Developing and validating a clinically actionable prediction tool for Parkinson disease using explainable machine...

Area of Science:

• Neuroscience and Public Health
• Computational Epidemiology

Background:

• Parkinson disease (PD) is a growing global health concern, especially in aging Chinese populations.
• Existing research on PD risk factors in China often overlooks multidimensional predictors like socioeconomic status, lifestyle, and early functional impairments.

Purpose of the Study:

• To develop and validate a machine learning model for predicting Parkinson disease (PD) risk in China.
• To identify key multidimensional predictors of PD within the Chinese population.

Main Methods:

• Utilized data from the China Health and Retirement Longitudinal Study (n=13,649).
• Employed ensemble feature selection (Boruta, LASSO, mRMR) followed by an extreme gradient boosting (XGBoost) model with Bayesian optimization.
• Model performance was assessed using AUC, sensitivity, specificity, PPV, NPV, and F1-score, with interpretability via SHAP values.

Main Results:

• The optimized XGBoost model demonstrated high predictive accuracy in the validation set (AUC=0.967).
• Key predictors identified include cognitive impairment (memory, executive deficits), physical inactivity, socioeconomic indicators (retirement, financial support), education, and comorbidities (liver disease, wrist pain).
• SHAP analyses confirmed the consistency and importance of these identified predictors.

Conclusions:

• An explainable XGBoost model integrating multidimensional data shows excellent performance for PD risk stratification in Chinese populations.
• This tool has the potential to support early identification of individuals at risk for Parkinson disease.
• Further external validation and prospective studies are recommended prior to clinical application.