Parkinson's disease in real life healthcare organization database: a medication-based algorithm.
Parkinson's disease in real life healthcare organization database: a medication-based algorithm.
Where did the research take place?
The study site has not been established. Author addresses may differ from where the research occurred.
IL · Author affiliation · country only
Data Research Center for Mental Health and Rehabilitation, Clalit Health Services, Petach Tikva, Israel. hila444@gmail.com.Location evidence
Tel Aviv, IL · Author affiliation
Neurological institute, division of movement disorders, Tel-Aviv Souraski Medical center, Tel Aviv, Israel. hila444@gmail.com.Location evidence
New York City, US · Author affiliation
Department of Neurology, Columbia University Irving Medical Center, New York, NY, USA.Location evidence
A plain-language reading has not been prepared for this paper yet.
Original abstract
BACKGROUND: Accurate identification of Parkinson’s disease (PD) in large electronic health record (EHR) population-based databases is challenging due to diagnostic heterogeneity in routine care, with a substantial proportion of individuals diagnosed with PD had not been diagnosed by a specialist. Our aim was to develop and validate a simplified rule-based medication algorithm to identify PD in a nationwide healthcare registry and apply it to estimate long-term incidence, prevalence, and pre-diagnostic diagnoses. METHODS: Using Clalit Health Services EHR data covering over five million individuals (2005–2025), we constructed a medication-based algorithm incorporating predefined inclusion and exclusion criteria and two levels of diagnostic certainty (probable/possible PD). Validation was performed against two independent specialist-confirmed PD cohorts and FDOPA PET/CT and a non-PD neurological cohort. Incidence rates per 100,000 were calculated annually with 95% confidence intervals (CIs) assuming a Poisson distribution. Age-adjusted incidence rates were computed using the WHO standard population. motor and non-motor diagnoses preceding PD were examined up to 18 years before the index date using matched controls. RESULTS: The algorithm identified 34,368 PD patients (56.5% male; mean age at index 75.2 ± 10.5 years). Sensitivity was 94.8% (95% CI 90.4–97.2) in the FDOPA PET/CT cohort, 94.8% (95% CI 92.1–96.6) in the private clinic cohort, and 94.7% (95% CI 90.9–96.9) in the movement disorder clinic cohort. Specificity was 85.2% (95% CI 77.8–90.6). Incidence increased markedly with age but declined significantly over time (overall annual percent change [APC] - 4.47%, 95% CI -4.90 – -4.03). Age-adjusted incidence rates (≥20 years) declined 2.4-fold between 2005 and 2024 (55 [95% CI 53–58] to 23 [95% CI 21–24] per 100,000). Overall prevalence declined modestly (APC -0.78%, 95% CI -0.84 – -0.72), with increases in younger age groups and declines in older groups. Constipation, depression, and tremor diagnoses were more frequent years before diagnosis, whereas smoking-related codes were less frequent among future PD patients. CONCLUSIONS: This validated medication-based algorithm provides a reproducible framework for PD identification in large registries. Applied over two decades in a nationwide cohort, it demonstrated high diagnostic performance and revealed age-dependent declines in PD incidence alongside heterogeneous prevalence trends.