The study was a single-center, open-label, one-armed interventional trial conducted at St. Olav’s University Hospital in Trondheim, Norway, from May 2022 to September 2022. The target population was adults with episodic migraine from all of Norway. The total duration of study participation for each participant was 12 weeks.

Inclusion criteria were (1) 18 years of age or older, (2) episodic migraine with or without aura diagnosed by a neurologist or physician per the International Classification of Headache Disorders, Third Edition, (3) having kept a headache diary with at least 80% adherence as part of routine clinical care in the last 28-days prior to inclusion, (4) 4 to 14 migraine attacks per 28-day period in the 3 months prior to screening, (5) at least three months of experience with smartphone and access to an iOS or Android phone at home, (6) onset of migraine before the age of 50 years, and (7) signed written informed consent.

Exclusion criteria were (1) having a continuous background headache that never disappears completely; (2) more than 14 days of headache (all types) per 28-day period; (3) participants diagnosed with trigeminal autonomic cephalalgias and neuralgias or (4) secondary headache conditions including medication overuse headache according to the International Classification of Headache Disorders 3rd edition; (5) participants with pathologies that inhibit use of the device according to the instructions for use (eg, blindness and deafness); (6) participants currently using migraine prophylaxis (pharmacological or nonpharmacological); (7) participants who have previously failed 3 prophylactic pharmacological treatments; (8) opioid use (>3 d per month) or use of barbiturates at time of screening; (9) alcohol overuse according to The International Classification of Diseases, Tenth Revision, Clinical Modification Code F10.1 or illicit drug use; (10) participation in another clinical investigation; and (11) participants unlikely to follow Clinical Investigation Plan or where treatment seems futile in the opinion of the Investigator or have demonstrated an inability to sufficiently adhere to headache diary entries (<80%).

The medical device tested in this study is a therapist-independent biofeedback system (Cerebri) consisting of a smartphone app (Figure 1) and 2 noninvasive wireless sensors (Figure 2). One sensor measures neck muscle tension as surface electromyographic (EMG) voltage from the skin over the upper trapezius muscle. The other sensor is placed on one index finger to measure HRV by photoplethysmography and peripheral skin temperature. As part of the product development, validation of sensor performance showed agreement with simultaneous control measurements made with stationary neurophysiological equipment [19]. The 12-week treatment phase with Cerebri consisted of daily biofeedback therapy sessions of a minimum of 10 minutes duration each.

Since the device is intended for independent use without the need for any supervision, no instructional sessions were offered upon receipt of the device or during the course of the study. A limited, written step-by-step on-screen guidance was provided to onboard the participant prior to the first 3 biofeedback sessions. The participants were introduced to 3 biofeedback modalities with real-time on-screen feedback: temperature biofeedback during the first session, HRV biofeedback in the second session, and EMG biofeedback in the third session. During the remaining sessions, the participants could freely alternate between the 3 modalities at their own discretion. Additional user instructions were provided in a user manual and in a library in the app, where the users could read more about biofeedback and relaxation techniques. During the biofeedback sessions, an adjustable breathing pacer was accessible to users.

All study visits were conducted remotely, and the sensor kit was sent by postal service. Participants accessed the app through a unique link provided via email, compatible with both Android and iOS devices. The app required an initial internet connection for download, but could subsequently operate without an active internet connection. Follow-up included 2 video consultations with a neurologist (IGW) at inclusion and the 12-week final visit; and 3 telephone follow-ups (wk 2, 4, and 8) with a study nurse. Participants were required to keep the daily eDiary integrated with the smartphone app throughout the study participation. Notifications for performing daily biofeedback sessions and diary entries were available for iPhone users, but did not work for Android users in the trial. Android users were therefore encouraged by study personnel to add a reminder manually in the integrated phone calendar to ensure this. The issue was persistent throughout the trial. Although participants were asked to register daily, the eDiary allowed registration within a 3-day interval (present day, and the 2 previous days). eDiary users were first asked if they had a headache or not. If a headache was reported, the app prompted the participant to answer additional questions about the type of headache (migraine or other), as well as the duration, intensity, and medication use. Questionnaires regarding usability and efficacy were collected via Helsenorge.

Descriptive data were summarized as means with SDs or medians with IQRs according to normality distributions. Assessment of normality was based on visual inspection of histograms. Adherence rates to the eDiary and biofeedback program were calculated for each 28-day period separately. Summary statistics for MAUQ were calculated for weeks 1 and 12. Changes in the number of days with migraine and headache in each 28-day period compared to the last 28 days prior to inclusion were analyzed with a mixed logistic regression model with participant ID included as a random variable. The model estimated the odds ratio for each 28-day period. This odds ratio represented the estimated daily odds of migraine or headache in each 28-day period compared to the baseline period. The estimated number of migraines or headaches in each 28-day period and the difference from baseline were calculated based on the regression coefficients using the nlcom command in Stata (version 17.0; StataCorp LLC). The use of a mixed logistic regression model allowed the inclusion of all available daily data, also from participants who had not completed all eDiary entries during each 28-day period. The change in MSQ domain scores was analyzed using mixed linear regression models with the mean change in each domain score estimated from these models.

Semistructured interviews were held through video or phone and recorded. Transcripts were stored on the Miro (RealtimeBoard Inc) platform during analysis, where they were grouped, mixed, and sorted in several rounds to identify patterns of themes that were relevant to the usability experience.

The study was approved by the regional ethics committee (457654) and the Norwegian Medical Products Agency (NoMA; 22/04658‐4; Eudamed: CIV-NO-22-03-039126), and registered at ClinicalTrials.gov (NCT05454319) prior to initiation. Trial information was disseminated via multiple communication channels. Interested parties made contact by email. All participants received written information via email prior to the first study visit. Eligibility assessment and a review of the study information were performed by a consultant neurologist (IGW) with headache expertise in a video meeting. If eligible, consent was collected by digital signature via the national government-operated electronic data entry application Helsenorge (Norsk helsenett SF). Data collection for the study was initiated after electronic consent. Data were deidentified through pseudonymization. No payment was made to the study participants for their participation.

A total of 20 participants were enrolled in the study intervention. All 20 participants contributed to the evaluation of the feasibility of the Cerebri biofeedback system. One participant withdrew consent, and one was lost to follow-up (Figure 3). The majority of participants were female and in full-time employment, with a mean of 6.7 (SD 2.8) migraine days per month (Table 1). One participant was pregnant during the entire study period.

Out of 20 participants, the majority continued to regularly record information in the daily eDiary throughout the study period, with 15 (75%) participants still recording information on at least 70% of days (20 d) in weeks 9 to 12 (Table 2). The proportion of participants completing at least 70% of biofeedback sessions decreased throughout the study, with 16 (80%) participants during weeks 1 to 4, and 4 (20%) participants during the final 4 weeks (Table 2).

The Cerebri app received favorable user perceptions, with MAUQ summary score for the ease of use subscale at week 1 (median score 6.2, IQR 5.6-6.6) and at week 12 (median score 6.0, IQR 5.2-6.8; Figure 4). Summary scores for the interface and satisfaction and usefulness subscales were 6.1 (IQR 5.7-6.7) and 5.0 (IQR 4.2-5.6) at week 1, and 5.8 (IQR 5.3-6.6) and 4.8 (IQR 3.5-5.2) at week 12, respectively.

The app was regarded as both easy to learn and to use (questions 1‐5), with stable median scores of 6 or 7 for almost all of these questions at both weeks 1 and 12. The exception was a notable decline in median scores from 6 (IQR 4-7) at week 1 to 4 (IQR 2.5-7) at week 12 for participants’ ability to recover after having made a mistake using the app (question 5).

MAUQ questions concerning the app’s interface and user satisfaction (questions 6‐12) also received favorable scores. However, there was a noticeable decline in the median score from 7 (IQR 6-7) at week 1 to 5 (IQR 5-7) at week 12 for question 11, which assessed users’ likelihood of using the app again. Nevertheless, overall satisfaction with the app (question 12) remained high with a median score of 7 (IQR 6-7) at week 1, and 6 (IQR 6-7) at week 12.

Furthermore, the scores for MAUQ question 13, which assessed the app’s perceived usefulness for health and well-being, remained consistently high with a median score of 6 (IQR 4-7) and 6 (IQR 4-6) at weeks 1 and 12, respectively. Meanwhile, MAUQ question 18, which evaluated whether the app provides an acceptable way to receive health care services, showed a median score of 5.5 (IQR 4-6) at week 1, decreasing slightly to 5 (IQR 4-6) by week 12.

MAUQ question 17, which addressed whether the app could be used without or with a poor internet connection, maintained a median score of 4 (IQR 4-6) at week 1 and 4 (IQR 3-5) at week 12. Notably, this item had fewer respondents compared to other questions, with 9 participants responding at week 1 and 11 participants at week 12, suggesting that not all users encountered situations where the app’s functionality under these conditions was tested.

A total of 6 participants were recruited to participate in the semistructured interviews. Five midway user interviews were completed in total. Three of these interviewees also partook in the final user interview. All interviewees reported that Cerebri was easy to both learn and use. Finding time for daily training sessions and implementing this as a daily routine activity seemed to be the most challenging part. However, despite this challenge, interviewees expressed a sense of satisfaction in being able to dedicate 10 minutes to achieving a state of complete relaxation, which they found rewarding in itself.

A concern was identified that the app focused excessively on achievements and earning high scores. Participants feared that they might not experience the intended treatment effects unless they attained the top achievements. When probed further about their reasoning, participants attributed significance to such score achievements, assuming a correlation between achieving them and experiencing a reduction in migraine symptoms.

The finger temperature parameter was reported to be preferred among users, as it was the easiest to understand whilst also providing the most concrete goal during biofeedback sessions. Output user data from all completed biofeedback sessions from all study participants identified the same preference, with a total collective screen time of 75 hours used within the finger temperature modality across a total of 132 hours in 795 sessions. In comparison, 22 and 35 hours were used in the EMG and HRV modalities, respectively. The EMG parameter appeared to provide little feedback for participants to observe and improve upon, except for one individual who reported having tense muscles and experiencing a strong connection between muscle tension and migraines. All 5 participants expressed difficulties understanding the goal of HRV measurement, how to achieve it, or both.

Several bugs in the app occurred during the second half of the study, resulting in difficulties launching the app or starting biofeedback sessions. Notably, all iPhone users were involuntarily logged out of the app at one point due to these issues. All 3 participants partaking in the final user interview mentioned bugs as a reason for reduced motivation. In total, 13 unique bugs were reported by one or more participants during the study (not limited to the interviewees). These were perceived as disruptive to the user experience and led to disruptions in use patterns, ultimately impacting user motivation and compliance. The degree to which the 3 participants experienced a drop in motivation due to technical errors was not quantified specifically during the interview, but addressed for the entire population in questions 4 and 5 in the MAQU questionnaire.

Three AEs were recorded during the trial period (Multimedia Appendix 2). All AEs were assessed as nonrelated to the medical device. Thus, no adverse device effects and no serious adverse events occurred. One participant was pregnant during trial participation. No AE were registered for that participant, and the child was born healthy at term.

We observed a statistically nonsignificant reduction in the number of migraine days compared to baseline in all three 28-day periods, but no clear change in the number of headache days (Table 3). At all timepoints, approximately 50% (8/16) of participants described overall health conditions as “unchanged” on the PGI-C questionnaire (Figure 5). However, 22% (4/18) and 24% (4/17) of participants described their condition as “much improved” in week 8 and 12, respectively. No reports of worsening conditions were reported after week 4. There was a trend toward improvements in all MSQ domains throughout the study, reaching statistical significance for the domain “role function-restrictive” in week 4 and the domain “emotional function” in week 12 (Table 4).

This study was designed to address the feasibility of the home-based, therapist-independent biofeedback system Cerebri as a preventive treatment in adults with episodic migraine to serve as a preparatory step toward a clinical efficacy trial. Additionally, major emphasis was given to assessing the usability and the safety of the invention.

Daily biofeedback session adherence declined throughout the course of the study from 80% in weeks 1‐4 to 20% in weeks 9‐12. The decline in diary adherence was much less pronounced, with a reduction from 90% to 75% in the same time interval. Low adherence is a common concern for nonpharmacological headache treatments [22]. This is also true for biofeedback treatment, as seen in a previous pilot on a precursor to the Cerebri system [18], as well as in another mHealth app for biofeedback intervention in migraine [23]. In this study, several causes for low adherence were identified.

First, software bugs in the last part of the study period hampered adherence, causing at one point all iPhone users to be logged off the app. Second, as eDiary adherence was better, software malfunction cannot be the sole cause for the declining biofeedback adherence. Usability assessments, including MAUQ scores, identified several other software elements that should be improved. Details were elucidated in semistructured interviews held in a subset of participants for product development purposes. The achievement features conveyed unrealistic goals for some participants, and failure to reach them was misinterpreted as a failed session by some. In addition, the goal of the HRV biofeedback training was difficult to understand.

This study was not designed to assess efficacy, and outcomes must be interpreted with caution. There was an observed tendency for improvements in several headache parameters, including quality of life measures. Notably, a not statistically significant average reduction in migraine days was observed across all three 28-day intervals compared to baseline, with an estimated reduction of 1 migraine day in weeks 1‐4 and weeks 5‐8. On PGI-C scoring, 24% (4/17) of participants described their overall health condition as “much improved” at week 12.

The intervention was safe, with no device-related or serious AE.

This study highlights the potential for therapist-independent, multimodal biofeedback systems to address unmet needs in migraine management, despite challenges in adherence. Adherence challenges, observed in this and prior studies [18,23], emphasize the need for robust technology platforms and user-centered design to secure long-term engagement, particularly in home-based interventions.

A possible explanation for the decreasing adherence to the biofeedback sessions, despite consistently high eDiary adherence, is that users experienced some sort of improvement from the therapy, and therefore, adjusted the frequency of treatment sessions while retaining the treatment effect. However, it may also be an expression of a lack of perceived effect, and either way, user interviews did not identify this as a major mechanism.

There is no consensus on the optimal treatment duration or frequency of home-based mHealth behavioral interventions in migraine. Traditional therapist-dependent biofeedback is typically provided as hour-long sessions once or twice weekly, sometimes with intermittent home-training of shorter duration. In a recent pilot study examining a home-based single-modality surface EMG biofeedback intervention for episodic migraine, participants engaged in weekly sessions lasting 40‐45 minutes under the remote guidance of a biofeedback therapist over a 6-week period [24]. Participants were instructed to do additional sessions of 30 minutes 3 times a week and keep track of their home practice. Participants attended on average 4.8 (SD 1.2) of the 6 sessions, whereas adherence to the therapist-independent sessions was not reported in the paper.

It can be hypothesized that increased treatment duration yields larger treatment effects. For home-based therapist-independent treatment, long sessions may hamper adherence, whereas short sessions may reduce potential efficacy. In a pilot study on a combined HRV-biofeedback-virtual reality device, participants attended an initial in-person session, after which they were instructed to perform sessions at home a minimum of 3 times per week, with each lasting at least 10 minutes [10]. There was an attrition rate of 9 out of 25 participants in the intervention group, compared to 3 out of 25 in the control group. No statistically significant reduction in headache days between the groups was observed. Results from this study suggest the possibility of significant dropout even with less frequent sessions. Therefore, increased session duration or frequency may not necessarily be determining factors for adherence to the intervention. This pilot study did identify adherence challenges, though other underlying mechanisms seem more important than session duration and frequency itself. Future work should empirically investigate optimal treatment duration and frequency.

Electronic behavioral interventions are dependent on sufficient user satisfaction to secure adherence. The overall positive perceptions of the app’s ease of use and utility underscore Cerebri’s potential to serve as an accessible, self-administered treatment option. However, the decline in median scores for certain usability aspects, particularly users’ likelihood of using the app again, could signal the need to revise and improve certain elements of the app’s design. On the other hand, a lack of perceived clinical effect could also explain such a finding. Regardless, addressing concerns related to usability features is crucial to maintaining patient compliance. To that end, revisions related to app achievement features should be a priority to mitigate misunderstandings that may affect participants’ motivation and perceived treatment effectiveness. Other strategies to reinforce habit formation should be explored.

Furthermore, our results suggest that the app interface of the HRV biofeedback modality should be revised to facilitate the correct performance of training within this modality. This could be achieved by including a visual guiding line matching the optimal sinus curve for the selected breathing frequency. In addition, improving HRV-related textual information in the app may increase user accessibility.

This pilot study has several limitations. First, the sample size of 20 participants makes the study underpowered to make claims regarding efficacy. Additionally, the absence of a control group limits our ability to attribute effects specifically to the intervention. A notable limitation of our qualitative analysis is the small sample size in the final user interviews, where only 3 participants participated. Additionally, the sequential selection of interview participants during the week 4 follow-up appointments could have introduced a selection bias as they were not randomly drawn. However, we aimed to include a diverse representation of age, gender, and adherence, and thus, opted for a sequential selection.

Participants were biofeedback naïve, but had already kept a headache diary with a minimum of 80% adherence in the 4 weeks prior to inclusion as part of routine clinical care, and were experienced smartphone users. Thus, study results may not be representative of a smartphone and eDiary naïve population.

Furthermore, as the study included only individuals with episodic migraine, caution is advised when inferring results to other patient groups, including chronic migraine or other primary headache disorders.

Cerebri is a unique multimodal, app-based, and therapist-independent biofeedback intervention that, in this study, underwent a rigorous feasibility and usability evaluation. The main feasibility obstacle was declining adherence to daily biofeedback sessions throughout a 12-week treatment period. On the other hand, eDiary adherence was consistently high. Usability issues and software bugs were identified and will be addressed in the final product development of Cerebri prior to a pivotal randomized controlled trial.