Record Linkage of secondary data in the NAKO: Overview of data sources with focus on health insurance claims data

Abstract

Abstract Background The collected primary data are enriched with secondary and registry data from different data owners to support morbidity follow-up and to obtain important exposure and progression data that could not be obtained otherwise. Goal A sophisticated data protection procedure had to be developed and agreed upon with hundreds of data owners. The outlines of this procedure and the various data bodies to be linked are presented, and consent behavior is described. Furthermore, it is discussed which hurdles still have to be overcome for a successful implementation of record linkage in Germany. Methods Study participants were asked for informed written consent to link their secondary and registry data. Confidentiality was ensured, among other things, by means of a complex pseudonymization procedure by an independent trust center. Legal and technical requirements for annual transmission and long-term use of their data were established with the data owners. Results At baseline, a total of 205,000 study participants were screened and interviewed in 18 study centers distributed across Germany starting in 2014. A second study wave followed from 2019, with a third starting in 2024. A high willingness to consent of 90.5% to use health insurance data was achieved. Willingness to consent is higher in the west than in the east of Germany, with 88.1% of consenters insured by one of the 113 statutory health insurance providers and 11.9% insured by one of the 46 private health insurance providers. Conclusions Large-scale record linkage of secondary data is a novelty in Germany. In addition to the identification of incident diseases, an almost complete medication history can be collected for the majority of study participants, independent of health insurance. However, linking is associated with a high administrative and technical effort, and some of the required data protection regulations will have a negative impact on data quality if improvements are not made here.