Epidemiological regressions for estimating risk ratios and risk differences with the R package rqlm: A practical guide

抄録

Risk ratios and risk differences express relative and absolute differences in the probability of a binary outcome. They are often useful in epidemiological and clinical studies, but obtaining adjusted estimates with appropriate confidence intervals can require several programming steps. The R package rqlm brings these steps together through modified Poisson regression, modified least-squares regression, and modified logistic regression using pseudo-observations. This tutorial is written for researchers who can specify an ordinary regression model and want to apply these methods without working through their mathematical derivations. We explain how to choose an effect measure, prepare data, fit the models, read their output, and save results. Complete example commands appear alongside the explanations. The worked examples use two fully synthetic datasets based on a target trial emulation comparing initiation of an SGLT2 inhibitor with initiation of a DPP-4 inhibitor in older adults with type 2 diabetes. Strong baseline prognostic differences illustrate why an unadjusted comparison can be misleading. We then explain the practical purpose and limitations of small-sample variance corrections and the Hadamard option for modified least-squares regression. An exactly unbiased variance estimator does not provide an exact 95% confidence interval. The main workflow uses standard robust inference, with selected additional options explained separately.