Accuracy and sensitivity of Parkinsonian disorder diagnoses in two Swedish national health registers.

BACKGROUND Swedish population-based national health registers are widely used data sources in epidemiological research. Register-based diagnoses of Parkinson's disease have not been validated against clinical information. METHODS Parkinson's disease (PD) and other parkinsonian disorder diagnoses were ascertained in two registers, i.e. the National Patient Register (NPR) and the Cause of Death Register (CDR). Diagnoses were validated in terms of accuracy (positive predictive value) and sensitivity against data from a population-based study of PD in 1998-2004 that screened more than 35,000 persons and identified 194 cases of parkinsonian disorders including 132 PD cases (the gold standard for the purposes of this study). RESULTS Accuracy for any parkinsonian disorder diagnoses was 88.0% in the NPR and 94.4% in the CDR. Accuracy of PD diagnoses was 70.8% in the NPR and 66.7% in the CDR. Misclassification between differential parkinsonian diagnoses was common. The accuracy of PD diagnoses in the NPR improved to 83.0% by restricting the definition to primary diagnoses only. The sensitivity of PD diagnoses in the NPR and CDR combined was 83.1%, with a mean time to detection of 6.9 years. CONCLUSIONS Population-based national health registers are valid data sources in epidemiological studies of PD or parkinsonian disorder etiology but are less suitable in studies of incidence or prevalence.