[Methodological Challenges when Using Claims Data of more than 70 Statutory Health Insurances - A Progress Report...

The adequate and need-based medical care of mentally ill patients places special demands on psychiatric care. The §64b Social Code Book (SGB) V enables mentoring mentally ill people through multiprofessional, cross-sectoral model projects across the treatment phase and implementing new forms of financing. These model projects have been evaluated in a prospective and retrospective claims data-based controlled cohort study (EVA64) since 2015.

In September 2016 and since then annually, the data transfer of all statutory health insurance funds (SHI) involved in this evaluation took place for the first time on the basis of a consented data set description. For later analysis, the clear identification of the index hospital admission and the assignment to the model or control group are important. The methodological challenges of data provision by the data owner, the formal and content-related data preparation as well as the subsequent establishing of an evaluation data set are discussed in detail.

So far, data from 71 SHI has been taken into account. In each case 20 tables with claims data from outpatient and inpatient care (including psychiatric institute outpatient departments [PIA]), drug and medical supplies as well as data from incapacity to work and personal data of the insurees. Not all tables could be filled completely by the SHIs. In addition, updates of the study designs require the adaptation of the data selection process. Even though data sets have been delievered regularly the data preparation process is still not routine.

The scientific use of claims data of numerous SHIs in the context of an evaluation study represents a great challenge. In the absence of reference values for abnormalities and implausibilities, an a priori determination of test algorithms was limited; instead they had to be updated every year. The individual examination of the data of all health insurance companies remains very complex. The detailed documentation of these algorithms provides support for future comparable studies.