< back to all PPMI Presentations

ACCURATE, NON-INVASIVE AND COST-EFFECTIVE PREDICTIVE MODEL FOR PARKINSON'S DISEASE.

Mike A Nalls and Andrew B Singleton on behalf of the Michael J. Fox foundation\'s Parkinson\'s Progression Markers Initiative (PPMI) study, the Parkinson Associated Risk Study (PARS), the NINDS Parkinson\'s Disease Biomarker Program (PDBP) and the International Parkinson\'s Disease Genomics Consortium (IPDGC).
 
2015-03-18
AD/PD: Nice, France
Download Presentation: abstractPPMI.pdf
 
Abstract:
OBJECTIVES: To identify non-invasive and cost-effective factors accurately predicting if participants in longitudinal studies were Parkinson's disease cases.

METHODS: We mined baseline biomarker, clinical and genetic data in the Michael J. Fox foundation's Parkinson?s Progression Markers Initiative (PPMI). Stepwise regression models were used to identify candidate factors for modeling based on Akaike information criterion. Factors remaining in the model included a genetic risk score incorporating 28 common risk loci as well as G2019S (LRRK2) and N370S (GBA1), University of Pennsylvania Smell Identification Test score (UPSIT), family history, gender and age. We developed the model using the PPMI cohort and then fit it to the Parkinson Associated Risk Study (PARS) cohort. We are pursuing further replication in the NINDS Parkinson's Disease Biomarker Program (PDBP) cohort.

RESULTS: Both PPMI and PARS showed > 93% predictive accuracy (area under the curve). Logistic regression examining predictive model fit showed significance at p-value < 2E-16. In the prospective PDBP cohort, we expect further replication and a retrospective estimate of accuracy relative to time from diagnosis. This second replication phase will also address possible recruitment bias in PARS.

CONCLUSIONS: This is an important model as we show we can accurately predict Parkinson's disease in populations using non-invasive and relatively inexpensive information costing no more than $150 USD per sample.