Reliability of administrative data for the identification of Parkinson's disease cohorts

Parkinson's disease (PD) is a major worldwide public health problem with a prevalence that is expected to increase dramatically in the coming decades. Because administrative data are useful for epidemiologic and health service studies, we aimed to define procedural algorithms to identify PD patients (on a regional basis) using these data. We built two a priori algorithms, respecting privacy laws, with increasing theoretical specificity for PD including: (1) a hospital discharge diagnosis of PD; (2) PD-specific exemption; (3) a minimum of two separate prescriptions of an antiparkinsonian drug. The two algorithms differed for drugs included. Sensitivities were tested on an opportunistic sample of 319 PD patients from the databases of 5 regional movement disorders clinics. The estimated prevalence of PD in the sample population from Tuscany was 0.49 % for algorithm 1 and 0.28 % for algorithm 2. Algorithm 1 correctly identified 291 PD patients (sensitivity 91.2 %), and algorithm 2 identified 242 PD patients (sensitivity 75.9 %). We developed two reproducible algorithms demonstrating increasing theoretical specificity with good sensitivity in identifying PD patients based on an evaluation of administrative data. This may represent a low-cost strategy to reliably follow up a large number of PD patients as a whole for evaluating the effects of therapies, disease progression and prevalence.