Linear-Sigmoidal modelling of accelerometer features and Tinetti score for automatic fall risk assessment

Area of Science:

• Gerontology
• Biomedical Engineering
• Data Science

Background:

• Falls in the elderly are a significant global health issue, leading to severe injuries and mortality.
• Existing home safety measures and lifestyle interventions are insufficient; at-home fall risk detection methods are needed.
• The Tinetti scale is a clinical tool for assessing fall risk, with scores ≤ 18 indicating high risk.

Purpose of the Study:

• To introduce a novel, simple, non-linear Linear-Sigmoidal (LS) model for predicting fall risk in the elderly.
• To evaluate the LS model's performance against standard linear regression (LR) using accelerometer data and clinical fall risk scores.
• To provide an accessible, interpretable method for at-home fall risk assessment.

Main Methods:

• One hundred twelve elderly subjects underwent a Tinetti test while wearing a 3D accelerometer.
• A Linear-Sigmoidal (LS) model was developed using seven accelerometer features and body mass index.
• Model performance was assessed on a training set (90 subjects) and a test set (22 subjects), comparing LS against standard linear regression (LR).

Main Results:

• The LS model demonstrated superior model agreement, achieving an R² of 0.76 compared to 0.72 for LR.
• The LS model exhibited higher classification accuracy (0.91) than LR (0.86) on the test set for predicting high fall risk.
• The study confirmed the LS model's effectiveness in modeling accelerometer features and predicting clinical fall risk scores.

Conclusions:

• The proposed Linear-Sigmoidal (LS) model is a simple and effective tool for predicting fall risk in the elderly.
• The LS model offers improved accuracy and agreement over standard linear regression for fall risk assessment using wearable sensor data.
• This methodology provides a promising approach for developing accessible, at-home fall risk detection systems.