The research team, led by clinicians and computational scientists at the University Hospital Würzburg together with collaborators at Charité – Universitätsmedizin Berlin, followed 100 people with Parkinson’s disease who underwent subthalamic deep brain stimulation between 2010 and 2020. The patients, whose average age at surgery was just over 60 years, underwent detailed assessments before the operation, again one year afterward, and in a subset of 33 patients, at three to five years after surgery. Rather than examining each symptom in isolation, the investigators deployed a battery of statistical tools, including unsupervised clustering, principal component analysis, and moderation models, to capture the joint structure of motor, cognitive, and mood symptoms as they evolved over time. The goal was to see whether deep brain stimulation preserves or reshapes the hidden architecture of the disease, and whether non-motor burden could explain why some patients regain mobility while others do not.

The technical pipeline was deliberately conservative. Missing clinical values were imputed using chained equations fitted only on pre-operative data, so that follow-up information could not leak backward and bias the baseline picture. Principal component analysis was fitted exclusively on the pre-surgery scores, and every follow-up visit was then projected into that fixed baseline space, allowing the researchers to track each patient’s trajectory along stable clinical dimensions rather than comparing shifting mathematical axes. Cluster stability was verified with 500 bootstrap resamples, and the robustness of the two-cluster solution was confirmed across 20 multiply-imputed datasets, with a mean adjusted Rand index of 0.945 against the primary solution. P values were corrected for false discovery across all simultaneous comparisons, and heteroscedasticity-consistent standard errors guarded the moderation models against uneven variance.

The first striking result concerned how patients group together before surgery. Unsupervised clustering of 22 clinical sub-domain scores revealed two reproducible profiles, a higher-impairment and a lower-impairment group, separated primarily by axial motor features: postural instability, gait, posture, and mobility, with smaller but meaningful differences in anxiety and depressive symptoms. The separation between clusters was large by conventional standards, with Cohen’s d values approaching or exceeding 1.9 for postural instability and gait. When the researchers projected the one-year post-operative assessments into this baseline cluster space, 84 percent of the patients who had started in the higher-impairment profile moved into the lower-impairment profile, while 95 percent of those who began in the lower-impairment group stayed there. Because the two profiles sit along a continuous severity dimension rather than representing truly distinct disease subtypes, the authors interpret this migration as improvement along a single axis of clinical burden, a continuous slide toward wellness that mirrors the well-documented efficacy of the stimulation itself.

The second major finding is perhaps the most sobering. Across every timepoint, the single largest source of between-patient clinical variance was a core motor triad: gait, mobility, and freezing of gait. This component explained 35 percent of the variance before surgery, 32 percent at one year, and still 28 percent at three to five years, and it was the only component to survive a rigorous permutation null, meaning that the other axes of variation were largely domain-specific and statistically fragile. In plain terms, even after electrodes have silenced tremor and rigidity, the way a patient walks remains the dominant thing that separates one person with Parkinson’s from another. This aligns with a growing body of evidence that axial symptoms, which depend on brainstem and cholinergic circuits as much as on the dopaminergic pathways that stimulation modulates, respond poorly to subthalamic stimulation and remain the leading source of unmet expectations after surgery.

The longitudinal projections added a temporal dimension to this picture. The motor severity axis dropped significantly in the first year after surgery, consistent with genuine motor improvement, but by three to five years it had drifted back toward baseline, echoing earlier reports of initial gait gains of more than 40 percent followed by substantial long-term decline. Meanwhile, a memory-dominated component, which captured 17 percent of between-patient variance at baseline, rose to 24 percent by the long-term follow-up, and within-patient projected memory scores increased significantly over the same interval. Mood features showed no significant change at one year but a small, statistically significant worsening at three to five years. The researchers are careful to note that these shifts reflect changes in relative heterogeneity and domain-specific burden rather than a wholesale reorganization of the disease, and that the covariance structure itself, tested with Procrustes analysis against 5,000 permutations, remained significantly aligned across timepoints.

The third and most clinically provocative finding emerged from the moderation analysis. Using analysis of covariance models that regressed each one-year motor outcome on its baseline value, a non-motor moderator, and their interaction, while adjusting for age and disease duration, the team asked whether pre-operative mood and cognition shaped the degree of motor recovery. The answer, in exploratory analyses, was yes. Higher baseline non-motor burden predicted poorer recovery of the composite gait-and-mobility triad, with an interaction coefficient of 1.45. Decomposed into individual features, the strongest culprits were psychomotor depressive symptoms, with a coefficient of 2.16, psychosomatic depressive symptoms, and language function, each associated with smaller mobility gains. Worse baseline orientation, by contrast, was associated with greater improvement. At the one-year follow-up, anxiety measured at the same visit showed a concurrent association with both the composite motor score and mobility, with coefficients of 2.30 and 2.81 respectively. Gait and freezing of gait themselves showed no significant moderators after correction, suggesting that the mood effect is concentrated in mobility rather than in freezing episodes.

Why would depression and anxiety written into a patient’s chart before surgery echo through their walking a year after electrodes are implanted? The authors point to the anatomy of the subthalamic nucleus itself, which is not a purely motor relay but a hub of striato-thalamo-cortical circuits serving cognitive and limbic functions as well. Gait and postural control demand constant cognitive-motor integration, the very faculty that frontal-executive dysfunction erodes, and symptom-specific tractography work has shown that axial motor symptoms are mediated by distinct white matter pathways connecting the subthalamic nucleus to the supplementary motor cortex and brainstem, rather than by the primary motor circuits that standard stimulation parameters most effectively engage. Shared prefrontal-subthalamic circuitry involved in both affective regulation and motor control offers a plausible substrate for the observed interactions, and some researchers have even proposed that patients implicitly trade cognitive performance for gait stability, a bargain that stimulation does not resolve.

The authors are appropriately cautious about how far these conclusions can be pushed. The study was retrospective, and strict inclusion criteria trimmed 180 consecutive patients down to 100, potentially selecting for more completely assessed cases. The long-term follow-up sample of 33 patients is modest, though comparable to other studies at equivalent timepoints, and that subset was enriched for higher baseline motor severity, which may have tempered the apparent long-term motor findings. The moderation analyses are explicitly exploratory, and no external cohort with harmonized multi-domain longitudinal data was available for validation, a scarcity the authors describe as a broader problem for the field. Medication changes were addressed through sensitivity analyses: levodopa equivalent daily dose fell by a median of 56.7 percent after surgery, but residualizing this change from every post-operative score left the component structure essentially untouched, with a Procrustes distance of just 0.004 between the original and adjusted solutions.

Even with those caveats, the clinical implications are hard to ignore. If pre-operative mood burden genuinely forecasts the size of a patient’s mobility recovery, then a psychiatric evaluation belongs in every deep brain stimulation candidacy discussion, not as a gatekeeping exercise but as a prognostic tool that helps patients and clinicians set realistic expectations. And if anxiety and psychosomatic symptoms at follow-up track concurrent motor burden, then post-operative mood monitoring, and interventions that combine motor rehabilitation with psychological support, such as dual-task training, become candidate levers for protecting the very gains that surgery was meant to deliver. The authors frame these as testable hypotheses for prospective multicenter trials rather than established determinants of outcome, but the direction of travel is clear: the future of deep brain stimulation may depend less on where the electrodes land than on how completely clinicians learn to read, and treat, the whole patient before and after the current flows.

Subject of Research: The relationship between non-motor symptom burden and motor recovery after subthalamic nucleus deep brain stimulation in Parkinson's disease

Article Title: Non-motor symptoms burden and motor recovery following STN–DBS in Parkinson's disease: a data-driven longitudinal analysis

Article References: Abbas, G., Peach, R., Temuulen, U., Tang, Y., Sil, T., Kufner, A., Endres, M., Volkmann, J., Lange, F., & Reich, M. (2026). Non-motor symptoms burden and motor recovery following STN–DBS in Parkinson's disease: a data-driven longitudinal analysis. Journal of Neurology, 273(10), Article 636. https://doi.org/10.1007/s00415-026-14160-x

Image Credits: AI Generated

DOI: 10.1007/s00415-026-14160-x

Keywords: Parkinson's disease, deep brain stimulation, subthalamic nucleus, non-motor symptoms, gait, freezing of gait, mobility, depression, anxiety, principal component analysis, longitudinal study, neurology