Clinical effectiveness of subsensory sacral neuromodulation in adults with faecal incontinence: the SUBSoNIC crossover RCT and mechanistic study

Randomised double-blind crossover design

Eligible participants were randomly allocated to two study arms after SNM implantation (see Figure 1). Both arms had two intervention periods of 16 weeks duration (T0–T16 and T16–T32). Efficacy outcomes were derived from assessments in the final 4 weeks of each cross-over period (T12–T16 and T28–T32) thus allowing for almost 3 months intervention before outcome assessments. In accord with usual clinical practice, a reprogramming (or sham reprogramming) session was conducted by the routine clinical care team at 6 weeks in both periods of both arms (T6 and T22). Time-points had an interval tolerance of ±1 week.

FIGURE 1.

Timeline.

Cohort study: 12-month outcomes

After completing the crossover section of the study, participants were followed up for a further 26 weeks. During this time, they had ‘open label’ stimulation, being able to choose between sub- or supra-sensory stimulation settings as would have been normal for routine clinical practice. Further efficacy outcomes were recorded at T54–T58. While it is accepted that these did not represent true 1-year outcomes (16 weeks had been sham treatment during the crossover), these were included to provide an indication of the short-term effectiveness of SNM within the rigor of a CTU-monitored prospective study.

Mechanistic studies

Mechanistic studies were performed in the final 2 weeks of the 4-week assessment periods in a subgroup of consecutively consenting patients from both arms until saturation (anticipated sample size: n = 20).

Inclusion criteria

• Adults aged 18–80 years.

• Meeting Rome III and ICI definitions of FI (recurrent involuntary loss of faecal material that is a social or hygienic problem and not a consequence of an acute diarrhoeal illness). 2,34

• Failure of non-surgical treatments to the NICE standard: diet, bowel habit and toilet access addressed; medication for example loperamide, advice on incontinence products, pelvic floor muscle training, biofeedback and rectal irrigation should be offered if appropriate. 7

• Minimum severity criteria of eight FI or faecal urgency episodes (including a minimum of four FI episodes) in a 4-week screening period.

• Ability to understand written and spoken English or relevant language in European centres (due to questionnaire validity).

• Ability and willingness to give informed consent.

Exclusion criteria

A standard list of exclusions (disease variants; surgical fitness, specific contraindications to implantation) were used. 10 Note that these are routine clinical exclusions to the use of SNM rather than participation in the research:

• Known communication between the anal and vaginal tracts.

• Prior diagnosis of congenital anorectal malformations.

• Previous rectal surgery (rectopexy/resection) performed < 12 months ago (24 months for cancer).

• Present evidence of full thickness rectal prolapse or a high-grade intussusception.

• Prior diagnosis of chronic inflammatory bowel diseases.

• Symptoms of chronic constipation with over-flow incontinence.

• Structural abnormality of the pelvic floor leading to clear evidence of obstructed defaecation based on examination and/or imaging.

• Presence of active perianal sepsis (including pilonidal sinus).

• Defunctioning loop or end stoma in situ.

• Diagnosed with neurological diseases, such as diabetic neuropathy, multiple sclerosis and Parkinson’s disease.

• Current or future need for MR imaging based on clinical history.

• Complete or partial spinal cord injury.

• Bleeding disorders for example haemophiliac, warfarin therapy.

• Pregnancy or intention to become pregnant during the study period.

• Not fit for surgeon preferred method of anaesthesia.

Brief description of SNM surgery

Patients meeting the mandated response using the test phase (monopolar temporary wire or quadripolar tined lead based on local preferred practice) underwent implantation of the permanent InterStim system under general or local anaesthesia (with sedation) by trained expert colorectal surgeons following recommended procedural steps35 in brief: fluoroscopic-aided percutaneous insertion of 3889, 978A1 or 978B1 lead using curved stylet and accepting position only when 3 of 4 electrodes provide low voltage (< 3 V) contraction of the anal sphincter and pelvic floor ± big toe. The implantable pulse generator (IPG; 3058; Medtronic or InterStim Micro rechargeable model 97810) were placed as pre-marked in the ipsilateral buttock.

The device was activated as per local policy, either on the same day of surgery or after a surgical stabilisation period of up to 2 weeks. General programming parameters followed a written algorithm based on best clinical practice and manufacturer’s guidelines. Prior to programming, an impedance check was performed and recorded to ensure integrity of the electrical system.

Active intervention (sub-sensory stimulation)

The clinical team set the electrode configuration to achieve sensory threshold defined as the stimulation amplitude where the patient felt the first sensation of stimulation in the anus or perineum at 14 Hz frequency, pulse width 210 μs. The amplitude was increased in 0.1 V increments from zero until the sensory threshold was reached for each electrode and optimal electrode configuration defined. Sub-sensory chronic stimulation was then initiated on the patient controller device at a level just below the habituated sensory threshold (for blinding). This process was repeated at the 6-week time point.

Sham intervention (stimulation off)

Sensory thresholds were recorded identically; however, the level was then adjusted to 0 V. An identical procedure (to active) was repeated at the 6-week time point.

Commercially available devices evolved during the course of the trail and were adopted in some centres. The new TH90P03 handset could not be blinded to allocation if the voltage was set to 0 V as participants only had the ability to turn the device on. Participants were unable to turn the stimulator off as the handset deemed 0 V as off. The sham setting for participants with this programmer was 0.05 V as this setting is considered well below the therapeutic dose.

Visit 0: screening (face-to-face in clinic or phone)

Patients were assessed for eligibility against the inclusion and exclusion criteria checklist. The indication for SNM implantation required approval by the pelvic floor multidisciplinary team (MDT) prior to visit one. Patients who were initially found to be ineligible but who became eligible prior to any surgery were rescreened. Eligible patients were given the study invitation letter and participant information sheet (PIS). Patients were given adequate time to review the PIS prior to consent. All patients screened were added to the screening log and were given a study ID.

Visit 1: baseline (face-to-face)

Eligibility against the inclusion/exclusion criteria was re-reviewed, the study and the PIS were discussed and those patients in agreement completed written informed consent. A maximum of 18 weeks before permanent implantation was allowed for this visit. Once a patient was consented, the following assessments were completed:

• demographics, standardised medical/surgical history including history of incontinence symptoms, gynaecological history and pregnancy test (females of childbearing potential)

• clinical exam of perineum, anus and rectum (if not documented within the previous 6 months)

• baseline outcome assessments: St Mark’s continence score, deferment time, OAB-q SF, International Consultation on Incontinence Bowel (SF-ICIQ-B) questionnaire, FI QoL score and EQ-5D-5L/visual analogue scale (VAS).

At this visit patients were given the 4-week paper bowel diary (which also recorded loperamide usage) and taught how to use the electronic diary, which started from this visit. A paper viscerosensory bowel diary was also provided with instructions for completion over 5 days.

Visit 2a: test stimulation (face-to-face)

Test stimulation was performed according to routine clinical practice and not considered a study intervention. Data (see below) were only collected at this visit if a tined lead was implanted.

Visit 2b: MEG study enrolment (face-to-face)

Before permanent device implantation, participants who met the locally agreed criterion for progress based on the test stimulation phase, or those who had a tined lead inserted with a high probability of going through to permanent stimulation, were selected for and consented to the magnetoencephalography (MEG) studies depending on geographical location and magnetic resonance imaging (MRI) eligibility.

Visit 3a: permanent SNM implantation (face-to-face)

Following test stimulation, participants were admitted as a day case for permanent device implantation. Eligibility for full enrolment was re-confirmed (including assessment baseline diary data). Participants were randomised prior to knife to skin to either one of the two trial arms:

• Arm 1: SNM/sham

• Arm 2: sham/SNM

Intraoperative data were collected including: (1) lead position (radiological side, foramen level, number of electrodes in foramina); (2) motor thresholds for each of the four electrodes on the quadripolar lead; (3) physiological motor (± sensory) response for chosen foramen for lead implantation; and (4) other intraoperative data including length of operation, type of anaesthesia (including use of any muscle relaxant agent), blood loss, any other complications. If the tined lead was inserted at the start of the test stimulation phase, these data were collected during this test stimulation visit (visit 2a).

Visit 3b: initial programming (T0) (face-to-face)

Post-operative baseline checks were performed including impedance measurements of the four electrodes to ensure integrity of the electrical system. Participants had their SNM programmed as per routine care. This was undertaken in the immediate post-operative recovery period or up to 2 weeks post-surgery. All further follow-up visits were counted from the initial programming day and not from the day of surgery.

To reduce selection bias, no consenting patient with an implant in situ was excluded from participation that is regardless of the surgeon’s views on success or otherwise of implantation. At each follow-up visit, impedance measurements were repeated to ensure maintained integrity of the electrical system. If a closed or open circuit was detected (suggesting possible neurostimulator or lead malfunction) then this was documented. If a better (stronger perception at lower amplitude) sensory response was achieved using an alternative electrode configuration, the device was re-programmed and so the patient could continue in the study. In the absence of a satisfactory sensory response with an abnormal impedance measurement the patient was still followed up as per ITT and any changes to treatment were recorded in the deviation log.

At each visit, any change in electrode configuration, sensory threshold and location of maximum bodily sensation were recorded. Any AEs were systematically questioned and recorded at this visit and all subsequent face-to-face visits.

Programming was performed either using the Model 8840 N’Vision^TM (Medtronic, Minneapolis, MN, USA) clinical programmer (InterStim II) or the Model A51200 Micro Clinician app for the InterStim Micro rechargeable neurostimulator or A510 Clinician app for Models 3023 and 3058 InterStim neurostimulators with the HH90 Handset and TM90 Communicator. Tamper-proof tape was applied to the areas of the screen that could unblind the patient, but leaving areas of the screen where they could access the apps to turn on/off and recharge if applicable. Following initial programming:

• Arm 1: the subsensory amplitude was recorded along with the electrode configuration used

• Arm 2: the subsensory amplitude was recorded along with the electrode configuration used before returning the amplitude to 0.05/0.00 V (depending on which programmer the patient has).

Visit 4: 6-week reprogramming visit (T + 6) (face-to-face)

This visit was only completed if this was part of routine care. The tamper-proof tape was left on the patient’s programmer, programming was done via the clinician’s programmer if the older device was used. If the Smart programmer was used the clinician removed the tape to be able to perform the programming via the clinician app and reapplied new tape once finished:

• Arm 1: the patient was assessed for sub-optimal efficacy or unwanted effects of stimulation. In the presence of sub-optimal efficacy or adverse effects the electrode configuration was changed as per reprogramming algorithm. The sensory threshold was once again recorded, and the device was returned to the sub-sensory setting.

• Arm 2: the sensory threshold was recorded; the electrode configuration could be changed if the site of stimulation appeared to be suboptimal (aim for anal stimulation) before the device was returned to 0.05 or 0.00 V.

Visit 5: diary assessment (T + 12 to +16)

All participants completed the 4-week paper bowel diary and 5-day viscerosensory diary. Participants undergoing mechanistic studies had their first follow-up MEG study.

Visit 6: crossover visit (T + 16) (face-to-face)

At crossover, the device was turned off for 20 minutes followed by re-evaluation of the sensory threshold and best electrode configuration in the manner outlined above. The intervention was then reversed for each arm. Paper diaries and follow-up assessment questionnaires (St Mark’s continence score, deferment time, OAB-q SF, International Consultation on Incontinence Bowel (SF-ICIQ-B) questionnaire, FI QoL score and EQ-5D-5L/VAS) were completed. Participants also recorded their satisfaction on a Likert scale.

Visit 7: 6-week reprogramming visit (T + 22) (face-to-face)

All participants had a further follow-up 6 weeks after crossover at T22 if this was part of routine care. This visit was identical to visit 4 (see above).

Visit 8: diary assessment (T + 28 to +32)

All participants completed the 4-week paper bowel diary and 5-day viscerosensory diary. Patients undergoing mechanistic studies had their first follow-up MEG study.

Visit 9: end of crossover (T + 32) (face-to-face)

A further full set of outcomes was collected including the 4-week paper bowel and 5-day viscerosensory diaries (as per visit 6). After collection of these final crossover study data, participants entered the follow-up phase with patient chosen stimulation (sub- or supra-sensory) as would be normal for routine clinical practice. A member of the clinical team reprogrammed the device accordingly. As blinding was no longer necessary, participants had the option of changing their patient programmer for the new Samsung patient programmer. Further programming and advice were provided as per routine care during the period 32–58 weeks. All visits or contact with the clinical team during this time was recorded.

Visit 10: final assessment (T + 54 to + 58)

Participants completed a further full set of outcome questionnaire assessments and diaries, including the 4-week bowel diary (T54–58). During the final visit both the e-diary and paper diaries were collected. Participants underwent final re-programming and were then discharged from the study into normal clinical care.

Primary clinical outcome

The primary clinical outcome was reduction in FI events per week (recorded on paper bowel diaries over a 4-week period) in SNM versus sham periods of crossover (16 and 32 weeks).

While the limitations of bowel diaries are well-established,36 they remain the gold-standard in FI trials. 10,23,37,38 Although 2-week bowel diaries were the norm in these studies, we elected to use a 4-week recording period following international guidance. 39

The measure of treatment effect was the reduction in FI events per week whilst undergoing SNM as compared with undergoing sham stimulation.

The paper diary (see Appendix 1) was completed prior to implantation, then at the end of each cross-over period, and again at the end of the cohort follow-up. The degree of faecal loss was not quantified. While this is an acknowledged (and regularly debated) limitation of all existing outcome instruments, we believed that simplicity would be sacrificed if participants were required to judge the semantic differences between ‘staining’, ‘leakage’ and ‘frank incontinence’. Patients were asked (in notes at top of bowel diary) to fill a zero on days where no events occurred.

Secondary clinical and mechanistic outcomes

Other bowel diary measures, e-event recording and a panel of summative questionnaires were recorded at 16, 32 and 58 weeks.

• E-event recorder including episodes of faecal material, leakage of flatus, urgency without incontinence, social and physical activity.

• Other bowel diary measures: urgency, urge and passive faecal incontinence episodes, use of loperamide and social functioning.

• Summative questionnaire assessments: St Mark’s continence score;40 OAB-q SF score, FI QoL score;41 International Consultation on Incontinence Bowel (SF-ICIQ-B) questionnaire. 42

• Viscerosensory bowel diary recording quality, site and intensity of defacatory urge. 43

• Generic QOL: EQ-5D-5L. 44

• Likert scale of patient’s global impression of treatment success (scale 0–10) and patient perception of treatment or sham allocation (blinding success).

• Electrode settings (inc. motor, first and habituated sensory thresholds), programming (and if applicable re-programming data).

• Adverse event reporting.

E-event recording

A simple touch screen electronic device (Figure 2) developed with Medtronic allowed participants to record real-time-indexed episodes of leakage of faecal material, leakage of flatus and urgency without incontinence. In addition to comparing fidelity of events recorded by the current gold-standard (paper), the Samsung device provided opportunities to analyse novel similarly time-indexed measures for example social and physical activity using embedded android hardware device (GPS, accelerometer). GPS data were recorded but not analysed due to problems of data processing/acquisition.

FIGURE 2.

Example photograph of touchscreen icons on e-recording device.

The same device was used as a touchscreen application for digitalisation of established and new (SF-ICIQ-B) summative scoring questionnaires. The touch screen was used from the baseline visit throughout the crossover and cohort follow-up studies.

The information collected in the touch screen electronic device was logged in real calendar time and stored as time-linked data. It was downloaded by hardwire (USB) connection. The app was not simply an e-version of a paper bowel diary. Rather we developed a new app that greatly simplified use whilst also improving the accuracy of data over paper bowel diaries which are acknowledged to have major insufficiencies due to patient compliance,36 retrospective completion,45 and interpretative bias (if unblinded assessors).

Baseline data

Baseline characteristics and questionnaires were summarised by sequence (those randomised to SNM/sham vs. those randomised to sham/SNM) using descriptive statistics. Baseline (pre-surgery) assessments of the outcome measures were used to screen patients for eligibility and to provide a baseline for the longer-term cohort analysis. They are not adjusted for in the analysis of the crossover trial as they would only provide a baseline for the first treatment period and not the second.

Primary analysis: randomised crossover

All outcomes were summarised according to treatment arm and period, as recommended in the CONSORT extension for crossover trials. 32

The planned analysis pre-specified in the SAP for the primary outcome was as follows: to compare sham and active therapy in both arms of the cross-over trial, at T12–T16 and T28–T32, using mixed Poisson regression analysis to adjust for a fixed effect of period and a random effect of individual. Including a random effect for participant accounts for correlation between observations in different periods within the same participant and provides unbiased estimates even if some participants only provide data at one of the two periods, under the missing at random assumption implied by the model. To allow observed numbers of events before and after activation in the same individual to have an over-dispersed Poisson distribution, a random effect of time within individual was included. To allow for varying completeness of the 28-day diaries between subjects, an offset for the (log) completed days would be included in the model. All non-missing data were included in the analysis (i.e. not only those with data in both periods of the crossover), adjusting for the stratification variables (random effect of centre and fixed effect of sex). This approach is unbiased if missingness is related to observed outcome data or stratification factors from the same participant (a ‘missing at random’ assumption).

Analyses of secondary outcomes used the same mixed models as described for the primary outcome: Poisson regression for outcomes that are counts, and linear regression for other quantitative outcomes.

However, when running the analyses it was found that the Poisson regression models for the count outcomes failed to converge, even after following the strategy pre-specified in the SAP (replace random effect for centre with fixed effect, not allow for overdispersion, remove stratification covariates). Therefore, an alternative simpler analysis was used whereby the comparison of FI rate between sham and active therapy was done using paired t-tests, and treatment effects summarised by mean differences in number of episodes per week (with 95% CIs). For these analyses, only participants with outcome data in both periods of the crossover could be included. The paired t-test analyses assume no period effect, as the within-participant difference in outcomes is calculated regardless of the order of sham versus active therapy received. For consistency and ease of interpretation, secondary outcomes such as questionnaire measures were also analysed using paired t-tests. Stratification variables (centre and sex) were not considered when using paired t-tests.

To assess the effect of incomplete paper diary completion on the analysis of the primary outcome, a sensitivity analysis was done using a best-case scenario imputing a zero for days when some (but not all) of the count outcomes were left blank.

E-event time-linked recordings of the number of faecal leakage and urgency episodes were intended to be analysed using the same Poisson regression mixed-effects model as the primary outcome, but without inclusion of an offset as the e-recordings are continuous. However, as the regression models also failed to converge for e-recordings, paired t-tests were adopted instead. Unlike paper bowel diary outcomes, due to the continuous nature of e-recordings zero episodes were assumed for days when no event was reported. Outcomes reported on the paper bowel diaries were compared with the e-recording equivalent to assess agreement between the two modes of outcome measurement and summarised in Bland–Altman plots.

Original plans to derive measures of social and physical activity from the geospatial data were discarded as the data were shown to be unreliable, with highly improbable measurements recorded.

Programming data were summarised descriptively for each time-point. Analyses followed the modified ITT principle, including all participants according to randomised allocation and with available outcome data. There was no imputation of missing data other than the for the sensitivity analysis of the primary outcome described above.

Cohort study analysis

Data for participants with outcome measurements at baseline and at the end of the study (58 weeks) were summarised descriptively at each time-point, with no formal statistical analysis.

Statistical considerations due to COVID-19

The assumption for the primary analysis was that for participants paused in one of the cross-over periods, the eventual outcome in that period is unaffected by the extra time spent in the allocated treatment condition. We hypothesised that after the scheduled 6-week interval between reprogramming and assessment, a participant’s outcomes would stabilise. Too few participants were randomised for sensitivity analysis to be performed to investigate this.

Mechanistic study: participant selection

A subgroup of participants underwent central nervous system mechanistic studies at the Institute of Health and Neurodevelopment (IHN) at Aston University. These patients were identified and offered participation based on the following: (1) geographical location (Midlands residents); (2) if they were known to be proceeding to implantation (and therefore participation in the main efficacy trial); and (3) if a standard NHS safety checklist indicated suitability for MRI. Separate (secondary) written, informed consent was obtained locally.

Mechanistic study: interventions

Transcutaneous posterior tibial nerve stimulation

Two disposable self-adhesive electrode pads were placed transcutaneously over the posterior tibial nerve just posterior to the right medial malleolus with the end of the leads connected to a stimulator. Stimulation levels (SLs) were slowly increased from zero in increments of 0.1 mA at a rate of 2 Hz until sensory level was reached when sensation was reported in their foot. This level was then reduced by 0.05 mA at a rate of 2Hz until the sensation disappeared, before increasing the intensity again by increments of 0.01 mA until a sensation was detected again. This level was determined as the sensory threshold (ST). The SL was calculated at 2.5 X ST and the SL was increased slowly in increments of 0.1 mA at a rate of 2 Hz. The patient experienced a ‘strong but not painful’ sensation. This level was reduced to a more tolerable level if the initial calculated SL was not tolerated by the patient. The maximum SL permitted was as high as three times the ST if the calculated SL was not able to produce a strong enough sensation. Stimulation was delivered at an average of 400 stimulations at an Interstimulus Interval (ISI) of 700–800 ms at a rate of 2 Hz.

Fist clenching and anal squeezing paradigm

Verbal instructions were given to the patient to squeeze onto the anal probe if a ‘down’ arrow was shown on the screen and to make a fist with their right hand if an ‘up’ arrow was shown on the screen. The patient was asked to perform each task for as long as the arrow appeared on the screen. Each trial lasted for 8 seconds, which comprised a baseline (pre-stimulus) phase (fixation cross for 3 seconds), followed by a stimulus phase (red arrow pointing up or down for 2 seconds) before the recovery phase (fixation cross for 3 seconds) (Figure 3). The ratio of anal squeeze: fist activity was 2 : 1. Electromyography (EMG) activity of the anal probe was concurrently recorded to confirm that anal squeezing activity was adequately performed in a timely manner when a ‘down’ arrow was shown on the screen.

FIGURE 3.

MEG protocol for induced activity: fist clench/anal squeezing paradigm: baseline (pre-stimulus) phase with fixation cross = 3 seconds, stimulus phase with ‘up’ (fist) or ‘down’ (anal squeeze) arrow = 2 seconds, recovery phase with fixation cross = 3 seconds.

Mechanistic study: visit schedule

Patients who expressed an interest were invited to three visits to the Aston University IHN. At the first visit, they were shown the clinical facilities and had the opportunity to enter the MRI scanner to exclude claustrophobia. A baseline MEG was acquired followed by a 3T MRI head scan (N.B. the order of this is important since the MR scanner can induce tiny levels of magnetism in materials such as make-up and hair dye that can affect MEG recordings). At the second and third visits (SNM or sham in random sequence), the patient had further MEG acquisitions only.

Mechanistic study: outcomes

Whole cortical data were obtained using standard methods on an Elekta Neuromag® (Elekta: Stockholm, Sweden) Triux 306 channel system utilising noise cancellation methods to eliminate implant and stimulator artefacts. A beam-former analysis methodology was employed to evaluate both evoked and induced changes in brain activity associated with SNM and anal stimulation. Brain sources were constructed using individual co-registered T1-weighted MRI brain volumes. The outcome of this process was a measure of the changes in brain oscillatory power and/or frequency changes computed from brain structures where maximum changes associated with anal stimulation were observed. These changes were depicted in statistical brain volumetric images.

Mechanistic study: data analysis

MEG data were analysed by the IHN using existing bespoke computer analysis packages [Graph (Elekta^TM Elekta: Stockholm, Sweden); Matlab^TM (The MathWorks, Inc., Natick, MA, USA) and FieldTrip^TM (Donders Institute for Brain, Cognition and Behaviour, Radboud University: Nijmegen, the Netherlands) and SPM8^TM (Functional Imaging Laboratory, UCL Queen Square Institute of Neurology: London, UK)]. A beam-former analysis methodology47 was employed to evaluate both evoked and induced changes in brain activity associated with SNM and anal stimulation.

Group analysis of these data allowed determination of any functional cortical changes associated with chronic SNM. This was achieved by the spatial normalisation of individual MRI volumes into a grid based on the Montreal Neurologic Institute standard template.

Statistical analysis employed a non-parametric cluster-based permutation test. 48 Firstly, an uncorrected dependent-samples t-test was performed on pre- and post-stimulus brain activity across the entire brain volume. All voxels exceeding a 5% significance threshold was grouped into clusters. A null distribution was obtained by randomising the condition label (pre- or post-stimulus data) 1000 times and calculating the largest cluster-level t-value for each permutation. This methodology has been shown to adequately control for issues of multiple comparisons. 49

Trial committees

The project fell under the auspices of the Chief Investigator and the PCTU. The project was overseen by a Trial Steering Committee (TSC). The role of the TSC was to provide overall supervision of the study on behalf of the sponsor and funder to ensure the study was conducted in accordance with the principles of Good Clinical Practice (GCP) and relevant regulations.

A Trial Management Group (TMG) met monthly initially during study set up and then less frequently, every 2 months. The TMG was responsible for day-to-day project delivery across participating centres, and reported to the TSC.

A Data Monitoring and Ethics Committee (DMEC) met at least 4 weeks prior to the TSC to enable recommendations to be fed forward. The DMEC reviewed unblinded comparative data, and made recommendations to the TSC on whether there were any ethical or safety reasons why the trial should not continue. The DMEC membership fell in accordance with NIHR/MRC as well as PCTU guidelines.

Participant recruitment and flow

A flow diagram is included as Figure 4. The first patient was recruited on 2 February 2018; the trial was terminated on 24 July 2022 on advice of the DMEC. This decision was reached on the basis of futility given the ongoing significant barriers to recruitment posed by COVID-19 (see barriers to recruitment).

FIGURE 4.

Flow diagram. a, Patient no longer wanted to be randomised and blinded to treatment, but remained in follow-up for cohort analysis.

In total, 220 patients were screened for eligibility at nine sites from the UK and one site from Ireland between February 2018 and July 2022 (see Appendix 2). A total of 49 patients declined study participation and 106 were ineligible due to study specific exclusion criteria (Figure 5). A total of 65 patients were pre-enrolled and consented to the study, of whom 26 did not meet the baseline minimum frequency criteria of faecal incontinence episodes or did not receive an implantation. The remaining 39 patients were randomised (arm 1: N = 17; arm 2: N = 22); however, only 16 completed the primary outcome during both cross-over periods (arm 1: N = 9; arm 2: N = 7). The remaining 23 patients withdrew from the study (N = 12), some were excluded on the basis of problems of eligibility (N = 5) (see Appendix 2) or did not complete the primary outcome data (N = 6: still included in the cohort follow-up period).

FIGURE 5.

Number of patients who declined study participation and reasons for ineligibility.

A total of 22 patients started the cohort follow-up period, although 3 of these patients did not complete the final follow-up visit, leaving 19 patients for the open label efficacy assessment.

Barriers to recruitment

The trial was affected by several barriers to recruitment, with the COVID-19 pandemic being the main reason for termination before the required sample size was achieved.

By the beginning of 2020 the number of patients screened and randomised had improved after the changes were made to the inclusion and exclusion criteria, bringing the study close to the trajectory needed to complete recruitment by August 2020. The imminent opening of two large volume sites (St Mark’s Hospital, London and Erlangen, Germany) would have added to the certainty that the study would have finished on time. The outbreak of COVID-19 quickly brought all study activities to a halt. The impact of COVID-19 on the study included:

• All face-to-face patient visits stopped. All study visits apart from initial screening required face-to-face contact, therefore almost all study activities stopped.

• Most of the clinical and research staff were redeployed to COVID-19 facing clinical roles.

• All non-urgent surgical procedures, including all SNM procedures, were cancelled.

• Multiple staff changes (due to redeployment) amongst the research teams resulted in loss of commitment to the study.

Once the first COVID-19 wave had subsided, sites remained unable to recruit further patients as UK incontinence surgery (including SNM) was graded in the lowest urgency of surgery category (priority 4) by the Joint Surgical Colleges guidance (www.rcseng.ac.uk/coronavirus/surgical-prioritisation-guidance/). Face-to-face visits restarted over the summer of 2020 at some sites (but certainly not all) allowing follow-up visits to gradually restart.

In June 2021, the decision to restart recruitment was made on the balance of some key sites having restarted SNM surgery with the large backlog of patients available. Unfortunately, the Omicron strain of COVID-19 caused further disruptions over the winter period (2021–2). SNM surgery remained low priority, behind the backlog of cancer related and other surgical procedures, resulting in many sites not having restarted in early 2022. Once recruitment restarted, patients appeared more reluctant to take part in the research as their SNM implantation would be delayed.

Participant retention and data completeness

The effects of COVID-19 were not limited to recruitment, but also played a major role in participant retention and data collection.

• Delays in patient follow-up led to an increase in patient withdrawals as they faced an unknown delay in the cross-over period as well as extra visits to hospital that increased the risk of becoming infected with COVID-19. The two participants randomised in Ireland became lost to follow-up due to the much stricter and longer lockdown rules imposed there. A further two withdrew in the UK.

• Data quality: across all sites research staff were redeployed to clinical care or COVID-19 research. Even after routine outpatient care resumed, research teams were not able to fully support data collection and checking. This affected the data collection in two ways. First, at one site, data collection became much less thorough as the research team were less able to support the clinical team who themselves were under pressure with the backlog of work. This meant that bowel diaries were poorly checked for patient completion. Secondly, data entry also became severely delayed. Centralised data entry was in place with sites emailing completed case report forms via NHS.net. This meant that any issues with protocol deviations such as missing baseline diaries were not detected in a timely manner so changes could not be made to further prevent any issues. COVID-19 also delayed rollout of the trial database thus further reducing the ability to detect and act on data completeness concerns.

Protocol violations

Two patients were found to have their devices switched off when they should have been receiving SNM. The first patient had turned off the device for driving but did not turn it back on again. This was discovered at visit 6, but the amount of time the patient had not been receiving treatment was unable to be determined. The second patient was found to have a defective programmer at visit 9 and not receiving treatment. This occurred during the period when visits were delayed due to COVID-19.

Statisticians were informed both times but not unblinded. The first patient was analysed as per ITT. The second patient had not completed the bowel diaries. As the visits were disordered due to COVID-19 and the site had made the decision to withdraw her from the study at visit 9. As these patients had not completed the bowel diaries they were not included in the final analysis of the primary outcome.

Baseline data

Implantation details and intraoperative data

Implantation details, including intraoperative sensory and motor responses are detailed in Appendix 4. Test stimulation was performed using a tined lead in 68.6% of the participants. The lead was positioned in the S3 foramina in most participants (91.4%). General anaesthesia was used in 70.6% of the procedures and median operating time was 36 minutes [interquartile range (IQR) 30–55 minutes]. The implantation was considered successful in all cases, and a median post-operative stay was 3 hours (2–4 hours).

Motor responses were used in most participants to guide correct lead placement. There were some variations in fidelity of siting based on individual electrode responses. Bellows contraction was observed in the majority of implantations (93.5%), followed by big toe flexion (80.0%) and anal sphincter contraction (62.5%). An ideal intraoperative motor response (all 4 electrodes < 1 V, bellows contraction, and big toe flexion) was only recorded in 18% of participants. Sensory responses were only used in a small number of participants (n = 3). A total of 50% lead placements achieved motor or sensory responses for 3 electrodes < 1 V. Initial programming data are shown in Appendix 5.

Primary outcome

Due to the under-recruitment it is important to interpret the findings as exploratory. The effect of SNM versus sham on total FI events per week is shown in Table 5 for the 16 participants with complete data. A table including these data and the effect by allocated arm are supplied in Appendix 6. Compared to sham, SNM led to a non-significant mean difference of < 1 episode per week (−0.7, 95% CI −1.5 to 0.0; p = 0.06; see Table 5). The number of days the paper bowel diaries were completed for the primary outcome throughout the entire study by all randomised participants is shown in Table 6. A sensitivity analysis (Table 7) using a best-case scenario in which missing counts were imputed as zero when at least one item had been completed for that day led to a slightly greater effect size (paired t-test: −0.9, −1.8 to 0.0; p = 0.04). The treatment effect was greater but less precise in the small number of participants (N = 7) who used e-event recordings as an alternative method of measurement of the primary outcome (paired t-test: −1.5, −3.5 to + 0.5; p = 0.12).

Secondary outcomes

As for the primary outcome, due to the under-recruitment, it is important to interpret the secondary outcomes as exploratory.

Comparison of crossover effect sizes with changes from baseline

Although baseline observations have no relevance to detection of changes in the cross-over design, the large differences observed between baseline and findings in both arms, regardless of period, merit documentation. For the primary outcome, in the subset of participants (N = 16) who contributed to the primary analysis, baseline FI events showed a mean of 6.4 per week (SD 6.2). This compares with a mean of 3.0 (SD 3.7) for sham stimulation. This effect, a halving of events, is clearly much greater than the within cross-over effect of SNM versus sham (−0.7, CI −1.5 to 0.0). Similar observations could be made for all other outcomes for example for St Mark’s incontinence score, there was almost no within cross-over effect (−0.15 points) but a reduction of about 5 points (19–14) with sham stimulation. This effect was less noticeable for viscerosensory bowel diaries.

Order effects and washout

Appendix 6 shows some variability between cross-over periods in terms of effect sizes for some variables (although not for the primary outcome: −0.8 SNM/sham vs. −0.7 sham/SNM). There was some evidence for an order effect with most variables following a trend to symptom reduction in period 2 versus period 1 (regardless of arm). This mitigates against a carryover effect from period 1.

Cohort follow-up

Table 10 summarises the bowel diary and e-event outcomes in the 19 participants with complete data at baseline and at the end of study (58 weeks follow-up). Symptom improvement was observed for all recorded outcomes. Number of FI episodes halved to a mean of three episodes of per week [baseline: 6.2 (SD 6) vs. T58: 3.2 (SD 3.3)]. Unlike during the crossover (where urgency episodes non-significantly increased), mean number of urgency episodes per week also decreased from 7.7 (6.0) at baseline to 2.8 (3.2) at 58 weeks. Loperamide usage decreased from 57% of days at baseline to 7% of days at follow-up. The percentage of days FI limited patient’s social activities also decreased from 26% at baseline to 0% at follow-up. Data on e-event outcomes at baseline and follow-up were available in three participants only.

Improvement was also observed for the secondary outcomes at follow-up (Table 11). There was a mean decrease on the St Mark’s incontinence score from 19 (baseline) to 13.5 (58 weeks). Faecal Incontinence Quality of Life Questionnaire improved on all domains, as did the SF-ICIQ-B and EQ5D-5L. Interestingly, in contrast to a numerical difference in OAB-q SF between both arms during the cross-over period (in favour of SNM), improvement at 58 weeks follow-up compared to baseline was less noticeable. Patient-reported treatment success at 58 weeks follow-up (recorded on a 0–100 Likert scale) was high [median 80.0 (IQR 65.0–87.5)].

Mechanistic study results

Findings in FI patients versus controls

Evoked afferent activity

Mean sensory threshold and SL for posterior tibial nerve stimulation (TNS) and AES in patients versus controls are shown in Table 13.

TABLE 13 - Comparing TNS and AES sensory thresholds and stimulation levels in patients vs. controls

Measurement	Patients (n = 25)	Controls (n = 20)	Significance
Mean (range) posterior tibial nerve sensory threshold (mA)	4.5 (2.3–7.8) (SEM 0.3, 95% CI 3.9 to 5.2)	3.4 (0.9–5.7) (SEM 0.3, 95% CI 2.6 to 4.0)	p = 0.025
Mean (range) posterior tibial nerve stimulation level (mA)	13.1 (6.0–42.000) (SEM 1.4, 95% CI 10.1 to 16.1)	8.4 (2.8–16.3) (SEM 0.8, 95% CI 6.7 to 10.1)	p = 0.013
Mean (range) anal electrical sensory threshold (mA)	9.5 (1.10–49.80) (SEM 2.2, 95% CI 5.0 to 14.0)	5.5 (2.3–11.4) (SEM 0.6, 95% CI 4.29 to 6.70)	p = 0.126
Mean (range) anal electrical stimulation level (mA)	20.8 (4.4–80.5) (SEM 20.8, 95% CI 13.6 to 27.9)	14.3 (5.8–28.5) (SEM 1.4, 95% CI 11.4 to 17.2)	p = 0.094

The data show some significant differences in sensory threshold (p = 0.025) and SLs (p = 0.013) between patients and controls although there were very wide variations, especially in the patient group. Overall, responses were ‘harder’ to elicit in patients than controls.

Table 14 shows the CEP latencies derived from right-sided posterior tibial nerve stimulation (TNS) for patients (at baseline) and controls based on cortical recording site (SI, SII left and right) and by sensor (all channels) and source (localised areas using beam former) level. Table 15 shows similar data for AES. An exemplar CEP is shown in Figure 6.

TABLE 14 - TNS evoked potential latencies in patients vs. controls

N,  number of subjects in which a CEP latency could be measured; NA, not applicable.

Measurement	Patients (n = 25)	Controls (n = 20)	Significance
Sensor level: mean (range) ms	N = 21 141.1 (67.0–233.0) (SEM 4.873, 95% CI 58.42 to 78.58)	N = 24 68.5 (45.0–145.0) (SEM 4.873, 95% CI 58.42 to 78.58)	p < 0.001
Source level: mean (range) ms
Left SI	N = 13 81.00 (64.00–142.00) (SEM 6.774, 95% CI 66.24 to 95.76)	N = 15 70.87 (56.00–87.00) (SEM 3.012, 95% CI 64.41 to 77.33)	p = 0.164
Left SII	N = 4 99.75 (65.0–120.0) (SEM 12.42, 95% CI 60.23 to 139.3)	N = 8 90.38 (56.0–126.0) (SEM 8.242, 95% CI 70.89 to 109.9)	p = 0.534
Right SI	N = 4 103.3 (63.00–156.00) (SEM 20.82, 95% CI 36.98 to 169.5)	N = 0	NA
Right SII	N = 7 115.2 (85.0–137.0) (SEM 7.984, 95% CI 95.75 to 134.8)	N = 0	NA

TABLE 15 - AES evoked potential latencies in patients vs. controls

N, number of subjects in which a CEP latency could be measured.

Measurement	Patients (n = 25)	Controls (n = 20)	Significance
Sensor level: mean (range) msec	N = 20 138.9 (67.0–233.0) (SEM 8.861, 95% CI 120.4 to 157.4)	N = 23 133.3 (92.0–206.0) (SEM 6.952, 118.9 to 147.8)	p = 0.620
Source level: mean (range) msec
Left SI	N = 8 164.5 (80.0–243.0) (SEM 20.73, 95% CI 115.5 to 213.5)	N = 12 160 (93–228) (SEM 160, 95% CI 131.1 to 188.9)	p = 0.849
Left SII	N = 12 161 (92–229) (SEM 12.64, 95% CI 133.2 to 188.8)	N = 15 179 (82–408) (SEM 19.16, 95% CI 137.9 to 220.1)	p = 0.465
Right SI	N = 8 150.4 (105.0–227.0) (SEM 14.54, 95% CI 116.0 to 184.8)	N = 12 160.0 (92.0–228.0) (SEM 13.13, 95% CI 131.1 to 188.9)	p = 0.948
Right SII	N = 13 153.9 (92.0–250.0) (12.03, 95% CI 127.7 to 180.1)	N = 2 132.5 (120.0–145.0) (SEM 12.5 95% CI −26.33 to 291.3)	p = 0.513

FIGURE 6.

Exemplar CEP recording from USM-009. The figure shows a CEP latency at the left SI at 84 ms as demonstrated by the red arrow. Time of stimulus is represented by the dotted red line.

The tables show that with the exception of TNS sensor level data, there were no statistically significant differences between patients and controls. Sensor data are difficult to interpret because they represent an average of all brain areas. As for stimulation thresholds and levels, there was a general observation that well defined CEPs were harder to elicit in the patient group. Also, AES CEP latencies were generally harder to elicit than tibial CEPs with higher ranges of values.

Induced motor activity

Induced motor MEG activity was demonstrated throughout the whole sensorimotor strips bilaterally during voluntary fist and anal squeezing activity in both patients and controls at 14–30 Hz (beta band) (Figure 7). The increase in beta oscillations was reflected by a decrease in activity (desynchronisation), also known as movement-related beta decrease (MRBD) or event related desynchronisation (ERD). This is later followed by an event related synchronisation (ERS) or post-movement beta rebound (PMBR). In both groups, cortical activity followed the same pattern as the EMG response although for patients this was generally less timely and less brisk. The apparently larger cortical area of activity seen in patients should not be overinterpreted. While this could reflect compensatory processes, without very detailed statistical analyses of individual brain coordinates, such changes in visual representation can reflect tiny variations in activity.

FIGURE 7.

Induced motor MEG activity was demonstrated throughout the whole sensorimotor strips bilaterally during voluntary squeezing activity in both patients (A) and controls (B) at 14–30 Hz (beta band). The relationship between MEG activity as demonstrated on time-frequency response (TFR) plot and EMG activity in patients (C) and controls (D).

Findings between randomised phases (SNM vs. sham)

Evoked afferent activity

Comparable tables for mean sensory threshold and SL for TNS and AES in SNM versus sham are shown in Table 16. No significant differences were observed although sham data were all numerically higher than for SNM.

TABLE 16 - Comparing TNS sensory thresholds and stimulation levels in SNM vs. sham groups

Measurement	SNM (n = 10)	Sham (n = 12)	Significance
Mean (range) posterior tibial nerve sensory threshold (mA)	3.6 (2.4–5.5) (SEM 0.3, 95% CI 2.8 to 4.4)	3.9 (1.8–6.0) (SEM 0.4, 95% CI 3.0 to 4.8)	p = 0.598
Mean (range) posterior tibial nerve stimulation level (mA)	9.2 (6.0–13.8) (SEM 0.8, 95% CI 7.3 to 11.0)	9.6 (4.4–15.0) (SEM 0.9, 95% CI 7.6 to 11.6)	p = 0.729
Mean (range) anal electrical sensory threshold (mA)	8.0 (2.5–17.2) (SEM 1.6, 95% CI 4.4 to 11.5)	8.0 (3.5–14.8) (SEM 1.3, 95% CI 5.2 to 10.8)	p = 0.986
Mean (range) anal electrical stimulation level (mA)	17.3 (5.4–43.0) (SEM 3.9, 95% CI 8.5 to 26.2)	18.5 (8.7–37.0) (SEM 2.7, 95% CI 12.6 to 24.4)	p = 0.809

Comparable tables for CEP latencies with TNS and AES are shown in Tables 17 and 18. The very small N numbers and large ranges of latency severely limit interpretation.

TABLE 17 - TNS evoked potential latencies in SNM vs. sham

N, Number of subjects in which a CEP latency could be measured.

Measurement	SNM (n = 10)	Sham (n = 12)	Significance
Sensor level: mean (range) msec	N = 6 55.17 (50.0–61.0) (SEM 1.740, 95% CI 50.69 to 59.64)	N = 8 58.50 (47.0–75.0) (SEM 3.094, 95% CI 51.18 to 65.82)	p = 0.410
Source level: mean (range) msec
Left SI	N = 6 71.5 (62.0–94.0) (SEM 5.078, 95% CI 58.45 to 84.55)	N = 7 71.9 (58.0–97.0) (SEM 5.106, 95% CI 59.36 to 84.35)	p = 0.962
Left SII	N = 3 131.3 (91.0–184.0) (SEM 27.55, 95% CI 12.81 to 249.9)	N = 5 94.6 (78.0–152.0) (SEM 14.92, 95% CI 53.18 to 136.0)	p = 0.242
Right SI	N = 2 127.5 (60.0–195.0) (SEM 67.5, 95% CI −730.2 to 985.2)	N = 2 70.0 (61.0–79.0) (SEM 9.0, 95% CI −44.36 to 184.4)	p = 0.487
Right SII	N = 4 115.5 (90.0–135.0 (SEM 9.836, 95% CI 84.2 to 146.8)	N = 5 110.2 (70.0–153.0) (SEM 14.44, 95% CI 70.1 to 150.3)	p = 0.783

TABLE 18 - AES evoked potential latencies in SNM vs. sham

N, Number of subjects in which a CEP latency could be measured.

Measurement	SNM (n = 10)	Sham (n = 12)	Significance
Sensor level: mean (range) ms	N = 7 138.4 (99.0–187.0) (SEM 11.96, 95% CI 109.2–167.7)	N = 4 143.8 (133.0–160.0) (SEM 5.865, 95% CI 125.1–162.4)	p = 0.758
Source level: mean (range) ms
Left SI	N = 3 157.7 (143.0–174.0) (SEM 8.988, 95% CI 119.0 to 196.3)	N = 4 132.5 (80.0–184.0) (SEM 21.7, 95% CI 63.44 to 201.6	p = 0.849
Left SII	N = 5 132.4 (98.0–178.0) (SEM 17.57, 95% CI 83.62 to 181.2	N = 5 148.0 (86.0–231.0) (SEM 24.91, 95% CI 78.85 to 217.1)	p = 0.465
Right SI	N = 12 160.0 (93.0–228.0) (SEM 13.13, 95% CI 131.1 to 188.9)	N = 8 150.4 (105.0–227.0) (SEM 14.54, 95% CI 116.0 to 184.8)	p = 0.948
Right SII	N = 13 153.9 (92.0–250.0) (SEM 12.03, 95% CI 127.7 to 180.1)	N = 2 132.5 (120.0–145.0) (SEM 12.5, 95% CI −26.33 to 291.3)	p = 0.513

Induced motor activity

Induced motor MEG activity was demonstrated throughout the whole sensorimotor strips bilaterally during voluntary fist clenching and anal squeezing activity in both patients and controls at 14–30 Hz (beta band) (Figure 8). The increase in beta oscillations was reflected by a decrease in activity (desynchronisation), also known as movement-related beta decrease or event related desynchronisation (ERD). This is later followed by an event related synchronisation (ERS) or post-movement beta rebound (PMBR). In both groups, cortical activity followed the same pattern as the EMG response. The apparent visual difference in distribution of cortical activity between SNM and sham (wider in sham) is subject to the same caveats as for patients and controls (above).

FIGURE 8.

Induced motor MEG activity was demonstrated throughout the whole sensorimotor strips bilaterally during voluntary squeezing activity in SNM (A) and sham (B) groups at 14–30 Hz (beta band). The relationship between MEG activity as demonstrated on time-frequency response (TFR) plot and EMG activity in SNM (C) and sham periods (D).

Recruitment failure and other limitations

The SUBSoNIC trial ended after about four years having failed to recruit the required number of participants (N = 90). Of the 39 recruited, only 16 contributed to analysis of the pre-defined primary outcome. The inference of the findings (discussed below) is therefore very limited. These failures can be in part attributed to the COVID-19 pandemic which hit about two years into the lifespan of the trial. The impact of COVID-19 (discussed in detail in the results) was felt on both recruitment and retention of participants but also greatly affected fidelity of data capture; the failure to recognise deficiencies of data capture (as might normally have been picked up by data input checks) meant that no corrective actions were taken. While Figure 9 demonstrates that at the time of the first COVID-19 lockdown, recruitment had picked up, there were other causes of poor recruitment that are common to many trials of complex interventions in the NHS. These included the very slow opening of sites due to contractual and governance processes with some sites never opening despite having had PI involvement (and enthusiasm) from the study’s inception. Also, the researchers predicted a higher conversion rate from screening than that observed. Of 220 patients screened, the number declining participation was lower than expected (N = 49, i.e. only about one quarter). The problem was the collective frequencies of a multitude of other excluding factors that led to only 39 being randomised. Most of these factors were anticipated but the influence of some was underestimated. In particular, it seemed as if the ‘archetypal’ female patient, with sphincter injury and urge FI, was much less common than assumed. Since the start of SUBSoNIC, the researchers and others have published studies that support the notion that problems of rectal evacuation (an exclusion criteria) are much more prevalent in the FI population than previously supposed. 50,51 Also, 50 patients failed to meet the minimum symptom severity criterion and other criteria that would normally be expected for clinical consideration of SNM like failure of first line treatments. This perhaps reflects the enduring nature of FI as a problem where even the slightest frequency of leakage has a severe effect on QOL leading to progression along a pathway of care toward specialist surgery.

FIGURE 9.

Recruitment graph showing the start of the COVID-19 pandemic in the UK (arrow).

Primary outcome

Excepting very important caveats of under-recruitment (39 of 90) and attrition (only 16 with complete data), SUBSoNIC is the first randomised study of SNM in a treatment naive population and probably the first with effective double blinding. The mean difference in effect between SNM and sham (−0.7 FI, CI −1.5 to 0.0, episodes per week) represents a mean percentage reduction of 23.3% when expressed with reference to sham frequency (0.7/3.0). This effect is less than that sought by the predetermined sample size calculation (0.77 vs. 0.70) and therefore if extrapolated (assuming that further population sampling derived similar patient responses) would not have met the primary trial endpoint. However, it was clear that the inadequate completion of bowel diaries, even within the 16 participants who provided data in both periods, had an influence on outcome. The problem arose from days (of the 28 total possible) where no entries were made for any symptom despite instruction to put a zero in days where no events occurred. This posed the question whether such days represented a true absence of any symptoms or a day of failed compliance. Such problems are well known40 and informed the clear written instructions provided to participants (see Appendix 1). When zeros were imputed for missing counts on days with at least one entry in a sensitivity analysis the estimate of effect increased slightly [−0.9 (−1.8 to 0.0)], which when compared to a mean sham frequency of 3.0 is equal to the 30% predicted in the sample size estimation. The finding of a very different effect size (−1.5, CI −3.5 to +0.5) using the e-event recording method invokes further uncertainty regarding the interpretation of the primary outcome. Figure 10 shows the difference between rating methods graphically.

FIGURE 10.

Comparison of effect estimates for the primary outcome (FI episodes per week) based on different methods of detection (outcomes 1, 2 and 3 in Figure). All show a directional effect (reduction in events) in favour of therapy. This is much more marked for open label use at 58 weeks (outcome 4 in Figure) than for blinded evaluations. Of blinded findings, the e-diary provides a greater magnitude of effect, but this is imprecise due to small numbers (N = 7). The paper diaries show small effects that vary depending on how missing data are treated. The sensitivity analysis is based on treating all unfilled days as 0 episodes whereas the primary analysis excluded these days based on the assumption of non-compliance rather than absence of symptoms. a, Result from sensitivity analysis.

A further important point from the cross-over analysis is that the overall magnitude of effect, regardless of how it was measured, is much less than the effect of being allocated to either arm or in either period. Overall, compared to baseline mean FI episode frequency, which was of the order of seven events per week (as correctly predicted in the sample size assumptions), the frequency of FI episodes in the sham periods had fallen to approximately three. This reduction in symptoms was also observed for the open label follow-up cohort of 19 participants where FI episodes reduced from a baseline of 6.2 (SD 5.9) to 3.2 (SD 3.3). Given that participants were unable to correctly predict their allocation (SNM vs. sham) and most (approx. 60%) thought they were receiving active stimulation regardless of a cross-over period, this points to a strong placebo response and is in keeping with other studies of neuromodulation in general (see below). However, it is unsure to what extend COVID-19 had an effect on this finding. A number of participants filled in their baseline questionnaires before the COVID-19pandemic. As participants had to stay at their homes during the lockdowns, it could be that FI episodes occurred less frequent (in both arms) as they stayed close to the toilet. Another possibility is that some participants inflated their symptoms at baseline to gain access to SNM therapy (and trial entry).

Secondary outcomes

As predetermined in the SAP, secondary outcomes were analysed for all randomised participants who completed at least one observation at one timepoint (N = 20–24 depending on outcome). Very small non-significant effects were observed for all secondary outcomes. These all favoured SNM on the basis of direction of change with the exception of urgency which increased with SNM. The increase in urgency may be explained by a genuine treatment effect whereby episodes of urgency that previously resulted in incontinence have been ‘reduced’ to urgency without incontinence because of increased perception. This is in keeping with extensive observational data that report improvements (lengthening) of deferment time with SNM25 and the restoration of rectal sensory function to stimuli such as balloon distension. 30,52 The application of new viscerosensory bowel diaries43 sought to further explore this phenomenon that is effects of SNM on enteroception, which are considered to reflect the dominant mechanism of action of SNM. Regrettably, despite the wealth of data for a small number of participants, the poor overall uptake (and incomplete pairing of data) limited the analysis. Results have only been documented descriptively (see Appendix 5) and show no obvious effect.

The only secondary outcome to show a potentially meaningful level of clinical effect was the International Consultation on Incontinence Questionnaire Overactive Bladder Module (ICIQ-OAB), a questionnaire for evaluating overactive bladder and related impact on QOL. 42 The ICIQ-OAB provides a measure to assess the impact of urinary frequency, urgency, urge incontinence and nocturia symptoms and has a range of 0–100 where higher scores represent more symptom burden. A reduction of 10 points or more as observed (−10.8, CI −23.0 to +1.14) is considered a minimally important clinical difference. Urinary symptoms are common in patients with FI53 and this finding is in keeping with SNM’s position as the dominant second line therapy for OAB. 54,55