A machine learning approach to quantifying fall conversion risk in fall-naïve Parkinson's patients.

Area of Science:

• Neurology
• Biomedical Engineering
• Data Science

Background:

• Recurrent falls significantly impact quality of life and increase healthcare costs for people with Parkinson's disease (PwPD).
• Existing fall prediction models struggle to identify individuals with Parkinson's disease (PD) who will transition from non-fallers to recurrent fallers.
• There is a critical need for improved methods to predict fall risk in PwPD.

Purpose of the Study:

• To develop and validate a machine learning model for predicting fall risk in people with Parkinson's disease (PwPD).
• To identify key factors contributing to the transition from non-faller to recurrent faller status in PwPD.

Main Methods:

• Utilized baseline clinical motor, biomechanical, cognitive, and quality of life data from 246 PwPD in a clinical trial.
• Developed a fall conversion prediction model using an XGBoost algorithm on data from 174 fall-naïve participants.
• Employed repeated cross-validation to assess model performance, with Area Under the Curve (AUC) as the primary metric.

Main Results:

• The XGBoost model achieved a mean cross-validated AUC of 0.63.
• The model demonstrated a mean sensitivity of 66% and specificity of 51%, with higher sensitivity (77%) for predicting multiple falls.
• Key predictive features included biomechanical balance assessments, NeuroQoL and MDS-UPDRS questionnaires, and cognitive tests (Trails Making Test A, Symbol Digit Matching Test).

Conclusions:

• Multifaceted data, including biomechanical and self-report measures, are valuable for identifying PwPD at risk of initial and recurrent falls.
• Enhanced precision in predicting falls can inform targeted interventions and behavioral modifications to prevent the progression to recurrent faller status.
• This approach holds potential for improving management strategies and mitigating the impact of falls in people with Parkinson's disease (PD).