A machine learning-based fall risk prediction model for Parkinson's disease considering ophthalmic disorders

Area of Science:

• Ophthalmology
• Neurology
• Data Science

Background:

• The link between ophthalmic disorders and fall risk in Parkinson's disease (PD) remains unclear.
• Previous research has not fully elucidated the predictive factors for falls in PD patients.

Purpose of the Study:

• To investigate the association between ophthalmic disorders and fall risk in PD patients.
• To develop and evaluate machine learning models for predicting one-year fall risk in PD using clinical and ocular data.

Main Methods:

• Utilized data from 543 individuals with PD from the Parkinson's Progression Markers Initiative (PPMI) database.
• Employed random forest and categorical data analysis program (CATDAP) for predictive modeling.
• Split data into 80% learning and 20% testing sets to build and validate models.

Main Results:

• The random forest model with CATDAP feature selection achieved 82.6% accuracy and an MCC of 0.456.
• Key predictors for fall risk included "fall history", "glaucoma", "arising from chair", and "gait".
• Glaucoma was identified as a significant predictor of fall risk.

Conclusions:

• Glaucoma is a notable risk factor for falls in PD patients, in addition to their fall history and neurological symptoms.
• Ophthalmologic assessments can contribute to fall prevention strategies in clinical practice for PD.
• Integrating ophthalmologic data into fall risk prediction models for PD is recommended.