Biostatistics of generalized estimating equations in developmental medicine and child neurology

This review provides clinicians and researchers in developmental medicine and paediatric neurology with a guide to using generalized estimating equations (GEEs) for longitudinal paediatric data. We present a concise primer on core GEE concepts for non-statistical audiences, emphasizing paediatric applications. Using a randomized trial of oestrogen versus placebo for postnatal depression, we provide a reproducible workflow (in R code) for continuous and binary outcomes. We compare exchangeable and autoregressive (first-order autoregressive model) working correlations and discuss implications for efficiency and interpretation. Because the data set is maternal and contains no child outcomes, we treat it as a perinatal case study relevant to child development and use it to illustrate marginal (population-averaged) inference in longitudinal clinical data. GEEs yielded stable marginal estimates across correlation structures when the mean model was correctly specified. Oestrogen was associated with significantly lower odds of postnatal depression than placebo, with negligible differences in model fit (correlation information criterion). Statistical choices mainly affected efficiency and standard errors rather than effect sizes. GEEs provide a robust, interpretable framework for analysing correlated outcomes in paediatric research. Paired with a reproducible example, this helps clinicians and researchers select appropriate models, report working correlations transparently, and interpret marginal effects in practice.