stratified methods; 2) generalized linear regression with a log link and binomial distribution; 3) generalized linear regression with a log link, normal distribution, and robust variance estimator; 4) Poisson regression with a robust variance estimator; 5) Cox proportional hazards regression with a robust variance estimator; 6) standardized risk ratios from logistic, probit, complementary log-log, and log-log regression; and 7) a substitution method. Advantages and drawbacks are noted for some methods.