The research team, led by Long Wang of Pukyong National University in Busan, South Korea, together with colleagues in China and Korea, employed a sophisticated analytical framework known as Bayesian network meta-analysis. Unlike a traditional meta-analysis, which can only compare two interventions at a time, a network meta-analysis weaves together direct and indirect comparisons across an entire network of trials, allowing researchers to rank many different exercise types even when they have never been tested against one another in a single study. By placing the analysis in a Bayesian framework, the authors could express their results as posterior probability distributions—essentially, a full picture of what the data suggest about each effect, complete with credible intervals that quantify uncertainty.

The team focused on four motor outcomes that matter enormously in daily life for people with Parkinson’s. The first was the motor section of the Unified Parkinson’s Disease Rating Scale, known as UPDRS part III, a clinician-rated measure of tremor, rigidity, bradykinesia, and postural instability. The second was balance, typically assessed with standardized clinical scales. The third was walking speed, a fundamental determinant of independence and fall risk. The fourth was the Timed Up and Go test, or TUG, which measures how long it takes a person to rise from a chair, walk three meters, turn, return, and sit down again—a deceptively simple task that integrates strength, balance, and gait into a single number.

The networks behind each outcome were substantial. The UPDRS III analysis drew on 110 trials, the balance analysis on 106, the walking speed analysis on 89, and the TUG analysis on 96. Across these networks, one category of exercise stood out for the clinician-rated motor scale: mind–body exercise, a category encompassing practices such as tai chi and yoga that combine physical movement with attentional, breathing, and meditative components. Compared with non-exercise control groups, mind–body exercise showed the largest posterior median effect on UPDRS III scores, with a standardized mean difference of 0.65 and a 95 percent credible interval running from 0.42 to 0.87. In plain terms, trials of mind–body approaches tended to show noticeably larger improvements in clinician-rated motor symptoms than trials in which participants did no structured exercise at all.

But the modality comparison was only half the story. The researchers also wanted to know whether the total planned dose of exercise—how much structured activity a trial actually scheduled each week—was associated with better outcomes. To do this, they converted each trial’s planned weekly exercise exposure into metabolic equivalent task minutes, or MET-minutes, a standard unit from the Adult Compendium of Physical Activities that expresses the energy cost of an activity relative to resting metabolism. A brisk walk might accrue around 300 MET-minutes in half an hour, while more vigorous activities accumulate faster. The team then built dose–response models, adjusted for exercise class, asking a deceptively simple question: for every additional 250 MET-minutes per week of planned structured exercise, how much better did the motor outcomes get, on average, at the trial level?

The answer was encouraging for three of the four outcomes. Each additional 250 MET-minutes per week was associated with an increase of 0.121 standardized mean difference units in UPDRS III improvement, with a 95 percent credible interval of 0.055 to 0.194. For balance, the corresponding estimate was 0.111, with a credible interval of 0.047 to 0.184, and for the Timed Up and Go test it was 0.124, spanning 0.058 to 0.190. In other words, trials that planned more weekly structured exercise tended to report larger average improvements in clinician-rated motor symptoms, balance performance, and functional mobility. Walking speed, however, told a different story: its association was essentially flat, at 0.008 with a credible interval stretching from −0.063 to 0.076, meaning the data provided no clear evidence that higher planned doses translated into faster walking.

Why might walking speed resist the dose–response pattern seen elsewhere? The authors and the broader literature suggest several possibilities. Walking speed in Parkinson’s disease is heavily influenced by gait-specific mechanisms, including stride length, freezing of gait, and the ability to respond to external cues. Some trials have found that walking training with auditory cueing improves speed more than walking training alone, hinting that the content of the intervention may matter more than its raw volume for this particular outcome. It is also possible that generic measures of planned exposure, expressed in MET-minutes, simply fail to capture the gait-specific ingredients that drive speed gains. A dose of tai chi and a dose of treadmill walking may carry the same MET-minute tally while engaging the neural circuits of gait in profoundly different ways.

Crucially, the authors did not oversell their findings, and the caveats deserve as much attention as the headline numbers. Every prediction interval for active exercise versus control crossed zero, meaning that for an individual new trial, the expected effect could plausibly be null or even negative. The confidence assessment, conducted with the CINeMA framework—Confidence in Network Meta-Analysis—rated the evidence as low or very low across outcomes. Within exercise classes, the randomized dose-contrast estimates were imprecise for all four outcomes, and the TUG association proved sensitive to the high-dose tail of the distribution: when the analysis truncated doses at the 95th percentile, the TUG estimate attenuated. These are not minor statistical footnotes; they signal that the apparent dose–response relationship could be shaped by a handful of trials prescribing unusually large amounts of exercise.

The team was explicit about what the evidence does not establish. The analysis did not identify a universally best exercise modality, an optimal dose, or an individual causal target. The dose associations are trial-level observations, not prescriptions for a specific patient, and they cannot rule out confounding by trial design, population characteristics, or publication dynamics. What the study does provide is a carefully quantified map of the evidential landscape: mind–body exercise currently holds the strongest posterior estimate for clinician-rated motor symptoms, planned weekly exposure shows positive trial-level associations with motor severity, balance, and mobility, and walking speed appears to march to its own drummer. For clinicians following guidelines such as those from the American Physical Therapy Association, which already recommend exercise as a core component of Parkinson’s care, the findings reinforce the principle that structured, sufficiently dosed exercise belongs in the treatment plan—while cautioning against claims that any single modality is definitively superior.

There is also a methodological lesson here for the field of exercise science itself. The authors made their frozen analysis data, code, model files, and figure source data publicly available on Zenodo, an unusual and welcome degree of transparency that allows other researchers to interrogate every modeling choice, from the covariance-aware random-effects structure to the sensitivity analyses. As Bayesian network meta-analyses proliferate across medicine, this study demonstrates both their power and their pitfalls: they can synthesize hundreds of trials into coherent rankings and dose–response curves, but the resulting estimates are only as trustworthy as the underlying trials, and honest uncertainty quantification—credible intervals, prediction intervals, and formal confidence grading—must travel alongside every headline number. For people with Parkinson’s and the clinicians who treat them, the message is refreshingly practical: move regularly, move deliberately, and consider mind–body practices as a serious option, while recognizing that science has not yet found the single perfect prescription.

Subject of Research: Effects of exercise modality and planned weekly exercise dose on motor outcomes in Parkinson's disease

Article Title: Exercise modality and planned weekly exercise exposure for motor outcomes in Parkinson’s disease: a systematic review, Bayesian network meta-analysis, and meta-regression

Article References: Wang, L., Wang, H., Hong, F., & Wan, Z. (2026). Exercise modality and planned weekly exercise exposure for motor outcomes in Parkinson’s disease: a systematic review, Bayesian network meta-analysis, and meta-regression. Journal of Neurology, 273(11), Article 654. https://doi.org/10.1007/s00415-026-14189-y

Image Credits: AI Generated

DOI: 10.1007/s00415-026-14189-y

Keywords: Parkinson's disease, exercise, network meta-analysis, Bayesian analysis, mind-body exercise, dose-response, motor symptoms, balance, walking speed, Timed Up and Go, UPDRS, physical therapy