Predicting Parkinson’s Disease in Rural America Based on Initial Examination Data

Neurological disorders are frequent causes of morbidity and mortality in the US, whereby Parkinson’s is one of the most common. Increasing incident rates and a continuous shortfall of neurologists is prevalent. Additionally, neurophobia has been identified in medical student cohorts, residents and among general practitioners, which gives concern regarding care delivery for neurological patients. Even more dramatic is the situation in rural areas. More patients have to be treated by physicians and primary care providers without formal neurologic training. The American Journal of Managed Care just reported results from a poll that 26% of Parkinson patients having been misdiagnosed. Little publications can be found on Parkinson’s from the perspective of Primary Care Physicians. This thesis closes this gap and takes the angle of a Primary Care Physician. In this study a prediction model is developed that addresses partially the need for better PD detection in primary care in rural settings, and the shortage of neurologists. The model uses machine learning and routinely accessible data to Primary Care Physicians and allows to discriminate between HC and PD with an accuracy of 92 to 96%. This model could assist PCPs with a specificity >96% (LR) and sensitivity >97% (RF) in the triaging process and potentially reduce the number of referrals to neurologists by giving PCPs greater confidence in whether a subject shows sufficient symptom to be diagnosed with Parkinson's.