Predicting minimum toe clearance in older adults on different surfaces from level ground data: A machine learning...

Trips are more likely on non-level, carpeted, or obstacle surfaces, whereas minimum toe clearance (MTC) is usually measured on level-ground. Current studies indicate that MTC differs between level and non-level walking, but it's unclear whether MTC data collected in a safer laboratory environment can predict MTC on these more challenging non-level surfaces, as this may have implications for predicting trip-falls in the community.

This study examined whether level-ground MTC and related gait variables in community-dwelling older adults can predict MTC on carpet and obstacle surfaces.

Twenty-one healthy older adults completed 25 gait trials at self-selected speeds on level-ground, carpet, and surfaces with 5 mm and 13 mm obstacles. Fifteen level-ground gait variables were used as predictors. Predictive models were developed using nested cross-validation, and principal component analysis (PCA) was applied within the modeling pipeline to address multicollinearity.

The k-nearest neighbors with PCA (kNN-PCA), using 15 level-ground variables reduced to 10 principal components, was the best-performing model. Toe-tip MTC was predicted with good agreement (CCC 0.79-0.90) and low error (MAE 2.30-3.84 mm). The model's prediction accuracy decreased as obstacle height increased, and right-foot outcomes were generally more predictable than left-foot outcomes. A reduced seven-variable model performed similarly.

MTC on carpet and obstacle surfaces can be reliably estimated from level-ground gait measurements in community-dwelling older adults. These findings support the use of level-ground laboratory assessments as a practical basis for estimating toe clearance in more challenging walking environments relevant to fall-risk assessment and rehabilitation.