Deep brain stimulation for Parkinson’s disease: systematic review with meta-analysis and trial sequential analysis

Search strategy and selection criteria

A full time search specialist searched the Cochrane Central Register of Controlled Trials (CENTRAL) in the Cochrane Library, Medical Literature Analysis and Retrieval System Online (Medline Ovid), Excerpta Medica database (Embase Ovid), Latin American and Caribbean Health Sciences Literature (LILACS; Bireme), CINAHL (EBSCO host), BIOSIS (Web of Science), Science Citation Index Expanded (Web of Science), Chinese Biomedical Literature Database (CBM), Conference Proceedings Citation Index-Science (CPCI-S; Web of Science), China Network Knowledge Information (CNKI), Chinese Science Journal Database (VIP), and Wanfang Database, to identify relevant randomised clinical trials. We searched all databases from their inception to 9 May 2023. Online supplemental text S2 shows details of the search strategies.

Reference lists of relevant trial publications were checked for unidentified randomised clinical trials. To identify unpublished trials, we searched clinical trial registries in Europe, the US, and China, and websites of pharmaceutical companies, the US Food and Drug Administration, and European Medicines Agency.

We included randomised clinical trials irrespective of publication year, status, and language. We included crossover trials with only data from the first period of the trial. Participants in all age groups with a diagnosis of Parkinson's disease (as defined by trialists) were included. Participants were included irrespective of sex, comorbidities, and risk factors.

We included any type of deep brain stimulation as the intervention (as defined by trialists), independent of the target of the electrodes, stimulation settings, unilateral or bilateral, and device used. As control interventions, we accepted no intervention, usual care, sham stimulation, medical treatment, or resective surgery. We accepted any cointerventions if these were planned to be delivered similarly in the experimental and control groups.

Data extraction and risk of bias assessment

Two authors (JJP and SJ) independently screened relevant trials. Two authors (JJP and PF) independently extracted data with a standardised data extraction sheet and assessed risk of bias based on the Cochrane Risk of Bias tool, version 2 (RoB 2).17 18 Discrepancies were resolved by discussion or, if required, by discussion with a third author (JCJ). We contacted corresponding authors if relevant data were unclear or missing.

Outcomes and subgroup analyses

Primary outcomes were all cause mortality, serious adverse events (as defined by the International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use Guideline for Good Clinical Practice (ICH-GCP)),19 and disease specific symptoms (online supplemental table S1).15 19 Secondary outcomes were quality of life, depressive symptoms, executive functioning, level of functioning, and non-serious adverse events (online supplemental table S1). We classified non-serious adverse events as any adverse event not assessed as serious, according to the ICH-GCP definition.19 For all outcomes, we used the trial results reported at maximum follow-up. We also planned subgroup analyses on overall risk of bias, vested interests, target nucleus, subtype of disease, and severity of disease.15 We chose to add follow-up time and disease duration as post hoc subgroup analyses.

Assessment of statistical and clinical significance

We performed our aggregate data meta-analyses according to Cochrane,17 Keus et al,20 and the eight step assessment by Jakobsen et al21 for better validation of meta-analytic results in systematic reviews. We used risk ratios for dichotomous outcomes. We assessed three primary outcomes for each comparison, and therefore adjusted our thresholds for significance21; we used a P value of ≤0.025 as the threshold for significance.21 Because we mainly considered the results of the secondary outcomes as hypothesis generating, we did not adjust the P value thresholds for secondary outcomes. We conducted random effects (inverse variance and DerSimonian-Laird) and fixed effect (Mantel-Haenszel) meta-analyses for all analyses and mainly reported the most conservative result (highest P value). Less conservative results were considered in a sensitivity analysis.17 21 22 Primary outcomes (all cause mortality and serious adverse events) were also analysed with beta binomial regression models because of multiple zero event arms.

We used trial sequential analysis to control random errors.23–31 Trial sequential analysis estimates the diversity adjusted required information size (DARIS), which is the number of participants needed in a meta-analysis to detect or reject a specific intervention effect. Statistical heterogeneity was quantified by calculating heterogeneity (I², τ²) for traditional meta-analyses, and diversity (D²) for trial sequential analysis.28 We used Grading of Recommendations Assessment, Development, and Evaluation (GRADE) to assess the certainty of evidence.32 We downgraded imprecision in GRADE by two levels if the accrued number of participants was <50% of DARIS and one level if 50-100% of DARIS.21 We did not downgrade if benefit, harm, futility, or DARIS was reached. We used Fisher's exact test to calculate P values if the results were from only one trial. Stata 17 was used for all statistical analyses.33

Patient and public involvement

Patients and/or the public were not involved in the design, or conduct, or reporting of this research. However, we are committed to a transparent and comprehensive dissemination of our research results. We aim to inform patients, caregivers, clinicians, and other relevant stakeholders through a variety of channels, including conferences and social media platforms. By actively sharing our findings through these channels, we aim to foster an informed community, encourage ongoing dialogue, and enhance the practical application of our research in clinical settings.

Trial characteristics

Our literature searches (up to 9 May 2023) identified 29 269 records after duplicates were removed. We included 12 clinical trials, with 1564 participants randomised to the intervention or control.34–45 The primary comparison was deep brain stimulation with antiparkinsonian drug treatment versus antiparkinsonian drug treatment only (seven trials).34–40 Other comparisons were deep brain stimulation with antiparkinsonian drugs versus surgery with sham stimulation and antiparkinsonian drugs (three trials)43–45 or resective surgery with antiparkinsonian drugs (two trials) (figure 1 and online supplemental table S2).41 42 We identified two more trials of deep brain stimulation versus surgery with sham stimulation for dementia46 and balance.47 In these trials, six participants were randomised but no relevant data were reported and we could not obtain data from the authors; these studies were not included in our analyses.46 47

Request permission

For participants receiving deep brain stimulation, the electrodes were placed bilaterally34–45 or unilaterally38 41 in the subthalamic nucleus,34 36 39 40 42–45 subthalamic nucleus or globus pallidus pars interna,37 38 caudal zona incerta,35 or ventral intermediate thalamic nucleus.41 The stimulation parameters (voltage, frequency, and pulse width) were continuously adjusted after implantation of the electrodes.34–45 In both the experimental and control groups, antiparkinsonian drug treatment was continuously adjusted for each participant.34–40

Online supplemental table S2 shows the characteristics of the included trials. All trials had a high risk of bias (online supplemental table S3). The maximum follow-up time ranged from three months43–45 to two years after randomisation.39 40 Figure 2 has a visual overview of all of the results, including the GRADE assessments.

Request permission

Most trials assessed symptom scores after antiparkinsonian drug treatment was paused in all participants.34–45 We mainly report assessment of symptom scores when participants received antiparkinsonian drugs because most patients attending everyday clinics will receive drugs continuously. The supplementary material (online supplemental text S3, online supplemental figures S1 and S2) shows the full results of subsections of the Unified Parkinson’s Disease Rating Scale (UPDRS) score, when participants received antiparkinsonian drug treatment and when drug treatment was paused, analyses of secondary outcomes, and overall analyses of all cause mortality and serious adverse events, including all comparisons.

Deep brain stimulation with antiparkinsonian drugs versus antiparkinsonian drugs

We identified seven trials where 1125 participants were randomised to receive deep brain stimulation with antiparkinsonian drug treatment versus antiparkinsonian drug treatment only.34–40 At follow-up, the difference in levodopa equivalent daily dosage was −190 to −650 mg/day in trials with stimulation in the subthalamic nucleus,34 36 39 40 −318 to −453 mg/day in trials with stimulation in the subthalamic nucleus or globus pallidus pars interna,37 38 and −126 mg/day in the trial with stimulation in the caudal zona incerta.35

Serious adverse events

Seven trials reported serious adverse events at six months,34 35 37 12 months,38 18 months,36 and two years39 40 after randomisation; 195 of 540 participants (36.1%) had a serious adverse event in the intervention groups versus 100 of 556 (18.0%) in the control groups (table 1). One trial reported no events in either group.36 Meta-analysis of the remaining trials indicated evidence of a harmful effect of deep brain stimulation with antiparkinsonian drugs (risk ratio 2.36, 95% CI 1.37 to 4.09; I²=73.7%; τ²=0.24; P<0.01; six trials; Bayes factor 3.43) (figure 3). Visual inspection of the forest plot and measures to quantify heterogeneity (I²=73.7%; τ²=0.24) indicated substantial heterogeneity. This heterogeneity was resolved by conducting the meta-analysis without the study of Schuepbach et a39 (risk ratio 2.92, 95% CI 2.15 to 3.97; I²=0.0%; τ²=0.00; P<0.01; five trials, online supplemental figure 8).

Table 1

View inline|View popup

Overview of serious adverse events for deep brain stimulation with antiparkinsonian drug treatment versus antiparkinsonian drug treatment only, occurring in more than one patient

Request permission

Figure 3

Meta-analysis (risk ratio with 95% confidence interval (CI)) of deep brain stimulation with antiparkinsonian drug treatment versus antiparkinsonian drug treatment only, on serious adverse events34 35 37–40

Trial sequential analysis including all trials showed that we did not have enough information to confirm or reject a relative risk reduction of 10% between groups (not shown). The increased risk of serious adverse events with deep brain stimulation was mainly caused by perioperative complications, such as cerebral haemorrhages and postoperative confusion, and events related to the hardware, such as infection at the stimulator site, dislocation of the device, or reoperations. Also, depression was more frequent in participants receiving deep brain stimulation (11/323 v 1/339, table 1). This outcome result was assessed as having a high risk of bias and the certainty of the evidence was very low (online supplemental table S3 and online supplemental table S4).

Subgroup analyses comparing the effects of different targets of the electrodes (P=0.43) or follow-up time (P=0.10) indicated no evidence of a difference in serious adverse events (online supplemental figures S9 and S10). Subgroup analysis comparing the effects of different lengths of disease indicated evidence of a difference (P<0.01) in serious adverse events (online supplemental figure S11). The remaining predefined subgroup analyses could not be performed because of lack of relevant data. In a sensitivity analysis, we performed beta binomial regression and the results were similar (risk ratio 2.91, 95% CI 2.07 to 4.08; P<0.01; online supplemental figure S12).

Deep brain stimulation versus surgery with sham stimulation

We identified three trials where 360 participants were randomised to receive deep brain stimulation with antiparkinsonian drugs versus surgery with sham stimulation and antiparkinsonian drugs.43–45 All participants in the control group underwent surgery and had electrodes implanted, but the stimulation was not turned on until three months after surgery.43–45 At follow-up, the difference in levodopa equivalent daily dosage was −307 to −408 mg/day in the two trials reporting these data.43 44

Serious adverse events

Two trials reported serious adverse events at three months after randomisation43 45; 28 of 222 participants (12.6%) had a serious adverse event in the intervention groups versus eight of 74 (10.8%) in the control groups (online supplemental table S6). Meta-analysis indicated no evidence of a difference in serious adverse events (risk ratio 1.17, 95% CI 0.56 to 2.45; I²=0.0%; τ²=0.00; P=0.68; two trials; Bayes factor 1.17) (figure 4). Visual inspection of the forest plot and measures to quantify heterogeneity (I²=0.0%, τ²=0.00) indicated no heterogeneity. Trial sequential analysis showed that we did not have enough information to confirm or reject a relative risk reduction of 10% between groups (not shown). This outcome result was assessed as having a high risk of bias and the certainty of the evidence was very low (online supplemental table S3 and online supplemental table S5).

Request permission

Figure 4

Meta-analysis (risk ratio with 95% confidence interval (CI)) of deep brain stimulation with antiparkinsonian drug treatment versus surgery with sham stimulation and antiparkinsonian drug treatment, on serious adverse events43 45

Deep brain stimulation versus resective surgery

We identified two trials where 79 participants were randomised to receive deep brain stimulation with antiparkinsonian drugs versus resective surgery with antiparkinsonian drugs.41 42 Participants receiving resective surgery underwent pallidotomy42 or thalamotomy of the ventral intermediate thalamic nucleus.41 At follow-up, the difference in levodopa equivalent daily dosage was −485 mg/day in the trial reporting these data.42–44

Serious adverse events

Two trials reported serious adverse events six months after randomisation41 42; three of 42 participants (7.1%) had a serious adverse event in the intervention groups versus one of 37 (2.7%) in the control groups (online supplemental table S8). Meta-analysis indicated no evidence of a difference in serious adverse events (risk ratio 1.72, 95% CI 0.23 to 12.67; I²=0.0%; τ²=0.00; P=0.59; two trials; Bayes factor 1.06) (figure 5). Visual inspection of the forest plot and measures to quantify heterogeneity (I²=0.0%, τ²=0.00) indicated no heterogeneity. Trial sequential analysis showed that we did not have enough information to confirm or reject a relative risk reduction of 10% between groups (not shown). This outcome result was assessed as having a high risk of bias and the certainty of the evidence was very low (online supplemental table S3 and online supplemental table S7).

Request permission

Figure 5

Meta-analysis (risk ratio with 95% confidence interval (CI)) of deep brain stimulation with antiparkinsonian drug treatment versus resective surgery with antiparkinsonian drug treatment, on serious adverse events41 42

Subgroup analysis comparing the effects of different targets of the electrodes indicated no evidence of a difference in serious adverse events (P=0.33) (online supplemental figure S48). The remaining predefined subgroup analyses could not be performed because of lack of relevant data. In a sensitivity analysis, we performed beta binomial regression and the results were similar (risk ratio 2.71, 95% CI 0.27 to 27.52; P=0.40; online supplemental figure S49).

Principal findings

In this systematic review assessing the benefits and harms of deep brain stimulation for Parkinson's disease, we identified seven trials where 1125 participants were randomised to receive deep brain stimulation with antiparkinsonian drugs versus antiparkinsonian drugs only (primary comparison). All results had a high risk of bias and the certainty of the evidence was very low for all primary outcomes. The information size was insufficient when assessing all cause mortality. Meta-analysis showed that deep brain stimulation increased the risk of serious adverse events, mainly because of an increased risk of perioperative complications, such as cerebral haemorrhages and postoperative confusion, and events related to hardware, such as infection at the stimulator site, dislocation of the device, or reoperations. Meta-analyses indicated that deep brain stimulation might reduce symptoms specific to Parkinson's disease, but the assessments of disease specific symptoms had several methodological limitations.

Strengths and limitations of this study

Our systematic review had several strengths. We used up-to-date methods to assess the methodological quality of the included trials.21 50 51 Our methodology was predefined and described in detail before starting the literature searches,15 and was based on the eight step assessment, suggested by Jakobsen et al,21 and trial sequential analysis.23 Risk of bias was assessed with the Cochrane RoB 250 and overall certainty of the evidence with GRADE.51 Therefore, both risks of random errors (play of chance) and systematic errors (bias) were taken into account in our review.

Our systematic review had several limitations. Firstly, the maximum follow-up period in most trials was less than a year, and therefore long term beneficial and harmful effects could not be assessed. Secondly, all of the included trials were assessed as having a high risk of bias, mainly because of missing data, lack of blinding, and poor reporting without prespecified protocols, clinical trial registrations, or statistical analysis plans. Hence our results presumably overestimated the beneficial effects of deep brain stimulation.52–54 Thirdly, the overall risk of type I errors was increased because of the large number of outcomes, comparisons, and subgroup analyses. We adjusted our thresholds for significance only according to the total number of primary outcomes. Fourthly, the included trials differed in participant characteristics (eg, age, sex, and length of disease) and trial design (eg, follow-up time and placement of electrodes), increasing the risk of statistical heterogeneity. We investigated these concerns by performing subgroup analyses, when possible, but lack of relevant data limited the power of these analyses. Only the subgroup analysis comparing different lengths of disease when assessing serious adverse events (online supplemental figure S11) indicated evidence of a difference. This finding was mainly influenced by the study of Schuepbach et al enrolling participants with early motor complications.39 Although the relatively reduced risk of serious adverse events with deep brain stimulation seen in this trial could be attributed to the shorter length of disease, two smaller trials included in the same subgroup did not indicate any evidence of this trend.35 40 Therefore, whether a longer length of disease increases the risk of serious adverse events is not known.

Measurement of symptoms specific to Parkinson's disease in the included trials also had methodological challenges. Only one trial reported total UPDRS score when participants received antiparkinsonian drugs. Many studies focused on UPDRS III scores, which only evaluated motor scores, and only a few trials39 40 45 had predefined disease specific symptoms as their focus. Also, most trials assessed symptoms when antiparkinsonian drug treatment was paused. This situation is challenging to apply to the clinic where most participants are expected to receive antiparkinsonian drugs, although the efficacy of deep brain stimulation is perhaps more precisely assessed when drug treatment is paused.

Our results showed that participants receiving deep brain stimulation were likely to reduce their dose of antiparkinsonian drugs. This finding could be a result of more frequent hospital visits for stimulation, and adjustments to drug treatment, in the intervention groups receiving deep brain stimulation, or an effect of deep brain stimulation itself. Because antiparkinsonian drugs are known to have adverse effects, some of the beneficial effects in subsections of the UPDRS might be linked to the decrease in drug dose because of deep brain stimulation (eg, reducing fluctuations and adverse effects of drugs, as seen in UPDRS IV). This finding could also explain the beneficial effects seen in some subsections of the UPDRS of deep brain stimulation with antiparkinsonian drugs versus antiparkinsonian drugs only, when drug treatment was paused, as seen in UPDRS II and UPDRS III. These results from the meta-analyses are difficult to interpret, however, as the assessment scales used by the trialists were not related to predefined and evidence based minimal important differences.55–58 These methodological limitations need to be considered when interpreting the analysis results based on disease specific symptoms.

Comparison with other studies

Previous reviews have assessed deep brain stimulation for Parkinson's disease.59–62 Generally, these reviews showed a beneficial effect of deep brain stimulation on UPDRS score, troublesome dyskinesias, and quality of life.59–62 As well as methodological limitations, these reviews15 mainly focused on the effects of deep brain stimulation on motor symptoms, dyskinesias, and quality of life, whereas little or no information on adverse effects was reported. Furthermore, previous reviews have not considered the implications of using scales without predefined and validated minimal important differences to assess motor symptoms.

Study implications

The use of deep brain stimulation for Parkinson's disease is increasing and will likely become part of standard care.63 When an intervention has already been approved and integrated as part of standard care, setting up randomised clinical trials is often difficult because clinicians might question whether patients should be randomised to a treatment that does not include the intervention in question. Hence the results of our review highlight the urgent need for well conducted, long term randomised trials with patient important outcomes, including assessing adverse effects, before deep brain stimulation has been even further integrated as part of the standard care. Observational studies could supplement randomised trials, especially for investigating rare adverse events.

Conclusions

In this review, we found that the certainty of evidence was very low for all primary outcomes, and based on the included evidence, the beneficial effects of deep brain stimulation were questionable because of methodological limitations. Compared with only antiparkinsonian drug treatment, deep brain stimulation with antiparkinsonian drugs seemed to increase the risk of serious adverse events, mainly because of perioperative complications and hardware related events. Conducting randomised clinical trials of adequate methodological quality to effectively evaluate the effects of deep brain stimulation is crucial.