Purpose: Analyses of resource use in economic evaluations of multinational clinical trials typically do not account for the underlying hierarchical structure of the data. Given that medical practice patterns can differ substantially across countries, use of a statistical modeling technique, such as multilevel modeling, that accounts for such variations, may provide more precise estimates of treatment effects. In addition, multilevel models can also provide a means to estimate and compare country-level findings. The aim of our analysis was to compare the findings from three regression models in the evaluation of a treatment effect on hospitalizations using data from a large, multinational trial of patients with cardiovascular disease. Methods: The data include information on hospitalizations from over 5000 patients in 16 countries. The number of hospitalizations that occurred over the follow-up period for each patient was modeled as a function of treatment group, age, gender and race. Estimates of the effect of treatment group on hospitalization were compared using three models. The first was a Poisson regression model that did not take into account the fact that patients were nested within countries. The second model was a random intercept Poisson regression model which allowed the intercept to vary across countries. The third was a random slope Poisson regression model in which the treatment effect was allowed to vary across countries. Results: The results demonstrate that the estimates of treatment effect are sensitive to the method used. The estimates of treatment effect for the Poisson regression and random intercept model were similar (b= -.088, p=.06 and b=-.075 ,p=.05, respectively). However, the effect of treatment on hospitalization was about two and half times larger when the effect of treatment was allowed to vary across countries (b=-.212, p<.01). Conclusions: Analysts comparing resource use between treatment groups with data from multinational trials should strongly consider the use of statistical models that reflect practice pattern variations across countries. Omitting country-level variation could lead to biased estimates of the effects of treatment and lead to misinformed local decision making. Future work should explore whether characteristics of countries involved in multinational trials systematically influence the direction and magnitude of bias with non-hierarchical modeling approaches.