OBJECTIVE: Prediction of future need for healthcare is an essential component of pharmacoeconomic models. However, recent developments in applied econometrics have demonstrated the importance of investigating for heterogeneity of risk within populations when using count data (e.g. hospitalisation). We tested the hypothesis that there is a healthy sub-population with very low risk of hospitalisation amongst patients with Type 2 diabetes. METHODS: The study population comprised 4625 type 2 diabetic patients diagnosed before 1st Jan. 1993 and still alive until 31 Dec. 1995 in Tayside, Scotland. Number of hospitalisations for each patient was calculated using a record linkage database. Risk factors age, gender and previous hospitalisation for cardiovascular diseases were obtained from the databases. We compared Poisson, two-part Poisson, Negative binomial (NB) and two-part NB models. The two-part models include a logistic regression model as the first part and the truncated Poisson or NB models as the second part. The maximum likelihood procedure was used to fit the models. The likelihood ratio test and the Akaikes Information Criterion were used for model comparison. RESULTS: The best model was the two-part NB model. The risk factors associated with at least one hospitalisation were age (log-OR=0.0245, S.E.=0.0026) and previous hospitalisation for MI (log-OR=0.5960, S.E.=0.1137), Stroke (log-OR=0.7181, S.E.=0.1915) and other cardiovascular diseases (log-OR=0.7488, S.E.=0.0912). The overdispersion parameters for the NB model and the two-part NB model were 0.3929 and 1.3796 respectively, indicating the necessity of applying a two part model. CONCLUSION: Our data support the hypothesis that there is individual heterogeneity for risk of hospitalisation and a “healthy” sub-population of patients with Type 2 diabetes. Consequently we recommend application of two-part models that first predicts the risk of any hospitalisation for each individual in the population and then predicts the likely number of hospitalisations separately for high and low risk sub-populations.