OBJECTIVE: The CODE-2 study provided costs for patients with diabetes mellitus type 2 in Germany. The aim of this analysis is to assess the accuracy of these estimates. METHODS: The German arm of CODE-2 was based on a representative sample of 809 diabetic patients. Treatment strategies, resource use and costs were investigated for the year 1998. Patients were grouped in 5 strata based on their complication status, which was assumed to be the main cost factor. Rare „cost driver“ groups were over-represented in order to obtain higher accuracy of estimates. For extrapolation on a population level, the results were weighted according to real prevalence data on complications estimated in a pre-study. Results were calculated as a weighted average from the mean in each stratum. Because the degree of precision of these estimates was not measurable with the usual statistical methods, bootstrap estimates were computed for lower and upper bounds of 95% confidence intervals taking 10,000 independent bootstrap samples. RESULTS: Bootstrap estimates were stable after 10,000 replicate samples. Average Bootstrapping 95% confidence interval Total costs per patient and year [DM] 5,539 5,184 - 5,894 Total costs for Germany [DM] 21 billion 19,5 -22,5 billion HbA1c 7.84 7.63 - 8.07 Treatment type [%] Diet OAD Insulin 19.4 53.0 27.6 16.7 - 22.1 49.6 - 56.4 24.5 - 30.7 CONCLUSION: Bootstrap confidence intervals show a remarkable accuracy of estimations performed in this study. So combination of a „weighted stratification by cost factor“ followed by bootstrapping estimation is an appropriate method for analysing the average of highly variable parameters such as “costs of diabetes“.