JPD: machine learning to assist Parkinson's disease prediction

Parkinson's disease (PD) is a chronic, progressive and disabling neurodegenerative disease that begins in middle and old age

PD is now recognized as a systemic disease.

There is no test to diagnose PD, and only after extensive loss of striatal dopamine does motor symptoms and signs that meet the diagnostic criteria appear

In fact, more than 30 years of clinical trials for early Parkinson's disease have failed to find interventions to change the disease

Only a few previous studies have prospectively collected a wide range of prodromal characteristics and risk factors, and the correlation between these risk factors and the pathology of Louis at autopsy has not been fully studied

Importantly, among clinically normal people, up to 20% have discovered accidental Lewy pathology, which many people believe reflects early PD

They use questionnaire surveys and simple non-invasive clinical tests to predict the future diagnosis of PD

During the period 1995-2017, participants in the prospective Kuakini Honolulu-Asia Aging Study (HAAS) were evaluated by PD experts using standard diagnostic criteria every six months

They input simple clinical and risk factor variables into a machine learning algorithm based on ensemble trees, and derive a model to predict the incidence of PD

The study sample included 292 subjects, of which 25 had PD within 3 years and 41 had PD within 5 years

Modeling 12 predictors, correctly classified individuals with PD within 3 years (area under the curve (AUC) 0.

A large proportion of the control group misjudged as PD had Lewis pathology at autopsy, including 79% of those who died within 3 years

The PD probability is estimated to be inversely proportional to the density of substantia nigra neurons, and is the strongest among autopsy performed within 3 years after the initial date (r = -0.