An unspecified multilevel model with a random treatment effect was used to examine treatment effectiveness after adjusting for baseline predictors (study centre, country, treatment, age, gender, height, weight, body mass index, diarrhoea duration before inclusion, body temperature, dehydration level, rotavirus presence, number of stools in the 24 and four hours prior to treatment) which were selected using a backwards stepwise strategy. Hazard ratios for diarrhoea duration were synthesised using Cox Proportional Hazards models stratified by study.