Dual-Task Training in Parkinson’s: What Gait It Changes

Dual-task gait training can help people with Parkinson's disease walk faster, with a steadier rhythm and less step-to-step variation. It does not appear to correct gait asymmetry or the postural-control deficits most closely tied to falls. Worse, the documented gains disappeared after training stopped.

That is a more useful finding than the usual headline that "exercise improves gait." Parkinson's gait is not one problem with one dial. Different features rely on different neural systems, and a program can improve some without touching others.

Three gait domains improved. Two did not.

The clearest map comes from Rennie et al.'s 2021 secondary analysis of a randomized trial. The trial enrolled 100 people with Hoehn and Yahr stage 2–3 Parkinson's disease and tested 10 weeks of highly challenging balance and gait training, including dual tasks, three times per week. Researchers analyzed walking with a five-domain framework: pace, rhythm, variability, asymmetry, and postural control.

Training improved pace. At normal walking speed, step velocity increased by 8.2 cm/s (P=0.04) and step length increased by 3.0 cm (P=0.05). At fast speed, step velocity increased by 10.8 cm/s (P<0.01) and step length by 2.3 cm (P=0.05).

Rhythm also changed. Stance time fell by 24.3 milliseconds during normal walking (P=0.01) and by 29.6 milliseconds during fast walking (P=0.02). During fast walking, step time fell by 19.3 milliseconds (P=0.02) and swing time by 8.7 milliseconds (P=0.04). Measures of variability improved as well: fast-walking step-time variability fell by 2.8 milliseconds and stance-time variability by 3.9 milliseconds, both with P=0.02.

Those are short-term, statistically significant changes in parameters that affect everyday mobility. But the same analysis found no significant improvement in gait asymmetry or the postural-control domain. The distinction matters because a person can cover ground more quickly without gaining the reactive balance responses needed to recover from a shove, trip, or missed curb.

A meta-analysis supports the mobility gains—but cannot settle the balance question

Du et al.'s 2023 systematic review and meta-analysis pooled seven randomized controlled trials involving 406 participants with Parkinson's disease. Dual-task training improved gait velocity with a standardized mean difference of 0.62 (95% CI, 0.37 to 0.87) and cadence with a standardized mean difference of 0.29 (95% CI, 0.05 to 0.53). Heterogeneity was modest for velocity (I²=31%) and zero for cadence. Timed Up and Go performance improved by 2.38 seconds versus comparison interventions (95% CI, −3.93 to −0.84).

The review also found a 2.04-point improvement in mini-BESTest total scores (95% CI, 1.05 to 3.03). That result should not be read as proof that reactive postural control improved. The mini-BESTest combines 14 items across anticipatory adjustments, reactive postural control, sensory orientation, and dynamic gait. A higher total could come from dynamic-gait or anticipatory items while reactive responses remain unchanged.

Du et al. did not report the relevant subscales. As a result, the pooled total score cannot resolve the more specific null result in Rennie et al.'s gait-domain analysis. The studies are not necessarily in conflict; they measured balance at different levels of detail.

A better mini-BESTest total after training may reflect better dynamic movement. It does not establish that the protective stepping and reactive responses involved in fall recovery changed.

Why might some features resist training?

A plausible explanation is that dual-task practice relies heavily on cortical compensation. Kvist et al. (2026) observed greater prefrontal-cortex recruitment during dual-task walking in people with Parkinson's disease than in healthy older adults. This observational study did not test a training program, so it cannot prove the mechanism behind training effects. It does, however, fit the idea that people with Parkinson's recruit executive resources when basal-ganglia support for automatic walking weakens.

Related evidence comes from a small randomized, double-blind, sham-controlled crossover study of 20 people with Parkinson's disease. Conceição et al. (2022) found that adding prefrontal transcranial direct-current stimulation to a 30-minute moderate-intensity cycling session affected cortical activity, gait, and cognition. It supports the prefrontal cortex as a modifiable rehabilitation target, but it does not show that cortical training repairs the circuitry behind reactive balance.

Postural instability and axial symptoms also involve progressively affected non-dopaminergic brainstem locomotor systems. Cognitive-executive loading is unlikely to recruit those systems in the same way. That remains a biologically coherent explanation, not a demonstrated reason why every individual fails to respond.

The asymmetry and postural-instability gap is broader than dual-task training

The failure to improve asymmetry and postural control may reflect limits of current exercise therapy rather than an obvious programming mistake. Qian et al.'s 2023 network meta-analysis compared 24 exercise modalities for postural instability in Parkinson's disease. Some approaches, including Tai Chi, dance, and combined training, showed encouraging signals for balance outcomes, but no modality demonstrated convincing long-term superiority for postural instability.

Even a method designed to address asymmetry has produced modest effects. Sasikumar et al. (2025) studied split-belt treadmill training, which aims to use cerebellar adaptation to alter step-length asymmetry in Parkinson's disease. Its modest asymmetry effects do not prove the problem is untreatable. They do make it hard to claim that ordinary dual-task training merely needs a cleverer playlist or a more complicated cognitive game.

Medication state and disease phenotype add another complication. Yu et al. (2025) found distinct gait profiles and fall-risk patterns across tremor-dominant, akinetic-rigid, and postural-instability/gait-disorder phenotypes, with differences related to dopaminergic treatment state. The available dual-task trials did not adequately stratify participants by these factors. A mean improvement may conceal meaningful responder and nonresponder groups.

The major problem is durability

In Rennie et al.'s trial, every observed gain in pace, rhythm, and variability had returned to baseline by the six-month follow-up after training cessation. The analysis found no evidence that one responsive domain decayed more slowly than another. No study in this evidence base demonstrates 12-month retention of dual-task gait-domain gains.

That result changes how a 10-week program should be framed. It can be a useful period of rehabilitation, but it should not be marketed as a completed fix. If a benefit depends on continued practice, maintenance is part of the intervention rather than an optional afterthought.

Training duration probably matters, although the dose-response curve remains unknown. Pelosin et al. (2022) found that duration influenced motor-cognitive outcomes from treadmill training with virtual reality. The study does not identify the duration needed to retain gains for six or 12 months, nor does it show that more training corrects asymmetry or reactive postural control.

Other cognitively demanding programs support the broader case for motor-cognitive rehabilitation. Shah et al. (2022) reported improved balance and gait after a cognitively challenging Agility Boot Camp program, including among participants with freezing of gait. Long-term retention was not reported.

What this means in practice

Dual-task balance and gait training has a defensible role for mild-to-moderate Parkinson's disease when the goal is better walking speed, rhythm, and consistency. The pooled evidence supports those short-term outcomes, and the 100-person randomized trial explains which gait domains changed.

It should not be treated as a proven fall-prevention solution. The trials did not establish improvement in the reactive postural-control component of balance, and no exercise type has convincingly solved long-term postural instability. Gait asymmetry may warrant interventions designed specifically for it, such as split-belt treadmill work, though evidence for that approach remains modest.

The next useful trial would pre-register domain-specific gait outcomes, account for medication state and motor phenotype, include a maintenance program, and follow participants for at least 12 months. Until then, dual-task training looks like ongoing rehabilitation for selected gait problems—not a one-time answer to Parkinson's falls.