PROMISE: first-trimester progesterone therapy in women with a history of unexplained recurrent miscarriages - a randomised, double-blind, placebo-controlled, international multicentre trial and economic evaluation

Progesterone in pregnancy

Progesterone is essential to achieve and maintain a healthy pregnancy. It is an endogenous hormone, secreted naturally by the corpus luteum (the remnants of the ovarian follicle that enclosed a developing ovum) during the second half of the menstrual cycle, and by the corpus luteum and placenta during early pregnancy. Progesterone prepares the tissue lining of the womb (endometrium) to allow implantation, and stimulates glands in the endometrium to secrete nutrients for the early embryo. During the first 8 weeks of pregnancy, progesterone is produced by the corpus luteum, but between 8 and 12 weeks the placenta takes over this role and maintains the pregnancy thereafter.

The importance of progesterone in pregnancy has prompted many clinicians to infer that progesterone deficiency may be aetiologically linked to recurrent miscarriage (RM), and that progesterone therapy in the first trimester of pregnancy may reduce the risk of miscarriage. In 2003 a Royal College of Obstetricians and Gynaecologists (RCOG) guideline1 and a Cochrane review2 called for a definitive trial to test whether or not progesterone therapy in the first trimester could reduce the risk of miscarriage in women with a history of unexplained RM.

Burden of disease

Miscarriage is the commonest complication of pregnancy: one in six clinically recognised pregnancies ends in a miscarriage. 3 RM, the loss of three or more pregnancies, is a distinct clinical entity. The prevalence of RM (1%) is significantly higher than that expected by chance alone (0.4%). Even after comprehensive investigations, a cause for RM is identified in fewer than 50% of cases. 3 The majority of couples are, therefore, labelled as having unexplained RM.

Recurrent miscarriages affect over 6000 couples in the UK every year, and frequently result in substantial adverse physical and psychological consequences for women as well as impacting their families. For example, miscarriage has the potential to cause physical harm, including severe haemorrhage, infection, perforation of the womb during surgery for miscarriage and, rarely, maternal death. The eighth triennial Confidential Enquiry into Maternal Deaths identified several women who had died from complications related to miscarriage. 4 Moreover, qualitative studies have shown the level of distress and the bereavement reaction associated with miscarriage to be equivalent to the impact of the stillbirth of a term baby. 3

Costs to the NHS

It is estimated that RM costs the NHS £28M per year. This value includes the costs of diagnosis (blood tests and ultrasonography), management of miscarriages (expectant, medical or surgical), investigations for causes of miscarriages (e.g. antiphospholipid syndrome, parental karyotype and uterine cavity tests) and hospital inpatient costs. However, it does not include the management of the complications following treatment of miscarriages (such as uterine perforation, infection, bleeding or visceral damage) or any long-term health consequences of miscarriages or miscarriage management (including complications of intrauterine infections and adhesions). Thus, the true NHS perspective costs are likely to be higher than the estimated £28M per year. The societal costs, including days lost from work and out-of-pocket expenses for patients and partners, can be expected to be far greater.

Progesterone in clinical use for recurrent miscarriages

The PROMISE study was conceived to address the possibility that progesterone therapy in the first trimester of pregnancy may reduce the risk of RM. In 2007 we conducted a clinician survey in the UK (n = 114; response rate 102/114; 89.5%), and found that 2% (2/102) of clinicians use progesterone routinely and 3% (3/102) use it selectively in pregnant women with a history of RM. Over 95% (97/102) reported that they do not use progesterone for this indication and the vast majority of these (92/102; 90.2%) were willing to recruit to a trial evaluating the role of progesterone treatment for the prevention of RM.

We also carried out separate systematic reviews to examine (a) the effectiveness of progesterone in RM5 and (b) the safety of progesterone in pregnancy.

Effectiveness of progesterone in recurrent miscarriages

Two previously conducted systematic reviews2,6 had examined the role of progesterone therapy in RM, but, since these reviews were published and at the time of designing the PROMISE trial, new evidence7 had emerged. Therefore, we conducted a fresh systematic review with a meta-analysis.

We searched the Cochrane Controlled Trials Register, Cochrane Database of Systematic Reviews, Database of Abstracts of Reviews of Effects, ISI Web of Science Proceedings, International Standard Randomised Controlled Trial Number Register, metaRegister of Controlled Trials database, MEDLINE and EMBASE resources, from database inception to January 2008, for the following search terms: (‘progesterone’ OR ‘progestagen’ OR ‘progestogen’ OR ‘progestin’ OR ‘progestational [hormone or agent] ‘ OR ‘progest$’). We considered outcomes of miscarriage (as defined by the primary authors), live birth, gestation at delivery, pregnancy and neonatal outcomes.

Four randomised trials7–10 were identified. There were 14 trials assessing the effects of progesterone in miscarriages including spontaneous (one-time) events2 but these trials should not be confused with trials assessing the effects of progesterone in RM. The quality of the four trials was poor (modified Jadad quality scores between 0/5 and 2/5; Table 1), and participant numbers were small even when the trials were combined in meta-analysis, with only 132 women actively treated with progestogens.

TABLE 1 - Characteristics of the four pre-existing trials of progestogen use in RM

ITT, intention to treat.

Features	El-Zibdeh 20057 (n = 130)	Goldzieher 19648 (n = 18)	Levine 19649 (n = 30)	Swyer and Daley 195310 (n = 47)
Population	Unexplained RM (three consecutive miscarriages, and conditions such as antiphospholipid syndrome excluded)	Analysis restricted to those with a history of three or more miscarriages	History of three consecutive miscarriages	Analysis restricted to those with a history of three or more miscarriages
Intervention	Dydrogesterone 10 mg twice daily (oral)	Medroxyprogesterone 10 mg daily (oral)	Hydroxyprogesterone caproate 500 mg per week (intramuscular)	Progesterone pellets 6 × 25 mg inserted into gluteal muscle
Comparison	No treatment	Placebo	Placebo	No treatment
Duration of treatment	From diagnosis of pregnancy to 12 weeks	Unclear	Until miscarriage or 36 weeks	Unclear
Randomisation method	‘Randomised’; method not given	‘Sequentially numbered bottles’	Alternation	Alternation
Allocation concealment	Unclear	Unclear	Unclear	Inadequate
Blinding	No	Double	Double	No
ITT analysis	Yes	Unreported	No	Unreported
Follow-up rates	100%	100%	54% (26/56 excluded)	100%
Jadad score	0/5	2/5	0/5	1/5

Our review and a subsequent review conducted for Cochrane in 201311 found that all four trials showed a trend towards benefit of progesterone, but confidence intervals (CIs) were wide and differences were not statistically significant for all but one of the four trials. A meta-analysis showed a statistically significant reduction in miscarriages (odds ratio 0.39, 95% CI 0.21 to 0.72; Figure 1). There was no evidence of statistical heterogeneity in the results (heterogeneity p-value 0.98).

FIGURE 1.

Meta-analysis of the four pre-existing trials of progestogen use in RM (outcome: miscarriage). df, degrees of freedom.

Although this evidence would be graded level 1a in the evidence hierarchy (because it is a systematic review of randomised trials), our survey of clinicians showed that it did not result in the use of progesterone for RM by clinical practitioners, owing to the weak methods and small sample sizes employed in the four published trials. One example of weak methodology was in lack of concealment, which has been shown to exaggerate effect sizes by up to 41%,12 although there is some evidence that this exaggeration may not be a concern when the outcome is objective. 13 Small sample sizes increase the likelihood of random error (generating the wide CIs in Figure 1). Nonetheless, the existing evidence presented a powerful reason to proceed with a trial of progesterone in RM, especially in consideration of the size of the effect observed and the low cost, widespread availability and convenience of the intervention, in addition to its safety profile.

Primary objective

• To test the hypothesis that, in women with unexplained RM, progesterone (400-mg vaginal capsules, twice daily), started as soon as possible after a positive urinary pregnancy test (and no later than 6 weeks of gestation) and continued to 12 weeks of gestation, compared with placebo, would increase live births beyond 24 completed weeks of pregnancy by at least 10%.

Secondary objectives

• To test the hypothesis that progesterone would improve various pregnancy and neonatal outcomes (such as reduced miscarriage rates and improvements in survival at 28 days of neonatal life).

• To test the hypothesis that progesterone, compared with placebo, would not incur serious adverse events (SAEs) in either the mother or the neonate (such as genital abnormalities in the neonate).

• To explore differential or subgroup effects of progesterone in various prognostic subgroups, including subgroups of:

  • maternal age (≤ 35 years or > 35 years)

  • number of previous miscarriages (3 or ≥ 4)

  • presence or absence of polycystic ovaries.

• To perform an economic evaluation for cost-effectiveness.

Inclusion criteria

In order to be eligible for the study, it was necessary for participants to meet all of the following criteria:

• having a diagnosis of unexplained RM (three or more consecutive or non-consecutive first-trimester losses)

• being aged 18–39 years at randomisation (the likelihood of miscarriages due to random chromosomal aberrations is higher in older women3,25 and such miscarriages are unlikely to be prevented by progesterone therapy)

• trying to conceive naturally

• willing and able to give informed consent.

Exclusion criteria

Participants could not be included in the study if any of the following criteria were applicable:

• they were unable to conceive naturally (as confirmed by urinary pregnancy tests) within 1 year of recruitment or before the end of the randomisation period in the trial, whichever came earlier

• they had antiphospholipid syndrome [lupus anticoagulant and/or anticardiolipin antibodies (immunoglobulin G or immunoglobulin M)]; other recognised thrombophilic conditions (testing according to usual clinic practice)

• they had uterine cavity abnormalities (as assessed by ultrasound, hysterosonography, hysterosalpingogram or hysteroscopy)

• they had abnormal parental karyotype

• they had other identifiable causes of RM (tests initiated only if clinically indicated) such as diabetes, thyroid disease or systemic lupus erythematosus

• they were on current heparin therapy

• they had any contraindications to progesterone use (see the following section).

Contraindications to progesterone use

Women to whom any of the following applied were not eligible to take part in the trial:

• they had a history of liver tumours

• they had severe liver impairment

• they had genital tract or breast cancer

• they had severe arterial disease

• they had undiagnosed vaginal bleeding

• they had acute porphyria

• they had a history during pregnancy of:

  • idiopathic jaundice

  • severe pruritus

  • pemphigoid gestationis

• they were taking any of the following drugs, because these products interact with progesterone:

  • bromocriptine

  • cyclosporine

  • rifamycin

  • ketoconazole.

Non-English speakers

We made provision for translation, if necessary, to communicate with non-English speakers and accommodate any special communications requirements of potential study participants. The PISs and consent forms (see Appendices 1 and 2) were translated from English into Dutch for use in the Netherlands.

Sequence generation

Computer-generated random numbers were used, and participants were randomised online via a secure internet facility. This third-party independent ITMS was designed, developed and delivered by MedSciNet^® (Stockholm, Sweden) according to standards of the International Organisation for Standardisation 9001:2000 and the requirements of the US Food and Drug Administration CFR21:11. 26

Participants were randomised to receive progesterone or placebo in a 1 : 1 ratio. A ‘minimisation’ procedure using a computer-based algorithm was used to avoid chance imbalances in important stratification variables. The stratification variables used for minimisation were as follows:

• number of previous miscarriages (3 or > 3)

• maternal age (≤ 35 or > 35 years)

• polycystic ovaries or not

• body mass index (BMI) (≤ 30.0 or > 30.0 kg/m²).

Allocation and minimisation

After all of the eligibility criteria and baseline data items were entered online, the ITMS generated a code number which took into account the minimisation variables recorded for the individual, and which was linked to a specific trial intervention pack. The code number was advised via e-mail to the local principal investigator (PI), the relevant trial pharmacist (see Distribution and Investigational medicinal product supply, Storage, dispensing and return) and the research nurse or midwife performing the randomisation.

Blinding

Participants, investigators, research nurses, midwives and other attending clinicians remained unaware of the trial drug allocation throughout the duration of the trial.

In the case of any SAE, the general recommendation was to initiate management and care of the participant as though the woman was taking progesterone. If the drug allocation was specifically requested to assist the medical management of a participant, clinicians could contact the trial manager or the trial co-ordinator for this purpose, 24 hours per day, 7 days per week (see Safety monitoring, Adverse events). Cases that were considered serious, unexpected and possibly, probably or definitely related to the trial intervention (see Appendix 3) were unblinded as appropriate. In any other circumstances, investigators and research nurses and midwives remained blind to drug allocation while the participant remained in the trial.

Distribution

On randomisation, the research nurse or midwife arranged the dispatch of a trial intervention pack to the local participating centre or the home address of the participant within 72 hours of receiving the telephone call informing of pregnancy. Nurses in the UK contacted the trial pharmacy at St Mary’s Hospital in London and nurses in the Netherlands contacted the trial pharmacy in Utrecht (see Investigational medicinal product supply, Storage, dispensing and return). Each trial intervention pack contained either progesterone or placebo.

Instructions to participants

The research nurse or midwife also provided the participant with instructions. In cases of delivery directly to the home address, the research nurse or midwife contacted the participant to ensure receipt and understanding of how to use the supplied capsules.

Each participant was expected to commence the trial intervention on the day it was received, and continue until it was finished, at around 12 completed weeks of gestation, unless the pregnancy ended before this time. The research nurse or midwife also telephoned each woman in the days immediately after the trial medicine was supplied, to ensure that the participant had started taking the medicine. In the event of the capsules being mislaid, the participant was instructed to telephone the research nurse or midwife, who would liaise with trial manager or the trial co-ordinator to arrange a further supply of the same type of intervention.

Each participant was asked for consent to notify the primary care provider (in the UK, the general practitioner) by letter that she was participating in the trial (see Appendix 4). Moreover, each participant was given a business card and a fridge magnet with contact details of local PROMISE investigators and the central Trial Co-ordinating Centre (TCC), to inform any directing clinicians, in case of potential drug interactions.

Progesterone capsules

The investigational medicinal product (IMP) was micronised progesterone at a dose of 400 mg (that is, two capsules of Utrogestan^® 200 mg) taken vaginally twice daily (every morning and every evening) for the duration of treatment.

The anatomical therapeutic chemical classification code for the pharmacotherapeutic group of the IMP was G03D and the chemical abstract service number was 57-83-0. The product had all the properties of endogenous progesterone, with induction of a full secretory endometrium and in particular gestagenic, antiestrogenic, slightly antiandrogenic and antialdosterone effects.

Besins Healthcare held the manufacturing authorisation (France) for Utrogestan^®, including for the indication of threatened miscarriage or prevention of habitual miscarriage due to luteal phase deficiency up until the 12th week of pregnancy.

Placebo capsules

Placebo capsules were vaginal capsules, composed of sunflower oil, soybean lecithin, gelatin, glycerol, titanium dioxide and purified water, encapsulated in the same form as the IMP, and identical in colour, shape and weight, for use in the control arm of the PROMISE trial. The dose, route and timing of administration were also identical to those in the active progesterone arm of the study.

Dose

The ideal dose of progesterone for the potential prevention of RM was unknown. The biologically effective dosage of micronised progesterone capsules ranged from 200 mg once daily to 400 mg twice daily according to the summary of product characteristics (SmPC)30 and the British National Formulary (BNF). 31 Our choice of 400 mg twice daily was made after a careful review of the existing literature and an extensive survey of clinicians in the UK (see Chapter 1, Existing knowledge, Progesterone in clinical use for recurrent miscarriages). We also reviewed other related evidence. For example, progesterone vaginal capsules are commonly used for luteal support in assisted conception at a treatment dose of 400 mg twice daily, with no specific safety concerns raised on this dose. 16,32

After evaluating the evidence, we considered the dosage of 400-mg vaginal progesterone twice daily to be an acceptable regimen to ensure a clinically effective dose, and to minimise the risk of a negative trial result from therapy with a suboptimal dose.

Route

An immunomodulatory effect of progesterone at the trophoblastic–decidual interface is the key presumed mechanism for preventing RM. 3,33–35 Our choice to use the vaginal route of administration was, therefore, rational to deliver a greater proportion of drug to the relevant site (the uterus) using the ‘first uterine pass’ effect. 36,37 Furthermore, studies that have used vaginal progesterone in the prevention of preterm birth have reported its effectiveness when given via this route. 19,21,22 For example, 14 out of 36 studies of second- and/or third-trimester progesterone to prevent preterm birth (identified by a recent systematic review) used vaginal progesterone, with significant improvements being observed for various clinical outcomes, confirming the biological effects of vaginal progesterone. 38

The acceptability and availability of interventional drugs were also important considerations supporting the vaginal route of drug delivery. Our discussions with consumer representatives confirmed that a vaginal formulation would be more acceptable to women than an intramuscular preparation. These findings were further supported by a study in which 12% of participants were unable to tolerate the intramuscular progesterone preparation and declined participation or withdrew from that trial. 20 Of those who did continue, 34% complained of localised soreness around the injection site. Finally, in our survey of women with RM conducted at St Mary’s Hospital and Guy’s and St Thomas’ Hospitals in London, a very high acceptability of the vaginal route (81/88; 92%) was identified (Arri Coomarasamy, University of Birmingham, 2008, unpublished data).

The capsule formulation of the PROMISE trial is widely available in the UK and worldwide.

Timing

Treatment commenced as soon as possible after a positive pregnancy test and no later than 6 weeks of pregnancy, and continued until the gestational age of 12 weeks. Our rationale to discontinue the treatment at 12 weeks was that production of progesterone by the corpus luteum becomes less important than the placental production of progesterone after 12 weeks of gestation. Furthermore, in the only clinical trial of progesterone treatment for RM to have shown a statistically significant reduction in miscarriage rates, progesterone was given until 12 weeks of gestation (odds ratio 0.37, 95% CI 0.15 to 0.90). 7

Compliance assessment

Our previous experience of research and clinical care for women with RM demonstrated that they would be highly motivated and compliant with therapy advice. However, compliance in the PROMISE trial was evaluated by ‘pill-counting’.

Participants were asked to return completed, partially used and unused treatment packs to the trial centres. The research nurses and midwives at each study centre documented the capsules returned by each participant, while the trial pharmacists kept their own accountability logs.

In an effort to improve compliance, women who failed to return the blister packs from the previous 4 weeks, whether or not these were empty, using the envelope provided were contacted by the local research nurse or midwife by telephone or e-mail to be offered advice and support.

Non-compliance was defined as missing more than 20% of trial medicines for the gestational age at randomisation. Non-compliant participants were interviewed (face to face or via telephone) in an attempt to establish the reason(s) for their non-compliance.

Manufacture, packaging and labelling

All arrangements for trial drug supply, labelling, storage and preparation were undertaken as per the requirements of the Medicines for Human Use (Clinical Trials) Regulations 2004. 39,40

Active (Utrogestan^® vaginal 200 mg) and placebo capsules were manufactured and packaged (assembled) by Besins Healthcare, in compliance with good manufacturing practice (EU Directive 2003/94/EC)27 and GCP (Clinical Trials Directive 2001/20/EC)28 requirements. Besins Healthcare also provided qualified person release of the trial drug under the requirements of the Medicines for Human Use (Clinical Trials) Regulations 2004. 39,40

Each trial intervention pack contained the entire supply required for the treatment period of up to 8 weeks. As the treatment regimen was two Utrogestan^® 200 mg or placebo capsules twice daily, the drug package for each participant contained 224 capsules: 2 (capsules) × 2 (twice daily) × 7 (7 days per week) × 8 (8 weeks) = 224.

The drug packages were labelled in compliance with the UK Medicines for Human Use (Clinical Trials) Regulations 2004,39,40 ensuring the protection of each participant, traceability and proper identification of the IMP and trial.

Storage, dispensing and return

At study initiation, supplies of progesterone and placebo capsules were delivered by Besins Healthcare to two study pharmacies, where the products were stored and whence they were dispensed to all participants.

The pharmacies were located at St Mary’s Hospital in London and the University Medical Centre of Utrecht. Both sites complied with the relevant guidelines and regulations, including (in the UK) the Duthie report41 and the Royal Pharmaceutical Society of Great Britiain’s practice guidance on pharmacy services for clinical trials42 as well as the appropriate standard operating procedures of Imperial College London. 43

At the pharmacies, the PROMISE trial medications were stored separately from other stock, in areas with restricted access. Drugs expired or returned by participants were stored separately from unallocated trial medicines. Dispensing was undertaken against prescription forms, each entitled ‘The PROMISE trial, EudraCT Number 2009-011208-42’ and labelled with the name, date of birth and study identification number of the relevant participant and a unique code number provided by the ITMS for randomisation (see Randomisation, Allocation and minimisation). Each trial pharmacy kept detailed dispensing records including participant study numbers and names, code numbers, batch numbers, expiry dates, doses and dates of dispensing.

The trial co-ordinator monitored the quantity of supplies held by each dispensing pharmacy in comparison with the number and rate of randomisations undertaken, and liaised with Besins Healthcare to ensure adequate supplies of the IMP in the UK and the Netherlands.

Each participant in the PROMISE trial was provided with freepost envelopes to return the unused or used packets to the local study centre. All unused study drugs, including undispensed supplies and supplies returned by participants, were returned to the trial pharmacies for accountability and destruction. Both dispensing pharmacies were involved in the reconciliation of medicines returned by trial participants and the disposal of unused medication, in compliance with appropriate regulatory guidance.

Primary outcome

Live birth beyond 24 completed weeks of gestation.

Secondary outcomes

• Clinical pregnancy at 6–8 weeks (defined as the presence of a gestational sac, with or without a yolk sac or fetal pole).

• Ongoing pregnancy at 12 weeks (range 11–13 weeks) (defined as the presence of a fetal heartbeat).

• Miscarriage (defined as loss of pregnancy before 24 weeks of gestation).

• Gestation at delivery.

• Survival at 28 days of neonatal life.

• Congenital anomalies, and specifically genital abnormalities.

Exploratory outcomes

• Antenatal complications such as pre-eclampsia, small for gestational age (< 10th birthweight centile), preterm prelabour rupture of membranes and antepartum haemorrhage.

• For live births at beyond 24 completed weeks of gestation: mode of delivery, birthweight, arterial and venous cord pH, APGAR (appearance, pulse, grimace, activity, respiration) score and resuscitation.

• For neonates: surfactant use, ventilation support (days on intermittent positive pressure ventilation, continuous positive airway pressure and oxygen, and discharge on oxygen) and neonatal complications (such as infection, respiratory distress syndrome, necrotising enterocolitis, intraventricular haemorrhages and pneumothorax).

Resource use outcomes

The resource use data listed below were collected to estimate the costs associated with the provision of progesterone for RM:

• antenatal, outpatient or emergency visits

• inpatient admissions (nights in hospital)

• maternal admissions to high-dependency units (HDUs) or intensive care units (ICUs) (nights)

• neonatal admissions to special care baby units (SCBUs) or neonatal units (NNUs) (nights).

Future outcomes

Women were asked to consent for future evaluation of themselves and their child and the health records of both. Although long-term follow-up was outside the scope of the PROMISE trial, we recognised the value of data collected in this study to inform further studies on outcomes such as the composite end point of death or neurodevelopmental impairment at 2 years of age, cognitive scale scores at 2 years of chronological age, and disability classified into domains according to professional consensus. The hospital number and (in the UK) NHS number of each baby within the PROMISE trial were recorded to facilitate future follow-up studies.

Outcome assessment

The ITMS was utilised to capture baseline and outcome data, for contemporaneous data cleaning, to produce reports for the independent Data Monitoring Committee (DMC) and to maintain an audit trail. Relevant trial data were transcribed directly into the ITMS. Source data comprised the research clinic notes, hospital notes, hand-held pregnancy notes, laboratory results and self-reports.

First outcome assessment (6–8 weeks of pregnancy)

The research nurse or midwife at each study site telephoned every participant at between 6 and 7 weeks of gestation, to ensure there were arrangements for an ultrasound appointment with her usual carers, before 8 weeks of gestation (Figure 4). If an appointment had not been booked, the research nurse or midwife assisted with booking. The research nurse or midwife telephoned each participant again between 3 and 5 days after the scheduled date of the ultrasound appointment, to obtain details of the observations of a gestational sac.

FIGURE 4.

Participant care pathway and outcome assessment.

Withdrawal

Following discussion with the Trial Management Group (TMG), participants in the PROMISE trial could be withdrawn from trial treatment if it became medically necessary in the opinion of the investigator(s) or clinician(s) providing patient care. In the event of such premature treatment cessation, study nurses and midwives made every effort to obtain and record information about the reasons for discontinuation, and to follow up all safety and efficacy outcomes as appropriate.

Participants in the PROMISE trial could voluntarily decide to cease taking the study treatment at any time. If a participant did not return for a scheduled visit, attempts were made to contact her and (where possible) to review compliance and AEs. We documented the reason(s) for self-withdrawal where possible. Each woman remained able to change her mind about withdrawal, and reconsent to participate in the trial, at any time. Clear distinction was made between withdrawals from trial treatments while allowing further follow-up, and any participants who refused any follow-up. If a woman withdrew from taking the trial treatment but permitted further data collection, she was followed up and outcome assessments were undertaken for the remainder of the study.

If a participant explicitly withdrew consent to any further data recording, this decision was respected and recorded via the ITMS. All communications surrounding the withdrawal were noted in the study records and no further data were collected for such participants.

Concomitant non-trial treatments

Concomitant therapy was provided at the discretion of the care-providing clinicians, and all concomitant treatment and medications were documented via the ITMS. Post-randomisation use of heparin was discouraged unless there was a clear and recognised indication for it (heparin therapy at the time of randomisation made a woman ineligible to participate in the trial). Other than identified contraindicated drugs (see Participants, Exclusion criteria) and other progestogen preparations, the initiation of treatment for another indication did not necessitate withdrawal from the PROMISE study.

Known side effects

The SmPC for progesterone (vaginal capsules)30 also states that ‘local intolerance (burning, pruritus or fatty discharge) has been observed during the different clinical trials and reported in the literature but incidences were extremely low’ (© Datapharm).

Overdose

The symptoms of progesterone overdose may include somnolence, dizziness and euphoria. In case of overdose, the care-providing clinicians of the PROMISE trial were prepared to undertake observation and reporting, and provide symptomatic and supportive measures, as required.

Dose modification for toxicity

The PROMISE trial included provision that, for participants experiencing non-serious side effects, the dosage could be reduced to 200 mg twice daily (one capsule in the morning and one at bedtime) at the discretion of the care-providing clinician, without unblinding treatment allocation (see the following section). The dose modification was noted on the case report form, along with the gestational age and the date on which such change was implemented.

Adverse events

The pharmacovigilance procedures of the PROMISE trial, including documentation, validation, evaluation and reporting, and responsibilities for the performance of these requirements, were based on the contemporaneously available literature to guide good practice. 28,45–47

Summarising trial data

We planned to summarise the recruitment numbers, those lost to follow-up, protocol violations and other relevant data, using a Consolidated Standards Of Reporting Trials (CONSORT) diagram. Baseline data and outcome data were separately summarised. For categorical data, we planned to provide proportions (or percentages). For continuous variables, we planned to examine the distribution of the observations and, if normally distributed, we planned to summarise them as means with standard deviations (SDs). If they were not normally distributed, we planned to report medians and interquartile ranges (IQRs); additionally, we planned to use geometric means and SDs for data where distributions appeared to be log-normal.

We planned to use diagnostic plots to assess the severity of deviations from normality, using log-transformations where necessary, assessing results as estimates with 95% CIs, and using bootstrapping with bias correction and acceleration where non-parametric methods were indicated.

Following previously published CONSORT recommendations,49–51 significance tests were not given between randomised treatment arms.

Intergroup comparisons

The primary analysis was undertaken on the basis of intention to treat. This approach was intended to avoid any potentially misleading artefacts of the study (such as non-random attrition). Every attempt was made to collect a complete data set about each pregnancy and women were encouraged to allow continued data collection even if withdrawing from trial treatment, because the exclusion of withdrawn participants from data analysis could bias results and reduce the power of the study to detect important differences. In particular, participants were followed up even after any protocol treatment deviation or violation. However, if a participant explicitly withdrew consent to any further data recording then all communications surrounding the withdrawal were noted and no further data were collected for such participants (see Outcomes, Withdrawal).

Participants were analysed according to the original randomised allocation, irrespective of compliance and crossovers. Binary regression with a log-link was used to assess relative risks (RRs) for the primary outcomes and for other binary outcomes such as clinical pregnancy at 6–8 weeks; ongoing pregnancy at 12 weeks (range 11–13 weeks), miscarriage rate and survival at 28 days of neonatal life, adjusting for minimisation variables.

Significance tests were, in general, carried out only for estimates of treatment effects, as separate tests for changes over time in the two groups might have resulted in entirely false and misleading conclusions about the differences between the groups when comparing p-values.

Notwithstanding our best efforts to collect a complete data set about each pregnancy, in a small number of cases it was not possible to determine the primary outcome of the study (see Chapter 3, Numbers analysed). We conducted an analysis whereby participants with missing primary outcome data were not included in the primary analysis. This presented a risk of bias, and secondary sensitivity analyses were undertaken to assess the possible impact of the risk, considering alternative assumptions both in favour and against the effect of the intervention. Other sensitivity analyses involved simulating missing responses using multiple imputation, using those baseline variables that are significantly related to the outcome as predictors.

Subgroup analysis

Three subgroup analyses were planned:

• number of previous miscarriages (3 or ≥ 4)

• maternal age (≤ 35 or > 35 years)

• polycystic ovaries or not.

Subgroup analyses were conducted only for the primary end point, and multivariate logistic regression was used. In each case, a test for interaction was first used to determine whether or not treatment was particularly effective in individual subgroups; our performance of subgroup analyses was dependent on sufficient data. Because of the well-known risk of false positives, both main effects and tests for interaction were performed and assessed before we considered results for subgroups. In addition, post-hoc subgroup analysis was performed only for the purpose of hypothesis generation.

Adjustments and sensitivity analyses

The process of randomisation with minimisation is designed to result in comparison groups that are highly similar at baseline, even more so than would be expected by chance. However, if randomisation failed to achieve balanced groups, we planned linear or logistic regression to adjust for the imbalance. We planned to adjust for missing data using multiple imputation. In cases of difference, we planned to give greater weighting to the primary analysis of intergroup comparisons than to sensitivity analyses.

Interim analyses

Interim analyses of principal safety and effectiveness end points were conducted on behalf of the independent DMC. These were considered together with scheduled reports of SAEs. The trial statistician was unblinded to the level of groups ‘A’ and ‘B’. The meaning of ‘A’ and ‘B’ was made known to the DMC separately. The first interim analysis was undertaken after the primary outcome data became available for the first 100 participants, and thereafter at annual intervals.

We were prepared to consider early termination of the PROMISE trial in case of interim analyses showing overwhelming evidence of effectiveness or significant harm (see Termination). Effectiveness and futility criteria were defined by the DMC (see Governance, Trial oversight bodies) with regard to the Peto principle that a trial should be stopped only when there is overwhelming evidence against one treatment or another52,53 with a nominal interim alpha set at 0.001 using O’Brien and Fleming alpha spending rules. 54 Under these conditions, no adjustment to the overall level of significance was needed. 55,56

Long-term analyses

Although the development of the babies born to participants in the PROMISE trial was of interest to the investigators, this was outside the scope and time frame of the study. Nevertheless, we recognised the value of data collected in this study to inform further studies on outcomes such as the composite end point of death or neurodevelopmental impairment at 2 years of age, cognitive scale scores at 2 years of chronological age, and disability classified into domains according to professional consensus. Women were asked to consent to future evaluation of themselves and their child and the health records of both, and could be traced through NHS Strategic Tracing Services. The hospital number and NHS number of each baby in the PROMISE trial were also recorded to facilitate future follow-up studies.

Health economic methods (see also Chapter 4)

An economic evaluation was conducted alongside the PROMISE trial, pragmatically comparing the costs and consequences of treatment using progesterone versus those of usual care. To examine the effect of treatment of each woman (and baby) on health resource use, data were collected from enrolment until the trial end point (hospital discharge). The analysis considered the number of days that treatment was received and three categories of heath service resources [antenatal contacts, how the pregnancy ended (preterm pregnancy loss management or mode of delivery) and postnatal admissions]. This perspective assumed the most salient costs to be those accrued by perinatal services; the assumption was tested by model-based extrapolation and a sensitivity analysis.

The primary outcome of our economic analysis was incremental cost per additional live birth after at least 24 weeks of gestation, with data collected up to 28 days of neonatal life. In order to provide additional information about the wider resource implications to the NHS and longer-term implications to health status, a systematic search identified models employed previously in similar trial-based economic evaluations to extend the time horizon of the cost and generic health gains of the two arms of the PROMISE trial end point. Furthermore, strategies were identified to attribute variation relating to the surrogate outcome of intervention (gestational age) to estimate health-care costs and associated generic health gains in the longer term.

Data management

The trial was designed and conducted to meet the requirements of:

• the Data Protection Act 199857

• the NHS Code of Confidentiality58

• the Caldicott Principles. 59

Information about participants in the PROMISE trial was collected directly from trial participants and hospital notes, and was considered confidential. The trial manager was responsible for overseeing data custody, but all of the staff involved in the study (clinical, academic and support personnel) shared the same duty of care to prevent any unauthorised disclosure of personal information. For this purpose, each trial participant was allocated a unique study number at recruitment, and all study documents used this reference as the identifier. Personal data and contact details were held in separate NHS or university password-protected databases at local sites (in compliance with local and national confidentiality and data protection standards), which were linked to the secure ITMS via the unique study number. All data were analysed and reported in summary format (individuals remaining unidentifiable).

Data were collected and stored on secure NHS or university computers. Access to data was restricted by usernames and passwords at two levels (to gain access to NHS and university computers and then to access the ITMS). Only when strictly necessary and after anonymisation were trial data encrypted and transmitted outside the NHS or university settings (e.g. to the DMC). No study data were retained in handheld media, laptops, personal computers or other similar media.

The online ITMS was maintained according to the security policies of the Women’s Health Unit of King’s College London. These policies included provision for password assignment, encryption, immediate back-up, off-site back-up and disaster recovery processes. Electronic data were backed up to both local and remote media in encrypted format every 24 hours. Paper-based data (such as signed consent forms) were kept in locked filing cabinets at each site.

The data generated during the trial were available for inspection on request by the participating physicians, representatives of the sponsor, the REC, host institutions and the regulatory authorities. These data-handling arrangements were clearly conveyed to participants in the PIS (see Appendix 1), and permission was obtained in the consent form (see Appendix 2).

On completion of data collection, the site files from each study centre were to be securely archived at the sites. Electronic study data remained securely stored within the ITMS. The trial master file was to be securely stored by the sponsor when all study activities were completed. In accordance with the Medicines for Human Use (Clinical Trials) Amendment Regulations 2006 (sections 18 and 28),40 all the study data will remain securely stored for 25 years, to enable review, reappraisal and resolution of any queries or concerns and to facilitate further follow-up research. After this time the data will be securely destroyed.

Data quality assurance

The PROMISE trial co-ordinator performed hospital site visits as part of trial monitoring activities (see Governance, Trial monitoring). This quality assurance activity occasionally involved source data verification. The research and development (R&D) departments of participating study centres also performed routine monitoring audits at least annually. The PROMISE trial was additionally selected for MHRA inspection at two locations (Liverpool Women’s Hospital and Luton and Dunstable Hospital).

The trial also adopted a centralised approach to monitoring data quality and compliance, using the ITMS.

Ethical implications

In designing the PROMISE trial and considering the ethical implications of the study, the issues indicated below were considered and resolved or safeguarded.

Ethical governance

Following a favourable opinion of the National Research Ethics Service via the West Midlands REC and before recruitment commenced at each participating centre, the TCC obtained favourable site-specific assessments and R&D approvals as required.

The PI of each participating study site was responsible for liaison with administrative and managerial representatives of the local institution, and obtaining any necessary permissions from trust authorities. On behalf of the local institution, the PI was also required to sign an Investigator’s Agreement in respect of accrual, compliance, GCP, confidentiality and publication. Deviations from the Investigator’s Agreement were monitored and remedial action taken as appropriate by the TMG.

In addition, and in compliance with the International Conference on the Harmonisation of Good Clinical Practice, all institutions participating in the trial completed a delegation log and supplied this document to the TCC. The delegation log listed the responsibilities of each member of the local study team, and an up-to-date copy was stored in the site file at the institution and also at the TCC. A curriculum vitae of the PI, and confirmation of GCP training and honorary or substantive employment with the participating trust, was also verified by the trial manager and retained.

Clinical trial authorisation

Clinical trial authorisation for the PROMISE trial was obtained from the MHRA before recruitment started.

Changes to the protocol

It was agreed that if any amendments to the study protocol required regulatory approval (from the MHRA, REC or local R&D offices), these changes would not be instituted until the amendment had been reviewed and received favourable opinion from the relevant bodies. However, a protocol amendment intended to eliminate an apparent immediate hazard to participants would be implemented immediately, with notification and request for approval to the MHRA, REC and R&D offices as soon as possible.

There were no significant changes to the methods after trial commencement. All amendments to the study protocol were in the domain of clarifications to wording and intentions (see Appendix 5, Table 16).

Trial monitoring

The PROMISE trial was monitored according to the standard operating procedures of Imperial College London/Imperial College Healthcare NHS Trust Joint Research Compliance Office. 43

The purpose of monitoring was to:

• protect the rights and well-being of trial participants

• ensure that the reported trial data were accurate, complete and verifiable from source documents

• ensure that the trial remained compliant with GCP and other regulatory and good practice guidance.

Participating centres were monitored by the TCC by checking incoming electronic forms for compliance with the protocol, consistency of data and missing data. The trial co-ordinator and trial manager remained in regular telephone or e-mail contact with centre personnel to check on progress and resolve any queries. In addition, periodic site monitoring was undertaken as required by the TCC or the sponsor to:

• review understanding of the protocol and trial procedures by the trial staff

• verify that the trial staff had access to the necessary documents

• verify the existence of participants against clinic records and other sources

• verify selected data items and SAEs recorded, compared with data in clinical records, to identify errors of omission as well as inaccuracies.

Monitoring visits were followed by a monitoring report, summarising the findings of the visit and recommending remedial actions as necessary. Investigator meetings were held at least annually for the purpose of learning, updating and sharing.

Trial oversight bodies

The PROMISE trial was overseen by the TMG, the DMC and the TSC (Figure 5).

FIGURE 5.

Reporting relationships of trial oversight bodies. HTA, Health Technology Assessment; NIHR, National Institute for Health Research.

Site responsibilities

To ensure the smooth running of the trial and to minimise the overall procedural workload, each participating centre designated appropriately trained and qualified local individuals to be responsible for the institutional co-ordination of clinical and administrative arrangements.

Information

From the RM clinics at St Mary’s Hospital and Guy’s and St Thomas’ Hospitals in London, 88 women with a history of RM were surveyed with a set of structured and semi-structured questions to identify their opinions regarding the rationale for the trial and its ethical implications (see Governance, Ethical implications), the route of progesterone administration (vaginal, rectal or intramuscular), the duration of therapy, the suitability of courier transport to deliver study medications directly to study participants, and the relative importance of different outcomes assessed in the PROMISE trial. The interviews were conducted on a one-to-one basis by physicians with experience in early pregnancy care and training in listening to patient concerns. The information collected enabled the PROMISE trial team to understand that self-administration of vaginal progesterone twice daily would be acceptable to most potential participants in the study, and to select meaningful and important outcomes for data collection.

Communication

Five champions from among the participants in our survey volunteered to help the central organisers of the study to develop clear and concise literature for participants (see Appendices 1 and 2). During the development, the provisional documents were reviewed by other patients with a history of RM, as well as the independent service user representatives listed below:

• Ms Ruth Bender Atik (National Director of the Miscarriage Association)

• Ms Liz Campbell (Director of the Wellbeing of Women research charity)

• Officers of the RCOG consumers’ forum.

These members of the group were selected to represent professional and charitable communities with an interest in pregnancy and miscarriage. They brought experience of working with not only women who had previously suffered RM, but also stakeholders such as women trying to conceive, women in pregnancy and women who had previously suffered miscarriage, and their family members, from across the UK.

All of the lay stakeholders expressed appreciation for their involvement in PROMISE activities and enthusiasm for ongoing involvement in the PROMISE trial, whether as a result of inherent personal interest in the topic under investigation or the value they placed on empowerment to influence service improvements. They expressed opinions about their previous experiences of clinical consultations and identified strategies to improve existing services. Their views were valued by researchers and directly informed the development of literature for participants (see Appendices 1 and 2).

Oversight

Ms Ruth Bender Atik, the National Director of the Miscarriage Association, was also involved in overseeing the PROMISE trial as a member of the TSC throughout the duration of the study (see Governance, Trial oversight bodies). Although Ms Bender Atik was enthusiastic and dedicated, it is important to note that her voluntary contribution to the PROMISE trial took her time and attention away from competing commitments; in retrospect, it would have been helpful to ensure that such a time requirement was recognised financially.

Dissemination

During the years of recruitment and follow-up, various research team members facilitated peer group sessions and seminars, and delivered presentations about study activities and developmental concepts of the intervention. The composition of these groups varied, for example from closed meetings of senior clinical practitioners at each of the participating study centres to a large public audience for an inaugural lecture that was also shown online. The contents of the presentations were tailored accordingly.

Monthly newsletters were circulated to boost the morale of the trial staff at study sites, highlight important tasks and encourage recruitment. When recruitment was complete, the newsletters were no longer disseminated, but targeted, site-specific communications were used instead.

Primary outcome

Overall, 533 out of 826 women (64.5%) experienced a live birth after at least 24 weeks of gestation. Table 4 shows that the live birth rate in the progesterone group was 65.8% (262/398), and in the placebo group it was 63.3% (271/428), giving a RR of 1.04 (95% CI 0.94 to 1.15; p = 0.45).

Of the 533 pregnancies that proceeded to live birth beyond 24 weeks, 524 (98.3%) were singleton pregnancies and nine (1.7%) were twin pregnancies. None of the women participating in the PROMISE trial delivered more than two babies, and, overall, 1.1% of the trial participants [1.0% (4/398) in the progesterone group and 1.2% (5/428) in the placebo group] experienced a twin birth.

Secondary outcomes

Gestational age at live birth

The distributions of gestational age at delivery for the progesterone and placebo groups are illustrated by Figure 8. They show a hazard ratio of 1.04 (95% CI 0.91 to 1.19; p = 0.59).

FIGURE 8.

Distribution of gestational age by randomised treatment: pregnancies continuing beyond 24 weeks only. Hazard ratio 1.04 (95% CI 0.91 to 1.19; p = 0.59).

Congenital anomalies

Congenital anomalies observed in babies born to PROMISE participants are listed in Table 5. These outcomes were rare, both overall and in each arm of the trial, so our power to identify truly significant rates of difference between the groups is low. Among the pregnancies randomised to receive active treatment, we observed two cases of talipes, two heart anomalies (one ventricular septal defect and one transposition of great arteries), one case of tongue tie, one case of hernia, two chromosomal anomalies (one case of Pierre Robin syndrome and one case of Down syndrome duodenal atresia) and one genital abnormality (hypospadias). Among those randomised to placebo, we observed one case of talipes, three heart anomalies (all ventricular septal defects), two cases of ventriculomegaly, two cases of tongue tie, one case of hernia, one chromosomal anomaly (Turner syndrome) and one genital abnormality (urachal cyst). Overall, 8 out of 266 (3.0%) babies in the progesterone group and 11 out of 276 (4.0%) babies in the placebo group were affected by congenital anomalies, giving a RR of 0.75 (95% CI 0.31 to 1.85), with a p-value of 0.54.

Subgroup analyses

We planned a priori to conduct subgroup analyses by maternal age (≤ 35 or > 35 years at recruitment), number of previous miscarriages (3 or > 3) and the presence (or not) of polycystic ovaries (Table 6). At the end of the trial, we performed additional post-hoc comparisons between the outcomes of women commencing treatment at earlier and later stages of the first trimester, and between women in the UK and in the Netherlands.

Exploratory analyses

Cost-effectiveness and uncertainty

The primary cost-effectiveness outcome of the PROMISE study was cost per additional live birth after at least 24 weeks of gestation. To inform the decision-maker, the incremental cost-effectiveness ratio (ICER) is the ratio of the difference in cost (C) to the differences in difference in effect (E), denoted by the formula:

where Δ represents change, C represents the costs, E represents the effects, and subscripts I and C refer to the intervention and control arms, respectively.

The cost to achieve an additional live birth after at least 24 weeks was calculated to be £18,053. In the UK, a value of £20,000–30,000 per quality-adjusted life-year (QALY) is considered to be an acceptable value for adopting a health-care technology. 70 However, there remains unsettled debate about the valuation of prenatal life in economic evaluations. 72 The generic health gains in terms of QALYs are reported in Sensitivity analyses, Estimation of quality-adjusted life-years and costs beyond the trial end point.

Uncertainty around the mean statistic for both costs and outcomes is a commonplace feature of economic evaluations. A key component of any evaluation is to demonstrate this uncertainty and the robustness of the cost-effectiveness ratio calculated from the initial main trial analysis. A non-parametric bootstrap (with 50,000 replications) was performed to demonstrate uncertainty in the joint distribution of cost and effects in the PROMISE trial. Figure 9 provides a visual representation of the degree of uncertainty in the ICER on a cost-effectiveness plane.

FIGURE 9.

Cost-effectiveness plane (50,000 bootstrap replications). GBP, Great British pound.

The bootstrap estimates that the 95% CI of the incremental probability of an additional live birth beyond 24 weeks of gestation ranges from –0.0477 to 0.0845. Based on the current level of uncertainty, it might also be concluded that there is a 29.30% chance that progesterone may result in a decrease in the probability of a live birth beyond 24 weeks of gestation, compared with usual care. Nevertheless, 17.16% of the bootstrap replications fell within the bottom-left quadrant, where less effect also resulted in lower cost.

The top-right quadrant of the cost-effectiveness plane contained 59.31% of all replications. Although overall there is substantial uncertainty, this may suggest that treatment was most likely to cost more and have some additional benefit.

Given the uncertainty in the current estimates of the cost per additional live birth beyond 24 weeks of gestation, a decision-maker may wish to examine how the probability that progesterone would be the most cost-effective option will vary with an increasing willingness to pay.

Figure 10 presents the CEAC to illustrate how the probability of treatment being cost-effective changes as the willingness to pay increases.

FIGURE 10.

Cost-effectiveness acceptability curve.

At a willingness to pay of £100,000 per additional live birth beyond 24 weeks of gestation, the probability that treatment would be cost-effective asymptotes [p(ICER < 100,000) = 0.7203]. This raises the question as to what society would be willing to pay to increase the likelihood of an additional live birth beyond 24 weeks of gestation.

Fixed treatment cost until 12 weeks

In the base-case cost-effectiveness analysis, the direct cost of treatment with progesterone accurately accounted for the total number of days of treatment received by individuals (assuming, as was the case during the trial, that surplus capsules would be returned either at 12 weeks or in the event of miscarriage). Figure 11 illustrates trends in the variation of stages of pregnancy that treatment started and stopped. The variable trialstop illustrates a bimodal distribution relating to whether pregnancy proceeded beyond 12 weeks or treatment was withdrawn as a result of miscarriage.

FIGURE 11.

Distribution in points of gestation where treatment (a) started; and (b) stopped.

If this trial was to be replicated in a real-world setting, it is unlikely that surplus medication prescribed up to an anticipated treatment end point (12 weeks) would be returned and reused. Our sensitivity analysis illustrates the implications that treatment would be prescribed up to a fixed end point (12 weeks of gestation) while still allowing the observed variation in treatment initiation (trialstart).

Repeating the non-parametric bootstrap procedure to re-estimate the ICER and CEAC with respect to this change in the expected cost of treatment generated a predictable increase in the ICER to £22,256 per additional live birth beyond 24 weeks of gestation. Figure 12 illustrates a marginal decrease in the likelihood that treatment would be cost-effective as willingness to pay increases.

FIGURE 12.

Cost-effectiveness acceptability curve: treatment costs fixed until 12 weeks.

Estimation of quality-adjusted life-years and costs beyond the trial end point

The base-case analysis adopted a condition-specific end point to inform clinical choices of alternative strategies intended to increase the likelihood of a viable birth (live birth beyond 24 weeks of gestation). However, there are limitations of condition-specific outcomes to inform health-care policy. Furthermore, it could be argued that the binary outcome of live birth (or not) beyond 24 weeks of gestation, although providing some indication of life-years gained, does not provide any information on the quality of life achieved from treatment.

Expressing the health benefits in terms of QALYs provides more information to guide the allocation of a limited pool of health-care resources based on explicit monetary values per QALY. To explore methods by which the clinical outcomes might be translated into generic health benefits (such as QALYs), a systematic search was undertaken to identify and project the longer-term costs and consequences, and several potential modelling strategies were explored.

A Finnish study provided the most suitable basis for a strategy to express gains in generic health terms (such as QALYs);73 to our knowledge this is the only study to date to evaluate the effects of gestational age on QALYs. It estimated QALYs up to 4 years of age using the 17-dimension parental questionnaire, reported with expected variations in the 4-year QALY by gestational age at birth. 74

By utilising gestational age at delivery within the PROMISE trial to serve as a surrogate outcome for QALYs, QALYs per participant were deduced. However, the estimates of QALYs reported by the Finnish study truncated QALYs at 32 weeks of gestation. Consultation with clinical content experts suggested that generic health gains (QALYs) could be most reasonably assumed to linearly increase up to a gestational age of 34 weeks at birth and, based on this assumption, linear predictors of the expected QALYs for weeks 24–34 were estimated using an ordinary least squares regression. QALYs after 34 weeks showed no additional increase. Table 13 summarises the data on expected QALYs that were used to inform projected health gains up to 4 years of age by gestational age at birth.

The expected QALYs were applied to PROMISE records of gestational age at birth. In a situation of postnatal infant mortality (up to 28 days), the trial captured mortality date and adjusted individual QALYs to the date of mortality.

Costs beyond initial hospitalisation (at birth) were projected by utilising a data set of individual records supplied from the Oxford Records Linkage Study by Dr Stavros Petrou. 75 This data set provided a sample of 95,635 births in the UK, and their associated costs for the first year after initial hospitalisation (excluding costs associated with the initial admission related to pregnancy). Price years recorded in the data set ranged between 1979 and 1988; a time discount rate of 3.5% was applied to obtain 2011 values. Dummy variables representing weeks of gestation at birth from 23 to 44 weeks were generated. The cost for the first year after initial hospitalisation was regressed against these variables to estimate the magnitude of first year costs by gestational age at birth.

Table 14 provides regression outputs in which coefficients of each gestational age at birth show the expected additional costs following initial hospitalisation.

Gestational age at delivery served as a surrogate outcome to estimate QALYs (over 4 years) and additional health-care costs (over the first year), allowing the treatment effect to be estimated using regression analysis. Cost and QALY distributions were jointly estimated using Zellner’s seemingly unrelated regression,76 thereby allowing analysis to also account for correlation in the residuals. Table 15 presents the output of the regression.

First, examining the regression coefficients for progesterone indicated the incremental cost to be £366.53 and incremental QALY to be 0.0987, generating an ICER of £3712 per QALY. Subject to certain assumptions required to estimate the utility function (discussed below), this ICER could be considered to be highly cost-effective with respect to explicit decision rules for the reimbursement of health technologies in the UK; it falls well below the benchmark of between £20,000 and £30,000 that is suggested as an appropriate threshold by the National Institute for Health and Care Excellence (NICE). 70

Furthermore, significant correlation was observed between cost and QALY distribution within the estimated utility function (0.6487; p < 0.0001). To an extent, this correlation could be attributed to the common surrogate outcome of gestational age at delivery; lower gestational age would decrease neonatal health status and that would be correlated to higher health-care costs.

Figure 13 illustrates changes in probability that treatment may be cost-effective as the willingness to pay increases. The CEAC indicates a probability of between 0.7145 and 0.7341 to express the likelihood that prescription progesterone therapy in women with a history of unexplained RM falls within the NICE threshold.

FIGURE 13.

Cost-effectiveness acceptability curve: QALYs and cost projected beyond the trial end point.

Although the estimation of QALYs and first-year cost consequences to the NHS may be useful to policy-makers and decision-makers allocating global health-care budgets,77 it should be noted that there are several unavoidable limitations to consider when interpreting these results. First, the extrapolation of QALYs utilised data from Finland without accounting for any variation in the preferences for health state that may exist between countries. Second, incorporation of QALYs into the utility function essentially places values on prenatal life (zero QALYs for miscarriage) that may raise moral and ethical debates. Finally, no reliable method was identified to model the implications of miscarriage on individual health, or indeed health-care, utilisation.

Internal validity

Randomisation and minimisation were effective in achieving balanced treatment allocations at baseline. Thus, the PROMISE trial optimised statistical power and eliminated selection and allocation bias. 78 A computer-generated allocation sequence, allocation concealment and blinding prevented investigators from knowing the assignment of the next participant based on prior treatment assignments. Possible confounding factors such as the number of previous miscarriages, maternal age and polycystic ovaries were similarly distributed between treatment groups. As a study with allocations blinded to participants and care providers, the PROMISE trial also avoided performance bias.

The research team believed that it was important to ensure that the study was large enough to detect clinically important treatment effects, and, therefore, the target sample size of the study was calculated to enable detection of a MID of 10% in rates of live birth after at least 24 weeks of gestation. We planned to randomise 790 women in total (395 participants in each of the progesterone and placebo arms) and we actually exceeded this number. Consequently, we consider the findings of this trial to be methodologically and statistically robust.

Our trial design offered a number of other strengths with respect to data collection and analysis. The treatment of participants by a large number of study centres and practitioners allowed intervention impact to be evaluated without confounding by individual variance in clinical practice and local technology. The outcome indicators that we measured were variables of routine familiarity in pregnancy care, so they were well understood and easily recorded by clinical researchers. Almost all of the outcome data recorded during the PROMISE study were objective outcomes (rather than subjective descriptions), and the study was blinded, so there was no risk of incurring assessor bias.

Contextual suitability

The PROMISE trial was built on a plausible biological basis and encouraging preliminary trial data, and directly answered calls from national bodies to address the possibility that progesterone therapy in early pregnancy could be beneficial to women with a history of unexplained RM. Although clinical equipoise existed, the lack of definitive knowledge or policy guidance for health services practitioners allowed considerable variations in care. The PROMISE trial answered an important question.

Other strengths

There is increasing recognition of the need for clinical research to embrace the views of both lay and professional stakeholders. 64 In the PROMISE trial, PPI occurred at stages of design, development and monitoring. For example, patients were surveyed to assess the acceptability of the intervention, and engaged in the development of literature for participants, and the TSC included representatives of patient advocacy organisations. We believe these roles were important to ensure appropriate communication with study participants and project oversight throughout the duration of the research.

The trial intervention was deliverable within the context of customary care without major impacts on health service structure. The mode of administration of IMP was designed to reflect the preferences expressed by patients, and most of our data collection could be performed during routine antenatal and postnatal appointments of the study participants.