Random-effects models in investigating the effect of vitamin A in childhood diarrhea

PURPOSE: By adopting more appropriate and powerful statistical methods that fully exploit longitudinal structure, we reanalyze and extend previously published results from a large community trial to investigate the effect of vitamin A supplementation on the prevalence and severity of diarrhea in young children. METHODS: Generalized linear mixed models were used to allow for repeated measures in a reanalysis of a double-blind, randomized, placebo-controlled community trial conducted in a cohort of children in northeastern Brazil during 1 year. The response variable was weekly number of days with diarrhea for each child, and Markov Chain Monte Carlo methods were used to estimate model parameters. RESULTS AND CONCLUSIONS: Random effects suitably accounted for the underlying heterogeneity between and within children, and our longitudinal analysis shows a significant beneficial effect of vitamin A supplementation that was inconclusive in previously reported simple summary analyses of these data. Risk for diarrhea infection was estimated to be 1.57 times greater for a child administered a placebo as opposed to vitamin A (95% credible interval, 1.17-2.12). Additionally, we identified previously unreported temporal effects in these data, showing a decreasing daily probability of diarrhea for both groups during the trial and treatment-time interaction.