Background

Parkinson’s disease (PD) is the second most common neurodegenerative disease with a global prevalence of more than 6 million individuals.1 It is characterised by motor2 and non-motor symptoms.3 While pharmacological treatments play a central role in managing motor symptoms, some patients may experience inadequate symptom control or medication-related complications, necessitating the adoption of advanced therapies.4–6

Deep brain stimulation (DBS) and infusion therapies have emerged as promising interventions for individuals with PD who experience refractory motor fluctuations and dyskinesias. DBS involves the surgical implantation of electrodes into specific brain regions, such as the subthalamic nucleus (STN) or globus pallidus pars interna (GPi).7 8 Infusion therapies, including levodopa-carbidopa intestinal gel (LCIG), levodopa-entacapone-carbidopa intestinal gel (LECIG), continuous subcutaneous apomorphine infusion (CSAI) and continuous subcutaneous infusion of foslevodopa or levodopa (CSCI), continuously deliver medications to provide consistent symptom control.9–11

Despite the potential benefits of these advanced therapies, studies have reported increased falls after these treatments.12–14 Falls are a common and debilitating complication of PD, often resulting in injuries, hospitalisations and decreased mobility.15

The objective of this systematic review and network meta-analysis (NMA) is to evaluate the risk of falls associated with advanced therapies in individuals with PD. By synthesising existing evidence from randomised controlled trials (RCTs), the review aims to provide comprehensive insights into the safety and efficacy of these advanced therapies in reducing the risk of falls. Such knowledge is essential for optimising treatment decisions and improving the overall care and outcomes of individuals living with PD.

ResultsSelection of studies

The search identified 1321 studies, of which 573 duplicates were removed. The remaining 748 studies were assessed for eligibility, and 605 studies were excluded by title and abstract screening. Eligibility of the remaining 143 studies was assessed by full-text review, resulting in 13 studies being included in this review. Furthermore, one study was incorporated through citation searching. Ultimately, a total of 14 studies were included (figure 1).

Figure 1

PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) flow diagram of the study selection process.

Study characteristics

Four of the studies compared DBS to best medical treatment (BMT),14 17–19 two compared active simulation to sham stimulation,20 21 two compared STN DBS to GPi DBS22 23 and six studies compared infusion therapy to BMT.11 24–28 Thirteen studies had randomised parallel group design and one study had randomised cross-over design. The characteristics of the studies included are summarised in table 1 with additional information in online supplemental table S3.

Table 1

Summary of included studies

Participants

This review encompassed a total of 2021 unique patients with PD enrolled in 24 countries. The mean age of the participants ranged from 46.5 to 69.3 years, with a male prevalence of 68.5%. Disease duration ranged from 7.3 to 13.6 years.

Risk of bias

Across the studies included in the meta-analysis, significant variability in risk of bias assessments was observed. Specifically, six studies were classified as having an overall low risk of bias, while seven studies were identified as having some concerns, and one study was determined to have a high risk of bias (figure 2).

Figure 2

(A) Risk of bias summary: Judgements about each risk of bias item for each included study. Risk of bias for each domain in each study is represented in green for low risk of bias, red for high risk of bias and yellow for some concerns. (B) Risk of bias graphs: Judgements about each risk of bias item presented as percentages across all included studies.

In terms of specific domains, all studies demonstrated a low risk of bias for the randomisation process. However, among the studies categorised with some concerns, these pertained to various domains such as deviations from intended interventions, missing outcome data, measurement of outcomes and selection of reported results. Notably, one study was flagged with a high risk of bias specifically for the measurement of outcomes.19 Furthermore, among the included studies, 12 used an intention-to-treat analysis approach, while two opted for a per-protocol analysis method.

Traditional pairwise meta-analyses

The traditional pairwise meta-analyses encompassed studies comparing DBS to BMT, active stimulation to sham stimulation, STN DBS to GPi DBS, and infusion therapy to BMT.

DBS versus BMT

Four RCTs compared DBS to BMT, with a total of 506 participants in the DBS group and 522 in the BMT group. Among these, 46 fall adverse events were reported in the DBS group and 24 in the BMT group (figure 3A).

Figure 3

(A) Forest plot depicting the meta-analysis of fall events in the DBS to BMT comparison. (B) Sensitivity analysis performed by excluding William 2010 study. (C) Forest plot depicting the meta-analysis of the number of patients with falls in the STN to GPi comparison. (D) Forest plot depicting the meta-analysis of the number of patients with falls in the active stimulation group to sham stimulation comparison. (E) Forest plot illustrating the overall meta-analysis of the number of patients with falls in the infusion therapy group compared with BMT and subgroup analyses of LCIG only and CSCI only compared with BMT. BMT, best medical therapy; CSCI, continuous subcutaneous infusion; DBS, deep brain stimulation; GPi, globus pallidus interna; LCIG, intraduodenal infusion of levodopa-carbidopa gel; STN, subthalamic nucleus.

The pooled data suggest a higher risk of fall events in the DBS group compared with BMT, although this difference was not statistically significant (Risk Ratio (RR)=1.60, 95% CI 0.64, 4.03, p=0.31). Heterogeneity analysis revealed significant variability among the studies, with Tau²=0.47, Chi²=7.21, p=0.07 and I²=58%.

Conducting a sensitivity analysis using the leave-one-out method, we identified that the study by William et al was a major contributor to the heterogeneity.19 On excluding this study, the pooled data demonstrated a significant difference (RR=2.74, 95% CI 1.60, 4.67, p=0.0002, I²=0%) (figure 3B). Further evaluation revealed that the study exhibited an overall high risk of bias due to methodological issues such as the measurement method and concerns regarding unblinded participants and raters.19

Using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) system approach, the certainty of the evidence for the higher risk of fall events in the DBS group compared with BMT was determined to be moderate (online supplemental table S4). However, this finding was downgraded primarily due to the risk of bias present in the included studies.

STN DBS versus GPi DBS

Two RCTs compared STN to GPi DBS, involving a total of 153 participants in the STN group and 156 in the GPi group. Among these participants, 76 reported falls in the STN group and 64 in the GPi group (figure 3C).

The pooled data suggest a slight increase in the risk of falls among the STN group compared with GPi, although the difference was not statistically significant (RR=1.09, 95% CI 0.44, 2.66, p=0.85, I²=15%). Heterogeneity analysis indicated low variability among the studies, with I² suggesting minimal heterogeneity among the studies. Overall, the findings indicate that there is no significant association between the intervention and the risk of falls, and the results are consistent across the included studies with low heterogeneity.

Using the GRADE system recommendation approach, the certainty of the evidence for this outcome was determined to be moderate. However, this finding was downgraded primarily due to the imprecision present in the included studies (online supplemental table S5).

Active versus sham stimulation

Two RCTs compared active DBS stimulation to sham stimulation, involving a total of 222 participants in the active stimulation group and 74 in the sham stimulation group. Among these participants, 76 reported falls in the active stimulation group and 64 in the sham stimulation group (figure 3D).

The pooled data suggest a moderately increased risk of falls among the active stimulation group compared with sham stimulation, although the difference was not statistically significant (RR=1.55, 95% CI 0.18, 13.28, p=0.69, I²=60%). Heterogeneity analysis indicated substantial heterogeneity among the studies. No sensitivity analysis was performed as there are only two studies.

Using the GRADE system recommendation approach, the certainty of the evidence for this outcome was determined to be low. However, this finding was downgraded primarily due to inconsistency and the imprecision present in the included studies (online supplemental table S6).

Infusion therapy versus BMT

Six RCTs were conducted to compare infusion therapy to BMT, involving a total of 334 participants in the infusion therapy group and 333 in the BMT group. Among these participants, 27 reported falls in the infusion therapy group and 34 in the BMT group (figure 3E).

The overall pooled data suggest a slight reduction in the risk of falls among the infusion therapy group compared with BMT, although no statistically significant overall effect was observed (RR=0.86, 95% CI 0.45, 1.65, p=0.65, I²=28%). The I² statistic indicated moderate heterogeneity.

Subgroup analyses were conducted based on the method of administration of the infusions, with four studies in the LCIG group and two studies in the CSCI group. The test for differences in effect sizes between subgroups revealed statistically significant differences (χ²=4.84, p=0.03, I²=79.3%), suggesting that the heterogeneity could be attributed to the method of infusion administration. Using the GRADE system recommendation approach, the certainty of the evidence for this outcome was determined to be moderate for the infusion to BMT comparison, with certainty downgraded due to inconsistency (online supplemental table S7).

LCIG only

The subgroup analysis of the LCIG only group compared with BMT showed a non-statistically significant increase in the risk of falls compared with the BMT group (RR=1.76, 95% CI 0.71, 4.38, p=0.22, I²=0%), with an I² value of 0% indicating no observed heterogeneity. Certainty of the evidence was determined to be moderate for the LCIG only to BMT comparison, with certainty downgraded due to imprecision (online supplemental table S7).

CSCI only

The subgroup analysis of the CSCI only group compared with BMT showed a statistically significant reduction in the risk of falls compared with the BMT group (RR=0.52, 95% CI 0.29, 0.94, p=0.03, I²=0%), with an I² value of 0% indicating no observed heterogeneity. Certainty of the evidence was determined to be high for the CSCI only to BMT comparison (online supplemental table S7).

Network meta-analyses

The NMA comprised 13 studies, featuring 5 direct comparisons: CSCI, DBS (STN+GPi), LCIG and STN DBS versus BMT/control, along with GPi DBS versus STN DBS.11 14 17–20 22–28 The remaining comparisons formed indirect assessments between treatments. One study was excluded since the number of participants reporting fall was not available.21

Heterogeneity within and between study designs was quantified, revealing Tau²=0.21, Tau=0.46 and I²=27.4% (95% CI 0.0% to 66.2%). Tests of heterogeneity and inconsistency resulted in a total Q-value of 11.02 (p=0.20), indicating no significant heterogeneity or inconsistency within or between study designs.

Overall, the NMA findings based on the surface under the cumulative ranking (SUCRA) suggest that CSCI is the most effective in reducing falls, followed by BMT. DBS (STN+GPi), GPi DBS, LCIG and STN DBS are comparatively less effective, with STN DBS ranked the least effective (figure 4). The league table and the ranking plots of NMA results are presented in online supplemental table S8 and online supplemental figure S1, respectively.

Figure 4

Network meta-analysis. (A) Network graph illustrating direct comparisons between treatments and (B) forest plot comparing treatments to BMT/control based on SUCRA ranking. BMT, best medical therapy; CSCI, continuous subcutaneous infusion; DBS, deep brain stimulation; GPi, globus pallidus interna; LCIG, intraduodenal infusion of levodopa-carbidopa gel; STN, subthalamic nucleus.

Discussion

This systematic review and NMA aimed to evaluate the risk of falls associated with device-aided advanced treatments (DBS and infusion therapies) in individuals with PD. Our meta-analysis incorporated data from RCTs comparing various treatment modalities, including DBS to BMT, active stimulation to sham stimulation, STN DBS to GPi DBS and infusion therapy to BMT. In addition to traditional pairwise meta-analysis, NMA was conducted to enable a comprehensive comparison of the relative efficacy and safety of multiple advanced therapies through direct and indirect comparison of treatments.

Overall, our study revealed that while the observed increase in fall risk within the DBS group compared with BMT did not attain statistical significance, it remains a clinically significant observation. For example, in the study conducted by Weaver et al, a significant number of falls were reported in the DBS group, frequently resulting in severe injuries such as fractures, dislocations and head trauma. These injuries often necessitated surgical interventions or other medical procedures.14

In response to the substantial heterogeneity observed, we conducted a sensitivity analysis using the leave-one-out method. This analysis revealed that the study by William et al was a significant contributor to the observed heterogeneity primarily due to the lack of masked assessors and unblinded patients.19 After excluding this study from our analysis, the meta-analysis indicated a statistically significant increase in the risk of falls within the DBS group. In addition, while the meta-analysis did not find a statistically significant difference in fall risk between STN and GPi DBS, the body of evidence from included studies suggests a trend towards increased falls with STN DBS. Noteworthy, other studies not included in the meta-analysis have also noted increased incidences of falls in the DBS group compared with BMT. For instance, in Hacker et al’s 11-year outcome analysis comparing STN DBS to BMT, a notable discrepancy in fall occurrences was observed. Specifically, among deceased participants, those in the DBS group reported a higher number of falls (n=7, 100%) compared with individuals receiving best medical therapy (n=3, 60%).29 This study could not be included in our analysis due to partial data availability.

Among the infusion therapies examined, only LCIG and CSCI RCTs reported falls as adverse events and were thus included in this review. Although some open-label CSAI studies reported falls, they were indistinguishable and merged with the BMT groups. Moreover, the few RCTs on LECIG did not report fall as adverse events, thus leading to their exclusion from the review. In contrast with DBS, the overall meta-analysis comparing infusion therapy (LCIG and CSCI) to BMT did not yield a significant difference in the risk of falls. However, subgroup analysis based on infusion method revealed contrasting results. LCIG administration exhibited a non-statistically significant increase in fall risk compared with BMT, supported by moderate certainty of evidence. Conversely, CSCI showed a statistically significant reduction in fall risk compared with BMT, with a high level of certainty.

Finally, our NMA indicates that CSCI shows greater effectiveness in reducing falls compared with both BMT/control and other advanced therapies such as LCIG and DBS. Notably, our analysis suggests that STN DBS may even elevate the risk of falls. These findings support our meta-analysis results, highlighting the potential benefits and risks associated with different treatment modalities for PD.

Underlying mechanisms

Exploring potential mechanisms underlying the increased fall risk in the DBS group reveals interesting possibilities. It is plausible that DBS directly influences neurological pathways involved in balance and gait, predisposing individuals to falls.30 Alternatively, the observed rise in fall incidence may be secondary to enhanced functional capacity and increased activity levels among DBS recipients,14 especially in the context of deconditioned patients gaining greater mobility in a relatively short period of time.

Another—not mutually exclusive—hypothesis is the induction of reckless behaviour in the context of the increased impulsivity reported in STN DBS. To better understand this scenario, a dedicated STN versus GPi comparison was conducted. Such analysis revealed a slight increase in the risk of falls among the STN group, although this difference did not reach statistical significance. This lack of statistical significance may be attributed, in part, to variations in the sample sizes of the two included studies. For instance, the study by Burchiel et al was a pilot trial with only 10 participants.22 Notably, in the study by Follett et al, the number of patients reporting falls as a serious adverse event was higher in the STN group compared with the GPi group, with borderline statistical significance 13 (8.8%) vs 5 (3.3%; p=0.05).23 Our findings are corroborated by various studies demonstrating an increased risk of falls with STN DBS. For example, Rodriguez-Oroz et al reported greater adverse events related to falls and balance disturbances in the STN DBS group compared with the GPi DBS group.31 Similarly, Celiker et al, in a study comparing the effects of STN DBS and GPi DBS on fall risk, found that while there was no significant change in fall risk postoperatively compared with preoperative results in the GPi DBS group, the fall risk in the STN group significantly increased to 100% after 6 months (p<0.05).32 St George et al investigated whether DBS randomised to the STN or GPi improved compensatory stepping to recover balance after a perturbation.33 They found that within the STN group, five subjects who did not fall during the experiment before surgery experienced at least one fall after surgery, whereas the number of falls in the GPi and PD-control groups remained unchanged. The same group performed a prospective double-blind study randomising 28 patients to either STN or GPi DBS and having a range of clinical balance measures as primary outcome. While balance was not different on medication and on stimulation for both sites, GPi cohort performed better in absence of treatment, possibly indicating a detrimental effect of STN surgery.34 To gain further insights into the role of surgery, we then compared active DBS stimulation with sham stimulation, overall finding a moderately increased risk of falls with active stimulation, although this difference did not reach statistical significance. Although participants in the active stimulation group predominantly had STN DBS, overall this finding is consistent with an increased risk of falls in the subgroup because of surgery itself.

One potential factor contributing to the increased fall risk in the STN group may be related to medication adjustments following DBS. In line with previous studies, we found that the average reduction in dopaminergic medication use was greater in the STN group compared with the GPi group, where medication needs often remain unchanged or undergo minimal adjustments.35 It is possible that this reduction in medication could exacerbate postural instability and contribute to an increased risk of falls in the STN group.36 Further research is warranted to explore this potential relationship.

Although the study populations included in the NMA were comparable across treatments, there are notable differences between patients selected for DBS and those receiving LCIG therapy in real-world clinical practice. Patients undergoing STN DBS tend to be younger and more cognitively intact than those receiving LCIG. These real-world differences in patient profiles further support our findings of an increased fall risk following DBS compared with LCIG.

Freezing of gait (FOG) is known to increase the risk of falls37 and studies have shown that STN DBS is associated with the induction or worsening of FOG, in some cases also in conjunction with levodopa (a phenomenon known as on-FOG).38 39 By contrast, the induction or exacerbation of FOG, including on-FOG, is rare with LCIG,40 41 although diphasic dyskinesias, a known complication of LCIG therapy,42 can impair gait and contribute to increased fall risk.

The role of increased mobility and deconditioning, although contributory, seems less important by looking at the effect of infusion therapies on fall rates comparable to BMT. Interestingly, CSCI performed better than LCIG and BMT and some hypotheses can be formulated in this regard. In the studies by Soileau et al and Espay et al, the CSCI group received a continuous subcutaneous infusion over a 24-hour period of either foslevodopa-foscarbidopa, a new soluble formulation of levodopa and carbidopa prodrugs, or ND0612, a solution of levodopa and carbidopa.11 28 In contrast, participants in the LCIG-only studies received the conventional 16-hour infusion of levodopa-carbidopa during their waking hours. Notably, in a prospective study involving six participants receiving 16-hour LCIG infusion, falls were reported as one of the most common adverse events in 33.3% of cases.13 By contrast, in an RCT comparing 24-hour LCIG to 14-hour LCIG, one participant in the 14-hour group reported a fall, while none did in the 24-hour group.43 Moreover, a prospective open-label study by Chang et al demonstrated that 24-hour LCIG infusion improved freezing of gait and reduced falls in patients.44 Several mechanisms might explain the beneficial effects of an infusion around the clock. For example, it is plausible that participants receiving a nocturnal infusion experience an improvement in sleep quality, as previous studies have indicated that sleep disturbances elevate the risk of falls.45

Finally, mechanisms other than 24-hour infusion should be taken into account when explaining the difference between CSCI and LCIG, such as a different effect on vitamin metabolism and occurrence of neuropathy.46 A more important difference relies on the type of patient enrolled for these studies, as CSCI candidates tend to be younger and with a shorter disease duration. Notwithstanding, no significant differences were observed when comparing age and disease duration of the patients allocated to DBS, LCIG and CSCI (table 1).

Clinical implications

The decision to pursue DBS or infusion therapy should be personalised based on patient characteristics, disease severity and treatment goals. Our meta-analysis’ main message is that DBS, especially targeting the STN, might increase the risk of falling, especially when compared with other non-DBS advanced procedures. With respect to the infusion-based therapies, while LCIG might not change the fall risk, preliminary evidence shows that CSCI has a positive effect on falling, probably resulting from an improvement of motor fluctuations and/or dyskinesias in absence of a superimposed detrimental effect on balance. Balance impairment and FOG can be levodopa-responsive and thus mainly present during the off states.37 On the other hand, dyskinesias can contribute to falling due to the destabilising effect on the centre of body mass of involvement of the lower limbs during locomotion.47

Limitations

Several limitations should be considered when interpreting our findings. Variability in fall reporting methods across studies introduces potential recall bias and may underestimate fall incidence. In fact, many studies did not provide detailed methodologies for recording falls, such as whether falls were logged through participant diaries or solely during study visits, reducing the reliability of this outcome. Moreover, some studies distinguished between falls with and without injury, while others did not, further limiting the comparability of fall data. Overall, this limitation highlights the need for standardised, prospective approaches to fall data collection in future studies. The review excluded articles not written in English, non-RCTs, conference abstracts and book chapters may have led to the omission of relevant results. Heterogeneity among studies and inherent biases in RCTs may affect result robustness, emphasising the importance of rigorous study designs and transparent reporting. Limited studies in certain comparisons and small sample sizes in some analyses may impact precision, calling for larger, multicentre trials with standardised methodologies. Notably, the comparison between active DBS stimulation and sham stimulation bears similarities to the DBS versus BMT comparison, as the sham group did not receive active DBS stimulation until after a delay period, similar to their pre-surgical state when they were solely treated with medication. Moreover, a substantial heterogeneity was observed, potentially stemming from variations in the protocols for sham stimulation across studies. The certainty of evidence for this comparison was deemed low, primarily due to inconsistencies and imprecision in the included studies. The analyses comparing active DBS stimulation vs sham stimulation, as well as GPi versus STN DBS, are based on only two reports. Therefore, caution should be exercised when interpreting the strength of these conclusions.

Although our study suggests a potential association between STN DBS and increased fall risk, the data did not reach statistical significance. Confounding factors such as patient selection, surgical techniques, DBS programming and variations in postoperative rehabilitation and specialised medical management complicate isolating the specific impact of STN DBS. A well-designed, adequately powered prospective study comparing different advanced device-based therapies in PD is needed to accurately assess the impact of each treatment on fall risk.

Furthermore, CSCI is a relatively new therapy and has the highest rate of discontinuation among advanced treatments, often due to issues such as skin reactions, hallucinations or confusion.48 These potential long-term complications highlight the need for further study into the durability and safety of CSCI over extended treatment periods.

Finally, studies included patients with different profile, especially disease duration and age, although not different when compared across studies. For example, one can argue that STN DBS can worsen balance only in frailer and older patients. On the other hand, the aforementioned study by Hacker et al 29 enrolled patients within 4 years from disease onset, once again supporting the hypothesis that fall risk might increase after STN DBS.

Since DBS can worsen gait and increase fall risk, it is conceivable that novel technologies can be beneficial as they can mitigate stimulation-related side effects. For example, although not yet confirmed, adaptive (closed-loop) DBS may offer a safer alternative to standard continuous (open-loop) DBS. This is currently being explored in clinical trials (eg, NCT05402163).