JMIR Publications

ABSTRACT

Background:

Falls among older adults are a growing and costly public health problem that often leads to mobility decline and loss of independence. Although clinical tools such as Disease Control and Prevention (CDC) Stopping Elderly Accidents, Deaths & Injuries (STEADI) initiative recommends multi-factor screening (gait, balance, strength, fear of falling, and fall history), most wearable fall risk assessment systems rely on a small set of risk factors (typically gait), which creates a gap between clinical practice and automated wearable assessment.

Objective:

To evaluate the importance of STEADI-based fall risk factors, assess whether gait is enough for wearable fall risk assessment, and provide design guidance for clinically compatible wearable fall risk assessment systems.

Methods:

A dataset of 24 older adults (10 low fall risk, 14 high fall risk) was created from a publicly available plantar pressure dataset collected with smart insoles from 47 subjects. Only subjects whose fall risk labels were consistent across both Berg Balance Scale (high risk: BBS ≤36) and Timed Up and Go (high risk: TUG ≥13.5 s) were retained in the final dataset. Eighteen features were extracted from 3556 gait cycles to quantify gait, strength, balance, fear of falling, and fall history. Specifically, a novel domain-knowledge-informed feature, the foot flat phase ratio, was introduced to quantify cautious gait patterns associated with fear of falling. Random Forest (RF) models were trained with leave-one-subject-out cross-validation (LOSO-CV) to assess fall risks. Importance of STEADI-based factors was assessed by two methods: (1) by estimating SHapley Additive exPlanations (SHAP) values based on a single RF model trained on all features; and (2) by training five separate RF models, each on one STEADI factor category, and comparing their fall risk prediction accuracies.

Results:

The RF model trained on all features achieved 84.62% gait-cycle-level accuracy and 87.5% subject-level accuracy. SHAP analysis ranked the right foot flat phase ratio (fear of falling related feature) as the most important feature, followed by maximum right forefoot ground reaction force (strength related feature), whereas traditional gait features did not show in the top ten. The accuracies of five separate RF models, each trained on one STEADI factor category, confirmed this ranking: fear of falling achieved the highest accuracy (75.68%), followed by strength (72.44%), balance (68.34%), gait (66.31%), and fall history (62.5%).

Conclusions:

Gait is not enough for wearable fall risk assessment. Commonly overlooked fall risk factors, such as fear of falling and strength, were more important than popularly used gait features. The novel foot flat phase ratio outperformed all other evaluated features, which showed the value of domain-knowledge-informed feature engineering. These findings provide design guidance for developing wearable fall risk assessment systems that are more compatible with clinical practice. Clinical Trial: N/A