Use of a medication-based algorithm to identify advanced Parkinson's disease in administrative claims data: Associations with claims-based indicators of disease severity

Introduction: Lack of a gold standard definition for advanced Parkinson's Disease (APD), coupled with absence of disease severity information in diagnostic codes, hinders use of large administrative databases for conducting population health and comparative effectiveness studies.

Methods: Using pharmacy claims data, we created an algorithm to identify APD: any 30-day average levodopa equivalent dose (LED) >1000 mg/day. Using 2013 100% U.S. Medicare claims, we applied this algorithm and used multivariate logistic regression to examine associations between assigned APD status and claims-based indicators of PD severity (any deep brain stimulation, fall, hallucinations, walker, wheelchair, specialty bed, dementia diagnosis, skilled nursing facility, hospice), adjusting for sociodemographic, clinical, and treatment characteristics. Levodopa >1000 mg/day, levodopa >800 mg/day and LED >800 mg/day were used in sensitivity analysis.

Results: In our sample (N = 144,703), 20% were assigned APD status based on the LED >1000 mg/day cut-off. This group had significantly higher odds of having each claims-based indicator, compared with those assigned mild-moderate PD status. Odds ratios were highest for indicators for any DBS (OR: 2.96; 95% CI:2.75-3.19) and specialty bed (OR:2.15, 95% CI: 1.99-2.32) and lowest for fall (OR:1.27; 95% CI:1.20-1.34) and dementia diagnosis (OR:1.21; 95% CI:1.18-1.25). Results based on alternative approaches were similar.

Conclusions: Medicare patients classified as having APD via a pharmacy claims-based algorithm had higher odds of having claims-based clinical markers of APD, compared with patients categorized as having mild-moderate PD. This proxy strategy could facilitate future claims-based studies and warrants further refinement and validation using medical records or other clinical sources.