Bayesian classification of falls risk

Area of Science:

• Biomechanical analysis of gait.
• Geriatric medicine and fall prevention.
• Statistical modeling and machine learning.

Background:

• Traditional falls risk prediction methods (qualitative/clinical) have limitations.
• Qualitative methods often rely on falls history.
• Clinical methods lack quantification of classification uncertainty.

Purpose of the Study:

• To classify older adults into low or high falls risk categories using gait variables.
• To quantify the uncertainty in falls risk classification decisions.
• To develop a more robust falls risk prediction model.

Main Methods:

• Collected biomechanical gait data from 854 adults aged 65+ using a pressure-sensitive walkway.
• Employed k-means clustering to label falls risk for gait variable vectors.
• Utilized a Gaussian mixture model (GMM) to determine posterior probabilities of risk class membership.
• Applied a threshold based on Youden's J statistic for classification.

Main Results:

• The Bayesian classifier achieved over 96% accuracy in predicting low or high falls risk.
• Demonstrated superior performance compared to k-means falls risk labels via Monte Carlo simulation and ROC analysis.
• Successfully quantified uncertainty in the classification decisions.

Conclusions:

• A Bayesian framework utilizing biomechanical gait data can effectively predict falls risk.
• The proposed method quantifies uncertainty, offering a more comprehensive risk assessment.
• This approach enhances the prediction of falls risk in older adults.