Preventing recurrence of endometriosis-related pain by means of long-acting progestogen therapy: the PRE-EMPT RCT

Survey of practice

With the evidence base unable to guide the specifics of study design for a trial, we turned to a survey of national practice in December 2011. Members of the British Society for Gynaecological Endoscopy were sent an online questionnaire asking:

• whether they prescribed postoperative hormonal treatments

• their most commonly used hormonal treatment and

• the most relevant comparison for any future trial.

Some 62 members responded, with 56 of them having experience of treating endometriosis. Of these members, 45 (80%) indicated that they prescribed hormonal treatments and 11 (20%) did not. GnRH analogues, LNG-IUS, COCP and DMPA were the most commonly used treatments but with none obviously preferred over the others (39, 38, 37 and 25 responses, respectively). Three comparisons of interest ranked higher than the others (40 responses): LNG-IUS versus no treatment (18, 45%); LNG-IUS versus COCP (17, 43%); and LNG-IUS versus DMPA (12, 30%), but again with no particular favourite. 45

Trial design considerations

With no clear LARC or comparator favoured, we needed to consider how two viable LARCs, the LNG-IUS and DMPA, and two non-LARC options, COCP and no treatment could be accommodated into a substantive trial. A four-arm trial was the obvious choice, but concerns were raised about whether patient views would prohibit recruitment to a design requiring consent to four different interventions – it was likely a large proportion of this population would have tried one or more of these treatments before as they are recommended by the National Institute for Health and Care Excellence (NICE) for the initial (presurgical) management of endometriosis. 46 Pragmatic designs,47 where the patient or clinician could select their choice of LARC or non-LARC, were considered, but concerns were raised about ultimately unsatisfying and underpowered comparisons. The non-LARC group was particularly problematic, being a mixture of active (COCP) and non-active (no treatment) options. Given these difficult design issues, we decided to use feasibility of recruitment to a particular randomisation scheme to guide the type of design which should be taken forward.

Internal pilot design

Our proposal was to include a flexible-entry design approach where participants could be randomised to two, three or four treatments provided one was a LARC and the other was a non-LARC. This meant that a patient could enter any one of nine randomisation schemes, shown in Figure 1. On completion of the pilot phase, a decision about the substantive study design based on feasibility of recruitment would be made. As no outcome assessment was proposed, inflation of type I error was not a concern. 48 The options for a definitive trial included continuing with a four-way randomisation design if acceptable to women, or alternatively to drop one or two treatment groups if randomisation proved difficult. The design would be fixed, together with an appropriate sample size target to ensure that we would have enough power to detect a minimally important difference. 49 Data collected from participants randomised in the pilot phase to designs that remain in the main phase would be taken forward and combined with subsequent data collected; however, all women would be followed up to study conclusion regardless of the randomisation options selected.

FIGURE 1.

Design of the internal pilot.

To ensure an appropriate design choice was made, a post-pilot phase report was scheduled to be prepared for a joint Trial Steering Committee (TSC) and Data Monitoring Committee (DMC) meeting to review at the end of the pilot phase. The pilot phase was intended to last for 1 year with a recruitment target of 100 participants. The TSC and DMC would have the final say on any proposed changes, which would also be communicated to the funding body for approval. The post-pilot report was to include data on which randomisation schemes had been selected by patients and the results of a qualitative assessment (see Qualitative assessment). A threshold of 10% of eligible participants was set for each of the four arms as a criterion for inclusion on the definitive trial. Any updated review of external evidence published since the grant application was submitted would also be taken into consideration. Apart from informing the design changes, the pilot phase also aimed to fine tune operational procedures, assess data capture forms and confirm initial assumptions around sample size.

Internal pilot general trial methods

The internal pilot phase followed the same methods detailed in Chapter 3 in terms of inclusion and exclusion criteria, recruitment, randomisation methods and intervention-specific procedures.

Qualitative assessment

A qualitative approach to the study was considered appropriate, as the aim of the study was to gain insight into women’s experiences and motivations and also to explore issues of importance to them rather than to adhere strictly to a script. Narratives are an important way for people to explain disruptive events in their lives, and the use of narrative interviews and a focus group in this study allowed women to reflect on living with endometriosis and to raise the issues that had greatest impact on their lives. 50 Three sites, Aberdeen Royal Infirmary, Birmingham Women’s Hospital and Edinburgh Royal Infirmary, all in the UK, took part, which comprised a focus group discussion and individual interviews (Appendix 1, Table 22). The focus group and semistructured interview topic guides were informed by available literature on women’s experiences of medical treatments for endometriosis symptoms, as well as the expertise of the PRE-EMPT Trial Management Group (TMG). As a means of establishing rapport, at the beginning of interviews and the focus group, some demographic data were collected. Favourable ethical approval for this study was obtained from the North of Scotland Research Ethics Committee and site-specific permission from the NHS trusts of each of the hospitals involved.

One focus group and 10 individual semistructured interviews were conducted to elicit women’s past experiences with the proposed treatments and to assess whether they constituted a barrier to participation. The focus group discussion took place in one of the centres and included four women. Three women were interviewed in their homes and seven were interviewed over the telephone. The focus group and interviews were recorded and transcribed verbatim. Content analysis51 was employed, with a qualitative lead and two assistants independently reading the transcripts and agreeing upon common themes. Dissident views were also considered. The Standards for Reporting Qualitative Research (SRQR) were adopted. 52 Women participating in both the focus group and individual interviews shared their views and experiences of medical treatments and their motivation for enrolling in the PRE-EMPT trial.

Recruitment to the pilot phase

Six centres in the UK were involved with staggered starts from April 2014 to the end of March 2015 (recruiting on average for 10.5 months). During this period, 504 patients were assessed for eligibility and 77 were recruited. The most common reasons for ineligibility included a plan to conceive in the near future (42 patients, 10%); contraindications to one or more treatments (35 patients, 8%) and no endometriosis identified at diagnostic laparoscopy (33 patients, 8%). The main reason for not wishing to take part was a preference for a particular treatment [94 patients, (22%); the most common being LNG-IUS (30) and DMPA (30), with COCP less favoured (6)]. Details of randomised participants are given in Table 2.

Randomisation options chosen in the pilot phase

Only 5 of the 77 participants (6%) were willing to be randomised to all four treatment options (Figure 2 and Table 3). Participants willing to be randomised to both LNG-IUS and DMPA were relatively low, with the vast majority (82%, 63/77) expressing a preference for one or the other in roughly even proportions (43% for LNG-IUS and 57% for DMPA). In a similar fashion, most (71%, 55/77) expressed a preference for their choice of comparator in even proportions (51% for COCP and 49% for no treatment). Forty-six of the participants (60%) expressed a preference for both a LARC and their comparator and hence opted for variations of two-way randomisations.

FIGURE 2.

Randomisation options chosen in the internal pilot phase – summary.

Findings of the qualitative study

Women who agreed to participate represented a range of symptomology, treatment histories and allocated trial treatment groups (Appendix 1, Table 23). As no novel treatment was on offer, many women had previous experience of treatments available as part of the trial (either as endometriosis treatment or for contraceptive purposes), which strongly influenced their acceptability.

Women found flexible randomisation acceptable as they had an element of choice over which treatment groups to which to be randomised. Half of participants (n = 7) reported that without the option to opt out of a particular treatment group (or groups) they would have declined trial participation. No single treatment group was found more or less acceptable to women. Women made decisions regarding which treatment arms were acceptable based on their past negative or positive treatment experiences and the treatment experiences of significant others (female friends and family). Women chose to participate in the trial for reasons of altruism and self-interest and found the 3-year length of their participation acceptable.

New external evidence

In the period between the initial study proposal and completion of the pilot, two systematic reviews examining the use of COCP in this population were published. 53,54 The first identified 15 randomised trials including 850 patients. The combined odds of recurrence were noted to be lower in the COCP group compared with surgery alone (OR 0.31, 95% CI 0.22 to 0.45; p < 0.001). The second evaluated the use of prolonged (at least 2 years) postoperative COCP and endometrioma recurrence in a total of 965 women (726 in cohort studies and 239 in one RCT). Recurrence was lower with COCP compared with no treatment (OR 0.12, 95% CI 0.05 to 0.29; p < 0.001). The data from these two systematic reviews showed that COCP was beneficial in preventing recurrence of endometriosis and was instrumental in changing equipoise among clinicians as well as challenging the ethical justification for including a no-treatment arm in the definitive trial.

Proposal

Four-way and three-way randomisation designs were ruled out due to low numbers selecting these randomisation options. We also ruled out a trial design involving solely the most commonly selected two-way randomisation option (DMPA vs. COCP) as this attracted only 14 participants (18% of all participants). Given the strong preferences noted (including in the results of the qualitative work), we decided to incorporate some elements of choice in the revised design (Figure 3). The main comparison proposed was LARC, considered as a class of treatments, versus COCP, with LARC selected before randomisation by the patient if a preference was apparent (or alternatively allocated randomly if there was no opinion). The choice of LARC would need to be decided prior to randomisation to enable unbiased stratified (subgroup) analyses of LNG-IUS versus COCP and DMPA versus COCP (e.g. only those selecting LNG-IUS pre randomisation would be included in the LNG-IUS vs. COCP comparison). COCP was chosen as the comparator over no treatment on the basis of the new external evidence. Participants randomised to combinations not taken forward through to the main phase (e.g. DMPA vs. no treatment) would still be followed up as per the main LARC versus COCP comparison.

FIGURE 3.

Revised trial design for the substantive phase.

Sample size considerations

The sample size in the definitive trial was revised to reflect a main two-arm comparison and took into consideration a revised estimate of the standard deviation (SD) of the primary outcome (from pooled baseline data). To detect an 8-point difference on the Endometriosis Health Profile-30 (EHP-30) pain domain with 90% power (p = 0.05) and assuming the SD is 22 points requires 160 participants per group, 320 in total (to account for any loss to follow-up – estimated 20% – this target was inflated to 400). Eight points is equivalent to 0.36 SD, which can be considered half-way between a small (0.2 SD) to moderate (0.5 SD) effect size. 55 This size of sample would have also given us good power (80%) to detect a 10-point difference in the two stratified analyses of LNG-IUS versus COCP and DMPA versus COCP provided the remaining recruits into the study have a roughly even split.

Health economic considerations

In terms of the health economic evaluation, we conducted a pretrial model-based economic evaluation (6 months at the beginning of the trial coinciding with the internal pilot). A decision analytic model based on the alternative treatment pathways outlined in the trial design was constructed and populated from a pragmatic review of the available evidence on resource use, associated costs, effectiveness of interventions and the health-related quality of life (HRQoL) for the resulting health states. This collated evidence was used to estimate a baseline decision model, which allowed identification of important elements of resource use, costs and issues and gaps relating to either the interventions or quality of life and the level of detail to be obtained from primary trial data.

Primary objective

To compare, in women undergoing conservative surgery for pain due to endometriosis, the effectiveness of LARCs compared with COCP in preventing the recurrence of endometriosis-related pain and improving quality of life.

Secondary clinical objectives

• To compare LARCs versus COCP as per the primary objective in those that selected LNG-IUS as their method of delivery.

• To compare LARCs versus COCP as per the primary objective in those that selected DMPA as their method of delivery.

• To compare LARCs versus COCP in terms of pain relief, serious adverse effects and repeat surgery.

Depot medroxyprogesterone acetate injection

Depot medroxyprogesterone acetate injection (only available as Depo-Provera®, Pfizer, Walton Oaks, Surrey) is a long-acting reversible progestogen contraceptive that is administered at a dose of 150 mg in an aqueous suspension by intramuscular injection every 3 months.

Levonorgestrel-releasing intrauterine system

The LNG-IUS is a contraceptive system that slowly releases a daily dose of 20 μg levonorgestrel into the uterine cavity. Bayer Pharma AG market their LNG-IUS under the name of Mirena® and Gedeon Richter plc under the name of Levosert®: either was permitted. They are long-acting reversible preparations that require removal and reinsertion every 5 years for Mirena and 3 years for Levosert.

Comparator – combined oral contraceptive pill

Participants allocated the COCP were prescribed a formulation containing 30 μg ethinylestradiol and 150 μg levonorgestrel. For the management of endometriosis-related pain, it is unclear whether combined oral contraceptives should be taken conventionally, continuously or in tricycle regimen, so their intended regimen was according to the clinician or participant’s decision and recorded at randomisation.

Initiation and repeat prescription for trial treatment

The intention was to initiate the allocated treatment as soon as possible, ideally before discharge, to minimise non-compliance and for the convenience of the participant. The fitting of the LNG-IUS was ideally done by the treating gynaecologist during the conservative surgery, or before discharge. Similarly, the first DMPA injection was ideally given before discharge after surgery. If this was not possible, the participant was provided with a prescription and asked to attend their general practitioner (GP) practice or a sexual health clinic for fitting or injection, ideally within a month. Repeat prescription of the COCP and repeat DMPA injections were undertaken by the participants’ GP or sexual health clinic.

Timing of assessments

Women who agreed to enter the study completed a baseline participant booklet before randomisation, consisting of disease specific and generic quality of life questionnaires, pain scores and resource use questions (see Primary outcome measures and Secondary outcome measures sections for full details of the outcome measures used). Participants were then followed up for a period of 3 years with a similar questionnaire booklet, initially solicited by post, with consent given for additional methods of contact; telephone and e-mail. Over this period, the booklets were also sought at 6 months, then 1 and 2 years post randomisation. Women who did not return the questionnaire after two postal reminders, at 2- and 3-year follow-up, were contacted by a member of the clinical team, with the aim of completing a shorter questionnaire (containing the most pertinent information: primary outcome, generic quality of life required for the economic evaluation, treatment changes, relevant surgical interventions and pregnancy status) over the telephone. This was in response to lower-than-anticipated postal returns. Questionnaires were considered to have been completed on time if they were completed within 6 months for assessments prior to 3 years or by 4 years post randomisation for the final 3-year assessment.

Information on the results of the surgical procedure was taken from the surgical notes prior to discharge from hospital; pregnancies and repeat procedures were monitored throughout the period of follow-up by the clinical team at the trial centre. If there was a failure to obtain patient-reported returns pertaining to further hospital-based treatment (outpatient visit, pregnancy or further surgery, this was obtained directly from hospital records from the principal investigator of the participant’s recruiting centre.

The end of trial was determined to be when the final participant recruited completed their 3-year assessment. Owing to the long-term nature of some of the interventions, participants may remain on treatment beyond the end of the trial and will be cared for by the GP as they would outside of the trial.

Primary outcome measures

The primary outcome was the recurrence of symptoms as evaluated by the pain domain of the EHP-30 questionnaire at 3 years post randomisation.

The EHP-30 is a disease-specific questionnaire to measure the health status of women with endometriosis. It demonstrates good reliability, validity, acceptability and responsiveness,58 with low floor and ceiling effects for the core questions. 59 There are 30 core items with five scales and six modular parts of 23 questions, which are dependent on the woman’s circumstances (e.g. impact on work, sexual activity and fertility). The pain domain has possible scores: 0 (best outcome) – 100 pain score (worst score).

Traditionally, recurrence of endometriosis was diagnosed objectively by laparoscopy. Here, we have recognised the need to avoid unnecessary invasive surgery while acknowledging the impact of symptoms of pain on women’s lives. Objective evidence to determine prevention of recurrence was deemed unethical, as it would expose the participants to the risks of a repeat surgical procedure and general anaesthetic, when the most important outcome was maintained reduction of pain and improvement in quality of life.

Secondary outcome measures

Secondary clinical outcome measures were as follows:

• The pain domain of the EHP-30 at the other assessment points.

• The remaining four core domains of the EHP-30 questionnaire:

  • Control and powerlessness (0 = best outcome, 100 = worst outcome).

  • Emotional well-being (0 = best outcome, 100 = worst outcome).

  • Social support (0 = best outcome, 100 = worst outcome).

  • Self-image (0 = best outcome, 100 = worst outcome).

• The six modular domains of the EHP-30 questionnaire:

  • Work (0 = best outcome, 100 = worst outcome).

  • Relationship with family (0 = best outcome, 100 = worst outcome).

  • Sexual relationship (0 = best outcome, 100 = worst outcome).

  • Feelings about medical profession (0 = best outcome, 100 = worst outcome).

  • Feelings about treatment (0 = best outcome, 100 = worst outcome).

  • Feelings about infertility (0 = best outcome, 100 = worst outcome).

• Pelvic pain measured by VAS [three scales: pelvic pain during periods (0 = best outcome, 10 = worst outcome), pelvic pain during intercourse (0 = best outcome, 10 = worst outcome), pelvic pain at any times (0 = best outcome, 10 = worst outcome)].

• Responses to the question ‘compared to 1 month ago, would you say your pelvic pain has “Got much better”, “Got a little better”, “Not changed much”, “Got worse”’.

• Fatigue, as measured by Fatigue Severity Scale (FSS)60 score (7 = best outcome, 63 = worst outcome), which is the sum of the responses of the nine statements contained in the questionnaire.

• Menstrual regularity. Patients will be asked whether they are still having periods and, if so, to rate how regular their cycle is in one of the four categories; ‘Regular’, ‘Fairly regular’, ‘Irregular’ and ‘I have bleeding on and off all the time’.

• Generic quality of life (EQ-5D-5L):61,62

  • EQ-5D index score (patient completed; −0.59 = worst outcome, 1.0 = best outcome).

  • EQ-5D health thermometer (patient completed; 0 = worst outcome, 100 = best outcome).

• Capabilities, as a measure of wellbeing (ICECAP-A: patient completed; 0 = worst outcome, 1.0 = best outcome). 61,63 A score will be calculated from the five ICECAP attributes (attachment, stability, achievement, enjoyment, autonomy).

• Further therapeutic surgery or second-line treatment for endometriosis as a proxy for recurrence or ‘treatment failure’, defined as having undergone hysterectomy, ‘surgery for endometriosis’, laparoscopy or taking GnRHa treatment.

• Discontinuation rates of randomised treatment (time to first treatment change – see Analysis section for further details), with reasons for change.

• Serious adverse events (SAEs).

Sample size

The rationale for the revised sample of 400 is provided in Chapter 2; in brief, we planned to detect an 8-point difference on the EHP-30 pain domain in our main comparison (LARC vs. COCP) with 90% power (p = 0.05).

Statistical analysis

A comprehensive statistical analysis plan was reviewed by independent DMC and TSCs prior to any analysis. Full details of the statistical analysis can be found in the statistical analysis plan, which can be requested from bctudatashare@adf.bham.ac.uk.

Categorical baseline data were summarised with frequencies and percentages. Normally distributed continuous variables were summarised with means with SDs. Participants were analysed in the treatment group to which they were randomised (intention to treat), irrespective of adherence with the treatment protocol. All participants recruited under versions 2.0–7.0 of the protocol (from 23 October 2015) were included in the final analysis population, as these participants were randomised directly to ‘LARC’ or ‘COCP’ (with pre-randomisation choice of LNG-IUS or DMPA if allocated LARC). For those randomised under v1.0 of the protocol (pilot phase, participants were included in the final analysis population provided they were randomised to combinations of treatments that only included LARCs and COCP (see Chapter 2). Participants randomised to combinations that did not involve COCP (LNG-IUS vs. no treatment; DMPA vs. no treatment; LNG-IUS vs. DMPA vs. no treatment) were not included in the final analysis population (a summary of the primary outcome results comparing LARC vs. no treatment is provided in Chapter 4).

For the primary outcome (EHP-30 pain scores at 3 years), a mixed-effects linear regression model for repeated measures64 was used to calculate an adjusted difference between group means, along with 95% CIs. Parameters for participant, treatment group, time and time by treatment interaction were included as well as baseline response (as a continuous variable) and the minimisation variables (see Randomisation). Time was assumed to be a categorical variable and to allow for a varying treatment effect over time, a time-by-treatment interaction parameter was also included. Centre was included as a random intercept in the model, and all other factors as fixed effects. A compound symmetry covariance structure was assumed. An F-test was used to test statistical significance (p-value produced) of the estimated treatment group parameter generated from the maximum likelihood estimate.

Secondary outcomes measured on a continuous scale (remaining four core domains and the six modular domains of the EHP-30, VAS, FSS, EQ-5D index score, EQ-5D Health thermometer and ICECAP-A) were analysed in a similar manner as the primary outcome. Cycle regularity was analysed using a generalised estimating equation model with logit link that took into account all assessment times (correlated longitudinal data) and adjusting for the minimisation parameters. An independent covariance structure was assumed and ORs with 95% CIs for the treatment group parameter were produced. Change in pelvic pain was also analysed using a similar generalised estimating equation model, this time using cumulative logit link for ordered categorical data. Responses to ‘Are you still having periods?’ were tabulated. Time to further therapeutic surgery or second-line treatment for endometriosis (defined as having undergone hysterectomy, ‘surgery for endometriosis’, laparoscopy or taking GnRHa treatment) was analysed using a Cox regression model. Adjusted hazard ratios (and 95% CIs) were generated and a Kaplan–Meier plot produced. Centre was regarded as a fixed effect in the model. Discontinuation rates of randomised treatment (time to first treatment change) were summarised by group and presented using Kaplan–Meier plots but not formally analysed; reasons for treatment changes were tabulated. The number and percentage of participants experiencing any SAEs and were presented by intervention group. Statistical significance was determined by a chi-squared test. All estimates of differences between groups were presented with two-sided, 95% CIs.

Sensitivity analysis was performed on the primary outcome to investigate the assumption that missing data were missing at random. This incorporated a delta-based approach, which assumes missing data is missing not at random. 65 Missing data were imputed using multiple imputation with chained equations; 50 imputations were generated and all variables that were included in the analysis of this outcome were included to ensure compatibility of approach. The average increase of observed data at each time was calculated; a value ‘delta’, which is equivalent to a proportion of this average increase, was then subtracted from the imputed value in all of the imputed sets (i.e. missing data responses were assumed to be worse than those returned). The delta values were taken in turn as the following in separate investigations: 20% of average increase in both groups at each time point, 20% of average increase in the LARC group at each time point; 10% in the COCP treatment group and 10% of average increase in the LARC group at each time point; 20% in the COCP treatment group. For each of these investigations, analysis was then performed as per the original approach on the imputed and manipulated sets, with results combined using Rubin’s rules. Further sensitivity analysis was also performed removing a small number of late returned responses.

Preplanned subgroup analyses on the primary outcome were completed on the pre-randomisation selection of LARC (LNG-IUS or DMPA). This initially included participants who had their LARC randomly allocated (as they were happy to use either LARC), but this was repeated excluding these participants and then further excluding those where the pre-randomisation selection was specifically chosen by the participant (as opposed to the clinician). In addition, subgroup analyses were carried out on the minimisation variables: stage and extent of endometriosis and age. The effects of these subgroups were examined by adding the subgroup by treatment group interaction parameters to the linear model described above; statistical significance of these interaction parameters was determined by F-tests/Wald tests. Differences between treatment groups within subgroups and 95% CIs were generated.

Interim analyses of effectiveness and safety endpoints were performed on behalf of the DMC on an approximately annual basis during the period of recruitment. These analyses were performed with the use of the Haybittle–Peto principle,66,67 hence, no adjustment was made in the final p-values to determine significance.

Adherence

Of the 205 women randomised to LARC, slightly more were offered treatment with DMPA compared with LNG-IUS [114 (56%) vs. 91 (44%)]. Approximately four-fifths [81% (254/313); see Table 4] of these treatment options were driven by patient preference. Of those offered LNG-IUS, 85% (77/91) had the system fitted at the time of their laparoscopic treatment to endometriosis and, of those offered DMPA, 51% (58/114) had the first dose administered before discharge (Table 26). Of the 200 women that were allocated to COCP, 52% (103/200) had this prescribed while in hospital (Appendix 2, Table 27). Approximately 65% of participants allocated LARC were still using a LARC treatment (either LNG-IUS or DMPA) at 1 year, reducing to 37% by 3 years (Figure 5). The equivalent figures in the COCP group were lower, at 53% and 25%. Self-reported reasons for treatment change included perceived lack of effectiveness and commonly known adverse effects of these treatments. Over half the treatment changes in both groups involved cessation of any treatment; some of these were due to a desire for pregnancy (see Secondary outcomes – further therapeutic surgery or second-line treatment for endometriosis) or had undergone further surgery for endometriosis or hysterectomy (Appendix 2, Tables 28 and 29). Switching from one LARC treatment to another (i.e. from LNG-IUS to DMPA or vice versa) or supplementation of (related) non-trial drug was also a relatively common occurrence. Adherence to the allocated intervention (without any treatment change at all) occurred in 56% and 48% of participants at 1 year and 26% and 24% at 3 years, in the LARC and COCP groups, respectively (Figure 6; details in Appendix 2, Tables 28 and 29).

FIGURE 5.

Time to first treatment change (no longer on assigned treatment – changes from one LARC to another, e.g. LNG-IUS to DMPA, are not classified as a change; see Appendix 2, Tables 28 and 29 for full details of changes).

FIGURE 6.

Time to first treatment change (any treatment change – includes any relevant treatment change, i.e. includes the addition of a trial or related non-trial treatment; see Appendix 2, Tables 28 and 29 for full details of changes).

Background

We incorporated a pretrial model into the design of the economic analysis for this project, in accordance with best practice. A model-based analysis has the advantage of collating data from a number of diverse sources to enable an economic evaluation in the absence or presence of robust data. Typically, the objective of a pretrial model-based analysis is to identify and anticipate areas of ambiguity in the decision recommendation that could arise in the full trial analysis and thus identify key areas of focus with respect to data collection in order to minimise anticipated uncertainty caused by inadequate data. In general, if the data available at this early stage are robust, but new intervention options become available, the model can provide some indication of the practice recommended while awaiting the trial results. If we deem the available data non-robust, the pretrial analysis can support the case for a trial.

Of course, ambiguity in the decision recommendation can still exist even when the most robust data are collected. This typically occurs when the options compared have strengths and weaknesses with respect to different outcomes to which available instruments for capturing outcomes may have some limited sensitivity.

Aims and objectives

We carried out a preliminary (pretrial) economic evaluation based on best currently available evidence that existed prior to the trial being undertaken. We compared alternative treatments LNG-IUS, DMPA, COCP and ‘no treatment’ to prevent recurrence of endometriosis after conservative surgery in primary care.

The RCT was designed to collect data to explore the most effective and cost-effective treatment for preventing endometriosis following surgery. Given that the RCT was not due to report for at least 4 years, the pretrial economic analysis was carried out to support immediate decisions, explore the likely decision uncertainty and particularly to inform areas of particular focus for data collection in the full RCT to aid the subsequent full trial-based economic evaluation.

Methods of data collection and analysis

The evaluation took the form of a cost–utility analysis, based on an outcome of cost per QALY. We used a UK NHS perspective in a primary care setting. We developed a state transition (Markov) model in Microsoft Excel® (Microsoft Corporation, Redmond, WA, USA). The model structure was based on the NIHR-funded (PRE-EMPT study) RCT design, which was informed by a review of existing evidence and clinical input. We refined the model structure through iterative discussion with expert clinical colleagues.

We carried out a pragmatic literature search to identify evidence on the effectiveness of the treatment strategies to inform parameter inputs and their distributions for the transition probabilities of the decision model. We used estimates for resource use from the literature and some were proxy estimated for the anticipated resource use for the trial. In the analyses, we compared the four treatment strategies. A probabilistic sensitivity analysis, which simultaneously changes all relevant parameters in the model and is repeated 1000 times, was carried out. The analysis provides an assessment of the difference in costs and outcomes in terms of QALYs between the potential interventions over a 36-month time horizon.

Limitations

The limitations of this model were that there were no reliable data available to populate it and give confidence that the model-based analysis could help determine treatment recommendation. We did not test the structural assumptions in a sensitivity analysis, as this analysis would not have improved the precision of our findings. However, these limitations do not undermine the purpose of the pretrial analysis to inform data collection. We assigned wide distributions to test the extent to which uncertainty in the parameters altered the results and whether a trial is required. We used some illustrative data given that the impact of the treatment on this particular condition is subjective, wide distributions around values seemed most suitable.

Key findings

The main findings from the pretrial analysis are that there is considerable uncertainty about the cost-effectiveness of the existing treatments to prevent endometriosis following conservative surgery. There is little difference between the probability of all existing treatment strategies (DMPA, COCP, LNG-IUS) being the most cost-effective, and none can yet be singled out as a potential clear contender. In terms of the direction of results, COCP had the greatest probability of being cost-effective when compared with DMPA and LNG-IUS. DMPA had the highest probability of being cost-effective when compared with LNG-IUS.

Interrelation with other parts of the project

We carried out the model-based analysis for the sole purpose of aiding the design and focus of the full trial-based analysis. We report the trial-based economic analysis in full in next section.

Introduction

This section reports the main economic evaluation conducted alongside the PRE-EMPT trial. The objective was to compare the relative cost-effectiveness of LARC with the COCP in preventing recurrence of endometriosis in women undergoing conservative surgery.

Methods

The primary evaluation was a within-trial cost–utility analysis, with the results expressed in terms of additional cost per QALY gained over 36 months. The base-case analysis was performed from the UK NHS perspective as per recommended practice,69 with an additional sensitivity analysis for the partial societal perspective incorporating productivity costs. Subgroup analyses were also conducted comparing LNG-IUS with COCP and DMPA injection with COCP. Secondary analyses took the form of cost-effectiveness analysis, with secondary endpoints of cost per year’s full capability, cost per EHP-30 pain score reduction and cost per treatment failure avoided, respectively. All statistical analyses were performed using Stata^® version 17 (StataCorp LP, College Station, TX, USA).

Measurement and valuation of outcomes

All the outcome measures were collected at baseline, 6, 12, 24 and 36 months after randomisation and the results were presented as means and SD. Differences between the groups were calculated using bootstrapped mean differences. The measurement and valuation of outcome measures is described below.

Instruments and outcomes

Resource use and costs

From the UK NHS perspective, only the direct cost that the health service provider had incurred within the time horizon of the trial were included in analysis. Healthcare resource use data were collected alongside the trial at baseline, 6, 12, 24, 36 months after randomisation using the PRE-EMPT follow-up questionnaires. Information was obtained for medications including type of hormonal treatment and painkiller used, hospital and primary consultations, investigation procedures (laparoscopy, hysteroscopy, ultrasound scan) and further surgical management (surgery for endometriosis, hysterectomy). For the societal perspective analysis, the questionnaires also captured indirect non-medical costs, referring to income or productivity loss. This included paid and unpaid time off work due to endometriosis. This was valued by using the human capital approach, where the time lost (measured in days) because of endometriosis symptoms was multiplied by the average gross wage estimates. 73 This approach was considered appropriate, given that the majority of work absences were of relatively shorter duration.

Assumptions related to resource use

A number of assumptions were required in the measurement of the healthcare resource use and costs within the trial:

• All healthcare visits, including those to a GP or hospital, were counted, although it is possible that some of them might not have been related to endometriosis symptoms.

• When a participant reported having ‘surgery for endometriosis’ or ‘hysterectomy’ and specified the use of ‘laparoscopy’ as the same time, this was counted as a single procedure (i.e. surgery for endometriosis with laparoscopy or laparoscopic hysterectomy, respectively).

• If a participant switched treatment, it was assumed that the switch occurred midway between follow-up points for costing purposes.

Valuation of resource use

Relevant unit costs were identified from the established national sources, including the NHS Reference Costs 2020–1,74 the Personal Social Services Research Unit costs,75 and the British National Formulary. 76 Unit costs related to productivity loss due to time-off work were obtained from the Office for National Statistics in 2022. The relevant unit costs were then multiplied by resource use data to calculate the total treatment costs. Table 11 presents the relevant items of resource use, their associated unit costs and the source from which these costs were obtained. All costs were reported in 2021–2 Great British pounds. Costs were inflated where necessary, using the Hospital and Community Health Services Pay and Prices Index. 75

Missing data

Multiple imputation techniques were used to handle missing costs and missing EQ-5D-5L, ICECAP-A and EHP-30 pain domain data at each follow-up time point. 67,77 Costs were imputed at the total cost level for each cost category. Resource use and therefore cost data were considered missing if participants did not complete and return their follow-up questionnaire.

A multiple imputation with chained equation technique was performed, together with predictive mean matching method to the closest neighbour across 20 imputations to account for the non-normality of the distribution of costs and the outcome values for missing total costs and missing outcomes;78 and Rubin’s rule was used to combine the imputed data sets into one final imputed variable. 77 Subsequently, Rubin’s rules were applied to pool the estimates obtained from the multiple imputed datasets. Rubin’s rules are a statistical method used to appropriately pool parameter estimates and their variances from multiple imputed datasets, enhancing the accuracy and reliability of the overall statistical analysis. This helps to account for the uncertainty introduced by the imputation process and provides a more accurate estimate of the missing values. 67,77 The model for imputing missing EQ-5D-5L, ICECAP-A and EHP-30 pain domain data included the trial arm and score baseline. The imputed data were used to inform the base-case and sensitivity analyses unless specified otherwise.

Some assumptions were also made to deal with some missing data in the trial,

1. For partially completed questionnaires:

   • At any particular time point, if participants had not reported a change to their hormonal treatment, they were assumed to still be using the last one they previously reported.

   • Women who did not state that they visited their GP were assumed to have not done so.

   • In the absence of any specific mention of surgery or investigations, it was assumed that participants did not undergo those procedures.

2. If participants reported a healthcare visit (such as a GP visit or gynaecologist follow-up after surgery) without specifying the number of visits, a conservative assumption was made that they had one visit.

3. If participants did not report any healthcare visits, it was assumed that they had one additional visit to GP if they changed treatment, stopped using LNG-IUS (due to removal) or received DMPA (since a GP visit is required every 3 months for injection).

Economic evaluation

The first step was to undertake a cost–consequence analysis. This describes all the important disaggregated results relating to resource use, costs, and outcomes. 79 The analyses were performed according to the intention-to-treat principle. The primary base-case economic analysis took the form of a cost–utility analysis from the perspective of the UK NHS. This method assesses the gains in QALY relative to cost of different interventions. The secondary economic evaluation was the cost-effectiveness analysis, where the health consequences were measured in a natural unit as YFCs, pain score reduction from baseline to 36 months and treatment failure avoided. An incremental cost–utility and effectiveness analyses were undertaken to calculate the incremental cost-effectiveness ratio (ICER) as the cost per outcome. This is the ratio of the mean difference in the cost and mean difference in outcome between the two trial arms.

Both cost and QALY were discounted by 3.5% per annum as recommended by NICE. 69 Because cost and QALY data were skewed, all estimates were presented as means with bootstrapped 95% CIs, each with 5000 replications.

Probabilistic sensitivity analysis

To represent the overall uncertainty in the trial cost and outcome data, a probabilistic sensitivity analysis was undertaken for the base-case analysis only, by jointly bootstrapping mean cost and outcome differences to generate 5000 paired ICER estimates. The 5000 paired bootstrap estimate pairs of the mean costs against mean outcomes (paired differences) provided a graphical display of a cost-effectiveness plane. 80 The plane is divided into four quadrants, each representing a different cost-effectiveness scenario. The north-east quadrant represented situations where the intervention is both more effective and more costly than the comparator, indicating that the intervention is cost-effective if the decision-maker is willing to pay more for the additional benefit. The south-east quadrant represented situations where the intervention is both more effective and cheaper than the comparator, indicating that the intervention is dominant. This meant that the intervention was the preferred option, as it provided better outcomes at a lower cost compared with the alternative. The south-west quadrant represented situations where the intervention is less effective and less costly than the comparator while the north-west quadrant represented situations where the intervention is less effective and more costly than the comparator, indicating that the intervention is dominated by comparator. This meant that the intervention is deemed to be not cost-effective and should not be chosen unless there are other compelling and important factors that justify the additional costs. 55,79

Uncertainty was also estimated by constructing cost-effectiveness acceptability curves (CEACs). 81 For the primary economic analysis, the CEAC shows the probability for COCP and/or LARC being cost-effective at different cost-per-QALY thresholds. In the UK, interventions were deemed cost-effective if the cost per QALY gained is ≤ £20,000. 69

Deterministic sensitivity analysis and subgroup analysis

To assess the robustness of the base-case results, additional deterministic sensitivity analyses were also carried out.

1. Complete-case analysis – the analysis was rerun using only observations with complete cost and outcome data.

2. Undiscounted analysis – this analysis presented the undiscounted costs and outcomes.

3. Partial societal perspective analysis – this analysis assessed the impact of including work-related costs of patients.

4. Additional analysis incorporating costs of other types of surgery mentioned by participants (removal of fibroids, removal of polyps and endometrial ablation).

5. Subgroup analysis – this analysis assessed the cost-effectiveness of the COCP with each of the LARC’s subgroups:

   • COCP versus LNG-IUS

   • COCP versus DMPA.

Participants

In the total of 405 participants included in the trial, 200 were randomised to COCP. In the trial 205 women were randomised to LARC, of which 91 were either allocated based on their preference or randomised to LNG-IUS and 114 to DMPA. Follow-up rate at 36 months was 86% of all groups.

Primary economic analysis

Resource use

Average resource use per trial participant is presented in Table 14. There were few variations in mean resource use between groups. On average, participants in the LARC group had fewer surgical procedures (and therefore fewer follow-up episodes); fewer instances of GnRHa and analgesic use and fewer ultrasound scans than the COCP group. The LARC group reported more primary care (GP) visits (which may reflect the fact that DMPA needs to be injected every 3 months) and an increased number of diagnostic procedures (laparoscopy and hysteroscopy) compared with the COCP arm. In terms of productivity loss, the LARC group reported fewer days off work due to endometriosis.

Costs

In order to derive the health service costs accruing from each intervention in Table 15, the mean resource use was then combined with the unit costs (Table 11). Average costs of healthcare service use differed between the two groups, given that difference in mean resource use. The most significant cost difference between the two groups was in the cost of further/extra surgeries. Specifically, the COCP group had to undergo more surgeries for endometriosis and hysterectomy for which the associated additional costs compared with the LARC group was approximately £328 and £177, respectively. Furthermore, the required follow-up visits after these additional surgeries cost an another additional £160 compared with the LARC group. In terms of the non-health service cost, LARC was associated with lower cost of productivity loss in terms of cost of days-off paid-work due to endometriosis. While the non-health-service costs were not incorporated into the base-case results, they were included in the sensitivity analysis to assess the CE from a broader perspective.

Mean total costs

Mean total costs for each group are presented in Table 16. Cost of healthcare visit, further surgical procedures were almost £133 and £505 less per woman, respectively, in LARC group, and almost £34 and £71 higher in terms of medication and investigation cost, respectively. Overall, the intervention of LARC was less costly.

Cost–utility analysis

Table 17 provides a summary of the cost–utility analysis using data from the base-case UK NHS perspective. In the base-case analysis, the COCP group was estimated to be £533 (95% CI 52 to 983) per woman more costly but offered slightly higher QALYs by 0.043 (95% CI −0.069 to 0.152) compared with the LARC group, over the 36 months of follow-up. This resulted in an ICER of almost £12,280 per QALY, which is within the £20,000 threshold recommended by NICE. 69 The cost-effectiveness plane in Figure 10 shows the majority of the points are in the north-east quadrant, indicating that COCP was costlier, but more effective than LARC. The CEAC shows the probability that the both interventions are cost-effective at different levels of willingness-to-pay for a QALY. At the £20,000 and £30,000 per QALY, the probability that the COCP intervention is cost-effective was 61% and 66%, respectively. This means that the probability of COCP being cost-effective is higher than the LARC group at these thresholds.

FIGURE 10.

Cost–utility analysis from the UK NHS perspective. (a) Cost-effectiveness plane showing 5000 bootstrapped replicates of the ICER. (b) cost-effectiveness acceptability curve for COCP compared with LARC.

Sensitivity analyses

Details of deterministic sensitivity analyses and subgroup analyses are presented in Table 18. In complete case-analysis scenarios, the COCP intervention was shown to be less effective and more costly in terms of QALYs gained. However, it is important to note that the complete case analysis was only based on 212 cases (n = 113 in the LARC arm and n = 99 in the COCP arm), and the missing QALY values in the COCP arm tended to be higher than those in the LARC arm. This explains why the mean QALYs for the COCP arm in the complete case analysis were lower than those of the LARC arm. In other deterministic sensitivity analysis (undiscounted analysis, partial societal perspective analysis, additional analysis incorporating the cost of other type of surgery), the use of COCP was consistently more costly but also more effective in terms of QALYs gained compared with LARC. These findings were also consistent with the subgroup analysis comparing the COCP group with the LARC subgroups. Specifically, when compared with the DMPA subgroup, the COCP group incurred an additional cost of £851 (95% CI 391.075 to 1269.53) and yielded 0.052 (95% CI −0.091 to 0.171) more QALYs. Similarly, when compared with the LNG-IUS subgroup, the COCP group was associated with an additional cost of almost £135 (95% CI −708.614 to 769.042) and yielded 0.032 (95% CI −0.090 to 0.164) more QALYs.

Secondary outcomes

Table 19 displays the response rates for the analysis based on secondary outcomes, which included the ICECAP-A and EHP-30 pain domain questionnaires, at each follow-up time point. However, missing data were significant for the secondary outcomes. Of the total number of participants, complete economic data from baseline to month 36 were available for 147 (36.23%) and 212 (52.34%) participants for the ICECAP-A and EHP-30 pain domain questionnaires, respectively.

The scores of each outcome at each follow-up time point are presented in Table 20. Participants in LARC group had a slightly higher ICECAP-A and EHP-30 pain domain score compared with the COCP arm both at the baseline and at month 36. This resulted in mean adjusted years of full capabilities difference for imputed data of 0.0034 (95% CI −0.0562 to 0.0525) and 0.0254 (95% CI −0.116 to 0.0648) for the complete case, both favouring the LARC group. The reduction in pain score between baseline and month 36 of these two groups was also in favour of LARC, where the mean difference of LARC and COCP arm for the imputed case was 0.145 (−4.509 to 4.182) and 0.969 (95% CI −6.276 to 3.635) for the complete case. Furthermore, fewer treatment failure was observed in LARC group than the COCP group (16.59% and 20.5%, respectively). This resulted in mean difference of 0.039 (95% CI −0.121 to 0.034).

Cost-effectiveness analysis

The cost-effectiveness analysis of secondary outcomes, such as YFC, EHP-30 pain score reductions from baseline to 36 months and treatment failure avoided, is presented in Table 21. The results suggested that the COCP intervention was dominated by the LARC intervention (i.e. the LARC intervention was both cheaper and more effective). Probabilistic sensitivity analyses were carried out but are not reported in terms of the cost-effectiveness plane and CEAC in this report. This is because the analysis based on the secondary outcomes produced findings that were less reliable due to the higher number of missing data observed in secondary outcomes. For instance, the ICECAP-A questionnaire, had ˂ 50% completeness at 36 months.

Principal findings

Our results suggest that post-surgical use of COCP is more costly, with an average cost per participant of £2470 per woman compared with £1937 for the LARC. The difference in costs (£533, 95% CI £52 to £983) was mainly attributable to the cost of further surgery to treat endometriosis or hysterectomy. The COCP resulted in a small increment in QALYs of 0.04 (95% CI −0.069 to 0.152) over 36 months. The estimated ICER for COCP compared with LARC is £12,280 per QALY, which is within the acceptable threshold of £20,000 per QALY recommended by NICE. 69 Thus, treatment with COCP is likely to be considered the most cost-effective treatment according to standard thresholds.

The results of almost all the (cost–utility) sensitivity analyses bar one, and the subgroup analyses, supported these base case results. They consistently suggest that COCP is more expensive but more effective in terms of QALYs gained compared with the LARC. The only sensitivity analysis carried out that did not support the base case findings was the complete case scenario, which indicated that the LARC was cheaper and more effective compared with COCP in terms of QALYs gained, but this analysis is challenged by the much smaller sample size and the missing data.

It is noteworthy that the QALY improvement with the COCP compared with the LARC for the primary base case analysis was very small at 0.04. In contrast, LARC dominated COCP for all of the secondary outcomes, which included YFC; EHP-30 pain domain score reduction and treatment failure avoided. For all these secondary outcome measures, the results suggest that the LARC group had better outcomes compared with the COCP group, but all are challenged in terms of a robust interpretation because of the extent of missing data.

Comparison with other studies

This study is unique in its focus on the cost-effectiveness of COCP versus LARC for preventing the recurrence of endometriosis in women who have undergone surgery. To the best of our knowledge, no previous study has directly compared these two options. While prior research has examined the cost-effectiveness of different treatment strategies for endometriosis, none has specifically evaluated LARC and COCP in this manner. 41,42

Strengths and limitations of the study

A strength of this study is that it is the first economic evaluation conducted alongside a RCT comparing the relative cost-effectiveness of COCP and LARC to prevent the recurrence of endometriosis in women who have undergone surgery for endometriosis.

In addition, the analysis is that the economic evaluation is based on the outcome of QALYs using the EQ-5D-5L, which had a relatively high degree of completeness of 83% at 36 months. A range of sensitivity analyses were also conducted to test the robustness of our findings and to explore the impact of different assumptions and inputs on the cost-effectiveness results.

A potential limitation is the duration of follow up of 3 years, given that endometriosis is a chronic condition that can recur until menopause82 and the average age of women in the study was 29 years. Also, healthcare resource use information was based on self-reported data, which can potentially be affected by under-reporting. 83

A further potential weakness is that the QALY may not capture all the outcomes that are important to women and could further inform their treatment pathway. Although we undertook analyses based on other secondary outcomes, such as pain score reductions using the EHP-30 questionnaire (which is specific to endometriosis), YFC gained from the ICECAP-A questionnaire and the treatment failure avoided that requires further and more invasive interventions; the high rate of missing data, for these other outcomes, only served to undermine attempts at robust interpretation.

Finally, the high rates of discontinuation and treatment switching among participants posed challenges when attributing outcomes to specific treatments. Nevertheless, an intention-to-treat analysis was employed, sensitivity analyses were carried out and data collection at various time points to address these challenges. This is the most pragmatic approach that can be adopted and provides insight into real-world clinical scenarios, reflecting patient preferences and responses that influence treatment decisions.

Implications for policy

This trial-based economic evaluation suggests that COCP has a 61% probability of being cost-effective compared with the LARC. Confidence in such a probability is likely to be subjective and other factors can impact this. For instance, we found that COCP is associated with a higher risk of further surgery compared with LARCs. In terms of outcomes, both LARC and COCP offer similar benefits in terms of QALYs, with only marginal differences between the two. Therefore, we suggest that both options should be discussed with women. For some, LARCs may become the preferred option, informed by past experiences, acceptability of their invasiveness balanced by lower risks of future surgery.

Recommendations for future research

The complexity of treating endometriosis and the various factors such as patient adherence, variability in clinical practice and comorbidities that can influence outcomes make it challenging to provide a single recommendation with respect to the economic impact of LARC and COCP.

To improve the understanding of the cost-effectiveness of COCP, further exploration of the potential for treatment failure is necessary. However, crucially the emphasis should be placed on improving patient outcomes and alleviating the overall burden of endometriosis on both patients and healthcare systems. Therefore, it is essential to continue researching and identifying opportunities for more effective and cost-effective treatments.

Clinical interpretation

Our results show that the two LARCs (LNG–IUS and DMPA), as well as the combined oral pill, are similar in terms of prevention of recurrence of pain symptoms at each follow-up interval up to 3 years following endometriosis surgery, irrespective of stage of disease. Both are associated with sustained and significant improvements in pain scores of around 40% compared with presurgical values. Although the primary outcome was no different, women randomised to LARCs underwent fewer repeat surgical procedures (particularly laparoscopic treatment of recurrent endometriosis and hysterectomy) and treatment with GnRH analogues – a fact that reduced the overall costs of treatment in this group. At the same time, COCPs had a marginal advantage in terms of generating QALYs, suggesting that there may be some room for flexibility in decision-making according to women’s choice. The fact that pregnancies occurred in participants in both randomised groups reflects the complex nature of endometriosis, where, alongside symptom relief, fertility is a desired outcome for many women.

After 3 years, 37% of those in the LARC arm were still using their allocated treatment, while 25% of those initially randomised to COCP continued on this treatment. Reasons for this attrition included desire for pregnancy, further surgery for endometriosis or hysterectomy and perceived ineffectiveness or adverse effects prompting a change in medication.

The outcomes of the trial, both clinical and economic, are supportive of instituting a policy of preventing recurrence of endometriosis symptoms after conservative surgery using either LARC or COCP.

Strengths and limitations

Major strengths of this trial include its focus on patient-centred outcomes, a follow-up period of 3 years and the availability of outcome data on over 80% of participants. The pragmatic nature of the trial is more likely to enhance the generalisability of our findings although the predominance of white women in the recruited sample limits our ability to be confident about how our results might apply to women from other ethnic backgrounds. The ability to corroborate self-reported episodes of further treatment through direct interrogation of hospital records was an added strength.

The 3-year follow-up period and the pragmatic design meant that relatively few women continued with their allocated medication and changed to other treatments or stopped altogether, depending on their clinical circumstances. Our strategy of using telephone calls to collect to obtain primary outcome data from those who did not return questionnaires at 3 years, meant that follow-up data on secondary outcomes were available on fewer women.

The prolonged time horizon of this trial, which started recruitment in 2014, means that some aspects of clinical practice have changed, and newer treatments such as modern oral progestogens (such as dienogest) and GnRH antagonist tablets containing add-back HRT were not available for evaluation.

The commissioning brief requested that hormonal treatments were compared, which would be expected to target oestrogen and progestogen-driven mechanisms of pain and suppress recurrence from residual endometriosis lesions. As a pragmatic trial, we did not restrict the use of any analgesics such as non-steroidal anti-inflammatory drugs or neuromodulator drugs, and did not record their use. Investigation of coexistent pain syndromes (e.g. irritable bowel syndrome or bladder pain) was not mandated by the protocol and could confound the results, although should be balanced by randomisation.

The broad nature of the commissioning brief prompted us to opt for an initial four-arm pilot in the first instance, before considerations of equipoise, choice and emerging evidence shaped our design for the definitive study; that is, a two-arm trial comparing two types of LARCs with the combined oral pill, with preplanned subgroup analyses to investigate the mode of delivery of LARC (LNG-IUS or DMPA). In retrospect, this plan seems justified in the light of subsequent clinical guidance recommending both treatments for preventing recurrence of endometriosis following conservative surgery. At the same time, the lack of a no-treatment arm has made it impossible to provide any conclusive data on the sustained impact of surgery alone while treating the two LARC preparations as a single group in any comparisons with COCP has limited the power needed for meaningful comparisons between LNG-IUS and DMPA and COCP.

The open-label nature of the hormonal treatments can cause biased responses to our participant-reported outcome measures. We explored the possibility of dummy interventions but could not ethically proceed with masking the allocation. Any bias could be considered to impact both randomised groups equally. We also dismissed a repeat laparoscopy as an objective outcome as it would involve unnecessary surgical risk and would be unlikely to be acceptable to eligible women.

While all patients were recruited prior to the COVID-19 pandemic, as the follow-up period was for 3 years, the number of women who required further surgery may be underestimated given the negative impact of COVID-19 on elective surgery throughout the UK. It is possible that this may also have led to an increase in the use of GnRHa treatment if recourse to surgical treatment was affected.

Patient and public involvement

Input from patients and the public was crucial in shaping the design of the pilot, as well as the main trial, and in the choice of the primary and secondary outcomes. The fatigue scale was added at the request of patient and public involvement (PPI) feedback. As co-applicant, our lead PPI representative was instrumental in providing a patient-centred perspective to all discussions and decisions on recruitment, follow-up and the use of language within documents aimed at participants. The appreciation that flexibility of LARC choice was critical to recruitment and that certain types of dummy interventions, like an inert intrauterine device, would be unacceptable to women informed our decision to opt for a pragmatic approach to the trial. The transition from the pilot phase to the main trial also involved several decisions which involved close interaction with PPI partners.

Patient and public involvement colleagues also influenced our recruitment and follow-up strategies, especially the decision to opt for telephone follow-up for participants after two unsuccessful attempts to contact them by mail. Finally, input from PPI colleagues has been invaluable in interpreting trial results. As we enter the dissemination phase of this project, we continue to work closely with PPI groups, including Endometriosis UK, and to ensure that we use several complementary routes of communication to engage with patients from all backgrounds and ensure that the key messages from this trial are available to all those with endometriosis, their families and all those who care for them.

Equality, diversity and inclusion

Although we worked closely with our PPI partners in initiating and designing this trial, our hospital-based recruitment policy did not include any specific measures targeted at hard-to-reach populations or racialised groups. This remains a weakness of the trial, along with the fact that we did not include patient information leaflets in languages other than English and relied on normal translation facilities available within the NHS.

Generalisability

As a multicentre trial recruiting from hospitals across the UK, the results of this trial are generalisable across the NHS. This is helped by its pragmatic approach, flexible design to accommodate preferences and the use of medications which are familiar to women and clinicians. These preparations have been used by millions of women worldwide as contraceptives and have an excellent safety record. The trial is unable to comment on drugs which have come into clinical use more recently like dienogest or oral GnRH antagonists containing add-back HRT.

Comparison with the literature

Our findings are consistent with the findings of a recent systematic review and meta-analysis,84 which is supportive of hormonal treatment in preventing post-surgical recurrence of endometriosis in women for whom fertility is not a priority. The literature has progressed significantly since a Cochrane review in 2004,40 which was unable to find any conclusive evidence of benefit of postoperative hormonal suppression compared to surgery alone, based on short duration of follow-up of around 3 months. The role of COCP preparations for post-surgical prevention of recurrence of endometriosis has since been extensively studied85 and COCP has been recommended as the first-line treatments in clinical practice in recent guidelines. Long-acting progestogens like DMPA or LNG-IUS have also been considered to be viable options for women keen to avoid recurrence of endometriosis, while more recent oral progestogen preparations are used more widely. One such is dienogest, a fourth-generation nortestosterone-derived oral progestogen, with anti-oestrogenic, antiproliferative, anti-inflammatory and anti-angiogenic effects, which make it particularly suited to treating endometriosis. 86

The clinical trials register indicates that a phase 3 trial evaluating the use dienogest for treatment of endometriosis is actively recruiting at the time of writing (NCT04256200).

Gonadotropin-releasing hormone analogues, which were used as a second-line medical treatment for endometriosis, have found greater acceptance following regimens involving the addition of add-back HRT. Relugolix, an oral GnRH antagonist that includes add-back HRT, is licensed for treatment of fibroids and offers another possibility for endometriosis treatment. A recently published large, randomised trial has confirmed its effectiveness and acceptability for managing pain from endometriosis. 87 Two other GnRH antagonists with add-back HRT are in phase 3 clinical trials for treatment of endometriosis according to the clinical trials.gov register: linzagolix (NCT03992846) has completed recruitment and elagolix (NCT04333576) is actively recruiting at the time of writing.

The efficacy of hormonal treatments in prevention of endometrioma recurrence has been assessed by two meta-analyses. The first54 demonstrated the superiority of long-term (> 12 months) use of either cyclic or continuous COCPs compared with no treatment. The second88 pooled evidence from three RCTs and one cohort study and showed a lower recurrence rate for dysmenorrhoea following continuous COCP use compared with cyclic regimens. Although these systematic reviews showed a possible benefit of COCPs on recurrence, evidence was based on small sample sizes and the focus was specifically on endometriomas rather recurrence of pain and other hormonal regimens such as dienogest, LNG-IUS and GnRHa were not considered. A more recent systematic review and network meta-analysis that explored the efficacy of hormonal regimens preventing the recurrence of endometrioma ranked LNG-IUS highest, followed by dienogest and GnRHa with LNG-IUS but also found that long-term COCP superior to expectant treatment.

The most recent systematic review84 of post-surgical hormonal suppression for endometriosis, focused on recurrence, either based on imaging or recurrence of symptoms. Follow-up data (median duration 18 months) from 2137 women (13 randomised trials and 4 cohort studies) show a significantly reduced risk of post-surgical recurrence in patients receiving any type of hormonal suppression compared to no treatment or placebo (relative risk 0.41, 95% CI 0.26 to 0.65). This difference was also apparent in subgroup analyses of women treated with COCP or LNG-IUS as well as sensitivity analyses limited to randomised trials.

Implications for health care

Our results support current guidance which recommends hormonal treatment in women who have undergone surgery for endometriosis. While the effects of surgery alone have not been demonstrated, a policy of prescribing either LARCs or the COCP after surgery results in reduced pain scores, maintained over a 3-year period. LARCs resulted in lower rates of further surgery, in particular, surgery for recurrence of endometriosis and hysterectomy equating to lower costs on average, although COCP achieved marginally higher rates of QALYs.