Understanding dopaminergic dose reduction following STN-DBS: mediation analysis

Abstract

Background

Levodopa equivalent dopaminergic dose (LEDD) reduction after subthalamic nucleus deep brain stimulation (STN-DBS) in Parkinson’s disease varies widely. Identifying predictors may guide patient selection and programming. Our objectives were to identify predictors of LEDD reduction and to test whether motor improvement mediates this association.

Methods

Data from 144 patients treated by STN-DBS were analysed. Predictors of LEDD reduction were selected using the Boruta algorithm, a machine-learning method comparing variable importance to randomised features and then tested in a structural equation model for direct and motor-mediated effects.

Results

Mean LEDD reduction was 41.7% (±38.2%) and motor improvement was 48.6% (±26.7%) at 1 year. Among the four predictors identified by Boruta, lower baseline LEDD (β=0.39, p=0.001), greater axial impairment (β=−0.25, p=0.003) and higher total volume of tissue activated (β=−0.17, p=0.031) were directly associated with lower LEDD reduction, independent of motor improvement. Sensorimotor STN overlap was not directly linked to LEDD reduction but was positively associated with motor improvement (β=0.34, p=0.001), which showed a trend-level effect on LEDD reduction (β=0.16, p=0.065). The total effect of sensorimotor STN overlap on LEDD reduction was not significant.

Discussion

Dopaminergic dose reduction after STN-DBS is constrained by preoperative axial symptoms and stimulation spread, independently of motor improvement, while sensorimotor STN overlap improves motor symptoms but not dose reduction. Integrating motor phenotype with anatomical guidance may enhance medication management post DBS.

Introduction

Deep brain stimulation of the subthalamic nucleus (STN-DBS) is an established treatment for advanced Parkinson’s disease (PD), improving motor complications and quality of life.¹ A reduction in levodopa equivalent daily dose (LEDD) usually follows STN-DBS,² contributing to the overall clinical benefit by decreasing the risk of dopaminergic complications such as dyskinesias.³

However, postoperative titration of dopaminergic medication remains highly variable,² influenced by both patient and stimulation factors. Anatomically, the STN is a small functionally organised structure comprising sensorimotor, associative and limbic subregions.⁴ Stimulation of the sensorimotor territory is most closely associated with motor improvement.⁵ Moreover, modulation of specific adjacent structures—particularly the zona incerta (ZI), which lies dorsally to the STN—has been implicated in enhanced LEDD reduction.⁶ Conversely, current spread to other neighbouring regions may result in side effects that limit stimulation efficacy and thus restrict the capacity to reduce dopaminergic medication.

We aimed to identify clinical and neurostimulation factors associated with LEDD reduction 1 year after STN-DBS and to test whether motor improvement mediates this association.

Methods

Population and study design

We used data from the PREDI-STIM cohort. Patients were recruited between November 2013 and September 2019 at 17 specialised PD centres from the French NS-PARK/FCRIN network.⁷ Patients with a diagnosis of PD according to the UK Brain Bank criteria⁸ who were eligible and planning to undergo STN-DBS (online supplemental methods) were included. Informed consent was obtained from all patients prior to inclusion.

In this study, we included only patients with available baseline clinical data, neurostimulation parameters and LEDD measurements at both baseline and 1 year after surgery (online supplemental figure S1).

Data collected

Demographic and clinical data were collected at baseline including MDS-UPDRS part I, II, III and IV scores. MDS-UPDRS III was assessed in the off state in the morning after at least 12 hours of cessation of all anti-parkinsonian drugs (off condition), and then at the best on after acute administration of a suprathreshold L-Dopa dose (on condition). For each condition, MDS-UPDRS III total score and four subscores were calculated: Rigidity, Bradykinesia, Tremor and Axial subscores (online supplemental Table S1). Motor fluctuations (MDS-UPDRS IV part 3–5) and dyskinesia (MDS-UPDRS IV part 1–2) subscores were computed. LEDD (milligrams) was calculated at baseline and 1 year post-DBS as previously described.⁹ Calculation of dopa responsiveness, percentage of LEDD reduction and motor improvement after DBS are described in the online supplemental methods.

Lead localisation and VTA were processed using Lead-Group v2.5.3¹⁰ (online supplemental methods, online supplemental figure S1). The DISTAL atlas was used to compute the percentage of overlap (PO) between VTA and STN subregions (sensorimotor, associative, limbic), internal capsule (IC) and ZI.¹¹ Total VTA and bilateral PO were calculated by summing right and left volumes.

Statistical analysis

Continuous variables were summarised as mean±SD, and categorical variables as counts and percentages.

We used a two-step analytical approach. First, the Boruta algorithm (500 iterations) was applied on all clinical and stimulation variables except motor improvement at 1 year to identify all-relevant predictors of LEDD reduction. Boruta is a wrapper algorithm built around a random forest classifier that iteratively selects features based on their importance compared with randomised shadow variables.¹² Variables confirmed as important by Boruta were then included in a structural equation model (SEM) to test both their direct effects on LEDD reduction and potential indirect effects mediated through motor improvement.¹³ Model parameters were estimated by maximum likelihood with 1000 bootstrap iterations, and model fit was evaluated using CFI and SRMR. No imputation was performed for missing data. All analyses were conducted in R, and statistical significance was set at p<0.05 (two-tailed).

Results

Baseline characteristics of the population

Of the 144 patients included (online supplemental figure S2), 69% were male. Mean (SD) age was 60.9 years (±7.0) and mean disease duration was 11.9 years (±4.1). At baseline, MDS-UPDRS III in off condition was 40.5 (±14.0), dopa responsiveness was 73.0% (±13.8) and LEDD was 1312.8 mg (±429.5). One year after STN-DBS, mean LEDD reduction was 639.9 mg (±507.4) corresponding to a 43.7% (±39.9) reduction (ranged from a 100% reduction to a 164.7% increase) (online supplemental table S2). Motor improvement after DBS was, on average, 48.6% (±26.7).

Factors associated with LEDD reduction and mediation analysis

The Boruta feature selection algorithm identified four variables robustly associated with LEDD reduction: LEDD at baseline, axial motor impairment (MDS-UPDRS III axial subscore) at baseline, total VTA and PO with the sensorimotor STN. All other clinical and neurostimulation variables were classified as non-informative by the algorithm (online supplemental figure S3); notably, PO with the limbic STN, associative STN, IC and ZI did not outperform noise in the Boruta feature selection.

We included in the SEM the key predictors identified previously. The model showed lower LEDD reduction in patients with greater axial motor impairment (β=–0.25, p=0.003) at baseline, lower LEDD at baseline (β=0.39, p=0.001) and larger total VTA (β=–0.17, p=0.031) that only result from direct effects. PO with the sensorimotor STN was strongly associated with motor improvement (β=0.34, p=0.001), but neither its direct (β=0.004, p=0.963) nor its indirect effect through motor improvement (β=0.056, p=0.146) on LEDD reduction was significant, resulting in a non-significant total effect (β=0.060, p=0.507) (table 1).

Table 1. Results of structural equation modelling.

Path	β (Standardised)	SE	P value
Direct effects
Total VTA → Motor improvement	0.034	0.090	0.708
PO with sensorimotor STN → Motor improvement	0.349	0.099	0.001
LEDD at baseline → Motor improvement	0.113	0.078	0.146
MDS-UPDRS III axial subscore → Motor improvement	–0.008	0.078	0.919
Motor improvement → LEDD reduction	0.160	0.088	0.065
Total VTA → LEDD reduction	–0.175	0.077	0.023
PO with sensorimotor STN → LEDD reduction	0.004	0.094	0.963
LEDD at baseline → LEDD reduction	0.373	0.110	0.001
MDS-UPDRS III axial subscore → LEDD reduction	–0.245	0.087	0.004
Indirect effects
Total VTA → Motor improvement → LEDD reduction	0.005	0.017	0.742
PO with sensorimotor STN → Motor improvement → LEDD reduction	0.056	0.038	0.146
LEDD at baseline → Motor improvement → LEDD reduction	0.018	0.018	0.305
MDS-UPDRS III axial subscore → Motor improvement → LEDD reduction	–0.001	0.014	0.929
Total effects
Total VTA → LEDD reduction	–0.169	0.079	0.031
PO with sensorimotor STN → LEDD reduction	0.060	0.090	0.507
LEDD at baseline → LEDD reduction	0.391	0.113	0.001
MDS-UPDRS III axial subscore → LEDD reduction	–0.246	0.085	0.003

Direct effects represent unmediated associations between predictors and outcomes in 139 patients. Indirect effects reflect the mediated contribution through motor improvement. Total effects are the sum of direct and indirect paths. Bolded p values indicate statistically significant effects (p<0.05).

LEDD, L dopa equivalent daily dose; MDS-UPDRS, Movement Disorder Society-Unified Parkinson’s Disease Rating Scale; PO, percentage of overlap; STN, subthalamic nucleus; VTA, volumes of tissue activate.

Discussion

This study is the first to investigate both patient and DBS-related factors associated with dopaminergic medication changes following subthalamic STN-DBS in a large real-world cohort, while testing motor improvement as a mediator.

Both the average motor improvement and the reduction in LEDD after STN-DBS were comparable to those reported in previous studies.² Our results give further insights into the factors driving this variability. First, lower LEDD reduction was associated with higher preoperative axial score. It is noteworthy that our cohort selected patients suitable for STN-DBS, thus without prominent dopa-resistant axial symptoms. However, some of the patients had mild (accounting for only 13% of all motor symptoms) and highly sensitive to L-dopa (70% of improvement at baseline) axial symptoms. Even if such subtle symptoms do not represent a contraindication to STN-DBS, they limit the LEDD reduction postoperatively. This reflects the limited responsiveness of axial symptoms to neurostimulation; axial symptoms were solely improved by 37% compared with the 52% motor improvement with axial exclusion. Thus, a higher dose of dopaminergic medication is needed to overcome the poorest response of axial signs to neurostimulation.¹⁴

A larger total VTA was associated with lower LEDD reduction independently of motor improvement and sensorimotor overlap, suggesting that wider current spread may engage non-motor STN regions or trigger side effects limiting dose reduction. Although neurostimulation of ZI has been previously associated with clinical improvement,⁶ neither overlap with the ZI, internal capsule nor limbic or associative STN subregions significantly influenced dopaminergic reduction in our cohort. As expected, sensorimotor STN overlap was strongly associated with motor improvement,⁵ but neither its direct nor indirect effect on LEDD reduction was significant. This result clarifies certain clinical situations in which dopaminergic treatment remains at a high level despite the lead and neurostimulation being well localised in the sensorimotor STN. Although our study was not specifically designed to explore this aspect, our results suggest the existence of an optimal LEDD reduction ‘sweet spot’, that is, an anatomical region to target for optimising clinical effects.

Our study presents several limitations. First, as an observational study, the postoperative adjustment of dopaminergic medication was not standardised across patients but rather left to the clinician’s judgement introducing variability in LEDD reduction outcomes. Nonetheless, our cohort remains representative of STN-DBS patients commonly reported in the literature.² Manual MRI segmentation is subject to anatomical variability, mitigated here by normalisation to MNI space and use of the DISTAL atlas.¹¹ Given the large number of clinical and stimulation variables, a strength of our approach is the use of the Boruta algorithm, an all-relevant feature selection method that aims to recover the full set of predictors carrying signal rather than selecting only a minimal subset.¹² Other confounding factors may also contribute to the reduction in dopaminergic treatment, such as the occurrence of apathy.¹⁵ Finally, the SEM observation-to-parameter ratio (~6:1) is acceptable for exploratory analysis, but may limit statistical power, particularly for detecting indirect or small effects.

Our findings have important clinical implications. Greater axial motor impairment, lower baseline LEDD and larger total VTA were independently associated with less dopaminergic dose reduction, regardless of motor improvement. Thus, LEDD reduction after STN-DBS is not solely driven by motor improvement but also limited by the poor responsiveness of axial symptoms to neurostimulation. Although neurostimulation of the sensorimotor STN had no significant effect on LEDD reduction, its strong association with motor improvement and trend-level mediation suggests a possible indirect contributor. This highlights the value of anatomically guided targeting strategies to optimise motor outcomes and facilitate medication reduction.

Data availability statement

Data are available upon reasonable request.