Long-Term Surgical Outcomes and Influential Factors of Subthalamic Nucleus Deep Brain Stimulation for Dyskinesia in Parkinson’s Disease: A 3-Year Longitudinal Cohort Study

Introduction

Parkinson’s disease (PD), the second most common neurodegenerative disorder, affects more than 10 million individuals^1,2 and imposes a significant burden on patients and healthcare systems globally.^3,4 While levodopa remains the cornerstone of symptomatic treatment,^2,5 its long-term use is frequently complicated by levodopa-induced dyskinesia (LID),⁶ a debilitating motor complication that affects approximately 50%–80% of patients with PD after 5–10 years of therapy initiation, causing involuntary movements.⁷ Symptoms such as LID severely impair quality of life, limit therapeutic efficacy, and represent a major challenge in PD management.^8,9 Subthalamic nucleus deep brain stimulation (STN-DBS) has emerged as an effective treatment for advanced PD^10–12 and has demonstrated robust efficacy in alleviating motor fluctuations and LID.^2,12–14 However, while the short-term benefits of STN-DBS are well documented, evidence of its long-term dyskinesia outcomes for patients with PD is limited,^15–17 and critical knowledge gaps persist regarding its prognostic factors.^15,18,19

The antidyskinetic effect of STN-DBS is thought to be mediated both indirectly, by enabling substantial reductions in dopaminergic medication,^20,21 and directly, through modulation of pathological basal ganglia oscillatory activity.^22,23 However, existing studies on STN-DBS for LID have focused predominantly on short-term outcomes (≤ 12 months).^15,21,24 Although these studies demonstrated significant improvements in motor symptoms and a reduction in LID, the durability of these benefits over extended periods remains understudied.^15,24 While some studies have suggested that the therapeutic effects remain,^25,26 others have raised concerns that STN-DBS may lose effectiveness for LID over long-term follow-up due to the progressive nature of PD, neurostimulation-related adaptations, and increasing levodopa dosage again,^16,27,28 necessitating longitudinal evaluation to inform clinical decision-making. In addition, while several prior studies have explored the long-term effects of STN-DBS on LID, the heterogeneity in assessment scales used to evaluate dyskinesia limits comparability.^16,25–28 Crucially, no large-scale studies have systematically analyzed the influential factors of both short- and long-term LID outcomes following STN-DBS,^10,18 and clinicians lack reliable evidence to optimize patient selection and postoperative counseling.¹⁸ Furthermore, while STN-DBS is known to improve motor symptoms, its long-term effects on neuropsychological outcomes and cognitive function, particularly in patients with preexisting LID, remain poorly characterized.^6,21,24 Current evidence is also constrained by methodological limitations, including small sample sizes, unblinded outcome assessments, and heterogeneous evaluation protocols, which compromise the generalizability and reliability of findings.^15,20,28

To address these knowledge gaps, we conducted a rigorous retrospective cohort study with extended follow-up (1-year and 3-year intervals) in 84 consecutive PD patients with medically refractory LID who underwent STN-DBS. Using validated standardized rating scales (Unified Dyskinesia Rating Scale [UDysRS]²⁴) with single-blind assessment and analyses (observer-blinded) to minimize rater bias,²⁹ we comprehensively evaluated longitudinal trajectories of LID symptoms, motor symptoms, quality of life neuropsychological status, and cognitive function and investigated potential prognostic factors that may influence short- and long-term dyskinesia outcomes. By combining a robust study design with a substantial sample size, this research offers valuable insights into the outcomes of STN-DBS for patients with PD and LID and identifies key prognostic factors, thereby contributing to informed clinical decision-making and future practice improvements.¹⁶

Methods

Patients and Eligibility Criteria

This single blind longitudinal retrospective cohort study analyzed prospectively collected clinical data that were entered into a structured database at the time of patient visits.¹⁴ All patients with PD and LID who underwent STN-DBS at Beijing Tiantan Hospital, Capital Medical University, from January 2019 to December 2021 were consecutively screened and evaluated. Patients were included in this study if they met the following conditions: (1) diagnosed with idiopathic PD according to the Movement Disorders Society (MDS) Clinical Diagnostic Criteria³⁰ and the UK Brain criteria³¹ with the presence of disabling LID symptoms (with typical clinical symptoms such as levodopa-induced involuntary chorea/choreoathetoid movements⁶ and a total score greater than 0 on the UDysRS²⁴) not optimized with anti-PD medications, confirmed by at least two qualified senior movement disorder specialists; (2) received bilateral STN-DBS between January 2019 and December 2021 at Beijing Tiantan Hospital, Capital Medical University; and (3) without dementia, severe psychiatric disorders, severe brain atrophy or cerebral ischemic lesions, or systemic diseases that interfered with the surgery. The exclusion criteria were (1) patients without written informed consent; (2) patients with major missing baseline or follow-up data; or (3) patients who finished less than 3 years of follow-up. Patients with missing data or insufficient follow-up were contacted first for further review; those unreachable due to changed contact details were classified as lost to follow-up. The data collection process was finished by December 2024.

Surgical Procedure

The surgical procedure for DBS implantation has been previously described in detail in our previous studies.^32,33 On the basis of the patients’ symptoms, all included patients were bilaterally implanted with electrodes during the same surgery. In general, on the day of surgery, patients were mounted with a Leksell-G stereotactic frame (Elekta Instrument AB, Stockholm, Sweden) for a 0.625-mm consecutive CT scan (Revolution CT Scanner, GE Healthcare, Chicago, Illinois), which was then merged with preoperative 3-T MR images (MAGNETOM Verio MRI Scanner, Siemens, Berlin, Germany) to visually target the STN coordinates and trajectory. After that, the stimulation electrodes (Medtronic-3389, Medtronic, Minneapolis, Minnesota; or PINS-L301, PINS Medical, Beijing, China; with the same designs as quadripolar contacts, length of 1.5 mm, spaced 0.5 mm apart) were implemented with microelectrode recordings and intraoperative high-frequency macrostimulation testing for further target refinement under local anesthesia. An implantable pulse generator (IPG; Activa RC, Medtronic, Minneapolis, Minnesota; or Model G102R, PINS, PINS Medical, Beijing, China) was then implanted and connected with the intracranial leads through subcutaneous extension wires positioned in the left subclavian region on the same day under general anesthesia.

A postoperative CT scan was performed to confirm the lead position and to evaluate potential surgical complications. The IPG was initiated approximately 1 month later to avoid the microthalamotomy effect.²⁹ Patients were tested in unipolar mode first (with lead contact as the cathode [-] and IPG as the anode [+]). In the case of intolerable side effects, it was switched into bipolar mode (different contacts at the same electrode as the cathode and anode). A specialized multidisciplinary team was responsible for the programming and further adjustments in the follow-ups as needed to achieve the best optimization of therapeutic and adverse effects.

Clinical Assessment

All patient baseline clinical records, including sex, age at surgery, disease duration, and age at onset (those who had an age at onset < 40 years were defined as having young-onset PD^34,35), were recorded. In addition, LID and PD phenotypes were further classified by experienced movement specialists. The LID symptoms were categorized as peak-dose dyskinesia, diphasic dyskinesia, or off-period dystonia according to the relationships between their occurrence time and levodopa treatment as well as their typical symptoms.^6,8,36 One or several different types of LID can occur in the same patient.^8,36 Motor subtypes of PD were classified as tremor dominant (TD), postural instability/gait difficulty (PIGD), or indeterminate on the basis of the MDS-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS; TD/PIGD score ratio ≥ 1.15 for TD, ratio ≤ 0.90 for PIGD, and others for indeterminate).^37,38

All postoperative evaluations were performed by independent, trained experienced neuropsychologists who were blinded to the patients’ baseline clinical data, stimulation parameters, and the specific aims of the study (observer-blinded design). Patients were evaluated with validated standardized Chinese versions of assessment scales at (preoperative) baseline and each year postoperatively. The outcomes of short-term (1 year) and long-term (3 years) follow-up were collected for analysis. For motor evaluations, movement disorder specialists evaluate the clinically defined “off-medicine” and “on-medication” conditions. The “off-medication” state was evaluated after at least a 12-hour period without medication, whereas the “on-medication” state was assessed approximately one hour after medication intake, when its benefits were at the peak.^12,18,39 The levodopa responsiveness (%) was calculated for the included patients as a percentage of part III of the MDS-UPDRS “off-medicine” score minus the “on-medicine” score divided by the “off-medicine” score at baseline.⁴⁰ Postoperative outcomes were assessed during a “stimulation-on” phase after a minimum of one week of continuous stimulation with no parameter adjustments under the abovementioned “off-medicine” and “on-medication” conditions.^18,39

Surgical Outcomes

The primary outcome of this study was dyskinesia outcomes, which were assessed by the UDysRS (maximum 104 points, higher scores for more severe symptoms of dyskinesia).^41,42 We also calculated patients’ time spent with LID on the basis of the sum of the MDS-UPDRS item 4.1 (total hours with dyskinesia other than off-period dystonia) and the MDS-UPDRS item 4.6 (total hours with off-period dystonia) and evaluated the functional impact of LID on the basis of the MDS-UPDRS item 4.2 (scores ranging from 0--4, with higher scores indicating a more severe impact of LID on the patient’s daily function).⁴³

We also evaluated other motor outcomes, quality of life, neuropsychological status, and cognitive function as secondary outcomes. Assessments for motor outcomes included part III (motor examination) of the MDS-UPDRS (MDS-UPDRS-III; maximum 132 points, higher scores for more severe motor symptoms) in the “off-medicine” and “on-medicine” conditions,^43,44 part II (daily living) of the MDS-UPDRS (MDS-UPDRS-II; maximum 52 points, higher scores for more severe impact on daily living activities),^43,44 part IV (motor complications) of the MDS-UPDRS (MDS-UPDRS-IV; maximum 24 points, higher scores for more severe motor complications),^43,44 the Hoehn & Yahr scale (stages 1–5, higher stages for more severe impairment) in the “off-medicine” condition,⁴⁵ and the levodopa equivalent daily dose (LEDD, calculated by the latest 2023 MDS conversion formulae).⁴⁶

The quality of life of the patients was evaluated via the Parkinson Disease Questionnaire-39 (PDQ-39; maximum 156 points, higher scores for worse health-related quality of life).⁴⁷ In addition, neuropsychological status was assessed by part I (nonmotor experiences) of the MDS-UPDRS (MDS-UPDRS-I; maximum 52 points, higher scores for more severe nonmotor symptoms),^43,44 the Hamilton Anxiety Scale (HAM-A; maximum 56 points, higher scores for more severe anxiety symptoms)⁴⁸ and the Hamilton Depression Scale (HAM-D; maximum 76 points, higher scores for more severe depression symptoms),⁴⁹ and cognitive function was evaluated by the Chinese Mini-Mental Status Exam (CMMS; maximum 30 points; lower scores for possible cognitive impairment)⁵⁰ and the Montreal Cognitive Assessment (MoCA; maximum 30 points; lower scores for possible cognitive impairment). Any adverse events (AEs), including those related to the surgery, stimulation, or device, and their prognoses were also recorded.

Statistical Analysis

The assumption of a normal distribution was tested with the Kolmogorov‒Smirnov test, and Levene’s test was used for homogeneity of variance. On the basis of the variable types and distributions, means (± standard deviations, SDs) or numbers (percentages) were used for descriptive statistics. Given the potential clinical relevance of all variables included in the regression model, we employed complete-case analysis (listwise deletion) rather than data imputation to maintain analytical integrity and avoid potential bias from estimated values.

A series of one-way repeated-measures ANOVAs (or Friedman test, for ordinal variables such as Hoehn & Yahr scales), including time as a within-subjects factor (3 levels: baseline [preoperative], 1 year postoperative [short-term follow-up], and 3 years postoperative [long-term follow-up]), with post hoc pairwise comparisons adjusted by Bonferroni correction, was performed to analyze whether there were significant differences among the different periods. Furthermore, to identify potential influential factors for short-term (1 year) and long-term (3 years) dyskinesia outcomes, multivariable linear regression was performed. Sphericity was tested with Mauchly’s test; when violated, the Greenhouse-Geisser correction was applied. As our dataset was balanced across the three time points, repeated-measures ANOVA was appropriate; however, future analyses could consider linear mixed-effects models to handle potential missing data and intra-subject variability more flexibly.

All observations per independent variable (EPV) were confirmed with an adequate sample size (all EPVs ≥ 10). Univariable linear regression was performed first, and those variables with P < 0.10 were then entered into the multivariable linear regression (backward). The model-correlated independent variables were excluded to avoid multicollinearity. The β coefficient and its 95% confidence interval (CI), as well as the standardized β coefficient, were reported for the independent factors. Sensitivity analysis was performed by switching the regression model into enter method. Collinearity was assessed using variance inflation factors (VIF); all VIF values were <5, confirming the absence of harmful multicollinearity. Model diagnostics included inspection of residual plots and normality tests.

All the statistical tests were two-tailed, and P (or Bonferroni correction adjusted P) < 0.05 was regarded as statistically significant. All the statistical analyses and figures were generated via SPSS (IBM, Armonk, New York; version 27) and GraphPad Prism (GraphPad Software, San Diego, California; version 10).

Results

Baseline Characteristics

A total of 105 patients who were diagnosed with PD and disabling LID and who underwent bilateral STN-DBS surgery within the enrollment period were retrieved and evaluated for eligibility. After the exclusion of 21 patients because of a lack of informed consent (n = 3), missing baseline data (n = 5), and less than 3 years of follow-up (n = 13; 10 patients with missing follow-up data, 1 patient who died with no relationship with the surgery, and 2 patients who were explanted from the IPG due to infection and finished less than 3 years of follow-up), 84 patients (39 females, 46.4%) were ultimately included in the analysis. All included patients were confirmed to have stimulation electrodes correctly placed in the bilateral STN through neuroimaging. No known surgical-related mortality or morbidity was reported during the follow-up period. Figure 1 shows the details of the selection and inclusion of the patients.

Figure 1 Flow chart of the study and selection and inclusion of patients. Abbreviations: PD, Parkinson’s disease; LID, levodopa-induced dyskinesia; STN-DBS, subthalamic nucleus deep brain stimulation.

The baseline clinical characteristics of the included patients are described in Table 1 and baseline assessments are shown in Table 2. After a mean disease duration of 10.30 ± 3.02 years, these patients underwent surgery at a mean age of 61.89 ± 6.69 years. The mean age at PD onset was 51.59 ± 6.82 years, and 15.5% (13) of the patients were considered to have young-onset PD. For LID symptoms, 79.8% (67), 21.4% (18), and 26.2% (22) of the patients experienced peak-dose dyskinesia, diphasic dyskinesia, and off-period dystonia, respectively. The baseline mean score of the UDysRS scale was 26.73 ± 17.02, the mean duration of LID was 5.36 ± 2.73 hours, and the mean score of the functional impact of LID was 2.44 ± 0.95. With respect to other motor symptoms of PD, 25.0%, 61.9%, and 13.1% of the patients were considered TD, PIGD, and indeterminate phenotypes, respectively. The mean scores of the MDS-UPDRS-III (off-medicine), MDS-UPDRS-III (on-medicine), MDS-UPDRS-II, and MDS-UPDRS-IV were 52.65 ± 14.84, 22.63 ± 8.81, 20.12 ± 5.14, and 8.38 ± 3.89, respectively. In addition, 7.1% [6], 21.4% [18], 67.9% [57], and 3.6% [3] of patients were assessed with Hoehn & Yahr stages 2, 2.5, 3, and 4, respectively. The mean LEDD was 854.16 ± 387.24. In addition, other baseline assessments suggested mean scores of 62.95 ± 25.68 for the PDQ-39 for quality of life; 15.14 ± 4.39, 17.27 ± 8.93, and 16.18 ± 8.51 for the MDS-UPDRS-I, HAM-A, and HAM-D for neuropsychological status; and 27.10 ± 2.05 and 26.30 ± 2.07 for the CMMS and MoCA for cognitive function, respectively. No severe psychiatric symptoms or cognitive decline were observed in the current cohort.

Table 1 Baseline Demographic and Clinical Characteristics of the Included Patients (n = 84) with Parkinson’s Disease (PD) and Levodopa-Induced Dyskinesia (LID) Who Underwent Subthalamic Nucleus Deep Brain Stimulation (STN-DBS)

Table 2 Short (1 Year) and Long (3 Years) Term Dyskinesia Outcome, Other Motor Outcomes, Quality of Life, Neuropsychological Outcomes, and Cognitive Functions of the Included Patients (n = 84) with Parkinson’s Disease (PD) and Levodopa-Induced Dyskinesia (LID) Who Underwent Subthalamic Nucleus Deep Brain Stimulation (STN-DBS)

Surgical Outcomes

The surgical outcomes are shown in Table 2 and are illustrated in Figure 2 (for the scores of the original scales) and Figure 3 (for the relative changes). Overall, there were significant differences among the baseline, 1-year, and 3-year follow-ups regarding dyskinesia, motor, neuropsychological outcomes, and quality of life (all P < 0.001), while cognitive functions were not significantly changed (all P > 0.05). The UDysRS, time spent with LID, and functional impact of LID improved significantly by 74.4%, 62.5%, and 64.3%, respectively, from baseline to short-term follow-up (all P < 0.001), and the improvement remained significant (all P < 0.001) after long-term follow-up (improved 64.9%, 48.0%, and 52.5% from baseline for the UDysRS, LID time, and LID impact, respectively). However, these benefits declined significantly from short- to long-term follow-up by 36.9% (P = 0.012), 38.8% (P = 0.007), and 33.3% (P < 0.001) in the scores/hours of UDysRS, LID time, and LID impact, respectively.

Figure 2 Short-term (1 year) and long-term (3 years) dyskinesia outcomes, other motor outcomes, quality of life, neuropsychological outcomes, and cognitive functions of the included patients (n = 84) with Parkinson’s disease (PD) and levodopa-induced dyskinesia (LID) who underwent subthalamic nucleus deep brain stimulation (STN-DBS). (A) Dyskinesia outcomes such as UDysRS (A-i), time spent with LID (MDS-UPDRS-4.1), hours (A-ii), and functional impact of LID (MDS-UPDRS-4.2; A-iii); (B) Other motor outcomes such as MDS-UPDRS-III (off-medicine; (B-i)), MDS-UPDRS-III (on-medicine; (B-ii)), MDS-UPDRS-II (daily living, (B-iii)), MDS-UPDRS-IV (motor complications; (B-iv)), Hoehn & Yahr (off-medicine; (B-v)), and LEDD, mg (B-vi); (C) Quality of life as PDQ-39; (D) Neuropsychological outcomes such as MDS-UPDRS-I (nonmotor experiences; (D-i), HAM-A (D-ii), and HAM-D (D-iii); and (E) Cognitive function as CMMS (E-i) and MoCA (E-ii). *P < 0.05 (one-way repeated-measures ANOVA, with post hoc pairwise comparisons adjusted by Bonferroni correction); **P < 0.01 (one-way repeated-measures ANOVA, with post hoc pairwise comparisons adjusted by Bonferroni correction). Abbreviations: n.s., not significant. PD, Parkinson’s disease; LID, levodopa-induced dyskinesia; STN-DBS, subthalamic nucleus deep brain stimulation; UDysRS, Unified Dyskinesia Rating Scale; MDS-UPDRS, Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale; LEDD, levodopa-equivalent daily dose; PDQ-39, Parkinson Disease Questionnaire-39; HAM-A, Hamilton Anxiety Rating Scale; HAM-D, Hamilton Depression Rating Scale; CMMS, Chinese Mini-Mental Status Examination; MoCA, Montreal Cognitive Assessment.

Figure 3 Relative changes (%) from baseline to short (1 year) term, baseline to long (3 years) term, and short (1 year) term to long (3 years) term dyskinesia outcomes; other motor outcomes; quality of life; neuropsychological outcomes; and cognitive functions of the included patients (n = 84) with Parkinson’s disease (PD) and levodopa-induced dyskinesia (LID) who underwent subthalamic nucleus deep brain stimulation (STN-DBS). (A) Dyskinesia outcomes such as UDysRS (A-i), time spent with LID (MDS-UPDRS-4.1), hours (A-ii), and functional impact of LID (MDS-UPDRS-4.2; (A-iii)); (B) Other motor outcomes such as MDS-UPDRS-III (off-medicine; (B-i)), MDS-UPDRS-III (on-medicine; (B-ii)), MDS-UPDRS-II (daily living, (B-iii)), MDS-UPDRS-IV (motor complications; (B-iv)), Hoehn & Yahr (off-medicine; (B-v)), and LEDD, mg (B-vi); (C) Quality of life as PDQ-39; (D) Neuropsychological outcomes such as MDS-UPDRS-I (nonmotor experiences; (D-i)), HAM-A (D-ii), and HAM-D (D-iii); and (E) Cognitive function as CMMS (E-i) and MoCA (E-ii). *P < 0.05 (one-way repeated-measures ANOVA or Friedman test, as appropriate, with post hoc pairwise comparisons adjusted by Bonferroni correction); **P < 0.01 (one-way repeated-measures ANOVA or Friedman test, as appropriate, with post hoc pairwise comparisons adjusted by Bonferroni correction). Abbreviations: n.s., not significant. PD, Parkinson’s disease; LID, levodopa-induced dyskinesia; STN-DBS, subthalamic nucleus deep brain stimulation; B/L, baseline; UDysRS, Unified Dyskinesia Rating Scale; MDS-UPDRS, Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale; LEDD, levodopa-equivalent daily dose; PDQ-39, Parkinson Disease Questionnaire-39; HAM-A, Hamilton Anxiety Rating Scale; HAM-D, Hamilton Depression Rating Scale; CMMS, Chinese Mini-Mental Status Examination; MoCA, Montreal Cognitive Assessment.

Similar findings were also revealed regarding other motor outcomes. While MDS-UPDRS-III (off-medicine), MDS-UPDRS-III (on-medicine), MDS-UPDRS-II, MDS-UPDRS-IV, and LEDD suggested significant improvements in both short-term (by 50.9%, 37.7%, 41.0%, 36.6%, and 53.5%, respectively) and long-term (37.7%, 18.2%, 26.8%, 19.0%, and 35.7%, respectively) follow-up, their effects also worsened significantly by 27.0%, 31.4%, 24.0%, 27.9%, and 38.4% in the scores (all P < 0.05), respectively. The Hoehn & Yahr stages also significantly improved from baseline to 1-year (P < 0.001) and 3-year (P = 0.004) follow-up, with also diminished benefits with time (P = 0.049). A post‑hoc analysis revealed a significant positive correlation between the increase in LEDD and the worsening of the UDysRS total score from 1 year to 3 years (r = 0.34, P = 0.002; Pearson’s correlation), suggesting that the requirement for higher levodopa doses partially contributes to the attenuation of antidyskinetic benefit over time.

In terms of quality of life, the PDQ-39 score significantly improved by 41.3% (P < 0.001) from baseline at the short-term follow-up, and after the score decreased by 22.4% (P = 0.005), it improved by 22.4% (P < 0.001) after long-term follow-up. The outcomes of neuropsychological status revealed significant improvements in nonmotor symptoms of PD (assessed by the MDS-UPDRS-I), anxiety symptoms (assessed by the HAM-A), and depression symptoms (assessed by the HAM-D) at both short-term (improved by 18.2%, 39.0%, and 43.3%, respectively) and long-term (improved by 13.6%, 27.7%, and 27.5%, respectively) follow-ups (all P < 0.05), and short- and long-term outcomes were comparable (all P > 0.05). With respect to cognitive function (assessed by the CMMS and MoCA), no significant differences were revealed among baseline and different follow-up periods (all P > 0.05).

All AEs are reported in Supplementary Table S1. During the follow-up period, the most common surgery-related AE was infection (3.6%, n = 3), whereas the most common stimulation-related AEs were dyskinesia (6.0%, n = 5) and insomnia (6.0%, n = 5). All AEs were successfully resolved; infections were managed with surgical debridement and, in two cases, IPG explantation, while stimulation induced AEs were controlled through reprogramming. One patient experienced a small intracerebral hemorrhage that resolved without sequelae. No surgical mortality or permanent morbidity occurred. Patients who underwent IPG explantation were not included in the 3-year efficacy analysis due to incomplete follow up.

Prognostic Factors

Univariate analyses revealed the following preoperative clinical factors that potentially influence short-term dyskinesia outcomes: LID type, LEDD, and levodopa responsiveness (Supplementary Table S2). After multivariable linear regression models were adjusted, having diphasic dyskinesia (adjusted β = −16.932, 95% CI: −28.331–5.533, standardized β = −0.296, P = 0.004) was a negative factor for short-term dyskinesia outcomes, whereas a higher LEDD (adjusted β = 0.017, 95% CI: 0.005 to 0.029, standardized β = 0.275, P = 0.008) and greater levodopa responsiveness (adjusted β = 0.227, 95% CI: 0.016 to 0.438, standardized β = 0.215, P = 0.035) at baseline were independent positive prognostic factors (Table 3).

Table 3 Multivariable Linear Regression for Influential Factors of Short (1 Year) Term Dyskinesia Outcome for the Included Patients (n = 84) with Parkinson’s Disease (PD) and Levodopa-Induced Dyskinesia (LID) Who Underwent Subthalamic Nucleus Deep Brain Stimulation (STN-DBS)

With respect to long-term follow-up, disease duration, LID type, LEDD, and levodopa responsiveness were identified as potential factors in the univariable analyses (Supplementary Table S3). A multivariate linear regression model confirmed that longer disease duration (adjusted β = −1.012, 95% CI: −1.995 to −0.029, standardized β = −0.212, P = 0.044) and diphasic dyskinesia (adjusted β = −13.607, 95% CI: −24.763 to −2.451, standardized β = −0.239, P = 0.017) were associated with worse dyskinesia outcomes, whereas higher LEDD (adjusted β = 0.016, 95% CI: 0.004 to 0.028, standardized β = 0.261, P = 0.009) and greater levodopa responsiveness (adjusted β = 0.226, 95% CI: 0.011 to 0.441, standardized β = 0.216, P = 0.040) were associated with better dyskinesia outcomes (Table 4). Collinearity diagnostics revealed VIF values ranging from 1.03 to 1.41. Sensitivity analyses using forced entry models yielded consistent results.

Table 4 Multivariable Linear Regression for Influential Factors of Long (3 Years) Term Dyskinesia Outcome for the Included Patients (n = 84) with Parkinson’s Disease (PD) and Levodopa-Induced Dyskinesia (LID) Who Underwent Subthalamic Nucleus Deep Brain Stimulation (STN-DBS)

Discussion

The UDysRS provides a comprehensive method for evaluating dyskinesia symptoms in PD patients with excellent clinical properties;^41,51 it has been recommended by the MDS and widely verified among different PD groups and for assessing surgical outcomes.^24,52,53 Juhász et al analyzed 71 patients with PD and LID who underwent bilateral STN-DBS and revealed that the total UDysRS score significantly decreased from 38.0 ± 17.8 to 10.8 ± 13.0 (improved by approximately 71.6%) at 1 year after follow-up and suggested that UDysRS could reliably detect surgical improvements.²⁴ Fan et al revealed that STN-DBS significantly improved the UDysRS score by 60.73 ± 40.29% after a mean 21.60 ± 8.79 months of follow-up in 20 PD patients with preoperative LID.²¹ A cohort study with a longer follow-up period suggested that dystonia and peak-dose dyskinesia could be improved even after 5 years of STN-DBS in 33 patients, as evaluated by the UPDRS part IV and the CAPSIT-PD scales.⁵⁴ This study documented 74.4% and 64.9% dyskinesia improvement at 1-year and 3-year follow-ups respectively in PD patients with LID, utilizing standardized UDysRS assessments to fill an important research gap in longitudinal STN-DBS outcomes. In summary, our results align with those of prior studies suggesting STN-DBS as a cornerstone therapy for patients with PD and preexisting LID in mitigating dyskinesia and motor fluctuations for a long-term period.

However, the observed decline in benefits over time, as evidenced by a 36.9% rebound in dyskinesia outcomes, echoes concerns raised by Limousin et al¹⁶ and others regarding the progressive nature of PD, adaptations of the stimulation, and increasing levodopa dosage again.^{16,27,28,55,56} Compared with the 5-year outcomes reported by Simonin et al⁵⁴ and the randomized comparison by Follett et al⁵⁷ our 3-year data show a comparable trend of early, substantial benefit with subsequent partial decline, underscoring the need for continuous multidisciplinary management. Notably, in addition to the improvements in dyskinesia outcomes over time, scores on the MDS-UPDRS-III scale worsened faster under on-medicine conditions than under off-medicine conditions (31.4% vs 27.0%, respectively), even with a 38.4% increase in LEDD. This may reflect that the resurgence of levodopa requirements could only partially explain the decline in dyskinesia outcomes. Another potential cause could be the degeneration of nondopaminergic pathways, such as the cholinergic system at the midbrain, forebrain, and cortical level,^58,59 in the development of levodopa refractory or stimulation-resistant symptoms with the progression of PD.¹⁶ This highlights the complex interplay between disease progression, neurostimulation adaptation, and pharmacological management.^16,27,28

Furthermore, the sustained improvements in anxiety and depression scores underscore the stable psychological benefits of STN-DBS, which are likely mediated through motor symptom alleviation and increased psychosocial functioning. However, although current study suggested non statistical significance in changes of cognitive function, these results must be interpreted with caution, as the CMMS and MoCA are screening tools that are susceptible to ceiling effects and may not detect subtle declines in specific cognitive domains such as verbal fluency, executive function, or working memory.^60,61 The absence of a significant change in these scores should therefore not be taken as definitive evidence of cognitive preservation. Dedicated neuropsychological batteries in future studies are essential to fully elucidate the cognitive effects of chronic STN DBS. Previous studies presented inconsistent findings regarding the cognitive effects of DBS in PD patients, with some studies reporting stable cognitive function^14,62,63 while others observe mild cognitive decline (especially long-term memory, verbal fluency, and phonemic fluency).^64,65 These discrepancies may be attributed to differences in assessment tools, potential ceiling effects, natural age-related cognitive deterioration, and the potential pathological changes associated with PD.^55,66 A prospective controlled study is still needed to better understand the cognitive effects of DBS for PD, which can be a future research direction.

Diphasic dyskinesia was identified as a negative prognostic marker for both short- and long-term dyskinesia outcomes. Managing medication for diphasic dyskinesia can be difficult because simply reducing the dose of levodopa may not be sufficient to alleviate the symptoms caused by fluctuations in drug concentration and changes in dopamine receptor sensitivity.^67,68 Therefore, for patients with diphasic dyskinesia, fractionation of the levodopa dosage and combination with other types of medications, such as COMT inhibitors, MAO-B inhibitors, dopamine agonists, and apomorphine, should be carefully considered after STN-DBS surgery.^5,8,67,69 In addition, long disease duration was also identified as a negative factor of long-term dyskinesia outcomes, which might be related to the long-term reduction in dopaminergic receptor sensitivity.⁶ Kim et al also suggested that the disease duration at surgery was longer in patients whose diphasic dyskinesia persisted after STN-DBS than in those whose symptoms disappeared.⁷⁰ They also suggested that younger patients at onset had worse outcomes,⁷⁰ which was not observed in the present study. This might be related to the fact that the present study included only a limited number of patients with young-onset PD (15.5%). Conversely, higher baseline LEDD and levodopa responsiveness emerged as positive predictors. Patients with higher baseline LEDDs tend to have greater postoperative decreases in medications, thus resulting in better remission of LID.²⁴ However, it should be reiterated that our analysis tracked total LEDD rather than the adjustments of individual medication classes (eg., amantadine, dopamine agonists) due to unavailable data, which limits the ability to disentangle the specific contribution of each drug to dyskinesia improvement; therefore, future studies should systematically record and analyze medication class specific changes to refine prognostic models.^6,8,67,68 Given the potential influence of medication class adjustments on dyskinesia outcomes,^6,8,67,68 future studies should implement comprehensive medication class tracking to better characterize their effects on individual patient outcomes. Levodopa responsiveness is also regarded as one of the most important indicators of motor outcomes in PD patients after STN-DBS,^13,18,19 which was also revealed in current findings on dyskinesia outcomes.

Furthermore, emerging evidence suggests that globus pallidus interna (GPi)-DBS may offer superior long-term dyskinesia control compared with STN-DBS,^15,16,20,28 potentially due to its direct modulation of pallidal output pathways implicated in hyperkinetic movement generation.^15,20 In contrast, STN-DBS primarily reduces dyskinesia indirectly by enabling significant levodopa dose reduction,^20,21 although some studies have also suggested additional direct antidyskinetic effects.^22,23 However, comparative meta-analyses remain inconclusive,^{15,16,20,28,71} with heterogeneity in patient selection and outcome measures complicating cross-study interpretations. The RCT by Follett et al comparing STN-DBS and GPi-DBS revealed similar improvements in motor function and dyskinesia at 24 months,⁵⁷ but longer follow-up periods are still warranted in the future. In addition, the activation of contact localization and tissue connectivity could influence dyskinesia outcomes,^72,73 suggesting that further refinement of stimulation subregions of the STN may further optimize dyskinesia outcomes.^22,23 For example, Li et al suggested potential direct suppression of LID reported for areas above the STN.²² Furthermore, closed-loop adaptive DBS systems, leveraging real-time biomarkers such as dyskinesia-associated gamma oscillations, could also hold promise for further improvements in dyskinesia outcomes and sustaining therapeutic benefits.^74–76 Future studies integrating target-specific neuroimaging, adaptive algorithms, and mechanistic biomarkers are critical to resolve these controversies and advance personalized neuromodulation strategies.

Limitations

While our study has advanced in the field, several limitations warrant acknowledgment. First, potential selection bias and residual confounding inherent to the observational design may influence the magnitude of the observed associations. Second, attrition caused by missing data or incomplete follow up could influence outcome interpretations; patients who underwent device explantation were not included, which may overestimate the apparent durability of benefit. Complete-case analysis, while transparent, may introduce bias if excluded patients differed systematically from those retained; future intention-to-treat analyses with sensitivity analyses employing multiple imputation methods are advisable in future confirmatory studies. Also, the use of complete-case repeated-measures ANOVA, while valid for our balanced dataset, could be complemented by mixed-effects models in future multicenter studies with missing data. Third, the single-center nature of the study may limit generalizability to other clinical settings with different patient demographics or surgical protocols. Fourth, the absence of a control group prevents direct comparison with natural disease progression under best medical therapy or with other surgical options such as GPi-DBS; therefore, the observed changes should be interpreted as longitudinal associations in a real-world surgical cohort rather than as definitive causal effects of STN-DBS. Additionally, detailed information on active contact locations, chronic stimulation parameters (pulse width, frequency, amplitude), and medication subclasses was not systematically analyzed, which limits our ability to investigate the mechanisms underlying individual differences in dyskinesia outcomes. Future prospective studies integrating imaging-based contact localization and pharmacological profiling are needed. Finally, the current follow-up period is limited to 3 years. Future multicenter, prospective studies with extended follow-up (>5 years) and randomized designs are needed to validate our prognostic models and explore the interplay between stimulation parameters, genetic subtypes, and disease progression.

Conclusions

In this longitudinal cohort, STN-DBS was associated with significant improvements in dyskinesia outcomes, motor function, quality of life, and neuropsychological status over a 3-year period. However, the gradual decline in the therapeutic effects on LID and motor outcomes, thereby worsening quality of life, should also be considered. Furthermore, several key factors, such as LID type, LEDD, levodopa responsiveness, and disease duration, were identified as influencing dyskinesia outcomes. This could improve patient selection for DBS surgery and better address patient counseling. Given the observational, single center design, future randomized controlled trials with extended follow up are warranted to confirm these findings.