Bootstrap confidence intervals for costs-of-illness of type 2 diabetes mellitus in Germany

Abstract

Objective: The Costs of Diabetes in Europe-Type 2 study (CODE-2®, SmithKline Beecham plc) measures costs of managing patients with type 2 diabetes mellitus in Germany. The aim of this analysis was to assess the uncertainty of these estimates. Design and Setting: The German study arm was based on a sample of 809 patients with type 2 diabetes registered in general practices. Information on socioeconomic data, medical resource use, and clinical data was collected retrospectively for 1998. Patients and Participants: Patients were grouped in five strata based on their complication status, because of the high impact of complications on costs. To obtain higher credibility of resulting estimates, rare complication groups were overrepresented. To be representative, results were weighted using real prevalence data on complications from a prestudy. Main Outcome Measures and Results: Within each stratum, results were calculated as arithmetic mean except for demographic data, where the median was applied as input for weighted averages. Because the degree of precision of calculated estimates was not accessible analytically, 95% confidence intervals (Cls) were computed via bootstrapping of 10,000 independent bootstrap samples for each of the calculated estimates. All costs are given for the payers' perspective in German Deutsche Mark (DM). Costs per patient and year for ambulatory care were DM 775 with 95% Cl (721-835), for hospitalizations DM 2771 (2242-3342), for drug treatment DM 1496 (1399-1598), and for rehabilitation DM 120 (70-177). The indirect cost was DM 372 (144-645). From the perspective of the sickness funds, cost per patient and year was DM 5539 (5184-5894). Mean HbA1c status was 7.51% (7.37-7.66) with the majority of patients not achieving glycemic control below 6.5%. Conclusion: Bootstrap Cls are remarkably narrow. Combining a weighted stratification with bootstrap estimation is an appropriate method for analyzing the weighted average of highly variable and skewed parameters such as costs of diabetes.