Older Adult Fall Risk Prediction with Deep Learning and Timed Up and Go (TUG) Test Data

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Falls pose a significant health risk to the aging population, necessitating effective prediction tools.
• Current fall risk assessments often involve clinical evaluations and functional mobility tests like the Timed Up and Go (TUG) test.
• Wearable Inertial Measurement Units (IMUs) offer a novel approach to capture motion data for fall risk estimation.

Purpose of the Study:

• To investigate the efficacy of using IMU data from the TUG test to predict prospective fall risk in older adults.
• To develop and validate deep learning models for estimating the probability of falls in the near future.

Main Methods:

• A cohort of 106 older adults wore wireless IMU sensors during TUG tests.
• Deep learning models, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), were applied to extracted IMU features.
• Supervised learning was used with a binary outcome variable indicating falls within a six-month follow-up period.

Main Results:

• A Bidirectional Long Short-Term Memory (BLSTM) model achieved the best performance.
• The BLSTM model demonstrated an accuracy of 0.83 and an Area Under the Curve (AUC) of 0.73.
• The model also exhibited good sensitivity and specificity in predicting fall risk.

Conclusions:

• IMU data collected during the TUG test, when analyzed with deep learning, can effectively predict prospective fall risk in older adults.
• This approach offers a promising, objective method for identifying individuals at high risk of falls.
• The findings support the integration of wearable sensor technology and AI in geriatric fall prevention strategies.