Accuracy of medicare claims data in identifying Parkinsonism cases: Comparison with the medicare current beneficiary survey

Study Purpose: Administrative databases are commonly used to examine use of healthcare service, with researchers relying on diagnostic codes to identify medical conditions. This study evaluates the accuracy of administrative claims in identifying Parkinsonism cases compared to the self-reported Parkinson's disease (PD). Methods: The reference cases were identified based on the self-reported PD status and the use of PD drugs collected by the 1992-2000 Medicare Current Beneficiary Survey that contained 72,922 observations from 30,469 individuals. Using ICD-9 CM, cases with PD were extracted from the corresponding Medicare claims. We compared prevalence of PD obtained using different types of claims. Results: The sensitivities were the highest when all claims were used (66%). All the specificities were greater than 99%. When drug use information was included in the gold standard, the sensitivities became lower, while the specificities and positive predictive values (PPVs) increased. Using more diagnostic codes improved the sensitivity of the identification process but reduced PPVs. Conclusions: Administrative claims can provide fairly accurate and practical approach to "rule in" patients with PD. Depending on the purpose of evaluation, researchers may consider using more categories of claims to improve the sensitivity of the identification algorithm or use fewer diagnoses to minimize number of false positive cases.