Exploring Responsiveness to Highly Challenging Balance and Gait Training in Parkinson's Disease.

Exercise programs for Parkinson's disease often help on average, but not everyone responds the same way, and it has been unclear who benefits most from intensive balance and gait training. This study looked at people with mild-to-moderate Parkinson's who completed a highly challenging balance and gait program called HiBalance, part of a larger randomized controlled trial, and asked whether their individual characteristics before starting could predict whether they would be high responders, low responders, or non-responders in three areas: balance, walking speed, and daily step counts measured with a wearable sensor. Using a statistical modeling approach, the researchers found they could moderately distinguish these response categories only for balance, not for walking speed or physical activity levels. Within the balance domain, slower baseline walking speed was linked to a stronger chance of being a high responder, while fall history mattered more for identifying low or non-responders.

The sample was small, at thirty-nine participants, so these patterns should be seen as exploratory rather than confirmed predictors.

If you or someone you know has Parkinson's disease and has tried a structured balance and walking exercise program, you may have wondered why some people seem to improve more than others. This study looked at exactly that question, following thirty-nine people with mild-to-moderate Parkinson's who took part in an intensive balance and gait training program as part of a larger clinical trial.

Researchers tried to figure out, based on how someone was doing before the program, whether they could predict who would show a big improvement, a small improvement, or none at all. They looked at three things: balance, walking speed, and how many steps people took in daily life. The one area where they could reasonably tell responders apart in advance was balance.

People who walked more slowly before starting tended to be the ones who improved the most in balance, while a history of falls was more relevant for spotting people who improved little or not at all. For walking speed and daily activity, the researchers were not able to reliably predict who would respond well. Because this involved a small group of participants, these findings are considered exploratory.

They may eventually help design more personalized exercise programs, but they do not yet tell us who should or should not try a particular type of training.

This is a summary of research, not medical advice. Talk to your own healthcare provider about anything affecting your care.

This exploratory analysis drew on thirty-nine participants with mild-to-moderate Parkinson's disease who completed the HiBalance intervention within a randomized controlled trial, examining whether pre-intervention characteristics could classify people into high responder, low responder, or non-responder categories across three domains: balance (Mini-BESTest), gait velocity, and physical activity (accelerometry-derived steps per day). Separate random forest models were built for each domain, with classification performance benchmarked against chance-level agreement using Cohen's kappa. Only the balance domain model classified all three responsiveness levels above chance level, described as slight agreement, meaning the model performed only modestly better than guessing rather than achieving strong predictive accuracy.

Within that domain, slower baseline gait velocity was associated with a higher probability of being a high responder, whereas fall history distinguished low responders and absence of falls distinguished non-responders, indicating that variable importance shifted depending on which responsiveness level was being classified. Neither the gait velocity domain nor the physical activity domain could be classified reliably above chance. Given the modest sample size and the exploratory, hypothesis-generating nature of this random forest analysis, these findings should be interpreted as preliminary.

They may inform inclusion criteria for future personalized balance-focused intervention trials, but they do not establish validated predictors of individual response and require replication in larger, independent samples before any clinical inference is drawn.