The Place of Adaptive Deep Brain Stimulation in Parkinson's Disease: Spatial before Temporal Optimization

Abstract

Abstract

Subthalamic nucleus (STN) deep brain stimulation (DBS) can deliver major gains in function and quality of life in Parkinson's disease (PD). Broadly, it aims to reproduce the patient's best medication-induced “on” state, but continuously. In addition, STN DBS can suppress tremor, including tremor refractory to medication. Therefore, dopaminergic medication is reduced substantially, together with associated motor fluctuations and dyskinesias. How well does STN DBS achieve this ideal? A useful measure is the motor response ratio (MRR), the proportion of the pre-operative levodopa Unified Parkinson's Disease Rating Scale (UPDRS) motor response reproduced by DBS alone. In the INTREPID study, performed at expert centers using directional devices capable of fine spatial tuning, the group-average MRR was 0.89 at 1 year.1 Interestingly, outcomes improved over that year, potentially with more opportunities to refine programming. However, MRR does not directly account for side effects, varies substantially across individuals, and lower values are reported in older studies. Achieving an optimal result from STN DBS depends on three key elements: patient selection, accurate and safe lead implantation, and idealized programming. For the latter two, spatial optimization is crucial. What then is the role of adaptive DBS (aDBS), which, in its β-oscillation-triggered form, has recently achieved regulatory approval in many jurisdictions? aDBS offers the possibility of addressing motor fluctuations that remain or emerge with continuous DBS (cDBS), by periodically adjusting stimulation based on biofeedback signals (eg, local field potentials).2 As many patients on cDBS do substantially less well than their ideal MRR, such temporal optimization would be welcome. However, disappointing outcomes from cDBS can often be improved by addressing the fundamentals, for example, by correcting lead location and re-programming.3, 4 Even when optimized, STN DBS can drift away from the best available “on state” with disease progression, especially in the axial domain and, over time, many patients do require increasing dopaminergic medication. Fluctuations may also be influenced by physiological state, including diurnal rhythm. Some patients treated with cDBS choose to endure side effects, such as soft speech, to maximize MRR and avoid medication-related fluctuations. The need for aDBS is, therefore, clear, although current consensus is that it is suitable only for highly selected and carefully managed patients.5 Even so, the therapy is evolving. The paper by Hammer et al6 in this issue of Movement Disorders will help drive that evolution. The authors are pioneers of β-triggered aDBS.7, 8 In this observational study, they examined β oscillations streamed from a sensing implantable pulse generator in chronically implanted PD patients with STN DBS. The investigational device allowed for comparisons between longer duration recordings collected at-home versus shorter duration recordings in-clinic, both off and on medication and stimulation. An important finding, confirming previous reports,9 is that the frequency of peak β activity was reduced during DBS. Commercial aDBS tracks signal power in a defined 5 Hz bandwidth (±2.5 Hz around a nominated sensing frequency)—that has previously been determined OFF-stimulation.2 In the Hammer study, stimulation shifted the β peak by −4.1 ± 0.8 Hz, potentially displacing it beyond the sensing bandwidth. By extension, aDBS performance might be improved if the sensing bandwidth were instead defined in the ON-stimulation state.10 A previous study recently also reported frequency shifts in peak β activity with a commercial aDBS device and proposed a clinical workflow to address the limited 5 Hz sensing range.10 Namely, that the sensing range and thresholds for aDBS should be derived from at-home recordings during cDBS and be further refined using assessments of stimulation- or medication-induced modulation.10 Expected findings in the paper by Hammer et al,6 were that β power, already a very small spontaneous signal, was further reduced during cDBS, and β power fluctuations were attenuated with less levodopa. β power appeared to reach a lower plateau whereas motor benefit continued to accrue with therapy. Taken together, these observations raise the question of whether β activity has inherent practical limits as a control signal. Whether alternative or additional control signals, such as γ oscillations7 or evoked resonant neural activity (ERNA),11 perform better remains to be tested. The patients studied by Hammer et al6 seemed exactly a cohort for whom aDBS might be most helpful. The average MRR in these patients was approximately 0.48. There are many possible explanations for this low MRR, and the authors cited disease progression over a mean follow-up of 22 months as one. Another, among many possibilities, is that DBS was being delivered, and β activity sampled, at some distance from the optimal site. All patients were receiving cDBS in monopolar mode from one of the two middle contacts on a quadripolar lead, matching a proposed “sense-friendly” configuration. The authors note that the middle contacts exhibited the widest therapeutic window on monopolar review. However, this is not the same as identifying the precise fractionation of current across contacts, and parameter optimization, that delivers maximal therapeutic efficacy. This raises the question of whether the reported results, including the observed β dynamics, can be generalized to patients who achieve a much better result from cDBS. This points to a key issue in the adoption of aDBS into clinical practice. It should not come at the expense of optimizing cDBS. How might this happen? There are substantial technical constraints imposed by the need to sample β activity close to the site of applied DBS. In the pivotal Adaptive DBS Algorithm for Personalized Therapy in Parkinson's Disease (ADAPT-PD) study, a prospective, multicenter study of a commercial aDBS platform,12 and similarly in the paper by Hammer et al,6 aDBS required bipolar “sandwich mode” sensing from contacts adjacent to those used for monopolar stimulation. On a quadripolar lead, or a directional lead with four tiers, ‘sensing-compatible’ configurations currently confine programming to one or both of the middle two tiers. Such constraints sit uneasily with the importance of delivering DBS within fractions of a millimeter of the ideal site. In clinical practice, adjusting current fractionation up and down and around the lead, in very small increments, can meaningfully shift outcomes.13, 14 In addition, the depth of lead implantation is variable, and so must be the chance that the two middle tiers will lie exactly at the optimal stimulation site. Current commercial platforms also direct attention to β power at individual contacts, which may naturally steer users toward contacts with larger sensed β signals. However, the relationship between OFF-stimulation β power, ON-stimulation β suppression, and the precise sublocation optimal for DBS appears to be more complex.15 Moreover, the spatial distribution of β power across contacts can vary, even well after the microlesion effect has subsided.16 Many patients are excluded from aDBS because local field potential recordings are inadequate for sensing.10, 17 In some, this may reflect technical considerations, for example, better sensing performance from newer sensing optimized leads than legacy leads. However, patients with less ideally placed leads are also more likely to show smaller, or absent, β oscillations. Yet these are precisely the patients expected to have lower MRRs, and therefore, the greatest need for what aDBS seeks to offer. An additional issue is that β-based aDBS may not deal adequately with tremor, which may require an additional or alternative classifier. At a minimum, these considerations considerably narrow the population suitable for aDBS.5 At worst, optimizing for aDBS could compromise the delivery of the foundational elements of cDBS that provide the greatest benefit. Whether ironing out fluctuations in β oscillations with aDBS translates into robust and generalizable quality-of-life gains remains to be established. Patients in ADAPT-PD treated with aDBS did seem improved in post-hoc exploratory analyses—with one of the aDBS modes yielding an extra hour or so “on time” per day.12 Almost all patients elected to remain on aDBS, rather than cDBS, at the end of the study (albeit they were not blinded to this choice). Alternatively, perhaps some of these patients could simply have benefited from further optimization of their existing cDBS through reprogramming, even if this rendered them ineligible for aDBS. It is often the case that the more programming sessions patients undergo, the better the efficacy.3, 4 Clinical programming has already become arduous given the high spatial resolution offered by modern devices. Early real-world experience with aDBS suggests that it places substantial additional demands on time and expertise, with a clear risk of distracting from optimization of the fundamentals.10, 15 For clinicians in the era of aDBS, an important lesson is that standard-of-care practice should remain focused on optimizing what we already know works for STN DBS in PD. For some patients—those who happen to be on a sensing-compatible configuration and have sufficiently robust β recordings—aDBS may have something to offer, perhaps a great deal in those with substantial residual motor fluctuations. aDBS remains highly exciting for the future. The study by Hammer et al6 shows how further refinements may be revealed and included in aDBS algorithms. After decades in the making, aDBS is ready for real-world refinement. (1) Research project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. Design, B. Execution, C. Review and Critique; (3) Manuscript: A. Writing of the First Draft, B. Review and Critique. W.T.: 3A, 3B M.A.P.: 3B All authors had full access to all the data and accept responsibility to submit for publication. Data sharing not applicable to this article as no datasets were generated or analysed during the current study.