Results

Clinical efficacy

Participant flow

The Consolidated Standards of Reporting Trials (CONSORT) diagram illustrates the recruitment process (see Figure 2). During trial recruitment, 685 individuals completed the initial telephone eligibility screening, and at this stage, N = 261 were excluded for not meeting inclusion criteria, and N = 107 declined to take part in the trial. This resulted in 317 participants who consented to the BRIGhTMIND trial and attended the baseline assessment, with 39 participants not meeting inclusion criteria. The most common reasons for exclusion at this stage were that they were not diagnosed as MDD with HDRS-17 ≥16, or not diagnosed as TRD. Other exclusion reasons at the baseline assessment stage included: history of bipolar disorder (due to risk of mania) or depression secondary to other mental disorder; risk of suicidality; and change in prescribed medication within 2 weeks of baseline assessment.

FIGURE 2 CONSORT diagram. (Figure not reproduced here; see the original article.)

Between baseline assessment and randomisation (post-consent), there were 23 withdrawals, with the reasons including: time commitments; concerns of treatment exacerbating physical illness; no capacity to deliver TMS treatment due to staffing and the COVID-19 pandemic; personal reasons, anxiety relating to TMS and unknown side effects; and unable to tolerate scan or not attending MRI scan visit. This left 255 participants who were randomised to cgiTBS (N = 128) or rTMS (N = 127). The first randomisation was on 22 January 2019, with the final randomisation on 31 January 2022. The last follow-up assessment was 3 August 2022.

The Nottingham, Newcastle, Northampton, and London sites were given permission to commence recruitment between November 2018 and March 2019. Recruitment to the entire study was suspended on 30 April 2020 because of the COVID-19 pandemic, with sites permitted to recommence recruitment on 1 August 2020. This was with the exception of the Northampton site, which did not reopen to recruitment in August 2020 and was informed of the decision to close the site in December 2020, owing to low recruitment numbers and lack of staff to deliver either the TMS or research assessments. To increase recruitment, Oldham (Pennine Care NHS Foundation Trust) was opened as a study site and commenced recruitment in August 2021. The total number of participants randomised to treatments per site included 113 participants for Nottingham, 59 for London, 47 for Newcastle, 29 for Northampton and 7 for Oldham (see Appendix 2, Figure 18).

Post-randomisation discontinuation to trial

Of the 255 participants randomised, only 3.2% (N = 4) of participants discontinued their involvement in the trial altogether, during the treatments. This included one participant who had a flare up of a pre-existing physical health condition, one participant who found the commitment to the treatments and study too great, one participant who had to reduce their percentage rMT in order to tolerate treatments but felt this would not be effective, and one participant who experienced a suspected seizure at the motor threshold testing at their first treatment and was lost to follow-up. At the 8-week follow-up, completion rates were 88.2%, N = 112 of 127 versus 86.7%, N = 111 of 128, 16-week follow-up 88.2%, N = 112 of 127 versus 87.5%, N = 112 of 128 and at the 26-week follow-up 80.3%, N = 102 of 127 versus 81.3%, N = 104 of 128, for the rTMS and cgiTBS groups, respectively. The most frequent reason for discontinuation was loss to follow-up. Reasons for discontinuation are presented in Table 5.

Adherence to treatment and protocol

A total of 235 major protocol deviations were recorded for 144 participants (56.5%). In the rTMS group there were 117 major protocol deviations for 68 participants and in the cgiTBS group 118 major protocol deviations for 76 participants (see Table 6). These numbers are high because: (1) participants accumulated a protocol deviation for every different break they had of 4 days or more missed consecutive dates for treatment; and (2) participants accumulated a protocol deviation for every different psychotropic medication or psychological intervention that was not kept stable for 16 weeks.

In terms of treatment adherence, 70 participants (27.5%) had a gap of more than 4 days of missed consecutive treatments with an average break length of 5.8 days. Most of these participants had 1 or 2 breaks (see Table 7). In addition, 48 participants’ (18.8%) psychotropic medications or psychological interventions were changed between the baseline assessment and 16-week follow-up assessment (not including participants classed as at risk to themselves or others), and 5 participants (1.9%) commenced a daily prescription of benzodiazepines or ‘Z’ drugs (i.e. zopiclone, zolpidem, or zaleplon) that exceeded the criteria of maximum dose allowed for the trial, between baseline and the end of TMS treatments.

Further major protocol deviations related to non-adherence with randomised treatment, included 20 participants (7.8%) who received < 20 TMS sessions, 8 participants (3.1%) stopping one TMS session part way through treatment, and 2 participants (0.8%) whose course of treatments exceeded 6 weeks. There were also 22 participants (8.6%) who received the wrong treatment as per randomised allocation. Of these participants, 12 had treatments delivered at the incorrect site due to their TMS co-ordinates being uploaded incorrectly (N = 3 for 1 session, N = 1 for 3 sessions, N = 1 for 4 sessions, N = 1 for 6 sessions, N = 1 for 7 sessions, N = 1 for 11 sessions, N = 2 for 14 sessions and N = 2 for all 20 sessions). Furthermore, there were four participants whose treatment dose was set incorrectly for a portion of one treatment session and one participant whose treatment dose was set incorrectly for an entire session. One participant also had a train of treatment delivered of the incorrect treatment, one participant had six trains of treatment delivered with a shorter gap than expected, and one participant whose treatment co-ordinates were not correct. All of these occurred at one session only. Finally, there were two participants for whom no additional details were provided (see Appendix 2, Table 35).

Additional trial-related major protocol deviations included one eligibility violation of a person who was judged to be at risk of suicidality at baseline assessment. This became evident after the participant completed the study, that their baseline HDRS-17 suicidality score was recorded as 4, and while at the time it was indicated that the participant had strong protective factors in place, the PI was unable to be sure that the patient was not suicidal, due to lack of information on duration of suicidal ideation. Additionally, 5 participants (1.9%) commenced lamotrigine, pregabalin or gabapentin, after their TMS treatments had commenced but prior to their 16-week follow-up, and 22 participants (8.6%) were missing primary outcome data. Minor protocol deviations are not reported in detail here and can be found in Appendix 2, Table 36.

Participant characteristics

Table 8 summarises the demographic and baseline characteristics of the sample (post-randomisation). Participants had a mean age of 43.7, SD = 14.0 years, with N = 132 (51.8%) female gender, and a total of N = 232 (91%) of white ethnicity and N = 23 (9%) from ethnic minority backgrounds. In addition, N = 138 (54.1%) were in full-time or other employment, N = 87 (34.1%) were unemployed, and N = 30 (11.8%) were retired.

In regard to depression characteristics, the mean duration of current episode was 113.9, SD = 119 months [median 6.1 (IQR 2.1–12.1) years], and mean number of depressive episodes was 3.2, SD = 5.2 [median 2 (IQR 1–4.2) years]. In terms of medication use at baseline, 198 participants (77.6%) were taking antidepressants, with 41 participants (16.1%) having a combination of two or more antidepressants. Augmentation medications at baseline included antipsychotics, buspirone, lithium, methylphenidate, modafinil, pramipexole, and triiodothyronine. In addition, 16 participants (6.3%%) were taking hypnotics/sleeping tablets and 14 participants (5.5%) taking anxiolytics. Regarding treatment resistance, 95 participants (37.3%) were categorised as high treatment resistance (non-response to more than approximately 6 treatments), 73 participants (28.6%) as medium treatment resistance (non-response to about 4–5 treatments), and 87 participants (34.1%) categorised as low treatment resistance (non-response to about 2–3 treatments), with 10 participants (3.9%) receiving ECT in their current episode of depression.

Primary outcome analysis

Primary hypothesis ‘cgiTBS is more efficacious in reducing the mean HDRS-17 score over 26 weeks compared to standard rTMS in patients with TRD’

Table 9 summarises the HDRS-17 descriptive statistics for the primary and secondary analyses. In the ITT analysis, the participants were analysed in the groups they were allocated to (rTMS n = 127, cgiTBS n = 128). Table 10 summarises the descriptive statistics for the cognition outcome variable. The breakdown of baseline and follow-up HDRS-17 by research site, and also for the sensitivity and moderator analyses is reported in Appendix 2, Table 37.

As shown in Table 11, the primary analysis of the primary outcome (HDRS-17 over 26 weeks) indicated the adjusted mean difference between the rTMS and cgiTBS treatment arms was non-significant [−0.31 points (95% CI −1.87 to 1.24)]. This is contrary to the clinical efficacy hypothesis, demonstrating that cgiTBS does not show superior clinical efficacy compared to MRI-neuronavigated rTMS. The primary analysis demonstrated a clinically substantial change in HDRS-17 scores141 at 26 weeks, with a 7.8-point and 8-point reduction from baseline, for rTMS and cgiTBS, respectively. Figure 3. also shows that depression symptom improvement was maintained between 8 and 26 weeks following rTMS or cgiTBS.

FIGURE 3 The HDRS-17 mean scores for the primary analysis ‒ ITT (multiple imputation). (Figure not reproduced here; see the original article.)

Secondary and sensitivity analyses of primary outcome

The per-protocol analysis included n = 59 (46.5%) and n = 63 (49.3%) participants for rTMS and cgiTBS treatment arms respectively, and the completers analysis included N = 110 (86.6%) and N = 108 (84.4%) participants. The pre-/post-COVID-19 analysis showed that 72% of the entire sample (N = 184) were classed as ‘Post-COVID’. Participants whose 26-week HDRS-17 assessment was due before or up to 18 March 2020 (pause to recruitment due to COVID-19) were classified as ‘Pre-COVID’ and those whose assessment was due after this date were classed as ‘Post-COVID’. Across all secondary and sensitivity analyses there were no statistically significant differences between the cgiTBS and rTMS treatment groups. The results of these analyses demonstrated that the reductions in HDRS-17 scores at 26 weeks for both treatment arms were comparable to what was observed in the primary outcome analysis.

Analyses of moderators

The moderator analyses demonstrated that the following moderator variables alone significantly predicted the primary outcome: baseline HDRS-17 score [adjusted mean = 0.69 (95% CIs 0.45 to 0.93); p < 0.001], baseline GAD-7 [adjusted mean = 0.36 (95% CIs 0.11 to 0.61); p = 0.004] and number of TMS sessions delivered as per protocol (i.e. completing 20 sessions as planned vs. completing < 20 sessions as planned with completing < 20 sessions as reference) [adjusted mean = −5.64 (95% CIs −9.48 to −1.80); p = 0.004]. The results demonstrating that higher baseline HDRS-17, higher baseline GAD-7, and completion of < 20 sessions as planned, predicted greater depression symptoms over 26-week follow-up. The following moderator variables alone did not significantly predict the primary outcome: age at baseline [adjusted mean = −0.04 (95% CIs −0.12 to 0.05); p = 0.391], baseline CTQ [adjusted mean = 0.004 (95% CIs −0.06 to 0.07); p = 0.915], and with baseline MGH-S reaching marginal significance [adjusted mean = 0.30 (95% CIs −0.01 to 0.61); p = 0.058]. The interactions between treatment arm and all moderator variables were not statistically significant. This indicates that the moderator variables did not influence the relationship between treatment arms and the primary outcome. Gender was also examined as an exploratory moderator variable (i.e. female vs. male, with male as reference), which did not predict the primary outcome alone [adjusted mean = 0.61 (95% CIs −1.57 to 2.80); p = 0.581] or by interaction [adjusted mean = −2.19 (95% CIs −5.32 to 0.94); p = 0.171].

Secondary outcome analyses

Secondary hypothesis ‘cgiTBS is more efficacious for secondary clinical outcomes compared to standard rTMS in patients with TRD’

Table 12 reports the results of the binary outcome results comparing rTMS versus cgiTBS for responders and remitters at 8, 16 and 26 weeks, and sustained responders at 16 and 26 weeks. Consistent with the primary outcome analysis, there were no significant differences found between the treatment arms, for any of the responder, remitters, or sustained responder time points. Descriptively the overall proportion of responders was N = 74 of 223 (33.2%), N = 77 of 224 (34.4%) and N = 67 of 206 (32.5%) for the 8-, 16- and 26-week follow-up time points, respectively. The overall proportion of remitters was N = 42 of 223 (18.8%), N = 46 of 224 (20.5%) and N = 47 of 206 (22.8%) for the 8-, 16- and 26-week follow-up time points, respectively. Finally, the overall proportion of sustained responders at both 16 weeks and 26 weeks was N = 51 of 255 (20%).

In addition, given that degree of treatment resistance reached marginal significance as a predictor of the primary outcome, we were interested in descriptively exploring the breakdown of the number of responders by category of treatment resistance at the 8-, 16- and 26-week follow-ups (see Appendix 2, Table 38). In the cgiTBS group, there were 14 out of 34 classed as responders for low treatment resistance, 13 out of 30 for medium treatment resistance and 9 out of 31 for high treatment resistance at the 26-week follow-up. For the rTMS group, there were 16 out of 34 classed as responders for low treatment resistance, 9 out of 26 for medium treatment resistance and 6 out of 42 for high treatment resistance at the 26-week follow-up. This suggests that there were a slightly higher proportion of responders in the medium and high treatment resistance categories for the cgiTBS group at the 26-week follow-up, compared to the rTMS group, and overall more responders with low and medium resistance compared to those with high treatment resistance.

The same picture was found for the continuous secondary outcome analyses, with none of these variables demonstrating significant differences between rTMS and cgiTBS (see Table 13). At 26 weeks there was an overall decrease in self-report depression and anxiety scores, including a 6.4-point reduction on the PHQ-9, 10.7-point reduction on the BDI-II, and a 3.8-point reduction on the GAD-7. There was also a 6.5-point decrease in impairment of functioning as measured by the WSAS total score, and an 11.6-point increase in perceived health as measured by the EQ-5D-5L VAS score. Prior research has indicated that minimum clinically important differences for the PHQ-9 is ≥ 3.7 points,142 BDI-II is ≥ 6.0 points,143 GAD-7 is ≥ 3.3 points,142 WSAS is ≥ 3.7 points144 and EQ-5D-5L VAS is ≥ 8.0 points.145 Therefore, the findings demonstrate minimum clinical important differences for all these outcomes. Overall, alongside improvements in depression, there are improvements in anxiety symptoms, quality of life and functioning, which also appear to be maintained over 6 months in line with the primary outcome analysis.

Cognition hypothesis 1: ‘Treatment with rTMS and cgiTBS will lead to an improvement in cognition between baseline and week 16 that correlates with improvement in mood’

As per the QC checks detailed in the methods chapter, outliers were removed for any scores on the CRT, DSST, or N-back (primary outcomes) that were four times the SD of the overall sample (baseline CRT = 32, 16-week CRT  25, baseline DSST = 14, 16-week DSST = 19 and 16-week N-back = 7). In addition, 1 participant’s TMT score was removed from baseline for taking longer than 5 minutes to complete, with five participants’ baseline data not having at least three of the four objective tasks scores meet QC or not completed.

In regard to hypothesis 1, there were significant main effects of time point on CRT [F(1, 155.486) = 11.275, p = 0.001], TMT [F(1, 152.446) = 5.499, p = 0.020] and N-back [F(1, 151.091) = 7.748, p = 0.006], indicating improvements over time for sustained attention, executive functioning and working memory (see Figure 4). Significant interactions of time point*HDRS-17 change were found on CRT [F(1, 151.349) = 6.321, p = 0.013], and PDQ-5-D [F(1, 161.111) = 55.602, p = 0.001]. This arose because improvement on the HDRS-17 was a predictor of sustained attention (CRT) and subjective cognitive performance (PDQ-5-D) at the 16-week time point only (see Figure 5). The only significant main effect of the treatment group was on the DSST [F(1, 211.880) = 4.080, p = 0.045]. Interaction terms: treatment group*time point and time point*change in HDRS-17 score*treatment group were non-significant for all cognitive variable outcomes (p’s > 0.05), indicating no differential effect of treatment (rTMS vs. cgiTBS) on cognition outcomes, with the main effect of the treatment group observed on the DSST related only to a between-group baseline difference (better baseline performance for cgiTBS participants compared to rTMS participants). Of note, age was retained as a confounder for all cognitive outcomes, gender retained as a confounder for the DSST analysis, site retained as a confounder for the N-back analysis, and baseline GAD-7 retained as a confounder for the PDQ-5-D analysis.

FIGURE 4 Clustered bar mean of cognition and IIV variables at baseline and 16-week time points for rTMS and cgiTBS group. (Figure not reproduced here; see the original article.)

FIGURE 5 Grouped scatter plots of relationship between HDRS-17 score and change in cognition and IIV variables. (Figure not reproduced here; see the original article.)

Hypothesis 2: ‘Treatment with rTMS and cgiTBS will lead to a reduction in IIV in CRT performance that correlates with improvement in mood’

The results of these mixed linear regression models demonstrated significant main effects of time point on tau [F(1, 182.523)  6.784, p = 0.001] and iSD [F(1, 176.622) = 13.033, p = 0.001], suggestive of a reduction in the number and/or length of abnormally slow responses and less variability in RT. Additionally, there were significant interactions of time point* HDRS-17 change on sigma [F(1, 158.255) = 12.909, p = 0.001], and mu [F(1, 149.396) = 11.645, p = 0.001], demonstrating improved speed (consistent with the finding seen for the mean CRT RT), and less variability in RT with improvement in HDRS-17 scores. No significant effects were observed on CoV. Furthermore, there were no significant main effects of treatment group, or group*time point interaction for any of the IIV variables (p’s > 0.05), consistent with the findings related to hypothesis 1, indicating no differential effect of treatment (rTMS vs. cgiTBS) on cognition outcomes. Of note, age was retained as a confounder for all IVV outcomes, and site retained as a confounder for the mu analysis.

Based on these results, an additional exploratory secondary analysis (see Appendix 2, Table 39) was conducted to examine the relationship between participants change in HDRS-17 score and change in the cognition and IVV variables (i.e. baseline to week 16). Significant moderate correlations were found between HDRS-17 change and change in PDQ-5-D (r = 0.48, N = 145, p = 0.001), change in mu (r = 0.31, N = 140, p < 0.001) and change in sigma (r = 0.30, N = 140, p < 0.001). Significant weak correlations were also found between HDRS-17 change and change in CRT (r = 0.21, N = 147, p < 0.008) and change in DSST (r = −0.180, N = 147, p < 0.029). The overall results were consistent with improvement in cognition with reduction in depression severity.

Patients’ global impression of change

Descriptive statistics regarding participants’ impression of change at each treatment session and follow-up time point are reported in Appendix 2, Tables 40–41, with Figure 6 showing the percentage of participants reporting feeling somewhat, or much, better after each session for illustrative purposes. Overall, at the end of the final 20th TMS treatment, N = 1 (0.4%) of participants rated themselves as much worse, N = 7 (3%) as somewhat worse, N = 74 (31.2%) as just the same, N = 111 (46.8%) as somewhat better, and N = 44 (18.6%) as much better. By session 10, 45 participants (37.2%) and 60 participants (48.3%) reported feeling somewhat, or much, better for rTMS and cgiTBS, respectively. By session 20, this was reported for 76 participants (65%) and 79 participants (65.8%). The relationship between session number and perceived improvement appeared linear across the 20 sessions for the rTMS group. The relationship for cgiTBS was generally similar but appeared to rise somewhat faster over sessions 7 to 10, before plateauing then returning to a similar linear trend as for rTMS. In both groups, the proportion experiencing a benefit continued to increase even at the 19th and 20th sessions.

FIGURE 6 Percentage of participants who felt somewhat or much better over the 20 TMS sessions. (Figure not reproduced here; see the original article.)

Across the 8-, 16- and 26-week time points, respectively, the number of participants rating themselves as much worse was N = 3 (1.4%), N = 6 (2.7%) and N = 4 (2%), and the number of participants rating themselves as somewhat worse was N = 17 (7.7%), N = 17 (7.7%) and N = 23 (11.3%). In addition, the proportion of participants that rated themselves as just the same was N = 73 (32.9%), N = 90 (40.7%) and N = 70 (34.5%). Finally, the number of participants reporting feeling somewhat better was N = 86 (38.7%), N = 63 (28.5%) and N = 63 (31%), and those reporting feeling much better was N = 43 (19.4%), N = 45 (20.4%) and N = 42 (21.2%). Further analysis will explore in more detail the trajectory of illness and its relationship to a range of clinical, sociodemographic, cognitive and imaging variables.

Study blinding and treatment arm prediction

In total there were three unintentional unblinding for two participants. One participant’s treatment was unblinded to both the site-specific PI and the site-specific RA, and another participant’s treatment unblinded to the site-specific RA only. In these instances, another blinded RA completed both of these participants’ follow-up assessments. In terms of researchers’ treatment allocation prediction, correct guesses for the cgiTBS group were at 5.4%, 9.8% and 16.3% across the 8-, 16- and 26-week follow-ups, respectively. For the rTMS group correct guesses were at 8%, 13.4% and 16.7% across the follow-up time points respectively. Most predictions were that of ‘don’t know’, with overall rates of 84.8%, 79.5% and 74.3% across the time points (see Appendix 2, Table 42).

Inter-rater reliability of primary outcome measure

To ensure high inter-rater reliability on the HDRS-17, specific training was provided by the chief investigator for all RAs. In addition, an inter-rater reliability assessment was performed by seven RAs from the Nottingham, Newcastle, Northampton and London sites, who conducted the HDRS-17 for the BRIGhTMIND study, and three LEAP members, who volunteered to be assessed by the researchers, drawing on their lived experience of depression, at a meeting in November 2019.The intra-class coefficient for the total HDRS-17 score was 0.94, with a 95% reference interval for the difference (between any pair of raters) in total HDRS-17 score of 0.66 to 0.99. Inter-rater reliability was not assessed towards to the end of recruitment as originally expected due to the COVID-19 pandemic.

Accuracy of transcranial magnetic stimulation targets

An exploratory analysis was conducted to test how close the TMS stimulus was delivered on the first session to the MRI treatment co-ordinates provided, and how close TMS stimulus was delivered at the rest of the sessions compared to the first session. This was done by computing the Euclidean/angular distance measured from the given target position/orientation to the mean position/orientation that was delivered on the first session, and by computing the Euclidean/angular distance measured from the mean position/orientation that was delivered on the first session to the mean position/orientation that was delivered on each subsequent session and averaged over the number of subsequent sessions. The angular distance is a geodesic measure that quantifies how far away on a sphere one orientation is from another, so the range is between 0 and 180 degrees. As shown in Table 14, the median distance between treatment targets to where treatment was delivered on the first session, and between the first session and subsequent sessions was under or about half a centimetre difference and with a median angle difference under or about seven degrees. Appendix 2, Figures 19–22 show some outliers. However, overall adherence to target stimulation sites at the first session and across all treatment sessions were acceptable and similar across both treatment groups.

Summary of findings

1. Connectivity-guided intermittent theta burst stimulation (iTBS) was not superior to rTMS on any of the primary or secondary outcome measures.
2. Nevertheless, in this treatment-resistant moderately severe depression population, MRI and neuronavigation personalised TMS (cgiTBS or rTMS) was associated with one in three participants achieving a clinical response and in a fifth this was sustained at 26 weeks. About 20% of TRD patients might be kept well with MRI-neuronavigated TMS delivered twice per year.
3. There were no clinical or demographic moderators of clinical response in terms of depression or cognition to cgiTBS or rTMS.
4. Clinical improvement with cgiTBS or rTMS was gradual and required at least 20 TMS sessions. After 20 TMS sessions some participants reported a slight improvement and might have benefited for further TMS sessions irrespective of receiving cgiTBS or rTMS.

Safety, acceptability, facilitators and barriers

All adverse events and serious adverse events

A total of 1850 AEs were reported in 204 (80.3%) participants who were both randomised into the BRIGhTMIND trial and included in the safety population (N = 254) (see Table 15). There were 153 participants (60.3%) with 1 to 10 AEs, 42 participants (16.7%) with 11–30 AEs, and 9 participants (3.6%) with 31 or more AEs (see Appendix 3, Table 43). The overall number of AEs was greater in the cgiTBS group (1008 AEs) compared to the rTMS group (842 AEs), but this was largely accounted for by three participants in the cgiTBS group with an exceptionally high number of AEs. Regarding relatedness of AEs to the study, 4.9% were classed as not related, 9.3% were unlikely, 33.8% were possible, 49.6% were probable and 2.4% were classed as definitely related, with the expectedness of AEs at 66.4%. Furthermore, the majority of AEs were classed as mild (91.8%), with most AEs requiring no treatment (91.7%). Listing of adverse events in the rTMS and cgiTBS arm are reported in Appendix 3, Tables 44 and 45.

A total of 18 SAEs events were reported in 13 (5.2%) participants who were randomised into the BRIGhTMIND trial (see Table 16). There were two deaths in the cgiTBS group. One participant had underlying cardiovascular health conditions and passed away from a heart attack. Additionally, one participant passed away from opiate poisoning, with coroner’s inquest concluding accidental death. Both participants had completed their course of TMS treatments and died close to the 26-week assessment, with both reported as unlikely to be related to TMS treatments.

All further SAEs events reported required hospital admission, except for the suspected seizure. Two SAEs were reported as being possibly related to TMS treatment. In the cgiTBS group, one participant was hospitalised due to a manic episode which occurred after their 14th TMS treatment session. In the rTMS group, one participant was hospitalised for a psychotic episode with severe anxiety/depression 1 month after their TMS treatments were completed. There were, however, situational factors and a life event to take into consideration for these two participants, and while TMS may have contributed to these events, they were unlikely to be the only contributing factor. One further participant in the rTMS group was also admitted to hospital for nausea and vomiting following their baseline MRI scan, with this event reported as probably related.

In addition, there was one further SAE of a suspected 2-second seizure that occurred during the participant’s motor threshold testing at their first TMS session. A consultant psychiatrist and nurse both witnessed the event and determined that this appeared to be a seizure, so the TMS session did not precede. The participant reported remembering prior to the event happening that they suddenly felt dizzy and in a dream-like state. The participant also reported that they had a history of previous collapsing events and reporting that their sibling developed epilepsy as an adolescent. The participant’s general practitioner (GP) was made aware of the event, and the participant was also told to report it themself to arrange a review by their GP. The participant’s treatment was discontinued, and the Research Ethics Committee (REC) deemed the outcome satisfactory with no further action required. All further SAEs were reported as unrelated to the study.

Finally, in the rTMS group, one participant also experienced a vasovagal episode during their first treatment, and one participant experienced syncope during motor threshold testing at the first treatment. Both were reported as AEs rather than SAEs as no treatment or hospital admission was required. Both participants treatment was suspended for that given day, with both participants approved to continue with the rest of their sessions with no further incidents reported. In addition, there was one participant in the cgiTBS group who was suspected of having multiple seizures with focal onset. These events occurred outside of the TMS sessions, and were confirmed by neurology as being dissociative seizures, rather than epileptic seizures, and so while this was initially classed as an SAE it was then downgraded to an AE. It was decided by the PI and participant to discontinue treatments due to these health issues, and no further TMS sessions took place.

Adverse events reported as possible, probable, or definitely related to the study

Regarding AEs that were classed as possible, probable, or definitely related to the study (see Table 17); there was a greater proportion of mild (809 AEs) and moderate (53 AEs) in the cgiTBS group compared to the rTMS group (mild AEs = 697; moderate AEs = 20). Severe AEs were comparable across rTMS (three severe AEs) and cgiTBS (five severe AEs). Of note of these 8 severe AEs, 5 were classed as severe but not an SAE, and 3 reported as severe and an SAE. All SAEs have been reported in the previous section. In terms of duration of possible, probable, or definitely related AEs, 1038 AEs were resolved the same day, 313 resolved within 1–6 days, 125 resolved within 1–4 weeks, 56 resolved within 4–8 weeks, 24 resolved in 8–16 weeks, 10 resolved within 16–26 weeks, 6 resolved in 26–52 weeks and 15 classed as unknown, continuing, or unobtainable.

Table 18 also provides a breakdown of outcome, treatment and expectedness by mild, moderate, and severe AEs for those that were classed as possible, probable, or definitely related to the study. Overall, mild AEs required more concomitant medication, non-drug therapy or a combination of both (93 AEs) when compared to moderate (16 AEs) and severe (5 AEs). In addition, there was a greater proportion of mild (62 AEs) and moderate (15 AEs) in the cgiTBS that required concomitant medication, non-drug therapy or a combination of both when compared to the rTMS group (mild AEs = 31; moderate AEs = 1). Overall, 73.3% of AEs in the rTMS group and 76.6% of AEs in the cgiTBS group were expected. In addition, 1568 AEs were resolved, 5 resolved with sequalae, and 14 reported as continuing, unknown or unobtainable. Of those that were resolved with sequalae, 3 were mild AEs (tinnitus, tongue discomfort, and neck pain), 1 was a moderate AE (migraine) and 1 reported as a severe AE (the hospital admission for psychotic episode with severe anxiety/depression), with resolved with sequalae meaning episodically recurring but not continuously and/or needing medication to prevent it happening again. Of those that were reported as continuing, unknown or unobtainable, 3 were headaches, 3 were tinnitus, 2 were cognitive difficulties, 1 was tiredness, 1 was disturbed sleep, 1 was anxiety, 1 was jaw ache, 1 was short periods of elevated mood, and 1 was neck pain.

Summary of common and uncommon side effects data collected after TMS sessions

The summary of common and uncommon side effects collected at each individual TMS session is reported in (see Appendix 3, Tables 46–67). Of the 254 participants included in the safety population, on average at an individual session: N = 38 (15.0%) reported headaches, N = 23 (9.1%) reported neck pain, N = 21 (8.3%) reported scalp discomfort, N = 13 (5.1%) reported tinnitus, N = 16 (6.3%) reported dizziness, N = 14 (5.6%) reported jaw ache, N = 11 (4.4%) reported nausea, N = 7 (2.8%) reported watering eyes, and N = 26 (10.2%) reporting uncommon side effects (see Table 19). These were all considered to be probably, possible or definitely related.

The most frequently reported common side effect at the TMS sessions and the least reported was watering eyes. The side effect profiles across the two treatment arms were comparable except for higher rates of tinnitus in the cgiTBS group (N = 9 of 128; 7.0%) versus rTMS group (N = 4 of 126; 3.2%). Uncommon side effects at each TMS session were recorded as verbatim comments, therefore exact numbers cannot be provided. However, during treatments there were reports of but not limited to: increased agitation/irritability; anxiety; fatigue; insomnia/nightmares; potential cognitive difficulties (such as brain fog, difficulty finding words, slight memory loss); facial tingling; pain in eyes; and toothache.

Tolerability

As reported in the minor protocol deviations (see Appendix 2, Table 36) there were a higher proportion of participants in the rTMS group (N = 29, 22.8%) compared to the cgiTBS group (N = 10, 7.8%) whose treatment was adjusted due to not being able to tolerate the TMS protocols. Table 20 provides a breakdown of these adjustments and shows there were a greater proportion of participants in the rTMS group (N = 11; 8.7%) who had to reduce their rMT when compared to the cgiTBS group (N = 4; 3.1%). The number of participants whose targets were moved or who had a combination of moving the target and reducing motor threshold was largely comparable across both TMS treatment groups. There was also a greater proportion of participants in the rTMS group (N = 11; 8.7%) that required more than one adjustment across the TMS sessions in order to tolerate the treatments, when compared to the cgiTBS group (N = 2; 1.6%). Participants would have accumulated a protocol deviation for every TMS session where an additional adjustment had to be made, hence the greater proportion of protocol deviations per participant for this deviation. Additional details can be found in Appendix 3, Table 68. Furthermore, motor threshold was tested at the first session and retested at the sixth session. Regarding the first session, there were 16 participants for the rTMS group and 4 participants for the cgiTBS group that reduced percentage rMT in order to tolerate treatments. For these participants, the TMS intensity at the first session was delivered on average at 87.8% rMT rather than the expected 120% rMT for the rTMS participants, and at 68.2% rMT rather than the expected 80% rMT for the cgiTBS participants.

Quantitative acceptability

Patient acceptability for each of the 20 TMS sessions and three follow-up time points are reported descriptively in Appendix 3, Tables 69 and 70. At the final 20th TMS session, 75.9% of participants (N = 180) found the treatments acceptable, 21.9% (N = 52) were neutral, unsure or waiting to find out, and 2.1% (N = 5) found the treatments unacceptable. Similar rates were also observed at the final 26-week follow-up, with 71.5% of participants (N = 145) rating the treatments as acceptable, 22.7% (N = 46) neutral, unsure or waiting to find out, and 5.9% (N = 12) finding treatments unacceptable. For illustrative purposes, Figure 7 shows that the percentage of participants who found TMS treatments to be acceptable increased over the course of TMS sessions, which appeared to be maintained until the final 26-week follow-up. There were a slightly higher percentage of participants in the rTMS group who found treatments unacceptable at the three follow-up time points (8.1%, 4.5% and 7%) compared to the cgiTBS group (3.6%, 1.8% and 4.8%). Equally, there was a slightly higher percentage of participants in the cgiTBS group who found treatments acceptable at the 8- and 16-week follow-up (68.4% and 71.5%) compared to the rTMS group (64.8% and 68.4%). The findings overall demonstrate that TMS is an acceptable treatment for this population in both the short term and longer term.

FIGURE 7 Acceptability of TMS treatments over the course of TMS sessions and three follow-up time points. (Figure not reproduced here; see the original article.)

Qualitative facilitators, barriers and acceptability

Nineteen participants completed qualitative interviews. This included 15 participants who completed all treatments and follow-ups, and four participants who declined to participate in the trial prior to being randomised into treatments. No participants who withdrew from the trial during or after treatments expressed an interest in the interviews. Demographic characteristics were consistent with the demographics of the main BRIGhTMIND trial. The participants were aged 29–67 years (M = 49.7, SD = 12.6 years), 53% male, 90% White British or other and 10% participants of Asian or Asian British ethnicity. A total of 42% of participants were in full-time or other employment, 37% were unemployed, and 21% were retired. Of those that completed treatments, seven participants received rTMS treatment and eight received cgiTBS treatment. A total of 582 verbatim comments were also recorded across TMS treatments end of session feedback sheet and the three follow-up time points.

Four main facilitators to treatment were identified: (1) hope; (2) influence of staff; (3) interest in new treatments; and (4) altruism. Two main barriers were identified: (1) commitment concerns; and (2) treatment-related concerns. Additionally, three main themes arose for the acceptability research question: (1) the TMS experience; (2) variation in symptom changes; and (3) lay theories of effectiveness (see Table 21). These themes are illustrated with example quotes below and further detailed quotes can be found in Appendix 3, Table 71.

Facilitators

Hope

Participants had long durations of depression and resistance to various treatments. As previous medication and psychotherapies had not been beneficial, they were keen to try an alternative hoping to improve depressive symptoms. When participants considered TMS treatments prior to their study involvement, receiving TMS privately was deemed unaffordable and the waiting times too long within NHS TMS clinics. As a result, they were grateful for the opportunity to access the treatments for free through the trial, particularly given that all participants received active treatments.

To give me a sense of hope, the way my psychiatrist described the treatment study, it can be quite effective in some cases and in some cases, it could even offer a cure for resistant depression, obviously if there was a chance that that could help me then I was very keen to give it a go.

Influence of staff

Participants felt that staff involved in the BRIGhTMIND study were fully informed, professional, compassionate, and sensitive, with all these characteristics contributing to participants feeling comfortable and knowing what to expect from the treatment. Additionally, participants felt they were able to freely discuss their concerns and fears with staff members, who listened and provided appropriate reassurance which helped alleviate their concerns.

They were very careful to check in how I was when I arrived and when we’d finished the treatment, so there was an opportunity if I needed it to talk about any concerns.

Interest in new treatments

The treatment-resistant nature of participants’ depression caused frustration but also a sense of desperation. This meant they were willing to try any new treatment that might help relieve their symptoms. Some participants reported that following the trial of several medications with notable side effects and no response, they were particularly open to try alternative non-pharmaceutical treatments. Despite expecting some side effects related to TMS, they felt these would be minimal in comparison to medication-related side effects. Therefore, participants were particularly keen to try TMS, due to its reported safeness and non-invasiveness. Furthermore, the location of the treatment site within the NHS sites and the clinical members of staff delivering the TMS treatment, further reassured participants of the closeness of medical contact in the event of an AE.

I was just looking at options for treatment of depression. non-antidepressants, this is one of the things that came up. it seemed pretty safe. and different to taking pills and I was very keen to try it.

Altruism

There was a good understanding of the need for research in order to introduce new treatments and advance knowledge about depression. As participants were keen for new treatments to be introduced, they felt motivated to take part in the trial beyond consideration of their own benefits and they expressed an interest in wanting to help others in a similar situation.

My participation in the research could well help other people in the future. […] I might possibly benefit and something that I would not normally be able to benefit from, so the direct potential benefits for me was also a sense putting something back into society about being part of the pool of research subjects.

Barriers

Barriers to participating in the BRIGhTMIND trial were experienced by those who were randomised and completed treatments, as well as those who declined to participate in the trial. For those that declined, the barriers and concerns outweighed the positives and potential benefits of taking part in the trial.

Commitment concerns

Transcranial magnetic stimulation was provided during office hours. It was reported that having to attend 20 TMS sessions over a 4- to 6-week period was a challenging commitment. Some participants had to build their daily routines around the treatments. It may be suggested that treatments were more accessible for those who are self-employed or not working, with several employed participants using annual or sick leave to attend appointments. Additionally, as there could only be a maximum of four consecutive missed treatment days, participants were worried about becoming too unwell due to their health to attend the daily sessions. Once out of the house, anxiety related to the prospect of commuting and finding parking was also highlighted as a barrier, particularly in participants who were not comfortable with crowds or relying on public transport.

The scheduling was quite intense, and it was every day, well five days a week.

Sometimes my depression does make me kind of anxious about travelling anyway, and I just thought it was unfair to take part in the study and then maybe not to fully attend.

Treatment-related concerns

Other concerns that acted as a barrier to participating in the trial were directly linked to the nature of the treatment. Participants initially drew comparisons between TMS and ECT, with fears of memory loss and seizures. However, information provided to participants, including the rate of TMS-associated seizures and that they would be able to drive home immediately after treatments. This reassured them of the differences between these types of treatments and the safety of TMS. Additionally, the location of the treatment room at one of the TMS research sites was next to an ECT suite evoking negative feelings, due to the negative preconceptions surrounding ECT treatments and negative memories for those who received or know of people who received ECT.

I’ve always been very wary of ECT, so you know, learning more about [TMS], that it wasn’t like ECT, was an important part of helping me to feel comfortable.

There were also concerns about not knowing what to expect from the treatment in terms of sensation and potential pain. These fears seemed to be linked to the neuromodulation treatment making direct changes to the brain, and some participants were concerned that the treatment could make their depression worse.

The idea of something directly going to your brain is quite a weird one.

Acceptability

The TMS experience

Participants who reported perceived improvements found TMS treatments acceptable, positive, would recommend to others, with TMS outweighing other treatments tried regarding improvements and side effects. Some participants who did not see improvement also found TMS acceptable and were glad to have taken part, whereas others found the TMS less acceptable due to the side effects and lack of response to treatments. Several individuals also suggested long sessions and significant treatment commitment would be acceptable if there were sustained improvements.

TMS has been the best treatment for anxiety and depression I have received from the NHS in the last 20 years. It has none of the side effects and negative response of talk therapy e.g. CBT.

The supportive, caring, and considerate nature of the TMS staff and the continuity of dealing with the same staff members also helped participants feel comfortable and at ease during the TMS sessions. They enjoyed the conversations and options of listening to music, reading books or doing quizzes.

Staff always been cheerful, shown compassion and understanding. Felt at ease and easy to go through the sessions.

Tolerability of TMS varied between individuals, with some describing the sensation the coil triggered as: (1) painful; (2) an unusual sensation; or (3) comfortable and relaxing with none or minimal discomfort. This experience appeared to improve over the course of the treatment sessions, with adjustments to the positioning of the coil and motor threshold also contributing to better tolerability. Some participants also expressed that the feeling of the TMS treatments were not as expected, whilst also articulating the sensations they felt during the treatment sessions, such as ‘tingling’ or ‘tapping’.

I mean the discomfort it is a weird, very strange sensation like a static electric shock going into your head. So, it gives you sort of pain, where it is not too bad once you process that it is going to be coming every 30 seconds. It was … not really nice, but not really painful.

Furthermore, the MRI scan was mostly well received as participants were keen to get the treatment and understood its necessity for the study. Some participants reported that while it was not a pleasant experience due to the feelings of claustrophobia; having the MRI scan was worth it to take part in the trial.

Thought that perhaps it might not be very nice being in a claustrophobic sort of situation during the period, but there was no great concern to me because I thought that you know the benefits of this would far outweigh just having to lie in a scanner. It did not really bother me.

Regarding side effects: a number of individuals reported none or minimal side effects during TMS treatments. Others reported common side effects including headaches, dizziness, nausea, and scalp tingling, which were largely transient and less frequent over the course of the TMS sessions. Feelings of tiredness and fatigue were also mentioned during the course of treatments, and several participants also mentioned that their anxiety was still prominent or worse, together with vivid nightmares, more agitation and poorer sleep.

Have had some dizziness but it hasn’t lasted too long.

Variation in symptom changes

Those who reported perceived treatment benefits experienced them in varied ways, with some noticing improvements during the course of treatments, others only after the course of treatment were complete. The perceived benefits ranging from subtle to significant. Regarding response duration, some participants reported sustained improvement at the final follow-up assessment, and others witnessed treatment effects wearing off. This wearing off ranged from immediately after treatments, to across the 8-, 16- and 26-week follow-up time points. Improvement in different depression symptoms was also reported, which included improvements in mood, sleep, concentration, cognition, anxiety, motivation, relaxation, suicidal thoughts, and physical symptoms such as migraines and headaches. Some participants who reported no therapeutic benefit commented that it was disappointing but understood that not all individuals would have a response to TMS.

I think during the treatment […]I wasn’t sure whether it was just like a placebo effect that I was having or whether it was really happening, and the week after I had finished, I noticed that the symptoms that I had had were subsiding.

Lay theories of effectiveness

This theme reflects the participants’ own explanations and interpretations of the perceived effectiveness of the treatment received. Some participants related the improvements in depressive symptoms especially to the TMS treatments, while others felt improvements could be attributed to a combination of TMS treatments and other factors including psychological therapy, medications, and environmental changes. Additionally, some individuals felt that the routine of having to leave the house, attend the appointments, or the interactions with the TMS staff may also have had therapeutic benefit.

It is hard to say what contributed to my recovery and what was like, the percentage of that, meaning, how much the medication helped, how much therapy helped, how much RTMS helped but even the fact of having a treatment and doing it made me feel better because I felt like I am actually doing something about my health issues.

Furthermore, those who reported experiencing environmental stressors during or after the TMS treatment suggested that without these, the TMS treatments may have been more beneficial. Some participants also reported a lack of confidence in the effectiveness of TMS due to variability across treatment sessions (despite a lack of objective evidence for this observation by the neuronavigation machine or staff account). This included the perception of inconsistent coil placement dependent on the TMS staff member delivering treatment, and concerns that the coil movement away from the scalp and target location during treatments may have impacted the level of treatment received.

As soon as my head touched [the coil], or there was a bit of pressure, it would kind of ping off it wasn’t very a lot, but you need obviously contact, and there was not contact, so I did wonder whether I was actually getting any treatment at all.

There was also the perception that receiving a longer course of treatment sessions may have led to more substantial improvements for a number of individuals. Participants with perceived short-lived benefits expressed the need for further top-up treatments.

I felt like something was happening with it on the last week so I felt like personally for me it like I could probably done another week or two […] so whether there was a way doing may be four weeks then a week’s rest and then another two weeks or something like that for people where it wasn’t really working until towards the end.

Summary of findings

1. Both cgiTBS and rTMS were safe with only 1 probable (mania) and 1 possible (psychosis with depression and anxiety) SAE that might be attributed to TMS.
2. Connectivity-guided ITBS may have been associated with more frequent AEs than rTMS but the vast majority of these were mild and lasted less than a day. Tinnitus was experienced twice as often with cgiTBS than rTMS.
3. Repetitive transcranial magnetic stimulation was associated with more reduction in motor threshold and changes in the site of stimulation than cgiTBS because of lack of tolerability within the TMS session.
4. Both cgiTBS and rTMS were regarded as equally highly acceptable in this population with similarly high rates of completion of all 20 TMS sessions.

Economic evaluation

Economic data

The characteristics of the BRIGhTMIND participants who informed the economic analysis can be found in Table 8. Missingness was generally modest and broadly consistent across variables of interest. Resource use data was available for approximately 84–88% of participants at week 16 and 79–81% at week 26. EQ-5D and informal care data was available for 85% and 87% of participants at week 16, and 79% and 80% of cases at week 26, respectively. All variables considered had complete information at baseline with only one exception (productivity). The degree of missing data for each variable incorporated into the economic analysis is presented in Appendix 4, Table 72.

Costs

Intervention costs

The average intervention costs observed for cgiTBS and rTMS are presented in Table 22. Participants underwent 1.08 structural MRI and rsfMRI scans on average, at a mean cost of £479.10. Those randomised to cgiTBS attended an average of 19.35 treatment sessions lasting 52 minutes each, totalling £770.97 in staff-related costs. Those randomised to rTMS attended 19.13 separate 56-minute sessions on average, equating to £816.11 in staff-related costs. Mean equipment costs were £1058.82 per trial participant. Total intervention costs in the trial amounted to £2308.90 and £2354.04 for cgiTBS and rTMS, respectively.

Resource use and costs

Table 23 reports average imputed health service and broader societal costs for rTMS and cgiTBS in total, and for each individual resource category assessed in the trial. The primary cost drivers in the trial were from the intervention costs, outpatient hospital attendances, primary care, and productivity costs. Inpatient hospitalisations were relatively rare (approximately only 2% experiencing any inpatient admission), while community service costs, travel costs, and over-the-counter medication costs were generally modest. From both a health service and a broader societal perspective (excluding health service costs), average total costs were found to be lower for cgiTBS relative to rTMS with cost savings of approximately £165 and £100, respectively. Individuals randomised to cgiTBS incurred higher medication and private service costs but incurred markedly lower outpatient and productivity costs relative to rTMS. Overall, cost differentials were in favour of cgiTBS and predominately driven by outpatient hospitalisation attendances, productivity losses, and private service utilisation. The analysis of complete cases found similar results, albeit with larger cost differences. Appendix 4, Tables 73 and 74 show service use for rTMS and cgiTBS groups. Appendix 4, Tables 75 and 76 provide a complete case analysis breakdown of all costs recorded in the trial for each arm by follow-up period. Imputed costs by follow-up period are also displayed in Appendix 4, Tables 77 and 78.

Outcomes

Figure 8 presents the proportion of participants observed as having a problem in each domain of the EQ-5D by treatment arm and assessment period. Over the course of the trial the proportion of participants experiencing problems, and the severity of those problems, were generally reduced. Improvements appeared largest in the self-care and usual activities domains; however, severe problems with anxiety and depression did develop in both arms compared to baseline. A further descriptive assessment found moderate improvements in the usual activities, pain/discomfort, and anxiety/depression severity scales with cgiTBS relative to rTMS (see Appendix 4, Figure 23).

FIGURE 8 The proportion of observed problems reported by treatment arm and follow-up assessment. (Figure not reproduced here; see the original article.)

Figure 9 presents average imputed EQ-5D-3L preference scores for cgiTBS and rTMS. Baseline scores were higher for cgiTBS relative to rTMS (+ 0.027). Scores markedly increased in both arms following treatment. cgiTBS extended and maintained higher scores relative to rTMS over the course of the trial. Differences in imputed EQ-5D-3L (mapped) and EQ-5D-5L scores translated into average QALY gains of approximately 0.013 and 0.017 for cgiTBS relative to rTMS, respectively. Appendix 4, Figure 24 presents the same analysis using imputed EQ-5D-5L preference scores. Full details of average imputed health-related quality of life and QALY outcomes are reported in Appendix 4, Table 79.

FIGURE 9 Average imputed EQ-5D-3L score by treatment arm. (Figure not reproduced here; see the original article.)

Baseline informal care requirements were highest for the cgiTBS group (those up to 3 months prior to baseline assessment), with an average of 107.5 hours of care versus 68.6 hours for rTMS, suggesting an imbalance in baseline care needs. The additional informal care hours associated with cgiTBS relative to rTMS were reduced to 6.3 hours (cgiTBS: 80.0; rTMS: 73.7) at week-16 follow-up and −2.4 hours (cgiTBS: 51.83; rTMS: 54.23) at week-26 follow-up. Informal carers taking time off work was relatively rare, predominantly due to participant depression and followed a comparable trajectory to total hours. Full details of the informal care requirements observed in the trial are displayed in Appendix 4, Table 80.

Cost-effectiveness analysis

Table 24 reports the unadjusted and adjusted base-case mean cost, QALY and cost-effectiveness results from health service and broader societal costing perspectives. Appendix 4, Tables 81 and 82 present the regression analyses used to estimate the QALYs and costs evaluated in the adjusted cost-effectiveness analyses.

Connectivity-guided iTBS was found to have marginally lower costs and to be modestly more effective compared to rTMS across all perspectives and approaches considered (i.e. cgiTBS dominated rTMS). The unadjusted analysis found cgiTBS was associated with an additional 0.0157 QALYs and £341.74 in health service cost savings. These cost savings were extended to £788.63 when including broader societal costs. The adjusted analysis found cgiTBS was associated with an additional 0.0091 QALYs, £180.19 in health service cost savings [QALYS gained cgiTBS 0.275 (95% CI 0.26 to 0.29); iTBS 0.265 (95% CI 0.25 to 0.28); total costs cgiTBS £3256.79 (95% CI £2994.69 to £3522.75); iTBS £3436.98 (95% CI £3148.63 to 3719.60)], and £336.00 when including broader societal savings. For the adjusted (unadjusted) analysis, the INMB associated with cgiTBS at a £20,000/QALY threshold equalled £362 (£656) from a health service perspective, and £451 (£1103) from a broader societal perspective. INMB values increased with the cost-effectiveness threshold. A graphical display of INMB for each approach is provided in Appendix 4, Figure 25.

Figure 10 displays the cost-effectiveness planes for the unadjusted and adjusted health service perspective cost-effectiveness analyses. Predicted incremental costs and QALYs in all analyses had considerable overlap between positive and negative values, suggesting a significant level of decision uncertainty. The probability of being cost-effective remained > 80% across a wide range of thresholds (see Figure 11). Expected value of perfect information (EVPI) was highest in the adjusted analysis, generally increased with the cost-effectiveness threshold and reached £17.11 per participant from a health service costing perspective at a £30,000/QALY threshold. EVPI is presented graphically in Appendix 4, Figure 26.

FIGURE 10 Cost-effectiveness planes for unadjusted and adjusted health service perspective cost-effectiveness analyses. (Figure not reproduced here; see the original article.)

FIGURE 11 Base-case health service perspective cost-effectiveness acceptability curves (CEACs). (Figure not reproduced here; see the original article.)

Sensitivity analysis

Table 25 presents cost-effectiveness findings from each scenario analysis using a health service costing perspective. Compared to base-case findings, EQ-5D-5L preference weights had minimal impact on incremental QALYs or cost-effectiveness findings but did increase average QALYs in each arm. Removing imaging costs for rTMS (£479) made cgiTBS the most expensive alternative in both unadjusted and adjusted analyses. In this scenario, cost-effectiveness was highly sensitive to the analytic method, with ICERs of £8750 and £32,920 for cgiTBS relative to rTMS in unadjusted and adjusted analyses, respectively. Applying a 30-minute time saving for cgiTBS appointments relative to those for rTMS, improved the cost-effectiveness of cgiTBS via extending its cost savings by approximately £385. The joint scenario considering both the removal of imaging costs for rTMS and 30-minute time savings for cgiTBS appointments resulted in a moderate reduction (£94) in cost savings associated with cgiTBS in both the unadjusted (from £342 to £248) and adjusted analyses (from £180 to £94 savings). In this joint scenario, cgiTBS remained the dominant strategy. When separately comparing cgiTBS and rTMS to artificial placebos, ICERs were above £30,000/QALY for both analytic methods. ICERs fell below £30,000/QALY in all cases when equipment costs were distributed across 98 people over the lifetime of the machine (base-case analysis: 51 people). At base-case settings, ICERs for cgiTBS relative to rTMS were within the £20,000-£30,000/QALY threshold range at incremental costs of £313–£470 (unadjusted) and £184–£276 (adjusted).

Summary of findings

1. Both cgiTBS and rTMS recorded significant improvements in HRQoL from baseline values. Connectivity-guided iTBS was associated with further improvements in HRQoL and lower health service and broader societal costs compared to rTMS.
2. Connectivity-guided iTBS was considered a cost-effective alternative to rTMS treatment in the base-case analysis of the BRIGhTMIND study (more effective, less costly), albeit with considerable uncertainty.
3. Cost-effectiveness findings were sensitive to the imaging and staffing costs associated with each treatment, examples of which include non-routine use of pre-screen imaging in rTMS.
4. Connectivity-guided iTBS was considered cost-effective relative to rTMS in all economic analyses, provided it could be delivered at a cost of £184 or less than rTMS treatment (cgiTBS was associated with cost savings in all base-case analyses).

Mechanism of action

Functional connectivity and effective connectivity

A total of 255 participants completed baseline structural and rsfMRI scans for the study. Of these participants, 18 had to repeat the baseline scans with 2 having to have a second repeat, due to scans failing the QC criteria in order to generate the TMS target co-ordinates (see Table 26).

In terms of the mechanism-of-action functional and effective connectivity analyses, 46 were excluded due to having no clinical follow-up time points (N = 24), receiving treatment to an unintended target (N = 9), incorrect scanning parameters (N = 5), completing fewer than 15 treatment sessions (N = 4), failing QC (N = 3) or there being more than 6 weeks between the first and final treatment (N = 1). A total of 127 participants completed follow-up scans. A total of 26 were excluded due to baseline scans not being usable (N = 13), incorrect scanning parameters (N = 10) or failing QC (N = 3). Statistical analyses were with LMMs, unless otherwise specified. The dependent variable of the LMM was change in depression scores (baseline depression minus depression scores at follow-up at 8, 16 or 26 weeks), with higher positive scores indicating better clinical improvement. The HDRS-17 was the primary depression outcome measure, and the HDRS-6, BDI-II and PHQ-9 were exploratory depression outcome measures.

Baseline brain connectivity as a predictor of clinical response

The primary hypothesis was that greater clinical improvement would be associated with more positive baseline EC from the rAI (a region of the brain’s ‘salience’ network37) to left dorsolateral prefrontal cortex (lDLPFC, a region of the ‘central executive network’38), and that this relationship would be stronger in the cgiTBS group.

As reported in the introduction, EC refers to the directed influence of one region on another, calculated using Granger causality (greater EC is thought to reflect greater influence of one region’s activity on another’s). Figure 12 illustrates the location of the right rAI region of interest (purple), as well as the lDLPFC regions of interest for individual participants (blue). In this analysis, the cgiTBS target was used as the location of lDLPFC for each participant, regardless of their treatment group. This was because the cgiTBS site was itself chosen as the site that maximised baseline EC from rAI.

FIGURE 12 Sites of TMS stimulation on the brain. (Figure not reproduced here; see the original article.)

The primary hypothesis was not supported (LMM: dependent variable of reduction in depression measure from baseline to follow-up; random effect of participant; fixed factors of treatment group, follow-up time point, baseline EC from right AI to cgiTBS site as covariate, confounders – site, age, gender, MGH treatment resistance category; and 200/201/185 participants included at follow-up time points 8, 16, and 26 weeks). Baseline EC from the rAI site to the cgiTBS target did not predict clinical improvement on any of the measures of depression (p > 0.1; HDRS-17; HDRS-6, BDI-II, and PHQ-9).

The rAI site (purple), individual cgiTBS sites (blue), mean cgiTBS site across participants (green), rTMS site (red), and sgACC site (yellow) are shown on a standard brain template.146

We pre-specified two secondary hypotheses concerning baseline connectivity as a predictor of response. One hypothesised that greater clinical improvement would be associated with more positive baseline net outflow from rAI to lDLPFC, and this relationship will be stronger in the cgiTBS group.

Net outflow from rAI to lDLPFC was calculated as: EC from rAI to lDLPFC minus EC from lDLPFC to rAI. This analysis did find that net outflow was a significant predictor of clinical improvement on the HDRS-17 [Figure 13; F(1195.992) = 4.035, p = 0.046; significant confounders retained in model: site, MGH treatment resistance group]. However, inconsistent with the pre-specified hypothesis, it was lower net outflow from rAI to lDLPFC which predicted greater improvement (i.e. improvement was greater if there was greater influence of lDLPFC on rAI, than of rAI influence on lDLPFC). This effect did not differ by treatment group or by post-treatment time point (p > 0.1). In terms of the exploratory depression outcome measures, the relationship approached significance for the HDRS-6 (p = 0.102) but was not significant for the BDI-II or PHQ-9 self-report depression measures.

FIGURE 13 Hamilton Rating Scale for Depression-17 improvement from baseline to follow-up (collapsed across follow-up time points) vs. baseline net outflow from rAI to lDLPFC. Each circle is a participant. (Figure not reproduced here; see the original article.)

Given the above findings, we also conducted an exploratory analysis with baseline EC from lDLPFC to rAI as a predictor (as opposed to EC from rAI to lDLPFC as in the primary analysis). There was no significant relationship with improvement on HDRS-17 (p > 0.1), however there was a significant relationship with clinical improvement on the HDRS-6 [F(1196.855) = 4.209, p = 0.042; confounders retained: site, MGH]. Greater baseline EC from lDLPFC to rAI was associated with greater clinical improvement (this effect did not differ by treatment group or by post-treatment time point, p > 0.1). The relationship also showed marginal significance for the BDI-II (p = 0.098) but was not significant for the PHQ-9.

The final secondary analysis in this section used FC (correlation between the fMRI time courses from two regions, with greater FC thought to reflect greater co-operation between regions), between the sgACC (yellow sphere in Figure 12) and the lDLPFC. It was hypothesised that greater clinical improvement would be associated with more negative baseline FC between the sgACC and lDLPFC across both treatment groups. This analysis used the intended stimulation target for each individual as the locus of lDLPFC (i.e. the personalised cgiTBS target for the cgiTBS group, and the rTMS target for the rTMS group; significant confounders retained: site, MGH).

While baseline FC between sgACC and lDLPFC did not predict improvement in observer-rated depression (17-/6-item HDRS), it did predict improvement on self-rated depression (BDI-II, with borderline significance for PHQ-9). Specifically, there was a significant interaction of the effects of baseline FC and post-treatment time point [BDI-II: F(2362.262) = 4.130, p = 0.017; PHQ-9: F(2378.800) = 2.944, p = 0.054]. Effects did not differ by treatment group (p > 0.1). The interaction arose because baseline FC initially showed a negative relationship with depression improvement at 8 weeks but showed a positive relationship at 16 and 26 weeks (see Figure 14).

FIGURE 14 Beck Depression Inventory-II improvement from baseline to each follow-up time point vs. baseline FC between sgACC and the lDLPFC target. Each circle is a participant. (Figure not reproduced here; see the original article.)

In exploratory analysis, while there were no significant group effects, it is notable that mean baseline FC with the sgACC was significantly more negative for the rTMS site than the personalised cgiTBS site [−0.124 vs. 0.033, paired-sample t-test, t(229) = 5.363, p < 0.001]. In addition, baseline FC between the rAI and sgACC was significantly positive [0.066, t(229) = 5.343, p < 0.001].

Clinical response and change in brain connectivity

The primary hypothesis was that reduction in FC between the lDLPFC and the lDMPFC, a region of the ‘default mode network’,36) from baseline to follow-up would be associated with greater clinical improvement. For this analysis, we pre-specified 1 lDMPFC site, and 2 lDLPFC sites (1 more posterior site, which lay closer to the rTMS target, and 1 more anterior site, which lay closer to the average cgiTBS target), from previous literature.

No significant effects were observed using the more anterior lDLPFC site (p > 0.1; 98/101/93 participants included at follow-up time points 8, 16, and 26 weeks; no confounders retained), on any of the depression measures.

Using the more posterior site, reduction in FC between lDLPFC and lDMPFC from baseline to follow-up was associated with significantly greater improvement on the self-reported depression measures [PHQ-9: F(1107.685) = 6.959, p = 0.010; BDI-II: F(1106.408) = 4.555, p = 0.035]. These effects were not significant for the 17- or 6-item HDRS (p > 0.1). The effects did not differ by treatment group or post-treatment time point (p > 0.1). The correlation between improvement on the PHQ-9 (collapsed across post-treatment time points) and reduction in FC between lDLPFC and lDMPFC was 0.252 (see Figure 15; df = 96, p = 0.013).

FIGURE 15 Patient Health Questionnaire-9 improvement from baseline to follow-up (collapsed across post-treatment type points) vs. reduction in FC between the lDLPFC (posterior site) and lDMPFC. Each circle is a participant. (Figure not reproduced here; see the original article.)

Pre-specified secondary analyses examined relationships between clinical improvement and change in: rAI to lDLPFC EC, rAI to lDLPFC net outflow, and FC between sgACC and lDLPFC. There were no significant effects on these analyses for any of the depression measures (p > 0.1).

Magnetic resonance spectroscopy analysis

A total of 190 baseline MRS scans, and 98 follow-up MRS scans were available for data analysis. All spectra GABA and Glx were reviewed by an MRI physicist with spectroscopy experience. Following this QC, 131 scans were included for the baseline GABA+/Cr analysis, and 64 paired scans (both baseline and follow-up) for the GABA+/Cr change analysis. The paired scans underwent further review by an MRI expert with > 20 years of MRS experience, whereby off spectra used for the Glx fitting were reviewed for issues with line width, SNR, water suppression, baseline, fit quality and voxel position, dimensions, and repositioning (aided by individual and MNI space voxel reconstruction and derived overlap score). Following this, 45 paired scans were available for both Glx/Cr change and GABA/Glx ratio change analysis (QC exclusion flowchart, Appendix 5, Figure 27).

Descriptive statistics for the baseline GABA+/Cr, GABA+/Cr change, Glx+/Cr change and GABA/Glx ratio change analyses, broken down by MRS scanner platform and treatment group can be found in Appendix 5, Table 83. Of note only one pair of Newcastle data was available for the GABA+/Cr and Glx/Cr change, therefore this was omitted from final statistical analysis. Sensitivity analysis indicated results remained unchanged when including this Newcastle dataset.

Baseline GABA+/Cr as a predictor of clinical improvement

The primary hypothesis was that ‘greater clinical improvement on 17-item HDRS (16 and 26-week vs. baseline) will be associated with lower GABA+/Cr at baseline across both treatment groups’. The LMM revealed that baseline GABA+/Cr did not predict clinical improvement over 16- and 26-week follow-up for the primary outcome measure – HDRS-17, nor the exploratory HDRS-6 measure (p > 0.1). However, there was a trend for greater baseline GABA predicting better clinical improvement on the self-report PHQ-9 [F (1, 118.651) = 2.945, p = 0.089; significant confounders retained in model: MGH treatment resistance category] that is, in the opposite direction to what we hypothesised (see Figure 16).

FIGURE 16 Depression improvement from baseline to follow-up vs. baseline GABA+/Cr. Each circle is a participant. Note: PHQ-9 (follow-up averaged over 16 and 26 weeks), HDRS-6 and HDRS-17 (follow-up averaged over 8, 16 and 26 weeks). (Figure not reproduced here; see the original article.)

Linear regression models for secondary hypotheses also showed that higher baseline GABA significantly predicted improvement when depression change was averaged over 8, 16 and 26 weeks for the HDRS-17 (β = 0.247, t = 2.556, p = 0.012, significant covariates retained in model: study centre, MRS platform, treatment resistance category and baseline HDRS-17) and HDRS-6 (β = 0.219, t = 2.136, p = 0.035, significant covariates retained in model: study centre, MRS platform, treatment resistance category and baseline HDRS-17), again opposite to the direction hypothesised. Exploratory correlation analysis comparing associations between baseline GABA+/Cr and clinical improvement by MRS scanner platform, suggest findings were driven by the GE scanners (see Appendix 5, Table 84).

GABA+/Cr change and clinical improvement

Primary hypotheses were:

Changes in depression symptoms on 17-item HDRS (baseline minus 16 weeks, and baseline minus 26 weeks) will be associated with change in GABA+/Cr between baseline and 16 weeks (post minus pre-treatment) in both treatment groups, with increase in GABA+/Cr expected to be associated with better clinical improvement.

Linear regression models revealed these hypotheses were not supported at either 16- or 26-week follow-up for the primary outcome (HDRS-17) or the exploratory depression measures (HDRS-6, BDI-II, PHQ-9), all p, s > 0.1. Furthermore, change in GABA+/Cr did not predict clinical improvement (baseline minus follow-up averaged over 8, 16 and 26 weeks) for any depression measure, as per secondary hypotheses, nor were there differences in GABA+/Cr between treatment groups.

Glx/tCr change, GABA/Glx ratio change and clinical improvement

For Glx/Cr change, secondary hypotheses were ‘change in depression symptoms on HDRS-17 (baseline minus 16 weeks minus baseline and baseline minus 26 weeks) will be associated with change in Glx/Cr (follow-up minus baseline)’, and the ‘increase in Glx/Cr will be higher in cgiTBS compared to rTMS group’.

Linear regression models revealed a trend for a decrease in Glx/Cr predicting better clinical improvement on the HDRS-6 at 16 weeks (β = −0.247, t = −1.793, p = 0.083, significant covariate retained: study centre) and at 26 weeks the HDRS-17 (β = −0.211, t = −1.697, p = 0.097 significant covariates retained: study centre), and a significant relationship with 26-week HDRS-6 change (β = −0.260, t = −2.036, p = 0.048, significant covariate study centre retained). These findings were in the opposite direction to what was hypothesised, and the scatterplot between Glx/Cr change and HDRS improvement indicated one relatively extreme value of Glx/Cr (see Figure 17). Analysis was re-ran excluding this outlier, which revealed the relationship between Glx/Cr change and HDRS-6 improvement at 26 weeks went from significant to trend-level significance (β = −0.223, t = −1.714, p = 0.094; study centre retained). Additionally, the p-values for the two trends, HDRS-6 at 16 weeks and HDRS-17 at 26 weeks increased to p = 0.111 and p = 0.108, respectively.

FIGURE 17 Hamilton Rating Scale for Depression-6 improvement (baseline minus 26-week follow-up) vs. change in Glx+/Cr. Each circle is a participant. (Figure not reproduced here; see the original article.)

No differences between treatment groups for Glx/tCr change were observed. Finally, an exploratory hypothesis was to examine whether GABA/Glx ratio change would predict improvement on the HDRS-17 at 16 weeks and 26 weeks. No significant results were identified for any depression measure at either follow-up time point.

Summary of findings

1. The balance of influence between lDLPFC and rAI at baseline predicted clinical improvement on the observer-rated HDRS-17. Specially, greater influence from lDLPFC to rAI at baseline was associated with greater improvement.
2. Baseline FC between sgACC and the intended treatment target within lDLPFC predicted self-reported clinical improvement. The relationship appeared to change across follow-up time points, with more negative baseline FC being associated with greater response when measured at the first time point, but poorer response when measured at subsequent time points.
3. Greater self-reported clinical improvement was associated with greater reduction in FC between lDMPFC and a posterior lDLPFC site from baseline to follow-up.
4. Higher baseline GABA+/Cr predicted greater clinical improvement on observer-rated depression measures, when averaged over 8, 16 and 26 weeks, with a trend for self-reported depression over 16 and 26 weeks. These findings may be MRS platform specific.
5. Decrease in Glx/Cr from baseline to follow-up predicted greater clinical improvement on HDRS-6 at 26 weeks, with trends for HDRS-17 at 26 weeks and HDRS-6 at 16 weeks, when 1 extreme Glx/tCr value was retained in the analysis. Once the outlier was removed the relationship between Glx/Cr change and HDRS-6 at 26 weeks was of trend-level significance.