Interne Validierung von Diagnosen in GKV-Routinedaten: Konzeption mit Beispielen und Falldefinition

Abstract

Over the course of the last few decades, statutory health insurance data have become increasingly important for health services research. Of particular interest in this context are diagnoses. Since all health insurance data are originally collected for billing purposes, secondary analyses should examine the completeness, plausibility, and validity of the information provided. While an external validation through, for example, a comparison with the physician's records or a second independent medical examination can be seen as a gold standard, this is often not feasible. For this reason, internal validation approaches are recommended for studies based upon diagnoses drawn from routine data. For such approaches, no established standards are currently available. The aim of this contribution is to introduce a generic internal validation concept for chronic diseases. Data employed in the present contribution stem from the health insuree sample of the AOK health insurance fund Hesse. Criteria for assessing the validity of diagnoses (e.g., repetitions, codes assigned by various physicians, prescriptions) are presented for three chronic diseases - heart failure, dementia, and tuberculosis. Building upon these criteria, algorithms for the definition of epidemiologically certain cases are developed and prevalence estimates formed on the basis of these algorithms are compared with other data sources (registers and surveys). Internal confirmation of the diagnoses of heart failure and dementia was possible in 97% and 80% of cases, respectively. The difference between the two diagnoses is due to the low rate of treatment with specific pharmaceuticals in the case of dementia. Prevalence estimates are comparable with those based on other sources. Inpatient discharge diagnoses of tuberculosis were internally confirmed in 100% and outpatient diagnoses in 40% of cases. For this reason, outpatient diagnoses were not considered for the case definition of tuberculosis. A comparison with tuberculosis surveillance data reveals a somewhat higher incidence in the insuree sample. In selecting and weighting criteria as well as employing a case definition, the research aim of the respective investigation must be taken into account. The adopted procedure is to be presented in a transparent manner.

Telephone-based health coaching for chronically ill patients: study protocol for a randomized controlled trial

“…Although we apply advanced statistical methods, potential confounding cannot be ruled out completely. Another limitation could be the unsure validity of the routine data, but an extensive literature research revealed that the quality of routine data in Germany is good [29]. The generalizability of the secondary patient reported outcomes may be limited by social desirability, because the statutory health insurance is sending out the questionnaires.…”

“…Routine data depicts accountable variables, but not clinical parameters. The validity of routine data in Germany is not proven, yet studies show that the ICD diagnoses are valid in about 97% for heart failure [29,30]. The routine data are read out by the KKH and securely delivered with pseudonymized codes.…”

Influenza und ambulant erworbene Pneumonie in hausärztlich tätigen Arztpraxen in Deutschland

“…Weiterhin können die im ambulanten Bereich aufgezeichneten ICD-10-Diagnosen nur begrenzt validiert werden [40][41][42]. Die Validität der kodierten Diagnosen hängt vom Kodierungsverhalten ab, was wiederum von der Erfahrung des/der kodierenden Arztes/Ärztin, von dem Krankheitsbild des einzelnen Patienten und nicht zuletzt von externen Faktoren wie Leistungsabrechnung oder Algorithmen in der Praxissoftware beeinflusst wird [43].…”

Health Among the Elderly in Germany. New Evidence on Disease, Disability and Care Need

“…Because an external validation of the diagnoses is not possible, an internal validation is needed Schubert et al 2010;Swart and Ihle 2005). Nevertheless, routine claims data are a suitable data source for the analysis of dementia, as previous studies have shown Schubert et al 2010;Schulz and Doblhammer 2012;. A diagnosis-free period of at least two years was chosen to distinguish prevalent cases from incident cases.…”