OBJECTIVES: Most cost-effectiveness analyses conducted alongside multinational randomized controlled clinical trials (RCTs) are carried out applying the unit costs from the country of interest to each resource item with the objective of estimating total healthcare by treatment group. An alternative is to model healthcare resource use (HCRU) directly rather than expressed in monetary units. This study aimed to model HCRU collected alongside RCTs, accounting for their specific distributions and the hierarchical structure of the data. METHODS: The analysis was conducted using data from multinational RCTs enrolling approximately 2000 patients suffering from a chronic disease. For each HCRU, appropriate distribution functions were identified based on the deviance of the univariate model (including treatment effect only). Standard models were extended to the Bayesian multi-level models (MLM) settings, whereby covariates at different levels (patient, centre and country) were introduced as predictors. RESULTS: Depending on the treatment group, 69% to 71% of patients had no GP visits. The Poisson distribution under-estimated the proportion of zeros by 18%, whereas the negative binomial (NB) and zero-inflated Poisson (ZIP) provided good matches. The greater flexibility of ZIP models provided significantly better fit than NB. ZIP was the best distribution to model healthcare resource contacts and the zero inflated Poisson overdispersed (ZIPO) function was best representing concomitant medications treatment days. GP visits presented the highest heterogeneity between countries (9% of the variance was explained by the country effect) and this was well captured by the MLMs. CONCLUSION: Misspecification of statistical models may result in biased parameters and misleading inference. This study proposed the development of ZIP and ZIPO MLMs to model HCRU alongside RCTs. To obtain more precise estimates, multivariate analyses of HCRU could be conducted and other sources of evidence could be used additionally, external to the clinical studies.