Microdiscectomy compared with transforaminal epidural steroid injection for persistent radicular pain caused by prolapsed intervertebral disc: the NERVES RCT

Previous studies of surgical microdiscectomy for sciatica

Surgical removal of the PID is believed to be the treatment of choice for symptomatic PID, with > 90% success rates, return to work within 1 month in the majority of cases and complications of around 2%. 10 Several trials have attempted to compare surgical microdiscectomy with non-operative treatments, but with notable methodological flaws. Population studies have suggested that surgery is more effective than non-surgical treatments, but these studies were affected by selection bias and lacked clear definitions of the non-operative treatment arm. 3,11–13 The two most notable studies are large, prospective case series studies, the Maine Lumbar Spine Study12 and the Spine Patient Outcomes Research Trial,11 and are worthy of specific mention.

Recruiting > 500 patients treated by spinal specialist teams across Maine, almost an equal number of surgical patients were compared with non-surgical patients with sciatica. 12 Surgery was deemed effective in 71% of patients compared with 43% of non-surgical patients. The non-surgical group was, however, quite heterogeneous, with only 18% of non-surgical patients receiving ESI. In addition, the study was observational and, therefore, suffered from selection bias.

The Spine Patient Outcomes Research Trial11 had two main arms: an observational reporting group in which patients selected their treatment (surgical or not) and a randomised arm. Owing to a large degree of treatment crossover in the randomised section, intention-to-treat (ITT) analyses showed no significant difference in outcomes. However, the as-treated analysis favoured surgery. Once again, however, the non-surgical treatments being compared with surgery were heterogeneous.

To our knowledge, only one previous study14 (n = 100) has directly compared surgery with ESI (interlaminar route) and this study reported that ESI could prevent surgery in approximately 50% of cases. This study was a single practitioner cohort, but randomisation was employed in design of the study.

Previous studies of epidural steroid injections for sciatica

Epidural steroid injections are known to improve patients’ sciatica, but their efficacy varies widely throughout the literature. 15 A wide variation in practice exists across the UK in the methods of administration of the ESI.

Epidural steroid injections involve the administration of a mixture of local anaesthetic and steroid into the epidural space via one of three main routes: (1) through the base of the spine (i.e. a caudal epidural), (2) through the back of the spine (i.e. a interlaminar injection) or (3) through the nerve tunnel directly adjacent to the prolapsed disc (i.e. transforaminal injection). In the past, the most widely used injection therapy route was ESI, by either the interlaminar route or the caudal route. However, placing the needle through the bony tunnel through which the lumbar nerve root exits the spine can accurately place the drug closer to the target site (i.e. TFESI). Currently, ESI is not commissioned by all local commissioning groups within the NHS in the UK for sciatica, whereas TFESI is. This technique routinely requires X-ray guidance or computerised tomography (CT) scanning guidance and most pain clinics in the UK are able to offer this treatment.

Although randomised controlled trials (RCTs) have tested ESI for acute sciatica, these trials have not included comparisons between TFESI and interlaminar ESI. However, prospective and case–control studies have compared the two techniques and demonstrated a superior efficacy of TFESI. 13,16,17 A comprehensive review of the literature has recently been published by the Health Technology Assessment (HTA) programme. 15 Only one small RCT14 (n = 100) directly compared interlaminar ESI with surgery for sciatica secondary with PID and suggested that ESI could prevent 50% of surgical interventions. One previous UK RCT, the Wessex Epidural Steroid Trial (WEST)18 (n = 228), funded by the HTA programme, compared interlaminar injection of steroid with placebo (i.e. injection of saline between the spinous processes) in patients with sciatica ranging in duration from 4 weeks to 18 months and found no benefit of steroid injections beyond 3 weeks of follow-up. However, in this study, MRI was not undertaken as part of the trial to confirm pathology, with this relying on clinical findings alone. These possibly could be some of the factors contributing to less than promising results for ESI. Various other studies have shown that ESIs have only a small short-term effect on leg pain and disability compared with placebo, and no effect in the long term. 19 These poor medium- to long-term results have given ESIs poor perceived efficacy and hence they are not commissioned or recommended in the treatment and clinical pathway of sciatica secondary to PID management.

A prospective randomised study20,21 reported that transforaminal administration of the drug mixture into the epidural space (i.e. TFESI) under fluoroscopic guidance is the most successful route (more so than injection of saline or local anaesthetic into the epidural space, or intramuscular steroid or saline injection), and this route was used in this study. Relief of pain was corroborated by significant improvements in function and disability, and reductions in use of other health care.

Transforaminal epidural steroid injection is believed to be superior in efficacy to interlaminar administration of ESI, as the drug is delivered more accurately and closer to the site of the pathology/disc prolapse. A prospective study17 of TFESI (n = 48) for acute sciatica suggests long-term pain reduction in > 80% of patients. One RCT published in 201120 (n = 150) compared the outcomes of selective nerve root injection and local anaesthetic, local anaesthetic alone, normal saline and intramuscular injection of steroid or normal saline. The only radiological feature associated with successful outcome was the grade of nerve root compression. Of patients with low-grade root compression (n = 71), 75% responded favourably to selective nerve root injection and avoided surgery by 54 weeks’ follow-up.

Although there are few data directly comparing TFESI with interlaminar steroid injections for sciatica, during the recruitment stage of this trial a number of ongoing studies throughout the world were specifically looking at this. However, these studies were experiencing recruitment difficulties because of the lack of a surgical treatment arm. 22,23 One recent study24 (n = 238) reported that 65% of injections were effective at follow-up of > 6 months (based on patient-reported measures).

Adverse events (AEs) associated with TFESI procedures are rare, typically < 1%, but can be severe and include paraplegia, infection, haematoma, intravascular injection of medication, direct nerve trauma, subdural injection of medication, air embolism, disc entry, urinary retention, radiation exposure and hypersensitivity reactions.

The advantages of spinal injections include:

• The injections are a relatively cheap and low-risk procedure compared with surgery.

• Success rates have been estimated to be as high as 75%.

• Injections are delivered as a day-case procedure and, therefore, require no hospital admission and can be easily repeated.

• There is a range of treatment providers, including radiologists, surgeons and pain physicians.

The disadvantages of spinal injections include:

• The true success rate of injections is largely unknown. The injections may work well in the short term, but pain may return some weeks later.

• Injections are not able to prevent physical nerve root compression and are inappropriate for massive disc prolapses that cause motor weakness or numbness in the leg.

Primary objective

• To compare the clinical effectiveness of microdiscectomy with TFESI at 18 weeks post randomisation.

Secondary objectives

• To compare the clinical effectiveness of microdiscectomy with TFESI up to 1 year post treatment.

• To compare the cost-effectiveness of microdiscectomy with TFESI.

• To compare quality-of-life (QoL) outcomes for both treatments.

Ethics considerations

The trial abided by the principles of the World Medical Association Declaration of Helsinki. 33

Both of the treatments offered as part of the trial are standard NHS practice. As such, there were no major ethics concerns. When treatment has been considered to be unsuccessful, participants had full access to additional treatment needed as per routine care. Participation in the trial did not prevent access to additional treatments needed.

The specific issues pertaining to this trial are:

• requirement for an additional visit

• patients were randomised and, therefore, were unable to choose their own treatment.

Funding was in place to allow reimbursement of financial costs incurred by the trial participant to attend an additional appointment (i.e. a 54-week follow-up appointment post randomisation).

Patients provided informed consent to participate, with information provided about the randomisation process, data collection and other trial processes.

Ethics approval

The trial protocol received the favourable opinion of a Research Ethics Committee prior to initiation at the Liverpool Clinical Trials Centre (LCTC) and underwent independent review at the research and development (R&D) offices of participating centres. Local R&D offices were sent the appropriate centre-specific information form to complete with the necessary authorisation signatures, plus any other documentation requested for review. A copy of local R&D approval was forwarded to the LCTC before the centre was initiated and patients recruited.

Consent from patients was obtained prior to participation in the trial and after a full explanation had been given of the treatment options, including the conventional and generally accepted methods of treatment. Patients were asked to read and review a patient information sheet and consent form (PISC) containing key information about the trial, and then complete, sign and date the consent form if they consented to take part in the trial. The right of the patient to refuse consent to participate in the trial without giving reasons was respected. After the patient entered the trial, the clinician remained free to give alternative treatment to that specified in the protocol, at any stage, if he/she felt it to be in the best interest of the participant. However, the reason for doing so was recorded and the participant remained within the trial for the purpose of follow-up and data analysis, according to the treatment option to which they have been allocated. Similarly, participants remained free to withdraw at any time from the protocol treatment and trial follow-up without giving reasons and without prejudicing further treatment.

Selection of centres/clinicians

The trial was run in NHS outpatient neurosurgical, pain and orthopaedic clinics and community-based services. Patients were recruited from units receiving patients from pooled tertiary referrals from general practitioners (GPs), allied health professionals and non-spinal consultants.

Participating centres were initiated once all regulatory approvals and trial-specific conditions (e.g. training requirements) had been met, and all necessary documents had been returned to the LCTC.

Centre/clinician inclusion criteria

1. TFESI performed according to protocol requirements (i.e. specified pharmaceutical agents available from pharmacy via local routine prescription routes).

2. Able to provide both treatments within 12 weeks of randomisation.

3. Principal investigator can be a representative of either neurosurgery or pain management (note that both specialties should be represented within the local research team).

4. Clinical equipoise.

5. Local R&D approval.

6. Completion and return of a ‘delegation of authority and signature log’ to the LCTC.

7. Completion and return of centre suitability assessment to the LCTC.

8. Signed contract between centre and sponsor.

9. Receipt of evidence of adherence to points (a) to (g) by the LCTC.

10. Complete progression through the green light checklist.

Centre/clinician exclusion criteria

• Not meeting the inclusion criteria listed above.

Inclusion criteria

Patients were eligible for inclusion in the trial if they met the following criteria:

• They had been diagnosed with lower extremity radiculopathy (sciatica).

• They had sciatica secondary to prolapsed intervertebral disc (proven by magnetic resonance imaging).

• The duration of their symptoms was between 6 weeks and 12 months. [Note that, if symptoms were episodic, then ‘duration of symptoms’ refers to the initial incidence of severe symptoms (i.e. the disc prolapse). It does not refer only to the most recent episode.]

• They had leg pain non-responsive to conservative, non-invasive management.

• They were aged 16–65 years.

• They had previously undergone at least one form of conservative (non-operative) treatment (including but not limited to medication, physiotherapy and modification of daily activities) but this had not provided adequate relief of pain/symptoms.

• They provided written, informed consent.

Exclusion criteria

Patients were excluded from the trial if they met any of the following criteria:

• They had a serious neurological deficit (e.g. foot drop/possible cauda equina compression).

• They had previously undergone spinal surgery at the level of the prolapsed intervertebral disc.

• Their current episode of sciatica had lasted longer than 12 months.

• They were aged < 16 years or > 65 years.

• They had not previously undergone any form of conservative treatment.

• Patients with a contraindication for surgery and/or injection.

• They were known to be pregnant.

Contraindications for both groups of treatment were assessed on a case-by-case basis by the health-care team, as per routine NHS practice and according to local policy.

Changes to the eligibility criteria

During the course of the trial the following changes were made to the eligibility criteria.

Screening

All patients who attended a participating trial centre following referral for sciatica secondary to PID (previously proven by MRI scanning) were prospectively screened for trial eligibility. Trial information was provided to patients at, or prior to, the clinic appointment. Potentially eligible patients (i.e. those who met the eligibility criteria listed in Participant inclusion and exclusion criteria) were invited to participate in the trial. At the clinic appointment, the patient was allowed time to discuss the trial, ask questions and decide whether or not to consent to take part in the trial. Owing to the pragmatic nature of the trial, patients provided written, informed consent at the initial visit without requiring further time to consider participation. Patients requiring additional time to consider consent were managed on a case-by-case basis at a centre level and an additional visit occurred if required.

A screening log was maintained at each trial centre to record all individuals screened for the trial and the eventual outcome. Reasons for non-recruitment were documented (e.g. not eligible, declined consent) and the information was used for monitoring purposes. Patients were asked if they would like to provide a reason for non-consent, although they were not obliged to do so. Reasons for non-participation that relate to patient preference were recorded with the undesired treatment listed when possible.

Baseline and eligibility

After obtaining written, informed consent, the baseline case report form (CRF) was completed to assess and confirm eligibility. The baseline CRF included a medical and neurosurgical history based on source data in the participant’s notes and eligibility was confirmed by an appropriately qualified doctor. The details of recruitment into the NERVES trial were recorded appropriately in the participant’s notes [i.e. details of eligibility confirmation (when and by whom), consent and entry into the trial].

Participants were also asked to complete a questionnaire booklet [incorporating the Oswestry Disability Questionnaire (ODQ); modified Roland–Morris (MRM); Core Outcome Measures Index (COMI); visual analogue scale (VAS) for leg and back pain; the EuroQol-5 Dimensions, five-level version (EQ-5D-5L); and a Resource Use Questionnaire (RUQ)], with support from a health-care professional if needed. The participant-completed questionnaires were completed prior to randomisation, but after provision of consent. The ODQ collected primary outcome data for the trial and so it was important that it was completed accurately. Therefore, it was checked by centres and assistance was provided in completing it if required.

Ionising radiation

In accordance with the Ionising Radiation (Medical Exposure) Regulations 2000,34 participants in the trial received a small exposure to ionising radiation in both groups of the trial. This was required to provide imaging for verification of the treatment level for both microdiscectomy and TFESI. The ionising radiation exposure required was part of the normal care pathway and the same exposure would be necessary outside this clinical trial context. There was no additional ionising radiation exposure to participants as a result of trial participation.

Group A: transforaminal epidural steroid injection

Standard nerve root blockade was completed, as per local policy/technique, using the lateral, foraminal portal of entry. All fluoroscopically guided techniques (e.g. CT or X-ray screening) were permitted to specify the correct level. Treating specialists included pain specialists, radiologists, anaesthetists, surgeons and other appropriately qualified medical professionals, as long as radiological level confirmation was incorporated into the procedure.

The NERVES trial is a pragmatic trial and, as such, the agents used were obtained and prescribed via normal NHS routes. The following injection regimen was followed when possible to minimise variability across the participating centres:

• Injectate:

  • steroid [20–60 mg of triamcinolone acetonide (KENALOG™; Bristol-Myers Squibb Pharmaceuticals, Uxbridge, UK)]

  • local anaesthetic [0.25% levobupivacaine hydrochloride (2 ml) (Chirocaine^®; AbbVie Inc., North Chicago, IL, USA)].

As the NERVES trial is a Clinical Trial of an Investigational Medicinal Product, information regarding the pharmaceutical products used was provided to the MHRA. The following active ingredients were notified to the MHRA and, therefore, are also accepted for use if appropriate:

• Steroid:

  • dexamethasone

  • methylprednisolone acetate (Depo-Medrone^®, Pfizer Inc., New York, NY, USA).

• Local anaesthetic:

  • bupivacaine hydrochloride

  • lidocaine hydrochloride.

For the purpose of participant safety, centres were instructed to ensure that maximum doses were not exceeded (Table 1).

Note that, if the maximum dose was exceeded, then a data query form was produced at the LCTC and sent to the centre with a request for justification.

All participants randomised to group A received at least one therapeutic injection. As per local policy, participants could receive another injection if there was a favourable but partial response that could be boosted by further injections. Information about any further injections was collected.

The steroid/anaesthetic combination used in the TFESI was distributed from pharmacy via routine processes and so specific trial labelling was not required as per MHRA Exemption Regulation 46 of the Medicines for Human Use (Clinical Trial) Regulations 2004. 35 It is an off-label use of steroid, but is commonly accepted practice within the NHS and in the further medical field.

Group B: microdiscectomy

Standard microdiscectomy was performed as per local treatment protocols.

Treatment specialists at centres identified the correct side (left or right) and level prior to treatment, with level localisation advised as per local treatment protocols. Information on site and level was collected.

Treatment specialists were an orthopaedic or neurosurgical consultant or consultant equivalent (i.e. associate specialist), or a specialist trainee directly supervised by a consultant.

For both groups, treatment was given within 6 weeks of randomisation when possible and centres were instructed that treatment should occur within 12 weeks of randomisation to ensure valid collection of primary outcome data at the 18-week follow-up visit.

Additional treatments

The NERVES trial protocol allocated only initial treatment for sciatica, either microdiscectomy or TFESI. During the course of follow-up, some participants required further intervention for sciatica, as per routine NHS practice. Further clinical intervention was permitted for trial participants without the participant having to withdraw from the trial.

If a participant received additional treatment, information on the type of intervention (i.e. microdiscectomy or TFESI), the details of the treatment received and the reason were collected, and they remained in the trial.

Trial participants were able to cross over prior to receiving their initial treatment allocation without withdrawing from the trial (e.g. if they became unsuitable for the treatment they were initially randomised to). This was recorded on the treatment CRF with the reason for crossover indicated.

Schedule for follow-up

All follow-up visits were scheduled from the date of randomisation.

Each participant was followed up for 54 weeks following randomisation. During this time, participants attended scheduled follow-up visits. Table 2 shows the follow-up schedule. Any additional procedures provided to the participant and completed at the trial centre during this period were documented.

Normal clinical practice would typically include a 3-month post-treatment follow-up. Therefore, participants were followed up at approximately 18 weeks post randomisation to align with routine clinical practice, and then again at 30, 42 and 54 weeks. To maintain feasibility, the 18-, 30- and 42-week visits could take place within a 2-week visit window on either side. The 54-week visit had an acceptable window of 54–62 weeks post randomisation. Participants could be seen at other times, as clinically indicated. Additional visits outside the trial protocol were recorded.

After randomisation, scheduled treatment and follow-up stages were as follows.

Data collection tools

Participant-reported outcomes

Participants were asked to complete the following participant-reported outcome measures at baseline, 18, 30, 42 and 54 weeks post randomisation:

• ODQ

• MRM outcome score for sciatica

• COMI score

• Likert scale for treatment satisfaction

• EQ-5D-5L

• VAS scores for leg and back pain

• health-care resource use and participant costs (including time off work).

The questionnaire booklet was provided to the participant at the scheduled clinic visits (at baseline and at the 18- and 54-week follow-ups) and completed in clinic. The RUQ booklet was completed at the treatment visit before treatment and was provided to the participant in preoperative assessment.

Completion of these questionnaires was an important part of the trial. Particular emphasis was given to part 1 of the questionnaire booklet (i.e. the ODQ) because it was used to collect primary outcome data for the trial. It was therefore crucial that research staff at centres offered any necessary support to participants to ensure the questionnaires were completed correctly and returned to the LCTC either by centres or by the participant in accordance with the schedule for follow-up. The questionnaires took approximately 15 minutes to complete and participants were advised of the extended visit time prior to their appointment.

All questionnaires completed at baseline were completed after consent had been provided and prior to randomisation.

Original trial sample size calculation

A total of 172 participants was required to detect a difference of 10 points between the two groups on the ODQ at a 5% significance level with 90% power. This assumed a standard deviation (SD) of 20 points based on similar populations in previous published trials. 18,36,37,39,40 The previous large and well-carried out WEST, based in the UK, suggested a baseline ODQ SD of between 16 and 18 points. 18 Baseline ODQ data were collected on 11 potentially eligible patients from the fast-track sciatica clinic at the Walton Centre, Liverpool, and this generated a SD of 14.4 points, well under the assumed value. The initial target sample size for the trial was 200 patients, which would allow for a 10% rate of missing outcome data. Of the originally planned seven centres involved, allowing for one to have difficulties opening, this would require recruitment of 30 patients in total from each participating centre and 50 patients from the lead centre.

Revised sample size calculation for the primary outcome

The original sample size calculation did not assume any correlation between baseline and follow-up ODQ scores, as no data were available to estimate this. The sample size calculation was revisited during the trial with the agreement of the trial oversight committees. Based on a blinded analysis of the correlation between baseline and follow-up ODQ scores in the first 47 trial participants to have outcome data available, the correlation was estimated as 0.49.

Using this estimate, our revised sample size to achieve 90% power was 66 participants per group. Allowing for 10% loss to follow-up gives a revised target of 74 participants per group (i.e. 148 participants total).

Internal pilot study

An internal pilot study was included in the trial design. The study targeted two centres to open first to collect 6 months of recruitment data before progressing to a full trial. These centres, the Walton Centre, Liverpool, and Salford Royal, Manchester, were identified to cover recruitment of participants within specialty and mixed care settings.

The aim of the internal pilot study was to assess the feasibility of recruitment and the rates of potential crossover due to patient preference or treatment failure. The criteria for progression to a full trial were:

• at least 30 patients recruited

• a consent rate of ≥ 40% or more

• < 10% of patients unhappy with allocation and receive the alternative treatment

• < 50% of patients in the injection group proceed to surgery.

Trial Management Group

The TMG was a multidisciplinary team that comprised the chief investigators, several co-investigators, patient and public involvement (PPI) representatives, a sponsor representative, health economists and members of the LCTC. The TMG was responsible for the day-to-day clinical and practical aspects of the trial.

Independent Data and Safety Monitoring Committee

The Independent Data and Safety Monitoring Committee (IDSMC) comprised an independent chairperson, an expert in the field of pain and two independent members (one was an expert in the field of neurosurgery and one was an expert in medical statistics). The main responsibilities of the IDSMC was to safeguard the interests of the NERVES trial participants, assess the safety and efficacy of the interventions during the course of the trial, and monitor the overall progress and conduct of the trial. The IDSMC met at least annually during the course of the trial and provided recommendations to the TSC. Reports to the IDSMC were produced by the statistical team at the LCTC.

Trial Steering Committee

The membership of the TSC included an independent chairperson, an independent expert in the field of pain, an independent expert in the field of neurosurgery and an independent statistician, as well as representatives from the TMG. Observers from the sponsor and the funder were also invited to meetings. The TSC met at least annually, shortly after the IDSMC met and their main role was to provide overall oversight of the trial.

Trial funder

The membership of the oversight committees was suggested by members of the TMG to the trial funders and appointed by the funders with their constitution following funder requirements.

Withdrawals

There were a total of six withdrawals from the trial (three from the microdiscectomy group and three from the TFESI group).

The reasons given for withdrawal in the TFESI group were:

• patient unwilling to have microdiscectomy or TFESI

• site exhausted all means of communication with participant

• chest infection and investigation for possible bowel cancer.

Reasons given for withdrawal from the microdiscectomy group were withdrawal of consent for follow-up, participant not happy with the process (one withdrawal) and withdrawal of consent for follow-up, no additional reason (two withdrawals).

Treatment compliance

A further 14 participants did not initially receive any treatment (seven randomised to microdiscectomy and seven to TFESI). Table 7 shows all combinations of treatments received by participants.

Eight participants did not receive their randomised treatment, but were given the alternative treatment instead (five randomised to microdiscectomy and three randomised to TFESI). Of the five participants randomised to microdiscectomy, three changed because of participant preference, one because of safety concerns from the anaesthetist and one because their symptoms had improved and microdiscectomy was thought to be unnecessary. Of the three participants randomised to TFESI, two changed to microdiscectomy because of participant preference and the other was reported as a surgeon decision.

Additional treatment

The rows in italics in Table 7 indicate participants who have received the alternative trial treatment after their initial treatment. The largest group (28 participants) includes those who received the TFESI initially and subsequently went on to receive microdiscectomy as an additional treatment. The first row in italics denotes those who received their allocated treatment first and then the alternative treatment. The second row in italics denotes those who initially received a treatment to which they were not allocated, but then subsequently additionally received the original planned allocation. Table 8 summarises the timing of treatments for those patients who received both TFESI and microdiscectomy.

Adjustment for additional covariates

An additional model was fitted, adjusting for other prespecified variables, age, sex, duration of symptoms, BMI and estimated volume of canal. This adjusted model gave an estimate of the effect of microdiscectomy compared with TFESI of –5.03 with a 95% CI of –12.76 to 2.70. An additional variable, level of disc prolapse, was included in a post hoc analysis. This model resulted in an estimate of –4.94 (95% CI –12.81 to 2.93). Table 11 shows the full parameter estimates for both models.

A further post hoc analysis was conducted to assess if there was an interaction between duration of symptoms at baseline and randomised treatment. The interaction term was not statistically signficant and the estimated difference in least squares means between treatment groups for microdiscectomy compared with TFESI (when an interaction with duration of symptoms was adjusted for) was –4.18 (95% CI –11.06 to 2.70; p = 0.231). Therefore, there is no statistically significant difference between treatment groups when an interaction with the duration of symptoms at baseline is considered. This supports the conclusion of the primary outcome analysis.

Sensitivity analysis

As > 10% of data on the primary outcome were missing, a sensitivity analysis was conducted using multiple imputation. This analysis used treatment group, centre and all available measurements of ODQ at various time points to impute data to create 50 complete data sets, which were then analysed and combined to give the imputed estimate.

The estimate of the effect of microdiscectomy compared with TFESI from the imputation analysis was –3.08 (95% CI –10.16 to 3.99).

A post hoc sensitivity analysis was conducted to using multiple imputation, as above, but using only baseline ODQ to impute a week 18 score. The estimate of the effect of microdiscectomy compared with TFESI from the post hoc imputation analysis was –3.26 points (95% CI –9.91 to 3.39 points).

Table 12 summarises the estimates from the primary outcome analyses. All estimates lead to the same conclusion, that is there is no statistically significant difference between the treatments, with all estimates showing slightly better outcomes in the microdiscectomy group, although smaller than the prespecified clinically important difference.

Longitudinal outcomes

For the longitudinal outcomes (i.e. ODQ, VAS leg pain, VAS back pain, MRM and COMI) the questionnaires were collected at four key time points (i.e. 18, 30, 42 and 54 weeks post randomisation). The 18- and 54-week follow-up visits were conducted face to face, whereas the 30- and 42-week follow-up time points were postal questionnaires. As a result, the rates of completion are lower at these time points.

For the change from baseline summaries in the following sections the prespecified windows from the protocol for a questionnaire to be included were ± 2 weeks for the 18-, 30- and 42-week follow-ups and between 54 and 62 weeks for the 54-week follow-up. However, in the longitudinal mixed-effects models the time (in weeks) was treated as continuous and all valid and completed questionnaires were included in the analyses.

Secondary outcome 1: Oswestry Disability Questionnaire at 18, 30, 42 and 54 weeks

The ODQ data are analysed using longitudinal models and include the primary outcome time point of 18 weeks.

Table 13 shows the mean and SDs of the change from baseline for the observations at each time point. Table 13 includes observations in the prespecified time windows from the protocol. The longitudinal analysis includes all ODQ measurements taken at any point during the trial. Table 13 shows that most improvement in both groups happens between baseline and week 18, with any further improvements being relatively small.

The longitudinal mixed-model analysis adjusted for baseline ODQ initially included a time–treatment interaction term, but as this was non-significant it was removed from the final model, as prespecified in the statistical analysis plan. The adjusted estimate for the effect of microdiscectomy compared with TFESI was –4.67 points (95% CI –10.61 to 1.28 points; p = 0.123). This estimate is consistent with that shown in the primary outcome analysis of ODQ at 18 weeks.

For the joint modelling post hoc analysis of ODQ scores, the adjusted estimate for the effect of microdiscectomy compared with TFESI was –4.62 points (95% CI –9.84 to 1.27 points; p = 0.108). The estimate is similar to the longitudinal mixed model and the conclusions remained unchanged once adjusted for informative dropout. Figure 4 shows profile plots of mean ODQ score at each scheduled time point.

FIGURE 4.

Mean profile plot of ODQ scores with standard error bars.

Secondary outcome 2: visual analogue scale score for leg pain at 18, 30, 42 and 54 weeks

Table 14 shows change from baseline at each time point for VAS leg pain rating, where valid measurements were available within the prespecified time window. After an initial large reduction between baseline and week 18, average VAS leg pain ratings were similar over the further follow-up visits. Figure 5 shows profile plots of mean VAS leg pain rating at each scheduled time point.

FIGURE 5.

Mean profile plot of VAS leg pain scores with standard error bars.

The longitudinal mixed-model analysis adjusted for baseline VAS leg pain initially included a time–treatment interaction term, but as this was non-significant it was removed from the final model, as prespecified in the statistical analysis plan. The adjusted estimate for the effect of microdiscectomy compared with TFESI was –7.04 (95% CI –15.81 to 1.73; p = 0.115).

For the joint modelling post hoc analysis of VAS leg pain scores the adjusted estimate for the effect of microdiscectomy compared with TFESI was –7.06 (95% CI –15.82 to 0.86; p = 0.098). The estimate is similar to the longitudinal mixed model and the conclusions remained unchanged once adjusted for informative dropout.

Secondary outcome 3: visual analogue scale score for back pain at 18, 30, 42 and 54 weeks

Table 15 shows change from baseline at each time point for VAS back pain rating, where valid measurements were available within the prespecified time window. The largest reduction was between baseline and week 18, with average VAS back pain ratings similar over the further follow-up visits. VAS back pain score reductions were smaller than VAS leg pain score reductions. Figure 6 shows profile plots of the mean VAS back pain rating at each scheduled time point.

FIGURE 6.

Mean profile plot of VAS back pain scores with standard error bars.

The longitudinal mixed-model analysis adjusted for baseline VAS back pain initially included a time–treatment interaction term, but as this was non-significant it was removed from the final model, as prespecified in the statistical analysis plan. The adjusted estimate for the effect of microdiscectomy compared with TFESI was –3.01 (95% CI –11.29 to 5.26; p = 0.473).

For the joint modelling post hoc analysis of VAS back pain scores, the adjusted estimate for the effect of microdiscectomy compared with TFESI was –2.87 (95% CI –10.58 to 3.16; p = 0.457). The estimate is similar to the longitudinal mixed model and the conclusions remained unchanged once adjusted for informative dropout.

Secondary outcome 4: satisfaction with care

The two questions on the satisfaction with care questionnaire were (1) ‘Over the course of treatment for your low back pain or leg pain (sciatica), how satisfied are you with your overall medical care?’ and (2) ‘How satisfied are you with the treatment outcome of your leg pain (sciatica)?’. A score of 1 on either question would be ‘very satisfied’ with care, a score of 2 ‘somewhat satisfied’, a score of 3 ‘neither satisfied nor dissatisfied’, a score of 4 ‘somewhat dissatisfied’ and a score of 5 ‘very dissatisfied’ with care. Therefore, an average score of 1 would imply being ‘very satisfied’ for both aspects of care and an average score of 5 would imply being ‘very dissatisfied’ with both aspects of care. Groups were compared using a Mann–Whitney U-test. The median score at 54 weeks in the microdiscectomy group was 1 (IQR 1–2) and in the TFESI group was 1.5 (IQR 1–3). Figure 7 shows the profile plot of median satisfaction with care scores at each scheduled time point.

FIGURE 7.

Median profile plot of satisfaction with care scores (Likert scores).

The difference between the groups was statistically significant (p = 0.021, Mann–Whitney U-test), with participants in the microdiscectomy group being more satisfied with their care than participants in the TFESI group.

Secondary outcome 5: modified Roland–Morris score at 18, 30, 42 and 54 weeks

Table 16 shows change from baseline at each time point for the MRM score, where valid measurements were available within the prespecified time window. After an initial large reduction between baseline and week 18, average MRM scores were similar over the further follow-up visits. Figure 8 shows the profile plots of mean MRM score at each scheduled time point.

FIGURE 8.

Mean profile plot of MRM scores with standard error bars.

The longitudinal mixed-model analysis, adjusted for MRM initially included a time–treatment interaction term, but as this was non-significant it was removed from the final model, as prespecified in the statistical analysis plan. The adjusted estimate for the effect of microdiscectomy compared with TFESI was –1.82 (95% CI –3.67 to 0.03; p = 0.054).

For the joint modelling post hoc analysis of MRM scores, the adjusted estimate for the effect of microdiscectomy compared with TFESI was –1.72 (95% CI –3.44 to 0.10; p = 0.063). The estimate is similar to the longitudinal mixed model and the conclusions remained unchanged once adjusted for informative dropout.

Secondary outcome 6: Core Outcome Measures Index score at 18, 30, 42 and 54 weeks

Table 17 shows change from baseline at each time point for COMI score, where valid measurements were available within the prespecified time window. As with other outcomes, the largest reduction was between baseline and week 18, although the scores continued to improve in both groups between weeks 18 and 54. Figure 9 shows profile plots of the mean COMI score at each scheduled time point.

FIGURE 9.

Mean profile plot of COMI scores with standard error bars.

The longitudinal mixed-model analysis adjusted for baseline COMI initially included a time–treatment interaction term, but as this was non-significant it was removed from the final model, as prespecified in the statistical analysis plan. The adjusted estimate for the effect of microdiscectomy compared with TFESI was –0.77 (95% CI –1.58 to 0.03; p = 0.059).

For the joint modelling post hoc analysis of COMI scores, the adjusted estimate for the effect of microdiscectomy compared with TFESI was –0.78 (95% CI –1.54 to –0.02; p = 0.046). The estimate for microdiscectomy compared with TFESI is similar to the longitudinal mixed model; however, the p-value and CIs suggest a statistically significant treatment effect once adjusted for informative dropout (but less than the MCID of 2.2).

Secondary outcome 7: work status

Table 18 shows the number of participants who were employed/not employed at baseline. Of those participants who were employed, the numbers of participants who were off work/at work are also presented by treatment group.

At baseline, the information collected was whether or not the participant was employed and, if so, whether or not they were able to work. However, at follow-up the information collected was whether or not the participant was currently employed and whether or not they had returned to work since the last visit. As there were extra visits between the key time points when this information could change but was not collected, based on these questions it was sometimes impossible to say whether or not the participant had returned to work and whether or not they were currently working.

Owing to the issues with the data collection we are unable to accurately calculate the number of days lost from work and the work status outcome, and so these data have not been analysed or presented.

Measurement of resource use data

Resource use categories that might differ between intervention groups were chosen a priori. These were further differentiated into sciatica related and non-sciatica related. The measurement of resource use required complementary approaches using data collected as part of the NERVES trial, as well as routine NHS data. Within-trial costs were estimated by measuring health-care resource use, including (1) microdiscectomy and TFESI procedure, and other hospitalisations (e.g. additional treatments), (2) outpatient appointments and contact with health-care professionals (e.g. GPs, physiotherapists, other health-care professionals), (3) concomitant medication use, (4) costs incurred in the management of AEs and (5) participant out-of-pocket costs. Trial participants’ use of health-care services and other costs were obtained from the following sources:

1. Routine HES data obtained from NHS Digital on secondary care admitted patient care (inpatient) and outpatient attendances.

2. CRFs completed by medical or nursing staff, relating to participant contact with hospital and health-care professionals, including treatments issued, additional treatments received and medications prescribed.

3. A RUQ43 completed by the participant at baseline and treatment visits, and at weeks 18, 30, 42 and 54 for information on the use of primary and secondary care services, visits to other health-care professionals, participant out-of-pocket expenditures on over-the-counter medications and travel costs, as well as productivity losses resulting from days off work, in the previous 3 months (6 weeks at treatment visit) or since the questionnaire was last completed.

4. AE and SAE forms completed by research staff when trial participants were admitted to hospital for events considered possibly related to the trial intervention.

Valuation of unit costs

All resource use was valued in monetary terms (Great British pounds) using appropriate UK unit costs for the cost year 2017/18. When necessary, adjustments were made for inflation using the new Hospital and Community Health Service (HCHS) Index and using the Consumer Prices Index. 44 No discount rate was applied as follow-up was approximately 1 year.

Healthcare Resource Groups (HRGs) were the main currency of secondary care attendances, including the trial interventions, additional treatments, other hospitalisations, and day case, outpatient and accident and emergency (A&E) visits (Tables 19–21). NHS National Schedule of Reference Costs 2017–1845 were applied, as these most closely reflect the actual cost of the provision of care by NHS service providers. Medication costs for steroid and local anaesthetic associated with TFESI were assumed to be included in the HRG costs, as were all costs relating to microdiscectomy. HRGs were further subcategorised according to whether the episode of care took place in the day case, elective or non-elective setting.

The unit costs for all other NHS primary health-care resource use items were obtained from the Personal Social Services Research Unit44 (Table 22). Visit costs to other health-care professionals (i.e. private physiotherapists, osteopathy and chiropracty, acupuncture) were obtained from publicly available sources. 48–51

Concomitant medications were valued in monetary terms using prescription cost analysis data from September 201752 and supplemented by the British National Formulary (BNF). 53 Over-the-counter medication costs were obtained from published retail pharmacy sources. 54 The unit costs of the 10 most commonly prescribed medications are presented in Table 23.

Lost productivity was based on participant-reported lost earnings. In SAs, Office for National Statistics (ONS) Annual Survey of Hours and Earnings median wage values were applied based on full-time employment55,56 (Table 24). Other out-of-pocket participant costs associated with public transport to attend medical appointments were based on participant self-report. If travel was undertaken in a private car, a cost per mile of £0.68, including vehicle running costs, was applied. 57,58

Cost analysis

Resource use data were collected from 3 months (84 days) prior to baseline to the end of the trial at 54 weeks. For completeness, all resource use was measured and costed irrespective of reason. However, to allow for SA, it was categorised as sciatica related or non-sciatica related, as reported in the CRFs or patient questionnaire. Medications were categorised in the CRFs based on reason for prescription, which allowed for identification of those prescribed for sciatica-related reasons. HES data were categorised by reason for admission, based on HRG code, to identify admissions and visits associated with sciatica. The trial design also allowed for further treatment at the discretion of the clinician in the event of incomplete symptom resolution (see Chapter 2, Trial treatments for further details) and costs for subsequent treatments within the trial period were collected via the HES data set and included in the analysis.

When multiple sources reported the same type of resource use (e.g. hospital inpatient admissions), a hierarchical approach was applied, with priority given to routine HES data, followed by researcher-completed CRFs and, last, participant self-reported questionnaires. The different data sources were compared to ensure that all resource use was captured, but also to avoid any overlapping of data and consequent double counting.

To cost secondary care episodes, HES data were sourced from NHS Digital in the form of a series of routinely available codes and dates, including participant diagnoses, admission dates and discharge dates, and then converted into HRG codes using the NHS Digital HRG4+ 2018/19 Payment Grouper. 59 When unavailable, hospital events were assigned an appropriate HRG code based on reason for admission, condition and any complications, by reference to baseline, treatment, additional treatment, AE CRFs and patient questionnaires. Appropriate HRGs were manually applied to unassignable HRG codes (e.g. UZ01C and WA14Z) appearing in the HES data using clinical codes. Costs were assigned according to HRG code, length of stay (incorporating any excess ward bed-days) and whether the case was a day case, elective or emergency, according to the National Schedule of Reference Costs 2017–2018. 45 Locally negotiated unbundled HRGs were similarly identified through the cost grouper and costs were assigned directly from the national schedule. Privately treated participants were assigned the corresponding national schedule NHS HRG code to ensure that costs were appropriately apportioned. Visits to A&E were costed based on participant self-report, and a mean cost per visit was applied according to the national schedule. Participants admitted from A&E to hospital were further identified and costed according to the HES admitted patient care data set. If a hospital admission spanned the period preceding randomisation or beyond the trial period, an adjustment was made to apportion costs to only those incurred during the 54-week trial. To allow for follow-up visits that occurred out of the trial window, all protocol-related visits that were scheduled identically in both intervention groups were included, but all other visits outside the 54-week trial were excluded.

The RUQ collected participant-reported resource use for the preceding 3 months (or 6 weeks at the treatment visit). All visits were face to face except weeks 30 and 42 when questionnaires were issued by post with a pre-paid return envelope. Quantities of resource use for primary care services, including visits to the GP (i.e. surgery, out of hours or home visit), physiotherapist, acupuncturist, chiropractor and osteopath and any other health-care professional, were taken from participant responses and multiplied by unit costs to estimate total costs. As the base-case analysis takes the NHS perspective, all visits to the GP were included, as were NHS physiotherapist visits, assumed on the basis of a participant-reported out-of-pocket cost of < £20, which was attributed to travel costs. Physiotherapy visits with costs of > £20 were assumed to be private physiotherapy visits and were included in the secondary analysis adopting a wider perspective. As some physiotherapy visits were also identified in the HES outpatient data set, a cross-check was performed to avoid double counting and to include NHS physiotherapy visits reported by participants that were not identified in the HES data set. The costs of visits to an acupuncturist, chiropractor, osteopath and any other health-care professionals, as well as any over-the-counter medications, were considered to be participant out-of-pocket costs and included only in the secondary analysis on the basis that the NHS does not routinely fund these treatments. Questionnaires not completed within the predefined visit windows were protocol deviations but, nonetheless, were included in the analysis using appropriate time-based adjustments.

Medication costs were calculated based on CRF-recorded medication usage at baseline, treatment and follow-up visits weeks 18 and 54. All tablets, capsules, oral liquids, sprays and inhalers were costed on a unit dose basis, and creams and gels were assumed to be supplied on a one pack per month basis. If a medication administration spanned the period preceding randomisation, or beyond the trial time period, an adjustment was made to apportion costs to those administered only during the 54-week trial. For medications with missing start and/or end dates, but recorded as ongoing treatments, the start date was assumed to be 3 months (84 days) prior to baseline and the end date the trial exit date. When doses and frequencies were omitted, estimated values based on participants’ medication histories or BNF-recommended doses were applied. Medications administered on an ‘as needed’ basis were costed assuming administration based on 50% of the standard dose recommended in the BNF. In cases where prescribed doses did not match formulation doses, costs were calculated by combining different strengths to form the appropriate dose.

Out-of-pocket travel costs and participant or carer productivity losses were also obtained from the RUQ. Participants were asked to report both the number of days off work and lost earnings. When the number of days off work was reported but lost earnings values were missing, it was assumed that there were no lost earnings (i.e. participants received sick pay).

Follow-up costs regression model

where C is follow-up total cost, ν is predicted follow-up total cost, α is the gamma shape parameter, γ₀ is the regression intercept, γ₁ is the coefficient of injection, γ₂ is the coefficient of baseline total cost, T is 1 for TFESI and 0 for microdiscectomy and C₀ is the baseline total cost.

Quality-adjusted life-years regression model

where Q are QALYs, µ is the predicted QALY, σ is the residual error SD, β₀ is the regression intercept, β₁ is the coefficient of treatment allocation, β₂ is the coefficient of baseline utility score, T is 1 for TFESI and 0 for microdiscectomy and U₀ is the baseline utility score.

Incremental cost-effectiveness ratio

The cost-effectiveness of microdiscectomy compared with TFESI was assessed by its ICER calculated according to the formula:

where Δcosts is the difference in mean total costs between interventions (i.e. cost microdiscectomy minus cost TFESI) and ΔQALYs is the difference in mean QALYs gained between interventions (i.e. QALYs microdiscectomy minus QALYs TFESI).

The INHB allows for further interpretation of cost-effectiveness in reporting the net health benefits that would arise further to adopting microdiscectomy rather than TFESI, conditional on a cost-effectiveness threshold of £20,000 per QALY, as specified by NICE. 42 This was calculated by:

where λ is the chosen cost-effectiveness threshold.

Uncertainties in QALYs, the incremental results and resulting cost-effectiveness metrics were evaluated using a non-parametric bootstrap of the patient-level data. The bootstrap used 10,000 replicates and empirical bootstrap CIs were obtained for the statistics of interest. The uncertainty in the ICER was represented graphically on the cost-effectiveness plane to illustrate the joint uncertainty in incremental costs and QALYs and also as a CEAC to illustrate the probability of either microdiscectomy or TFESI being cost-effective for given cost-effectiveness thresholds. 65 Estimates of ICERs were compared with a threshold of £20,000 per QALY.

Base-case analysis

The base-case analysis was defined as pertaining to the 54-week trial period, based on the imputed data set to account for missing data and subject to the regression analysis. The mean QALYs and costs and incremental values were reported for each treatment group with 95% CIs in accordance with the bootstrap analysis.

The main assumptions made in the base-case scenario are detailed in Table 25.

Sensitivity and scenario analyses

A number of sensitivity and SAs were conducted to assess the impact of considering alternative perspectives, inputs and costs on the ICER (Table 26). Scenario 1 reflects the wider perspective of the secondary analysis and follows the same approach as the base-case analysis, but in approximating to a societal perspective included participant costs associated with time off work and other out-of-pocket costs (Table 27).

A complete-case SA included participants presenting complete-cost data for all cost variables at all time points and outcome data at baseline and at weeks 18 and 54. To assess the impact of alternative value sets, the more recent but yet to be widely adopted value set based on the EQ-5D-5L66 was used in a SA, as was a further evaluation using the EQ-VAS, also collected in the NERVES trial, which records the respondent’s self-rated health on a vertical VAS. The EQ-VAS requires a participant to report how they are feeling on the day of questionnaire completion, by reporting a score between 0 and 100 (anchored at the respondent’s worst and best health they can imagine) and, as such, provides the participant’s own overall assessment of their health. This contrasts to the EQ-5D-3L or EQ-5D-5L utility scores, which are based on public preference valuation of EQ-5D health states.

The wording of the RUQ asked participants to report their resource use over the last 3 months or since the questionnaire was last completed (approximately 3 months). In the base case it was assumed that participants consistently reported resources use over the last 3 months. An alternative approach assumed that participants reported resource use since they last completed a questionnaire, which may exceed 3 months in cases where observations were missed. This reduced the extent of missing data in assuming that later observations include resource use that cover previous missing observations. Further analyses included evaluating the cost-effectiveness for costs related to only sciatica, as well as scenarios to account for uncertainty in dosing regimens for ‘when-needed’ medication.

A number of alternative costing approaches for societal costs associated with productivity losses were also conducted. To account for missing values in patient questionnaires relating to costs of time off work, but when the number of days off work were reported, a substitute value of a day off work was applied using ONS median gross weekly pay by sex and age for full-time employees55 and converted to daily net take-home pay. For carers, for whom age and sex were unknown, the cost of missing a day’s work was based on the median full-time gross weekly earnings by place of work, assuming that this was in the same local authority district as the hospital centre where the participant’s sciatica treatment was undertaken. 56 To supplement the RUQ data, an alternative scenario was conducted whereby missing patient questionnaires were supplemented, if feasible, with employment status as reported in the CRF. In this scenario, all patients with reported days off work but no costs were costed using ONS median wage by age and sex. Finally, a further SA accounted for implausible reported lost productivity cost values, by ignoring all participant-reported productivity losses costs and recosting based on number of reported days off work alone, using the median ONS median wage by age and sex as the cost basis for participants and median wage by postcode for carers.

Data completeness

Resource use and EQ-5D-5L questionnaires were not completed by all participants. Missing data were observed when whole questionnaires were not returned or when questionnaires were returned with some questions incomplete and were therefore non-evaluable. Completion rates were highest at baseline, treatment and week 18 visits, but high levels of missingness were observed for the 30- and 42-week postal questionnaires (Table 28).

Resource use and cost analysis

Observed participant use of health-care resources for all randomised patients (n = 163) and corresponding NHS HES admitted patient care and outpatient data [available only for patients who did not withdraw from the study (n = 157)] were comparable at baseline in both intervention groups for the 3 months prior to randomisation (Table 29). The main cost drivers for the pre-baseline period were related to admitted patient care (£353 for microdiscectomy vs. £422 for TFESI), outpatient visits (£208 for microdiscectomy vs. £251 for TFESI) and lost productivity (£809 for microdiscectomy vs. £614 for TFESI), accounting for 22%, 13% and 41% of all costs, respectively.

Tables 30 and 31 present the observed disaggregated health-care resource use and costs, respectively, over the 54-week trial period. HRG code HC64C was the most frequent intervention observed in the microdiscectomy group, with a reported occurrence of 66.2 per 100 participant-years, whereas HRG code AB20Z was the predominant TFESI intervention, with a rate of occurrence of 86.3 per 100 participant-years. The observation in the TFESI group of an occurrence rate for HRG code HC64C of 28.8 per 100 participant-years reflects the trial protocol, which allowed for participants to receive additional treatments as necessary. Mean observed total NHS costs were higher for microdiscectomy (£6683, 95% CI £5632 to £8074) than for TFESI (£4422, 95% CI £3682 to £5291; difference in mean £2261, 95% CI £706 to £3589), but no difference was observed in participant out-of-pockets costs for microdiscectomy (£878, 95% CI £538 to £1204) compared with TFESI (£1307, 95% CI £708 to £2092). The main cost difference between groups related to admitted patient care, with microdiscectomy costing £1926 (95% CI £467 to £3128) more than TFESI. The cost of outpatient attendances also differed between groups, with costs for microdiscectomy being £237 (95% CI £50 to £414) greater than TFESI. Total combined NHS and societal costs were higher for the microdiscectomy group (difference in mean £1832, 95% CI £53 to £3555).

Outcomes

Utility and quality-adjusted life-years

Observed participant responses by trial visit to each of the EQ-5D dimensions (i.e. anxiety and depression, mobility, pain and discomfort, self-care and usual activities) indicated less impairment (more lower scores) in the microdiscectomy group (Figures 10–14). Figure 15 presents the utility scores for the observed data using the three-level crosswalk valuation set. For participants with complete responses to the EQ-5D-5L at baseline and weeks 18 and 54, baseline mean utility values using the three-level value set were 0.328 (95% CI 0.259 to 0.392; n = 55) for the microdiscectomy group and 0.276 (95% CI 0.188 to 0.366; n = 48) for the TFESI group (Table 32), and were comparable between groups (difference in mean 0.052, 95% CI –0.060 to 0.157). Similarly, over the duration of the trial, comparable QALYs were observed in the complete-case groups (i.e. participants who completed the EQ-5D at baseline and weeks 18 and 54), with 0.654 QALYs in the microdiscectomy group and 0.591 QALYs in the TFESI group (difference in mean 0.062, 95% CI –0.033 to 0.155). The five-level value sets and VAS results exhibited a tendency to generate higher utility scores than the three-level value set, but showed minimal differences in QALY gains over the 54-week trial period, which were also not statistically different.

FIGURE 10.

Responses by EQ-5D dimension for depression/anxiety: observed data.

FIGURE 11.

Responses by EQ-5D dimension for mobility: observed data.

FIGURE 12.

Responses by EQ-5D dimension for pain/discomfort: observed data.

FIGURE 13.

Responses by EQ-5D dimension for self-care: observed data.

FIGURE 14.

Responses by EQ-5D dimension for usual activities: observed data.

FIGURE 15.

Utility scores by visit, using three-level crosswalk value set: observed data.

TABLE 32 - Health outcomes for participants with observations at baseline, week 18 and week 54, adjusted for visit time deviations

Analysis	Health outcome	Microdiscectomy (n = 55), mean (95% CI)	TFESI (n = 48), mean (95% CI)	Difference in mean (95% CI)
EQ-5D-3L value set	Baseline utility	0.328 (0.259 to 0.392)	0.276 (0.188 to 0.366)	0.052 (–0.060 to 0.157)
	54-week utility	0.718 (0.649 to 0.784)	0.659 (0.573 to 0.739)	0.059 (–0.051 to 0.165)
	QALYs over 54 weeks	0.654 (0.588 to 0.709)	0.591 (0.518 to 0.658)	0.062 (–0.033 to 0.155)
EQ-5D-5L value set	Baseline utility	0.443 (0.377 to 0.513)	0.409 (0.322 to 0.486)	0.034 (–0.071 to 0.137)
	54-week utility	0.794 (0.728 to 0.851)	0.737 (0.660 to 0.807)	0.057 (–0.041 to 0.157)
	QALYs over 54 weeks	0.746 (0.686 to 0.801)	0.683 (0.614 to 0.746)	0.063 (–0.024 to 0.148)
		Microdiscectomy (n = 60), mean (95% CI)	TFESI (n = 55), mean (95% CI)
EQ-VAS	Baseline utility	0.490 (0.428 to 0.550)	0.473 (0.414 to 0.529)	0.017 (–0.068 to 0.099)
	54-week utility	0.744 (0.682 to 0.801)	0.677 (0.606 to 0.741)	0.067 (–0.029 to 0.161)
	QALYs over 54 weeks	0.706 (0.653 to 0.752)	0.645 (0.593 to 0.692)	0.061 (–0.011 to 0.130)

Cost-effectiveness analysis results

Base-case analysis

The results following multiple imputation and statistical analysis, averaged over 10,000 bootstrap replicates and for the 54-week trial time horizon, reported mean total costs for the microdiscectomy group of £6919 (95% CI £5503 to £8046) and for the TFESI group of £4706 (95% CI £3821 to £5516). The mean total QALYs were 0.616 (95% CI 0.570 to 0.671) and 0.559 (95% CI 0.503 to 0.620) for the microdiscectomy and TFESI groups, respectively. The mean incremental costs and QALYs were £2212 (95% CI £629 to £3677) and 0.057 (95% CI –0.009 to 0.124), respectively, resulting in an ICER of £38,737 per QALY gained and, at a threshold of £20,000 per QALY, an INHB loss of 0.054 QALYs (Table 33). The cost-effectiveness plane (Figure 16) similarly indicated that the mean and the density of the joint distribution of incremental costs and QALYs was located in the north-east quadrant, indicating microdiscectomy to be more costly but with greater health benefits than TFESI. The CEAC (Figure 17) shows that the probability of microdiscectomy being cost-effective at £20,000 per QALY is 0.17 and the probability of microdiscectomy being cost-effective at a higher threshold of £30,000 per QALY is 0.37. These results are consistent with those produced in the deterministic analysis, which generated a mean incremental cost of £2240, a mean incremental QALY of 0.057 and an ICER of £39,334 per QALY gained (Table 34).

TABLE 33 - Cost-effectiveness at 54 weeks, QALYs, costs (£) and ICER, base-case bootstrapped analysis

Cost-effectiveness	Mean values (95% CI)			Cost-effectiveness threshold (95% CI)
	Microdiscectomy	TFESI	Incremental effect	£20,000/QALY	£30,000/QALY
Cost (£)	6919 (5503 to 8046)	4706 (3821 to 5516)	2212 (629 to 3677)
QALY	0.616 (0.570 to 0.671)	0.559 (0.503 to 0.620)	0.057 (–0.009 to 0.124)
ICER			38,737
INHB (QALYs)				–0.054 (–0.166 to 0.060)	–0.017 (–0.110 to 0.077)
Probability cost-effective				0.17	0.37

FIGURE 16.

Cost-effectiveness plane: base-case analysis.

FIGURE 17.

Cost-effectiveness acceptability curve: base-case analysis. Vertical dotted lines represent the NICE threshold range of £20,000–30,000 per QALY. 42

TABLE 34 - Scenario analysis, including deterministic results

Base-case analysis, non-parametric bootstrapped approach.

Scenario	Treatment allocation, total cost (£)		Treatment allocation, total QALYs		Incremental		ICER (£/QALY)	INHB
	Microdiscectomy	TFESI	Microdiscectomy	TFESI	Cost (£)	QALY		£20,000/QALY	£30,000/QALY
Base casea	6919	4706	0.616	0.559	2212	0.057	38,737	–0.054	–0.017
Base case, deterministic	6941	4701	0.617	0.560	2240	0.057	39,344	–0.055	–0.018
SA 1	8353	6856	0.611	0.554	1497	0.057	26,290	–0.018	0.007
SA 2	5816	3948	0.730	0.617	1868	0.113	16,512	0.020	0.051
SA 3	6913	4689	0.709	0.653	2224	0.056	39,392	–0.055	–0.018
SA 4	6925	4702	0.690	0.644	2222	0.046	48,113	–0.065	–0.028
SA 5	6901	4669	0.615	0.553	2232	0.062	35,717	–0.049	–0.012
SA 6	5920	4299	0.616	0.558	1620	0.057	28,251	–0.024	0.003
SA 7	6886	4693	0.610	0.557	2193	0.053	41,422	–0.057	–0.020
SA 8	6935	4690	0.621	0.559	2245	0.062	36,163	–0.050	–0.013
SA 9	7115	5824	0.608	0.551	1291	0.056	22,923	–0.008	0.013
SA 10	8566	6967	0.613	0.555	1600	0.057	27,981	–0.023	0.004
SA 11	8147	6363	0.613	0.559	1784	0.054	32,807	–0.035	–0.005
SA 12	10,194	8239	0.608	0.555	1956	0.053	36,621	–0.044	–0.012

Scenario analyses

For all scenarios that utilised imputed data sets to address the missing data (i.e. all except the complete-case analysis, SA 2), none resulted in an ICER < £20,000 per QALY gained (see Table 34).

In scenarios in which only cost inputs differ, mean QALYs are also observed to differ slightly, relative to the base case. This can occur because the cost variables, among others, are used in the multiple imputation procedure to predict missing values in utility scores and, therefore, different costs can affect the value of imputed utilities. In the complete-case analysis (i.e. SA 2) an ICER of £16,512 per QALY gained was estimated, but there were only 35 participants providing complete data and, therefore, owing to this small sample of participants, these results should be interpreted with caution.

When the approximating to societal perspective was adopted (i.e. SA 1), mean total costs increased to £8353 for microdiscectomy and £6856 for TFESI, but the mean incremental cost difference reduced to £1497. As there was no difference in incremental QALYs between this scenario and the base case, including productivity losses lowers the ICER to £26,290 per QALY gained. Alternative costing approaches for productivity losses (i.e. SAs 10–12) had an impact on the results and effectively reduced the cost-effectiveness of microdiscectomy by increasing the ICER to £27,981, £32,807 and £36,621 per QALY gained, respectively.

Scenario analyses 3 and 4 considered the impact of applying alternative utility scoring methods (i.e. the EQ-5D-5L tariffs and EQ-VAS). Using the five-level valuation set had minimal effect, but applying the EQ-VAS scores increased the ICER to £48,113 per QALY gained. The EQ-VAS, however, is not generally used to inform decision-making.

Taking an alternative approach to costing resource use and assuming that respondents completed the questionnaire ‘since last completed’ (i.e. SA 5) reduced the ICER to £35,717 per QALY gained, whereas considering only sciatica-related costs (i.e. SA 6) had a larger effect, reducing the ICER to £28,251 per QALY gained. The effect of combining only sciatica-related costs in a wider perspective (i.e. SA 9) resulted in a lower ICER of £22,923 per QALY gained. Varying the dose of ‘when-needed’ medications to 25% and 75% of the time (i.e. SAs 7 and 8) resulted in ICERs of £41,422 and £36,163 per QALY gained, respectively.

Design

To the best of our knowledge, this study is the first to directly compare two nationally recommended treatments for persistent sciatica secondary to PID. Previous studies comparing surgery with conservative care have included multiple treatment options, including physiotherapy, ESI and pharmacological treatments, and have not been a direct comparison of two treatments. Hitherto, there is little evidence supporting ESI for > 3–4 weeks after the onset of sciatica outside single cohort studies. The pragmatic nature of this study allowed local NHS treatment and probably contributed to successful recruitment and ‘buy-in’ for the trial by regional units. Eleven centres contributed to recruitment, and this improved the representation and validity of the results. Previously, the only comparable RCT comparing surgery with ESI had been a single-centre study14 of > 100 patients that utilised a posterior interlaminar injection route.

We wanted to exclude patients with a rapid positive natural history, and therefore the minimal length of symptoms was arbitrarily taken as 6 weeks, in keeping with the NICE back pain and radiculopathy guidelines recommending investigations after 6 weeks of symptoms. Initially, the authors wanted to minimise duration of radiculopathy up to 6 months in an attempt to coincide with statutory sick pay. However, this negatively affected recruitment rates because of long NHS waiting times, hence, this inclusion criterion was modified to allow recruitment of patients with up to 12 months of radicular symptoms. Post hoc analysis of interaction between duration of symptoms and treatment allocation indicated that duration of symptoms does not appear to be a treatment effect modifier.

The ODQ was chosen as the primary outcome measure because of its long history of clinical validation and in accordance with previous recommendations of core outcome sets. 36 Although this study was devised prior to a recent publication from Chiarotto et al. ,38 the core outcome sets recommended have been collected within the NERVES trial. For physical functioning, the ODQ or MRM is recommended, and we collected both. A numerical pain intensity score and health-related QoL score is also recommended, and we recorded VAS and EQ-5D-5L scores. Finally, recording mortality rates is recommended and, again, this was captured by our safety reporting and no participants died during the trial period. The COMI has the advantage of being easily tolerated by participants because of its small number of questions and so this was included in the NERVES trial. The COMI is gaining popularity across Europe and may allow wider comparison with large registry data sets of routinely collected outcomes for sciatica. This study found that there was no significant difference between microdiscectomy and TFESI for the primary outcome. Future studies may utilise a single physical function outcome domain, perhaps COMI, for participant co-operation.

The study suffered from missing data, and patients whose data were incomplete were excluded from the primary analysis. Participants with missing data were, however, incorporated into secondary analyses and sensitivity analyses of the primary outcome indicated robustness to the assumptions made about the missing data. The number of missing data may reflect the large numbers of questionnaires being presented to participants, but this was an accepted risk to capture as much clinical data as possible. Future studies could use fewer outcome measures, such as those recommended by Chiarotto et al. 38 (i.e. ODQ, VAS and EQ-5D).

Conduct

Overall, 163 participants consented to taking part in the study; however, 168 patients declined or were not asked, giving an overall consent rate of 49%. Some units declined to take part in the study, reflecting pre-existing clinical preference for one treatment over another, with some clinicians favouring surgery and others favouring TFESI. The difficulty of obtaining consent may reflect the stark difference between the two treatments. Of those participants who were asked to provide consent, the reasons given for not providing consent were related to treatment preference. The reasons provided were not wanting to be randomised (n = 60), not wanting surgery (n = 33 patients) and not wanting TFESI (n = 22 patients). This is typical of surgical trials, but may be considered slightly lower than expected, possibly because of the difference between the two treatments (i.e. one being a day-case procedure carried out under local anaesthetic and the other a surgical procedure carried out under general anaesthetic and necessitating an overnight or short hospitalisation). Given the current study findings, it would be difficult to justify ethically comparing the same two treatments in future studies because of the lack of clinical difference and the clear difference in safety profiles of the two.

Clinical effectiveness analysis

The trial found no evidence of a significant difference in ODQ scores between the two treatment groups at 18 weeks post randomisation. Furthermore, no difference in any of the other secondary outcome domains (i.e. COMI or MRM) was found.

Overall, 28 out of 80 patients (35%) in the TFESI group received microdiscectomy as an additional treatment and, therefore, TFESI was effective in avoiding microdiscectomy in 65% of participants randomised to TFESI. Of these patients, half received the additional treatment following the primary outcome assessment, thereby having only a minimal impact on interpretation/validity of the results. The authors feel that this level of additional treatment is an improvement compared with previous studies of microdiscectomy versus conservative care, in which the rate was > 40%. 11

One additional possibility regarding the interpretation of these data is that neither treatment was effective, given the lack of a third, control, group receiving no treatment. The TMG did not think that this was an ethics design option as all participants recruited had disabling sciatica (i.e. a baseline VAS score for leg pain of > 7). A recent paper by Bailey et al. 68 explored an almost identical cohort of participants, treating sciatica secondary to PID of between 4 and 12 months’ duration either conservatively or with surgery. The fact that we included only patients with symptoms of > 4 months’ duration was important, as previous studies have focused on patients with symptoms of typically < 3 months’ duration. 11,12,14,68 The majority of symptoms may well improve naturally in this time frame, but beyond 3 months symptoms are less likely to resolve quickly and it may take up to 12–24 months for natural improvement to manifest. The NERVES trial baseline data recorded an average duration of sciatica of > 4 months at recruitment and, therefore, a similarly matched group to the study by Bailey et al. 68 In the study by Bailey et al. ,68 the conservative treatment arm received medication and physiotherapy with the possibility of also receiving ESI, but it is not clear what proportion of participants received injections. Importantly, Bailey et al. 68 found a clear difference between surgery and conservative management, with surgery improving VAS scores for leg pain more effectively than conservative treatment. This is not the same result reported by our data. We speculate that a rigorously controlled injection group, as presented in this study, accounts for the difference in reported efficacy and therefore the benefit of TFESI is real. Furthermore, the precipitous drop in outcome scores post treatment seen in both groups does suggest a treatment effect. Finally, because the participants recruited had significant sciatica (i.e. a baseline VAS score for leg pain of > 7) with an average duration of > 4 months, we do not believe that spontaneous resolution of symptoms would have occurred as quickly as it did without treatment. For these reasons, we believe that both TFESI and surgery are effective treatment options for this group of patients.

Economic analysis

The economic analysis has strengths in that it utilised routine patient-level NHS data sets and nationally reported costs collected within a pragmatic RCT that is designed to reflect current management and NHS practice for the care of the trial population. Unlike previous economic analyses, the NERVES trial used patient-level data and, therefore, avoided the use of non-trial data and any subsequent bias that might be introduced as a consequence. Participant-reported health outcomes were measured at key time points using the EQ-5D-5L questionnaire, which offered greater sensitivity than the three-level version to assess the impact of treatment on each health domain. However, because of presently unresolved issues regarding the five-level value set for deriving health state utility scores, NICE currently recommends mapping from the five-level to a three-level value set, which was the approach applied in the base case. Alternative valuation sets, explored in a SA, demonstrated that the five-level valuation set had minimal effect on the ICER and would not affect decision-making. This is contrary to other reported findings comparing the five- and three-level value sets, which suggest that in most cases ICERs are substantially higher when the five-level value set is applied. 69 Deriving EQ-5D scores from the NERVES trial primary outcome measure, the ODQ, was not deemed feasible as no robust relationship exists between these measures. 70

There were also some weaknesses to the economic analysis. Given the pragmatic nature of the trial and reliance on participant-reported outcomes, it was inevitable that there would be a degree of missing data. In particular, completion rates were lower for the two data collections that were reliant on postal questionnaires (i.e. weeks 30 and 42). When possible, assumptions around resource use, costs and QoL were made to maximise the use of data acquired within the trial. When this was not possible, multiple imputation was used to impute missing values, conditional on observed data that were considered informative to the imputation. 71 Imputing missing data associated with health state utilities and resource use may act as an additional source of uncertainty. It is also acknowledged that the participant-completed RUQ may be subject to recall bias; however, patient questionnaires are an important supplement to clinical records and provide resource use and costing data that would otherwise not be readily available. The 3-month recall period used in the NERVES trial is representative of the median recall time frame employed in UK National Institute for Health Research (NIHR) HTA programme-funded trials. 72 A further limitation was to limit the time horizon of analysis to 54 weeks, consistent with the clinical trial follow-up period. No extrapolation or modelling was performed to include a longer-term assessment of costs and consequences. Given that recurrence rates of lower back pain after discectomy of 65% at 3 years have been reported,73 a longer analytical time horizon might be more informative, but liable to bias given the absence of longer-term follow-up of NERVES trial participants.

Our estimation of productivity losses was subject to both missing data and incomplete questionnaire reporting of time off work. To address the gaps in question completion, a number of assumptions were made and alternative scenarios were constructed, including replacing partially completed questions with missing costs based on ONS data on reported median salaries, as well as applying median salary data to account for questionable reported lost productivity costs. However, the wider perspective that approximated societal costs remains subject to bias, and limited in its robustness and generalisability, but nonetheless corresponds with findings from Koenig et al. 29 (i.e. the ICER is reduced when loss of productivity costs are taken into account).

Previous economic evaluations were found not to be generalisable to the present analysis because of differences in setting, perspective and interventions tested. Although, to the best of our knowledge, this is the first UK study to compare microdiscectomy with TFESI, Price et al. ,18 in a comparison of ESI with placebo, reported an ICER of £44,701 per QALY gained from a provider perspective, increasing to £354,171 per QALY gained from a purchaser perspective. However, it is not feasible to draw comparisons between this analysis and the present NERVES trial for a number of reasons. The WEST,18 conducted between 1999 and 2002, compared blind ESI (see Chapter 1, Scientific background), whereas the NERVES trial has used the transforaminal route, which delivers the injectate much nearer to the site of pathology (i.e. the prolapsed disc and involved nerve root).

In terms of costing, Price et al. 18 utilised a bottom-up approach based on unit costs from a single NHS trust. This differs from the approach undertaken in the NERVES trial, which follows the NICE reference case,42 which recommends a HRG costing approach and utilises NHS reference costs. 45 These costs represent national average unit costs from all NHS organisations in England. Moreover, Price et al. 18 considered only the direct medical costs that were associated with the epidural technique and its follow-up, whereas the greatest cost component – resource use relating to clinical staff time – was calculated from non-RCT-based survey data. Regarding health outcomes, Price et al. 18 derived utility scores from the Short Form questionnaire-36 items, which differs from the NICE-preferred EQ-5D, and QALYs gained were based on only completed 12-week data, despite follow-up extending to 52-weeks. Small incremental differences (< 0.006) in QALYs were observed between ESI and placebo over the 12 weeks, which have the effect of making the ICER highly unstable, as reflected in the wide variation in ICERs reported. In contrast to Price et al. 18 who use a placebo control, the NERVES trial was unable to estimate a value for cost-effectiveness of TFESI per se, as we did not feel ethically justified in including a ‘no-treatment’ group in this study. From our inclusion criteria, all participants had previously exhausted simple conservative care and required further treatment.