Diabetes mellitus is a disease of high public health relevance. To estimate the temporal development of prevalence, routine data of statutory health insurances (SHI) are being increasingly used. However, these data are primarily collected for billing purposes and the case definition of specific diseases remains challenging. In this study, we present an algorithm for differentiation of diabetes types analyzing SHI routine data. The basis for the analysis was an age and sex-stratified random sample of persons of the Barmer SHI with a continuous insurance duration from 2010 to 2018 in the magnitude of 1% of the German population. Diabetes was defined in the reporting year 2018, as documentation of (1) a "confirmed" ICD diagnosis E10.- to E14.- in at least two quarters, (2) a "confirmed" ICD diagnosis E10.- to E14.- in one quarter with an additional prescription of an antidiabetic drug (ATC codes A10), or (3) an ICD diagnosis E10.- to E14.- in the inpatient sector, outpatient surgery, or work disability. Individuals were assigned to a diabetes type based on the specific ICD diagnosis E10.- to E14.- and prescribed medications, differentiated by insulin and other antidiabetics. Still unclear or conflicting constellations were assigned on the basis of the persons' age or the frequency and observation of the diagnosis documentation over more than one year. The participation in a disease management program was considered in a sensitivity analysis. The prevalence of documented diabetes in the Barmer sample was 8.8% in 2018. Applying the algorithm, 98.5% of individuals with diabetes could be classified as having type 1 diabetes (5.5%), type 2 diabetes (92.6%), or another specific form of diabetes (0.43%). Thus, the prevalence was 0.48% for type 1 diabetes and 8.1% for type 2 diabetes in 2018. The vast majority of people with diabetes can be classified by their diabetes type on the basis of just a few characteristics, such as diagnoses, drug prescription, and age. Further studies should assess the external validity by comparing the results with primary data. The algorithm enables the analysis of important epidemiological indicators and the frequency of comorbidities based on routine data differentiated by type 1 and type 2 diabetes, which should be considered in the surveillance of diabetes in the future.
Read full abstract