A machine learning approach to identifying important features for achieving step thresholds in individuals with...

Area of Science:

• Neurorehabilitation
• Physical Therapy
• Stroke Recovery

Background:

• Stepping activity post-stroke shows significant variability across studies.
• Identifying specific predictors for achieving different stepping activity levels is crucial.

Purpose of the Study:

• To determine the most important predictors for achieving two distinct daily step thresholds: 2500 steps (home/community ambulation) and 5500 steps (physical activity guidelines).
• To differentiate between primary and ancillary characteristics influencing stepping activity post-stroke.

Main Methods:

• Analysis of data from 268 stroke participants, including 25 demographic, performance-based, and self-report variables.
• Application of lasso regularization for dimensionality reduction followed by drop column feature importance analysis.
• Assessment of predictor importance across three unique algorithms to identify consistent factors.

Main Results:

• Predictors were reduced to 7 for the 2500-step threshold and 16 for the 5500-step threshold.
• The 6-Minute Walk Test and speed modulation were identified as primary characteristics essential for both thresholds.
• Readiness to change activity behavior and physical health were identified as important ancillary characteristics.

Conclusions:

• Physical capacity (e.g., 6-Minute Walk Test, speed) is necessary but not sufficient for achieving stepping activity goals post-stroke.
• Ancillary characteristics, including behavioral readiness and physical health, must also be targeted for comprehensive improvement.
• This distinction may help explain the heterogeneity observed in previous studies on stroke stepping activity predictors.