Safety and efficacy of occipital nerve stimulation as treatment of chronic cluster headache: an investigator-initiated, double-blind, randomized, placebo-controlled study

Abstract

Background

Occipital nerve stimulation has shown promising results in attack prevention in patients with chronic cluster headache, but evidence from controlled trials is scarce. Conventional (tonic) occipital nerve stimulation elicits paresthesia, hampering blinded comparison to placebo. Using paresthesia-free burst occipital nerve stimulation, we conducted a randomized, placebo-controlled trial.

Methods

The study is an investigator-initiated, double-blind, randomized, placebo-controlled clinical trial involving participants with chronic cluster headache. It comprised a four-week baseline, a 12-week trial with transcutaneous electrical nerve stimulation, occipital nerve stimulator implantation, a 12-week randomized, double-blind burst occipital nerve stimulation treatment period, and a 12-week open-label tonic occipital nerve stimulation treatment period. The primary outcomes were safety and the proportion of participants reporting a ≥30% reduction in attack frequency in the randomized and open-label trial phases.

Results

Participants were enrolled between August 2021 and October 2023. 116 chronic cluster headache patients were assessed for eligibility. Sixty-five were excluded, and 51 entered baseline observation. Before occipital nerve stimulator implantation and randomization, an additional 13 were excluded. Thirty-eight participants underwent occipital nerve stimulator implantation and were randomly assigned to burst occipital nerve stimulation (n = 19) or placebo (n = 19). After the randomized trial phase, the proportion of ≥30% responders was 18.81% (95%CI 0.28%–37.87%) in the burst occipital nerve stimulation group and 50.02% (95%CI 26.87%–73.09%) in the placebo group. The likelihood of reaching the primary endpoint was 31.20% (95%CI 1.29%–61.23%, p = 0.042) higher in the placebo group. After the open-label phase, 42.09% (95%CI 19.91%–64.34%) in the burst occipital nerve stimulation group and 51.11% (95%CI 27.32%–74.88%) in the placebo group had ≥30% frequency reduction. In total, 20 adverse events were registered; eight were treatment-related serious adverse events. The most common adverse event was temporary occipital dysesthesia in eight participants.

Discussion

Occipital nerve stimulation reduced attack frequency but was not superior to placebo. The treatment had an acceptable safety profile and was well-tolerated. The results call for attention to sufficient placebo control when planning further studies.

Trial registration

The trial is registered at clinicaltrials.gov (identifier NCT05023460).Study registration date: July 27th 2021.

Introduction

Cluster headache (CH) is a primary headache disorder characterized by excruciatingly painful attacks [1] with a strictly unilateral localization in the orbital area [2]. The headache attacks are usually accompanied by ipsilateral autonomic facial symptoms such as lacrimation and conjunctival injection and can last up to three hours when untreated [2]. Classically, attacks of CH occur in bouts with a duration of weeks to months, followed by longer periods of remission. However, 15–20% of CH patients have a chronic form (CCH) where the remission periods – if any - do not exceed three consecutive months per year; additionally, CCH can be refractory to conventional medical treatment options [3], making it a potentially debilitating disorder. Patients with CCH refractory to other therapies may be considered for invasive neuromodulation therapies such as occipital nerve stimulation (ONS).

ONS is a minimally invasive, non-destructive, reversible neuromodulation therapy that involves electrical stimulation of the greater occipital nerves (GON).

ONS has been employed as a treatment for various severe chronic headaches, among which one of the best-supported indications for ONS is CH [4]. In CH, ONS is believed to elicit central neuromodulatory effects via a convergence of cervical and trigeminal afferents on second-order nociceptive neurons in the trigeminocervical complex (TCC) in the brainstem. From here, projections reach higher centers believed to be involved in pain processing [5].

A growing body of evidence indicates that ONS can function as a preventive treatment in patients with CCH. The existing literature is primarily based on uncontrolled, open-label studies, but all studies agree that ONS significantly reduces the attack frequency in patients with refractory CCH [6–10]. A recent large, randomized, dose-controlled trial supports these findings [11]. Importantly, the preventive effect seems to be sustained at long-term follow-up [12, 13]. The proportion of treatment responders varies among studies, but a pooled 50% response rate of nearly 60% of participants has been described in a meta-analysis [14].

Stimulation of the GON has traditionally been performed using a square-wave stimulation pattern (often termed tonic stimulation) that evokes paresthesias in the scalp. This has hindered studies including placebo control because the paresthesias preclude blinded comparison.

Within the last decade, a different stimulation paradigm, burst stimulation, has been introduced in related therapy spinal cord stimulation. Burst stimulation, designed to mimic the naturally occurring firing pattern of the neurons, is intended to work at sub-perception amplitudes, resulting in paresthesia-free stimulation. A small study indicated that burst ONS has a significant preventive effect on CCH [15], and a recent case series from our center has described its application in clinical practice [16]. Hence, we hypothesized that burst ONS could be a relevant comparator to placebo in a blinded trial evaluating ONS for CCH.

In this study, we have investigated the efficacy and safety of burst ONS in a double-blind, randomized, placebo-controlled trial followed by an open-label conventional (tonic) ONS treatment period.

Materials and methods

Study design and participants

We conducted an investigator-initiated, double-blinded, randomized, placebo-controlled clinical trial at the Department of Neurosurgery, Aarhus University Hospital, Denmark. The department has extensive experience with invasive neuromodulation and has performed ONS since 2013 [16].

The study comprised a baseline period (four weeks), an open-label treatment period with transcutaneous electrical nerve stimulation (TENS, 12 weeks), ONS device implantation followed by a two-week grace period, a randomized, double-blinded ONS treatment period with either burst ONS or placebo (12 weeks) and an open-label ONS treatment period with tonic ONS (12 weeks) [17] (Fig. 1).

Fig. 1.

Trial profile. TENS = transcutaneous electrical nerve stimulation. ONS = occipital nerve stimulation

During the TENS trial phase, the participants stimulated the GON transcutaneously. This trial phase was included to examine the preventive effect of non-invasive GON stimulation. The details on the application of the stimulation technique and the results from the TENS trial are reported elsewhere [18].

We recruited adults with a diagnosis of CCH, as defined by the International Classification of Headache Disorders 3^rd edition (ICHD-3). Participants were recruited from the Danish Headache Center (a tertiary headache center, Rigshospitalet-Glostrup) and from other neurological departments and regional headache clinics in Denmark. Eligible participants had a minimum of 15 CH attacks per month. Participants receiving preventive medication for CCH had to remain on a stable dose and refrain from starting other preventive treatments, including transitional treatment, from at least 30 days before enrolling and throughout the entire trial. Exclusion criteria included other ongoing neuromodulation therapy, substance abuse, severe psychiatric disorders, pregnancy, previous major posterior surgery at C3-level or above, and a concomitant second type of chronic headache.

Procedures

In a weekly electronic headache registration, the participants logged information about the number of attacks, mean duration of attacks (minutes), and pain intensity of the attacks on a 0–10 numeric rating scale (NRS) for the past seven days as well as their self-reported sleep quality on a four-point Likert scale.

All forms and questionnaires were electronic and were collected in the Research Electronic Data Capture (REDCap) database system (Vanderbilt University) hosted at Aarhus University in a dedicated project database. Headache registrations and other questionnaires were entered via a personal link sent automatically by e-mail every Sunday. Reminder e-mails were generated if the headache registrations and questionnaires were not completed within 24 hours and repeated up to five times [17]. If a headache registration was still not filled out, the principal investigator (ISFA) contacted the participant by phone and completed the registration after a structured interview.

All surgical procedures were performed by the same two experienced implanters (authors KM and JCHS).

Implantations were performed in a sleep-awake procedure using a single percutaneous electrode lead (SC-2366-70 Linear 3–6 Percutaneous Lead, Boston Scientific, Marlborough MA, USA) placed transversely across the GONs at the level of the inion, connected to a non-rechargeable impulse generator (IPG, SC-1416 WaveWriter Alpha Prime 16, Boston Scientific) placed below the right clavicle. An on-table test stimulation based on the participant’s verbal feedback was performed to secure the correct positioning of the implanted lead, eliciting bilateral occipital paresthesia. The procedure is described in detail elsewhere [19].

After implantation, a 14-day grace period was employed to ensure healing of the surgical wounds and reduce tissue swelling, achieving optimal conditions for programming the ONS system before randomization and blinding.

After the grace period, an experienced nurse performed a new programming session, ensuring satisfactory paresthesia coverage. Based on this, a paresthesia-free burst program was created. The parameters for burst ONS (MicroBurst, Boston Scientific) were set according to manufacturer instructions with an intra-burst frequency of 450 Hz (six pulses), an inter-burst frequency of 40 Hz, and a pulse width of 300 μs. The stimulation amplitude was set to 50% of the individual perception threshold.

If randomized to placebo, the ONS device was turned off after the burst programming.

When entering the open-label phase, three different tonic stimulation programs were created, with frequencies of 10, 30, and 100 Hz, respectively. Pulse width (250–500 μs) and configuration (activation and polarity) of the eight contacts on the lead were adjusted to achieve the best possible bilateral occipital paresthesia based on patient feedback. The stimulation amplitude was determined by the participant’s individual perception and comfort and could be adjusted using a personal remote control given at the start of the open-label phase. The participants could change freely between the three tonic ONS programs and adjust stimulation amplitude. They were encouraged to use the stimulation continuously but were allowed intermittent use if preferred.

Randomization and blinding

After the burst programming session, participants were randomly assigned (1:1) to burst ONS or placebo. The randomization was performed as a randomly varying block design, with each block containing four, six, or eight allocations. The allocation sequence was generated by an independent data manager who had no further role in the study. The participants were assigned to treatment groups using an automated electronic randomization tool embedded in the REDCap database dedicated to the study. The randomization was performed by the implanter or device programmer, who was unblinded throughout the study and could handle any ONS-related events during the randomized trial phase without mitigating the blinding. The participants, the principal investigator (PI), and other study personnel involved in assessing outcome parameters were blinded to treatment allocation.

The used IPGs were non-rechargeable in order to avoid the need for recharging, which would jeopardize the blinding.

Outcomes

The primary outcomes were the proportion of responders to ONS treatment – both burst and tonic ONS – with a positive response defined as a reduction of at least 30% in attack frequency and safety with ONS treatment.

The secondary outcomes were the proportion of participants with a response of much improved or very much improved on the Patient Global Impression of Change scale (PGIC), treatment efficacy of burst versus tonic ONS as a preventive treatment for CCH, the proportion of participants with at least 30% reduction in pain intensity evaluated using the NRS, health-related quality of life evaluated with the EuroQoL-5D-5L questionnaire (EQ-5D-5L), self-reported sleep quality, and symptoms of anxiety and depression evaluated by the Hospital Anxiety and Depression Scale (HADS).

Safety assessment included registration and evaluation of adverse events.

Data from the last four weeks of each trial phase were used to assess the endpoints. Adverse events were continuously registered.

Statistical analysis

When calculating sample size, we hypothesized that 50% of the participants in the active treatment group would reach the primary endpoint. We hypothesized that no more than 10% [20] in the placebo group would reach the primary endpoint. With a power of β = 80%, a significance level of 0.05, a sampling ratio of 1, and using a χ² test, the trial would require a sample size of 40 participants, 20 in each group.

Baseline demographics and CH characteristics were summarized in numbers and percentages for categorical variables, means (SD) for normally distributed continuous variables, and medians [IQR] for non-normally distributed continuous variables.

Statistical analyses were conducted in collaboration with two independent, blinded statisticians not otherwise involved in the study. Based on advice from the statisticians, the prespecified analysis plan [17] was amended to better handle the multiple, repeated measurements for each participant and assess and depict the change in outcomes over time. Hence, a repeated measures mixed-effects model analysis was applied to ensure more precise estimates of the outcome measures. Allocation group, time, and interaction between allocation group and time were modeled as fixed effects, and participant ID was modeled as the random effect using an unstructured residual variance-covariance structure.

The data points were grouped into four-week time periods, resulting in one baseline time period and three time periods for each of the 12-week trial phases, respectively.

The data on attack frequency, duration, and intensity had a substantially skewed distribution and were log-transformed before analysis. Because the mean value of the outcome variables in some time periods was equal to zero, a constant with the value of one was added to all values to allow for the use of logarithm. After data was back-transformed to the original scale, this constant was subtracted from the estimated medians and the corresponding confidence intervals (CI).

The 95% CI for the estimated medians was calculated using the delta method.

Analyses were done as intention to treat. Additionally, a per-protocol analysis was performed to assess the effects of protocol violators.

The four-week period where baseline data is registered and the time of surgery is separated by a 12-week period of TENS (Fig. 1). As a notable change in headache patterns occurred during the TENS period, we decided post-hoc to include an additional set of analyses for the primary outcome in which the last data collection point of the TENS period (i.e., immediately before ONS surgery; here termed “immediate preoperative status”) was used for comparison.

The proportion of participants in each group reaching the primary endpoint of a minimum 30% reduction in attack frequency was estimated using a binomial regression model.

A marginal linear prediction model was used to calculate the likelihood of reaching the primary endpoint with burst ONS and placebo, respectively. The masking was assessed using a χ² test.

All statistical analyses were done using Stata 18.0 (StataCorp, College Station, TX, USA).

Data was unblinded simultaneously, and all analyses were done after all participants completed the randomized trial phase. There were no interim analyses.

Standard protocol approvals, registrations, and patient consents

The participants provided signed informed consent before entering the baseline period. The trial was conducted in accordance with the Declaration of Helsinki and local regulations. The protocol was approved by the independent regional scientific ethics committee (permit number VEK: 1-10-72-36-21). The Central Denmark Region authorized the acquisition, storage, and analysis of data (permit number 1-16-02-577-20). The full study protocol has been previously published and is available open-access [17].

The trial was registered on clinicaltrials.gov (NCT05023460, registered on July 27^th, 2021).

The study was funded by the Novo Nordisk Foundation. The foundation had no influence on the study design, data collection, data analysis, data interpretation, or writing of the report.

Results

Between August 27, 2021, and October 18, 2023, 116 patients were assessed for suitability. Sixty-five were excluded, and 51 entered the four-week baseline period. Before ONS implantation and randomization, 13 participants were excluded or withdrew consent. In total, 38 participants underwent ONS implantation and were randomly assigned to burst ONS (n = 19) or placebo (n = 19; Fig. 1). No trial participants or blinded investigators were unblinded during the study. One participant was excluded from the trial two weeks after the randomization due to non-compliance. Three participants (two burst ONS, one placebo) entered the open-label phase after eight weeks instead of the planned 12 weeks but remained blinded. This decision was made by the PI due to patients reporting severe concerns because of no clinical change, leading the PI to suspect an imminent risk of dropout. Their data for the last time period in the randomized trial phase were analyzed using last observation carried forward.

The baseline demographics, clinical characteristics, and CH medication of the burst ONS and placebo groups are listed in Table 1.

Table 1.

Baseline characteristics of the intention to treat population

	Burst ONS (n = 19)	Placebo (n = 19)
Demographics
Age, years	45.8 (12.8)	47.0 (10.5)
Sex
Males	11 (58%)	13 (68%)
Females	8 (42%)	6 (32%)
Smokers, current	11 (58%)	8 (42%)
Clinical characteristics
Duration of cluster headache, years	6.0 [4.5–10.0]	10.0 [7.0–16.0]
Localization of pain (primary site)
Right-sided pain	11 (58%)	14 (74%)
Left-sided pain	8 (42%)	5 (26%)
Number of attacks, weekly	20.5 [10.0–50.0]	16 [6.0–28.3]
Maximum pain intensity (NRS)	10.0 [9.0–10.0]	10.0 [8.3–10.0)
Mean pain intensity (NRS)	8.5 [7.2–10.0]	7.8 [6.0–8.3]
Attack duration, minutes	48 [35–90]	72 [25–120]
Treatment
Current use of abortive treatment
Oxygen users	13 (68%)	12 (63%)
Triptan users	10 (53%)	11 (58%)
History of transitional treatment
GON block	17 (89%)	17 (89%)
Oral prednisolone	10 (53%)	11 (58%)
Current use of preventive medication
Verapamil	5 (26%)	9 (47%)
Lithium	1 (5%)	-
Topiramate	3 (16%)	-
Other	7 (37%)	10 (53%)
No current preventives	6 (32%)	5 (26%)
Two or more previous treatment failures	18 (95%)	18 (95%)
SPG stimulation (previous)	3 (16%)	4 (21%)

Data are mean (SD), n (%), or median [IQR]

ONS = occipital nerve stimulation. NRS = numeric rating scale. GON = greater occipital nerve

SPG = sphenopalatine ganglion

We present median weekly attack frequency, responder rates, and changes from baseline in both the TENS trial phase, the randomized trial phase, and the open-label tonic ONS phase (Table 2).

Table 2.

Primary outcome: change in attack frequency

	Burst ONS (n = 19)	Placebo (n = 18)
≥30% responders (attack frequency), % (95%CI)
Last four weeks of randomized phase	18.81 (0.28 to 37.89)	50.02 (26.87 to 73.09)
Last four weeks of open-label phase	42.09 (19.91 to 64.34)	51.11 (27.32 to 74.88)
Weekly mean attack frequency, median (95% CI)
Baseline
Number of attacks	20.57 (11.17 to 29.97)	12.75 (7.25 to 18.27)
Immediate preoperative status
Number of attacks	15.33 (6.30 to 24.36)	7.52 (4.30 to 10.75)
Relative change (%) from baseline	−24.33 (−45.41 to 4.92)	−38.07 (−56.88 to −10.92)
Last four weeks of randomized phase
Number of attacks	13.87 (5.16 to 22.58)	6.67 (3.07 to 10.27)
Relative change (%) from baseline	−31.10 (−56.78 to 10.14)	−44.32 (−64.69 to −12.13)
Relative change (%) from immediate preoperative status	−11.49 (−43.78 to 37.51)	−12.50 (−43.02 to 34.53)
Last four weeks of open-label phase
Number of attacks	12.10 (4.76 to 19.45)	5.78 (1.84 to 9.72)
Relative change (%) from baseline	−39.34 (−58.62 to −11.04)	−50.10 (−70.09 to −18.33)
Relative change (%) from immediate preoperative status	−20.29 (−48.88 to 24,02)	−20.43 (−48.71 to 23.54)

Data is presented as the estimated medians from the mixed model analysis; note that the values do not equal the numeric medians

Immediate preoperative status (n = 30): Due to scheduling of their implantations, two participants (one from each group) received only two weeks of TENS, and four participants (three in burst ONS group) received eight weeks of TENS

We observed a large heterogenicity in baseline attack frequency between the two groups (Table 1). To account for this difference, we specified a random intercept in the mixed-effects model, capturing the variability across the groups. This adjustment did not affect the outcome (data not shown).

In the last four weeks of the randomized trial phase, the proportion of ≥30% responders was 18.81% (0.28%–37.89%) in the burst group and 50.02% (26.87%–73.09%) in the placebo group. Hence, in this period, the chance of achieving a ≥30% reduction in attack frequency compared to baseline was 31.20% (1.29%–61.23%, p = 0.042) higher in the placebo group compared to burst ONS.

In the last four weeks of the open-label phase, where both groups received tonic ONS, the proportion of ≥30% responders was 42.09% (19.91%–64.34%, p < 0.001) in the burst ONS group and 51.11% (27.32%-74.88% p < 0.001) in the placebo group. During the open-label phase, there was no statistical difference in the likelihood of achieving a ≥30% reduction in attack frequency between the groups (p = 0.63).

Figure 2 shows the relative number of weekly attacks (%) from baseline over time during all trial phases.

Fig. 2.

Change in attack frequency. (A) Relative change in attack frequency from baseline (100%) and throughout the TENS, randomized, and open-label trial phases. (B) Change in attack frequency throughout the entire study period. The dots represent the estimated median in each time period in the mixed-effects model; please note that this value is not equal to the numeric change in medians. Note that between the TENS trial phase and the randomized trial phase, the participants underwent ONS implantation and a 14-day grace period, which is not indicated in the figure as no data were collected during this period. *In the TENS trial, two participants (one from each group) had only two weeks of TENS and did not contribute with data (n = 36). ONS = occipital nerve stimulation. TENS = transcutaneous electrical nerve stimulation

The median attack intensity, duration, and changes in these parameters are presented in Table 3 and Fig. 3A–C. In the last four weeks of the randomized trial phase, 18.75% (0.34% to 37.83%, p = 0.055) in the burst ONS group and 33.33% (11.56% to 55.11%, p = 0.003) in the placebo group achieved at least 30% reduction in maximum pain intensity compared to baseline. In the last four weeks of open-label, the proportions were 31.58% (10.69% to 52.51%, p = 0.003) and 55.56% (32.60% to 78.51%, p < 0.001) for the burst ONS group and placebo group, respectively.

Table 3.

Secondary outcome measures

	Burst ONS (n = 19)	Placebo (n = 18)
Maximum pain intensity (NRS), median (95% CI)
Baseline
Max. NRS	9.39 (8.94 to 9.84)	8.74 (7.78 to 9.70)
Immediate preoperative status
Max. NRS	8.37 (5.34 to 11.39)	7.13 (5.80 to 8.47)
Relative change (%) from baseline	−9.92 (−35.03 to 24.91)	−16.49 (−28.02 to −3.24)
Last four weeks of randomized phase
Max. NRS	6.97 (4.59 to 9.35)	6.21 (4.72 to 7.70)
Relative change (%) from baseline	−23.33(−42.51 to 2.43)	−26.04 (−37.71 to −12.13)
Last four weeks of open-label phase
Max. NRS	6.89 (5.07 to 8.72)	4.88 (2.83 to 6.94)
Relative change (%) from baseline	−24.02 (−38.04 to −6.78)	−39.61 (−54.49 to −19.82)
≥30% responders, % (95% CI)
Last four weeks of randomized phase	18.75 (0.34 to 37.83)	33.33 (11.56 to 55.11)
Last four weeks of open-label phase	31.58 (10.69 to 52.51)	55.56 (32.60 to 78.51)
Mean pain intensity (NRS), median (95% CI)
Baseline
Mean NRS	8.11 (7.34 to 8.88)	7.11 (6.30 to 7.92)
Immediate preoperative status
Mean NRS	6.33 (4.35 to 8.31)	5.61 (4.61 to 6.61)
Relative change (%) from baseline	−19.54 (−38.93 to 6.18)	−18.51 (−29.33 to −6.04)
Last four weeks of randomized phase
Mean NRS	6.36 (4.23 to 8.49)	5.00 (3.79 to 6.21)
Relative change (%) from baseline	−19.20 (−38.18 to 5.71)	−26.01 (−37.52 to −12.49)
Last four weeks of open-label phase
Mean NRS	6.00 (4.37 to 7.72)	4.00 (2.35 to 5.66)
Relative change (%) from baseline	−22.59 (−35.31 to −7.47)	−38.31 (−52.72 to −19.59)
Attack duration (minutes), median (95% CI)
Baseline
Duration	53.83 (38.34 to 69.33)	55.67 (35.78 to 75.57)
Immediate preoperative status
Duration	29.45 (11.37 to 47.53)	44.62 (29.48 to 59.76)
Relative change (%) from baseline	−44.67 (−68.19 to −3.04)	−19.50 (−37.48 to 3.63)
Last four weeks of randomized phase
Duration	38.46 (16.71 to 60.20)	40.83 (26.13 to 55.54)
Relative change (%) from baseline	−28.02 (−55.41 to 16.08)	−26.24 (−42.89 to −4.51)
Last four weeks of open-label phase
Duration	38.72 (21.75 to 55.71)	26.20 (6.17- to 6.23)
Relative change (%) from baseline	−27.53 (−48.82 to 2.54)	−52.00 (−75.88 to −4.46)
Proportion of responders on the PGIC scale*
Last four weeks of randomized phase	4 (21%)	9 (50%)
Last four weeks of open-label phase	11 (58%)	13 (72%)
EQ-5D-5 L VAS score, mean (SD)
Baseline	34.63 (23.40)	45.5 (21.32)
Last four weeks of randomized phase	51.46 (27.26)	63.82 (23.05)
Last four weeks of open-label phase	46.88 (30.31)	66.33 (26.33)
HADS-A subscale score < 8, n (%)**
Baseline	5 (26%)	8 (44%)
Last four weeks of randomized phase	13 (68%)	10 (56%)
Last four weeks of open-label phase	11 (58%)	11 (61%)
HADS-D subscale score < 8, n (%)**
Baseline	10 (53%)	9 (50%)
Last four weeks of randomized phase	14 (74%)	17 (94%)
Last four weeks of open-label phase	11 (58%)	12 (67%)
Self-rated sleep quality, n (%)***
Baseline	1 (5%)	2 (11%)
Last four weeks of randomized phase	3 (16%)	7 (39%)
Last four weeks of open-label phase	6 (32%)	6 (33%)

Data is presented as n (%), mean (SD) or as the estimated medians (95% CI) from the mixed model analysis; note that these values do not equal the numeric medians

*A response of “much improved” or”very much improved” was considered a responder on the PGIC scale

**Subscale score < 8 = low risk of developing anxiety (HADS-A) or depression (HADS-D)

***The proportion of participants reporting a good or very good sleep quality

NRS = numeric rating scale. PGIC = patient global impression of change. EQ-5D-5 L = EuroQoL5-Dimensions at 5 Levels. VAS = visual analog scale. HADS = hospital anxiety and depression scale

Fig. 3.

Change in attack intensity and duration. Relative change in maximum attack pain intensity (A), in mean attack pain intensity (B), and in duration (C) from baseline (100%) and throughout the TENS, randomized, and open-label trial phases. *In the TENS trial, two participants (one from each group) had only two weeks of TENS and did not contribute with data (n = 36). TENS = transcutaneous electrical nerve stimulation

The per-protocol analysis of the primary outcome showed similar results as the intention-to-treat-analysis (data not shown).

After the randomized trial phase, 4 (21%) in the burst ONS group and 9 (50%) in the placebo group stated to be much or very much improved as evaluated on the PGIC scale. After the open-label trial, this was 11 (58%) and 13 (72%) in the burst ONS and placebo group, respectively.

At the end of the 12-week double-blind, randomized trial phase, the participants were asked to guess their assigned treatment group. Eight out of 19 (42%) participants in the burst ONS group correctly guessed their allocation. This was the case for nine out of 18 (50%) in the placebo group (p = 0.63). The PI correctly guessed the treatment assignment in 17 out of 38 (45%) participants (p = 0.87).

Eight participants experienced temporary, self-limiting occipital dysesthesia, resolving within weeks, which was the most frequent adverse event. Eleven serious adverse events were registered (Table 4). Five were hardware-related—i.e., IPG dislocation, preterm battery depletion, or revision of stimulation lead due to insufficient paresthesia coverage. Three of the serious adverse events were considered unrelated to the treatment. All surgical revisions were performed before randomization or after entering the open-label phase.

Table 4.

Adverse events

	Burst ONS (n = 19)	Placebo (n = 18)
Serious adverse events
Serious adverse events, hardware related
Total	4	1
IPG displacement	2 (11%)	0
Preterm IPG depletion	1 (5%)	0
Insufficient paresthesia coverage	1 (5%)	1 (6%)
Serious adverse events, biological complications
Total	3	0
Infection, profound	1 (5%)	0
Impaired wound healing	1 (5%)	0
Local IPG pain (requiring revision)	1 (5%)	0
Serious adverse events, not related to the trial	2 (11%)	1 (6%)
Adverse events
Adverse events, hardware related
Total	1	2
Local IPG pain (sustained)	1 (5%)	2 (11%)
Adverse events, biological complications
Total	3	6
Occipital dysethesia (spontaneous remission)	3 (16%)	5 (28%)
Infection, superficial	0	1 (6%)

Data are presented as n (%). The safety population included all randomly allocated participants

Participants could have had more than one adverse event

IPG: Implantable pulse generator

Discussion

To our knowledge, this is the first randomized, placebo-controlled, double-blind trial assessing the efficacy of ONS as a preventive treatment for CCH. We found that both burst and tonic ONS reduced attack frequency, but contrary to what we expected, both based on previously published data and our own clinical experience, burst ONS was not superior to placebo.

Overall, the ONS treatment was well tolerated and had an acceptable safety profile. In particular, no lead migration or breakage was registered, which has previously been described as a frequently recurring cause of revision surgery.

Our study has several strengths. The placebo used is of very high fidelity: Before treatment allocation, all participants underwent the same surgical procedure using identical, fully functioning ONS systems and subsequent device programming. Hence, the treatment of the participants differed only in the group to which they were randomly allocated.

Introducing a 14-day postoperative grace period before the burst programming session and randomization improved the conditions for optimal programming and fine-tuning of the stimulation settings.

Each trial phase was kept at 12 weeks, which has previously been shown to be enough time to detect a treatment response [11]. A 12-week phase also ensured that the participants were not kept on a potentially ineffective treatment (placebo) longer than necessary. Conversely, shorter trial periods could potentially increase the risk of a carry-over effect from the preceding treatment modality, thereby increasing the risk of bias. For the same reason, we focused the analyses on the last four weeks of each trial phase.

The overall trial design, including endpoints and the length of baseline and treatment periods, followed the Guidelines of the International Headache Society for Controlled Clinical Trials in CH [21].

The automatic electronic headache registration with a reminder function reduced the risk of missing data.

Furthermore, we achieved a successful masking of both study participants and outcome assessors.

Several ONS implantation techniques have been proposed, and the approach used in this study, having been used successfully in our department for years [16], differs somewhat from the approaches often described by others. Importantly, postoperatively and at the programming sessions, our patients reported upward spreading bilateral paresthesia, which has been shown to increase the probability of clinical effectiveness [22]. Because the IPG is implanted under the clavicle with a short distance to the lead implantation site, the lead is subject to less strain compared to abdominal or gluteal placements, which greatly reduces the risk of lead migration and eliminates the need for an extension.

In the study, we used the paresthesia-free stimulation paradigm MicroBurst (Boston Scientific) as the verum in the randomized trial phase. The proportion of treatment responders was larger when the participants were treated with tonic ONS than with burst ONS. Although the study was not designed as a non-inferiority study and cannot directly compare the two ONS paradigms, this observation raises the question of whether MicroBurst, in fact, has sufficient efficacy for CCH prevention. Furthermore, it seems paradoxical that burst ONS was, in fact, less efficient in attack prevention than the placebo with no stimulation.

Regardless of the difference in effect between burst and tonic ONS, the substantial and sustained treatment response in the placebo group cannot be explained by the possible inefficiency of MicroBurst as verum.

The exact mechanism of action for tonic ONS and burst ONS is unknown. Open-label studies with small sample sizes have suggested a clinical outcome of burst ONS on par with tonic ONS [15, 16], but no data from large trials are available to confirm this. A single-case imaging study on a healthy individual who volunteered to have ONS electrodes implanted indicated that burst ONS induces activation and deactivation in the same brain areas as tonic ONS, however, less pronounced [23].

Differences in stimulation design between various paradigms marketed as “burst” have been debated [24]. Yet, MicroBurst has the desired characteristic of being paresthesia-free and has been proven to reduce pain in clinical studies on spinal cord stimulation [25].

Even though ONS has shown potential effects as a last-resort preventive therapy in patients with severe and otherwise refractory CCH, the results of this controlled study indicate that a substantial part of the treatment effect may be attributed to a placebo response. This finding must be suspected to apply to all previous studies on ONS as a preventive therapy for CCH, which did not include a placebo control.

The only other randomized controlled trial evaluating ONS for CCH is the 2021 ICON trial [11]. Because paresthesia-free stimulation was not available at the time of the ICON study initiation, only tonic stimulation was employed, and the study could not include a true placebo control. Instead, two different tonic stimulation conditions were applied: high (100%) as the active comparator and low (30%) as the supposedly inactive control, assuming no perceptible difference in the paresthesia elicited by 30 and 100% stimulation. The results showed no difference between the two groups.

The authors argue that placebo is an unlikely explanation of the findings due to a pronounced effect in severely burdened refractory CCH patients and a sustained treatment effect; this finding is supported by a long-term follow-up study including the same group of ICON study participants [12] and by a French multicenter register study [13]. There is, however, no evidence supporting the idea that placebo responses are only short-term. Placebo-controlled surgical trials have documented a significant response to sham surgery lasting for years [26, 27]. Furthermore, a placebo response can likely be maintained over time by control visits and adjustment of stimulation parameters if the patients have reason to believe they are receiving an effective treatment. Trials involving surgery [28] and device implantation [29] are associated with a larger placebo response regardless of the number of surgical interventions during the trial period. The same applies to trials involving chronic pain patients, and when the outcome measure is subjective, such as pain [30]. All these are relevant potential drivers of placebo response in studies of ONS for CCH, and may very well be a part of the explanation for our finding Furthermore, a placebo response is not an unknown phenomenon in the CCH patient population, and trials on otherwise promising new therapies that have shown an effect on episodic CH have not resulted in treatment recommendations for CCH as an indication because the sham/placebo was found to be non-inferior to the active treatment [31, 32].

In this study, only burst ONS was placebo-controlled. Hence, we cannot directly compare the placebo group to the tonic ONS. However, it is noticeable that the proportion of responders in the placebo group resembles that in the open-label trial phase. Thus it is unknown whether a placebo response of similar magnitude is present in other studies evaluating tonic ONS for CCH and further studies are needed.

Limitations

The study turned out to be underpowered. In the sample size calculation, we predicted a 10% placebo response; however, the placebo response was above 40%. The high placebo response rate could affect the sensitivity of the sample size calculation of this trial, given its relatively small sample size compared with a large multicenter trial, and must be taken into consideration when interpreting the results. However, a sufficiently marked treatment response in the active group should still be expected to manifest in the data.

Our data showed a marked imbalance in baseline characteristics between the two groups, with the burst ONS group having a notably higher weekly attack frequency at baseline than the placebo group, which is likely to be explained by the modest sample size. This heterogeneity in the burst group was larger during the entire trial.

Even though almost all the study participants had severe, long-lasting, and medically refractory CCH, we cannot rule out periodical fluctuations in attack frequency that could possibly affect the outcome. Still, we have no reason to believe such factors would differ markedly between the groups.

However, factors such as baseline heterogeneity and potential fluctuation in headache frequency increase the risk of introducing bias, such as regression towards the mean. The mixed-effects model was adjusted to condition on initial headache frequency and thereby accommodate the difference in baseline attack frequency between groups. This did not affect the outcome (unadjusted data not shown). However, although including randomization in the trial design and introducing baseline adjustment in the statistical model reduces the influence of regression towards the mean, the modest sample size limits precision and precludes ruling out all residual bias.

Participating in a study setting that involves weekly symptom registration, regular clinical follow-up visits, and easy access to contact with the PI are all contextual factors that may contribute to an increased placebo response [33], but the marked difference between the two groups in favor of the placebo group was unexpected. Our study did not include an untreated control group to evaluate the natural course of the disorder and other contextual factors, but a significant outcome influence was deemed less likely in our severely affected cohort.

Notably, both groups experienced a significant reduction in attack frequency during the TENS trial, which was sustained through both the randomized and open-label trial phases. We cannot rule out a possible carry-over effect from the TENS treatment lasting into the randomized trial phase, but we took measures to minimize such a risk. Only data from the last four weeks of each trial phase were used in the analysis. Along with the 14-day postoperative grace period, the participants in the placebo group had been without active treatment for eight to 10 weeks before the time interval in which data was included in the analysis. Hence, the stable and sustained effect in the placebo group during the entire 12-week randomized trial phase is unlikely to be explained by a carry-over effect from the TENS treatment in this group of participants with severe, unremitting CCH who have previously failed several treatment options.

With respect to the limitations of this study, we wish to exercise caution in drawing too firm conclusions from its results, and we do not find that the evidence produced in this study is a sufficient basis for revising current treatment guidelines (where, indeed, neither MicroBurst nor other burst stimulation paradigms are mentioned). However, the findings of this study indicate that a placebo response may constitute a considerable part of the therapeutic effect of ONS for CCH and cannot be ignored. Therefore, we would point out the importance of shifting focus from uncontrolled studies to rigorously conducted randomized trials, preferably with a placebo group, and we encourage our colleagues in the field to work towards this aim.

Also, we encourage devising trial designs to investigate the placebo effect in studies with patients already on established ONS treatment, for instance, using an open-hidden design.

Conclusion

In conclusion, ONS was safe and well-tolerated. Both burst and tonic stimulation reduced attack frequency in patients with severe CCH, but the preventive effect of burst stimulation was not superior to placebo. Although not directly comparable, the same seemed to apply to tonic ONS.

The results of this study call attention to ensuring sufficient placebo control when planning further studies.

Abbreviations

AE

Adverse event

CH

Cluster headache

CCH

Chronic cluster headache

GON

Greater occipital nerve

HADS

Hospital Anxiety and Depression Scale

IPG

Implantable pulse generator

ONS

Occipital nerve stimulation

PI

Principal investigator

PGIC

Patients’ Global Impressions of Change

NRS

Numeric Rating Scale

TENS

Transcutaneous electrical nerve stimulation

TCC

Trigeminocervical complex

Author contributions

KM, JCHS, and RHJ designed the study. ISFA searched the relevant literature. The study data was collected by ISFA and ASP, analyzed by ISFA, and interpreted primarily by ISFA and KM. ISFA wrote the first draft of the manuscript, with a significant contribution from KM. ISFA made tables and figures. KM verified the underlying data. All authors reviewed and critically revised the manuscript and approved the final version before submission.

Funding

The study was funded by a grant from the Novo Nordisk Foundation (NNF19OC0058805).

Data availability

The datasets generated and analyzed during the current study are not publicly available but are available from the corresponding author on reasonable request.

Declarations

Ethics approval

The protocol was approved by the independent regional scientific ethics committee (permit number VEK: 1-10-72-36-21). The Central Denmark Region authorized the acquisition, storage, and analysis of data (permit number 1-16-02-577-20). All participants provided signed informed consent before enrolling in the study.

Consent for publication

Not applicable.

Competing interests

JCHS has received a restricted research grant (for the institution) from the Novo Nordisk Foundation. KM has received teaching fees from Medtronic and consulting fees from Salvia BioElectronics. KM and JCHS are co-owners and co-founders of the neuromodulation database company Neurizon. RJ received restricted research grants (for the institution) from Lundbeck Pharma and the Novo Nordisk Foundation. RJ has received personal fees for educational and teaching activities from Pfizer, Teva, Novartis, Abbvie, Lundbeck Pharma, and Eli-Lilly, and a fee (for the institution) for serving on the Lundbeck Pharma Advisory Board. RJ is the chair of the Master of Headache Disorders, Director of the Danish Headache Center, and unpaid activities as Director in Lifting the Burden. ASP has received a restricted research grant (for the institution), conference attendance from Lundbeck Pharma, and personal fees from Pfizer for teaching activities. ISFA has nothing to declare.