Transfer of thawed frozen embryo versus fresh embryo to improve the healthy baby rate in women undergoing IVF: the E-Freeze RCT

Static success rates

In vitro fertilisation success rates remain modest, with a mean live birth rate of 25% per treatment involving a fresh-embryo transfer. Data for three consecutive years (2010 to 2012) from the American4 and European registries5 suggest that there was no improvement in IVF live birth rates over the 3-year period.

Ovarian hyperstimulation syndrome

Exogenous hormones used for ovarian stimulation are associated with a risk of ovarian hyperstimulation syndrome (OHSS), which is exacerbated if a woman becomes pregnant following fresh-embryo transfer. Moderate to severe OHSS is a complication unique to IVF treatment, occurring in around 1–5% of treatments,6 and often requiring in-patient care, resulting in significant NHS costs. Severe OHSS is associated with significant morbidity (including ascites, pleural and pericardial effusion, respiratory failure and intensive care admission) and, rarely, death.

Poor obstetric and perinatal outcomes

Pregnancies resulting from IVF are associated with a higher rate of maternal and perinatal complications than pregnancies resulting from spontaneous conception. A systematic review7 has shown that babies, even singletons, conceived following IVF are more likely than babies conceived without IVF treatment to die during the perinatal period [risk ratio (RR) 1.87, 95% confidence interval (CI) 1.48 to 2.37], to be delivered preterm (RR 1.54, 95% CI 1.47 to 1.62), to have a low birthweight (RR 1.65, 95% CI 1.56 to 1.75) and to have congenital anomalies (RR 1.67, 95% CI 1.33 to 2.09). Women who become pregnant as a result of IVF are more likely than those who become pregnant as a result of spontaneous conception to develop pre-eclampsia (RR 1.49, 95% CI 1.39 to 1.59), bleeding in pregnancy (RR 2.49, 95% CI 2.30 to 2.69) and diabetes (RR 1.48, 95% CI 1.33 to 1.66) and to require a caesarean section (RR 1.56, 95% CI 1.51 to 1.60).

Although the absolute number of women with OHSS and pregnancy-related complications associated with IVF is relatively small, the increasing number of women receiving IVF4 has meant that the NHS burden of dealing with its short- and long-term complications is a serious and growing problem.

A possible cause of suboptimal live birth rates, as well as adverse maternal and perinatal outcomes, following IVF is the impact of the exogenous hormones used for ovarian stimulation on the lining of the uterine cavity. High levels of oestrogen produced by the ovary in response to this treatment affect uterine receptivity, reducing the chances of successful implantation and placentation. Suboptimal placentation may lead to obstetric and perinatal complications. It has been suggested that avoiding embryo transfer when the uterus is less receptive could improve success rates and reduce complications in pregnancy and delivery. Such a strategy also reduces the risk of OHSS by ensuring that a pregnancy does not occur in the presence of hyperstimulated ovaries.

Primary objective

The primary objective of the trial was to determine if a policy of freezing embryos, followed by thawed frozen-embryo transfer, results in a higher healthy baby rate than the current policy of transferring fresh embryos.

Secondary objectives

The secondary objectives of the trial were to assess if a policy of freezing embryos, followed by thawed frozen-embryo transfer, compared with the current policy of transferring fresh embryos, results in:

• fewer complications associated with IVF treatment and pregnancy

• greater cost-effectiveness from a health service and broader societal perspective.

Inclusion criteria

• The female partner was aged between ≥ 18 and < 42 years at the start of treatment (i.e. start of ovarian stimulation).

• Couples were undergoing their first, second or third cycle of IVF/ICSI treatment, where a cycle is defined as egg collection following ovarian stimulation.

• Both partners were resident in the UK.

• Both partners were able to provide written informed consent.

• They had at least three good-quality embryos [as defined by the Association of Clinical Embryologists (ACE)14] on day 3 after egg collection (note that the day of egg collection is counted as day 0). Good-quality embryos on day 3 were defined as those with 6–8 cells of grade 3/3 or above using the agreed national grading scheme. 14

At the start of the trial, only the first cycle was included. However, after discussion with the National Institute for Health and Care Research (NIHR) Health Technology Assessment (HTA) Monitoring Board, it was agreed that couples having their second or third cycle could also be included. This change to the eligibility criteria took effect from 12 April 2018.

A list of all amendments are described in Appendix 1.

Exclusion criteria

• Couples were using donor gametes.

• Pre-implantation genetic testing was planned.

• Elective freezing of all embryos was planned for medical reasons (e.g. severe risk of OHSS).

• Couples had been previously randomised to E-Freeze.

Identifying participants

Potentially eligible couples were identified from clinic case notes. An invitation letter and participant information leaflet (PIL) were mailed to eligible couples prior to their clinic appointment. A PIL was also provided at patient information/open evenings attended by couples preparing for their IVF/ICSI treatment. This was usually at least 24 hours prior to their clinic appointment. Eligible couples were approached by a clinician involved in their care and were invited to participate in the trial. Those interested in participating were able to discuss the study with a research nurse on the same day or at a later date.

Consenting participants

Informed consent from both partners was obtained by an appropriately delegated member of the study team. Contact details and baseline characteristics that were necessary for randomisation were recorded by the research nurse immediately after consent was obtained. Consent forms were signed by both partners; however, this could be undertaken at two different time points, as not all appointments were attended by both partners. This could be undertaken at their clinic appointment or at a subsequent visit until the procedure of egg collection; all consent forms had to be signed before the procedure of egg collection took place (Figure 1).

FIGURE 1.

Process and timescale for consent and randomisation.

Couples who may have previously consented to take part in E-Freeze during their first or second cycle of IVF were still eligible to participate in E-Freeze if they had not been previously randomised into the trial. For couples who had previously consented but had not been randomised onto the trial, informed consent was reobtained for any participation during future cycles and a new study number was generated.

After consent, each partner completed a short questionnaire on how they were feeling emotionally (see Report Supplementary Material 1). Each participant sealed their questionnaire in an envelope after completion and questionnaires were destroyed (unopened) if the couple did not proceed to randomisation.

The data needed for randomisation were recorded in the bespoke consent and randomisation program developed by the National Perinatal Epidemiology Unit (NPEU) Clinical Trials Unit (CTU) at the University of Oxford (Oxford, UK).

Confirmation of consent

A routine telephone call was made to couples 1 day after egg collection to inform them of the outcome of fertilisation (see Figure 1). Consent was confirmed during this routine telephone call from the embryologist or research delegate.

Screening for final eligibility

A final eligibility check was carried out on day 3 post egg retrieval. Couples with a minimum of three good-quality embryos were eligible for randomisation to receive either fresh-embryo transfer (i.e. the fresh-embryo transfer arm) or freezing of all good-quality embryos, followed by subsequent transfer of thawed embryos within 3 months (i.e. the freeze-all arm) (Figure 2).

FIGURE 2.

Final eligibility criteria.

Good-quality embryos on day 3 were defined as those with 6–8 cells grade 3/3 or above using the agreed national grading scheme based on guidance from ACE in the UK. 14

Randomisation

Randomisation was performed after the creation of embryos, 3 days post egg collection, once all eligibility criteria were established, including ensuring that three or more good-quality embryos were available. This minimised the randomisation-to-intervention time interval as embryos were transferred at either the cleavage or the blastocyst stage (i.e. day 3 or 5 after egg collection, respectively). Couples were randomised (in an allocation ratio of 1 : 1) to a strategy of either fresh-embryo transfer or freezing of embryos, followed by thawing and replacement at a later date (typically 4–6 weeks later and almost always within 3 months of egg collection). Randomisation was undertaken by the research nurse or a delegated member of the research team using a secure web-based centralised system [with 24 hours per day, 7 days per week (24/7) telephone back-up, 365 days per year] hosted by the NPEU CTU (University of Oxford), ensuring allocation concealment. The randomisation employed a minimisation algorithm to balance across the following factors: fertility clinic, woman’s age (at the time of start of treatment, i.e. ovarian stimulation), primary/secondary infertility, self-reported duration of infertility, method of insemination (IVF, ICSI or a combination of both) and number of previous egg collections (i.e. cycles).

Communication of randomisation to couples

As part of routine practice, the embryologist contacted the couple by telephone to let them know the quality of their embryos (on day 3 after egg collection). The embryologist or research delegate confirmed to couples whether or not they fulfilled the final inclusion criteria (three or more good-quality embryos on day 3) and which arm they had been randomised to at the time of their routine telephone call on day 3. The research nurse then contacted the couple if they had not fulfilled the inclusion criteria to answer any queries and offer follow-up in the clinic.

Standard-care arm

Women underwent fresh-embryo transfer at the cleavage or blastocyst stage in accordance with local protocols.

Intervention arm

All good-quality embryos were frozen in accordance with local protocols. Couples who were randomised to the freeze-all arm were contacted by the research nurse or research delegate within 3 working days post randomisation and arrangements were made for frozen-embryo transfer within 3 months of the egg retrieval process. This could involve a few visits to hospital for blood tests and ultrasounds to prepare the endometrium prior to embryo transfer.

At embryo transfer (in both arms), couples were asked to complete a short questionnaire to assess the additional costs related to the treatment (see Report Supplementary Material 2) and to repeat the emotions questionnaire (see Report Supplementary Material 1) that they filled in when they provided consent.

Ineligible and non-recruited participants

Details of all consenting couples were entered in a dedicated secure online database. It was anticipated that a proportion of those consented may not proceed to randomisation; the reasons for this were recorded (if available) and included the non-availability of three good-quality embryos on day 3. Couples not proceeding to randomisation were offered the most appropriate standard treatment. All clinics have access to supportive counselling as a requirement of the regulatory authority.

Primary outcome

The primary outcome was a healthy baby. A healthy baby was defined as a live, singleton baby born at term (between 37 and 42 completed weeks of gestation), with an appropriate weight for gestation (i.e. weight between the 10th and the 90th centile for that gestation, based on standardised charts).

Secondary outcomes

The secondary outcomes were separated into maternal safety, complications of pregnancy and delivery, measures of clinical effectiveness, measures of effectiveness of the process of freezing embryos, and health economic outcome measures.

Sample size calculation

The proposed primary outcome for this trial was novel and is not currently reported by IVF clinics or national regulatory bodies. This meant that a number of assumptions were made to determine the expected event rate in the fresh-embryo transfer arm (receiving current standard treatment, i.e. fresh-embryo transfer).

Prior to commencing the trial, the most recent data from the HFEA,4 which collects data on all IVF cycles from all clinics in the UK, showed that 25% of all women undergoing one episode of IVF treatment involving a fresh-embryo transfer have a live birth and 20% have singleton live births. These values were for women of all age groups, not necessarily for women fulfilling the inclusion criteria for this trial in terms of the number of good-quality embryos in their IVF cycle. The live birth rate for first, second and third cycles was similar. 4 No data were available regarding the primary outcome for this study: the healthy baby rate (i.e. live singletons born between 37 and 42 weeks, with appropriate weight for gestation). For our trial population, we anticipated that the fresh-embryo transfer arm event rate was likely to be < 25% and possibly as low as 17%.

To provide relevant information regarding the event rate expected in the fresh-embryo transfer arm, we surveyed 10 IVF centres that expressed an interest in the study, collecting data on the number of live births in women aged < 42 years who were undergoing their first IVF treatment in 2012. The average live birth episode rate from this survey was 31% (95% CI 25% to 37%). Accurate data on the healthy baby rate in those with at least three good-quality embryos were not available. Although the live birth rate is expected to be higher in women with at least three good-quality embryos (who are likely to have a better prognosis), we anticipated that the healthy baby rate in our trial population would be towards the lower end of the CI, that is around 25%, taking into account the higher risk of preterm delivery and babies who are small for their gestational age following IVF. 9

The following assumptions were made for the sample size calculation.

We assumed a healthy baby rate of between 17% and 25% in women who were eligible for the trial (i.e. aged < 42 years, with three good-quality embryos) undergoing standard care (i.e. fresh-embryo transfer). Taking into account the extra time, effort and potential expense involved in freezing embryos, and the delay in embryo transfer of up to 3 months, a difference of at least 8% in absolute terms was considered to be clinically important by an expert panel of clinicians to recommend a change in clinical practice. With 90% power and using a two-sided, 5% level of statistical significance, a total of 1086 couples (i.e. 543 couples in each arm) would be required to be able to detect an absolute difference of 8% (from 17% to 25%) and 9% (from 25% to 34%) in the healthy baby rate in the fresh-embryo transfer arm and the freeze-all arm, respectively.

It is a regulatory requirement for clinics in the UK to report live birth outcomes (including number, weight and gestation) after all embryo transfers; hence, loss to follow-up was not anticipated. Therefore, we did not take into account loss to follow-up for these sample size calculations.

It was anticipated that a proportion of those who consented may not reach randomisation (e.g. those who did not have three good-quality, day 3 embryos or who required all embryos to be frozen for medical reasons); therefore, a larger number of participants would need to be consented. It was anticipated that the number of participants who did not have three good-quality embryos would be 50% out of those consented.

Descriptive analysis

The flow of participants through each stage of the trial was summarised by trial arm. Demographic factors and clinical characteristics were summarised for all couples at trial entry, and separately for couples who delivered. Counts and percentages were reported for categorical variables, means [with standard deviations (SDs)] were reported for normally distributed continuous variables, and medians [with interquartile ranges (IQRs)] were reported for other continuous variables. No tests of statistical significance were performed and CIs were not calculated for differences between randomised arms on any baseline variable.

Primary analysis

All participants were analysed in the arms to which they were assigned, regardless of deviation from the protocol or treatment received under the ITT analysis principle. To perform the analyses for all outcomes on the ITT analysis population, the couple was included in the denominator once for all outcomes regardless of whether a pregnancy or a live birth occurred. Where this was a perinatal outcome, the woman was included once in the denominator. For neonatal secondary outcomes, the unit of analysis in the ITT analysis was the mother, and in cases of multiple pregnancy for which the infants’ outcomes differ, the worst outcome was reported.

Binary outcomes were analysed using a log-binomial regression model or a Poisson regression model with a robust variance estimator if the binomial model failed to converge. Linear regression was used for normally distributed continuous outcomes and quantile regression was used for skewed continuous outcomes. All comparative analyses were adjusted for the minimisation factors where possible. Fertility clinic was treated as a random effect in the models, where possible, and all other factors were treated as fixed effects. Both unadjusted and adjusted estimates are presented, but the primary inference is based on the adjusted analyses.

Comparative analyses entailed calculating the adjusted RR and 95% CI for the primary outcome, adjusted RRs and 99% CIs for all binary secondary outcomes, adjusted mean differences (MDs) and 99% CIs for normally distributed continuous secondary outcomes, or median differences and 99% CIs for skewed continuous secondary outcome variables. To account for the number of hypothesis tests performed, 99% CIs were used for all analyses of the secondary outcomes.

Customised birthweight centiles to calculate low weight for gestational age and high weight for gestational age were based on an existing, published, British model. 18

The following secondary outcomes were described only, and no formal statistical analysis comparing the arms was conducted:

• chronic hypertension, pregnancy-induced hypertension, pre-eclampsia and eclampsia

• most severe hypertensive disorder experienced (from least to worst severe: chronic hypertension, pregnancy-induced hypertension, pre-eclampsia and eclampsia).

Secondary analysis

The primary analysis for all primary and secondary outcomes was by ITT. Secondary analyses were performed to include clinically relevant denominators, such as the total number of women with a positive pregnancy test at 2 weeks ± 3 days after embryo transfer (for miscarriage), the total number of pregnant women with an ongoing pregnancy resulting in delivery (for pregnancy complications) and the total number of babies born (for birthweight and congenital anomalies). The adjusted analyses per total number of babies also accounted for the anticipated correlation in outcomes between multiple births. The rate of embryos not surviving after thawing (per embryo thawed) was reported for the intervention arm only.

Subgroup analysis

We performed subgroup analyses of the primary outcome on the following, as prespecified in the SAP:13

• fertility clinic

• woman’s age (at the time of start of treatment, i.e. ovarian stimulation): < 35, 35 to < 40 and ≥ 40 years

• blastocyst-stage compared with cleavage-stage embryo transfer

• single embryo transfer compared with multiple embryo transfer

• number of previous embryo transfers: 0 compared with ≥ 1 (the groups 0, 1–3 and ≥ 4 were prespecified, but were reduced to two groups in the analysis because of low frequencies).

The consistency of the effect of type of embryo transfer across specific subgroups of couples was assessed for the primary outcome using the statistical test of interaction, in addition to the adjusted model. The results are presented in forest plots, with RRs, 95% CIs and the results of the interaction test.

In addition, for those receiving frozen-embryo transfer, the primary outcome was summarised for the following subgroups using numbers and percentages only:

• natural cycles compared with hormone replacement cycles

• vitrification compared with slow freezing.

Additional analysis

The following prespecified analyses were carried out for the primary outcome only:

• per protocol – restricted to those who complied with the allocated intervention

• as treated – grouping couples according to the intervention that they actually received (i.e. those randomised to frozen-embryo transfer but who received fresh-embryo transfer were in the fresh-embryo transfer arm in this analysis)

• complier-average causal effect (CACE) analysis.

To assess the impact of non-compliance with the randomised allocation, that is women randomised to the frozen arm receiving fresh-embryo transfer (non-compliers), a CACE analysis was conducted. This analytic technique provides a robust estimate of the treatment effect among compliant participants. 19,20

The baseline characteristics of women randomised to the freeze-all arm were reported by compliance status, and the unadjusted event rate for the primary outcome was calculated for the observed compliers and non-compliers in the freeze-all arm. The CACE analysis assumed that the proportion of non-compliers in the fresh-embryo transfer arm (i.e. couples in the fresh-embryo transfer arm who would not have complied had they been randomised to the freeze-all arm) was the same as the proportion of non-compliers in the freeze-all arm. It also assumed that the event rate among the non-compliers in the freeze-all arm was the same as the event rate among the non-compliers in the fresh-embryo transfer arm. Applying these two assumptions, the unadjusted event rate for the primary outcome was calculated for the would-be compliers and would-be non-compliers in the fresh-embryo transfer arm. The unadjusted CACE RR with 95% CIs for the primary outcome was calculated using the event rates for compliant groups only (i.e. the observed compliers in the freeze-all arm and the would-be compliers in the fresh-embryo transfer arm). The CIs for the CACE-estimated RRs were calculated using bootstrapping methods. 21

Analysis of emotions questionnaire

The emotions questionnaires at randomisation and post embryo transfer captured responses to the STAI. 16 The response at randomisation was used as a covariate in an analysis of covariance (ANCOVA) model. Hypothesis testing investigated if there was a post-embryo transfer difference in the means of the two treatment arms after adjustment for responses at randomisation. To avoid bias, maximise the power of the study and obey the ITT principle, the missing-indicator method22 was used to replace missing baseline scores. This method replaces all missing baseline observations with the same value and an extra indicator variable is added to the model to indicate whether or not the value for that variable is missing. For any partially completed questionnaires, if one or two items were omitted, then the prorated score was obtained by calculating the mean weighted score for the completed items, multiplying by 20 and rounding to the next whole number. If three or more items were omitted, then the whole questionnaire was analysed as missing. Models were fitted separately for the female and male partners.

Adverse events

As per the protocols of the trial unit, an adverse event (AE) was defined as any untoward medical occurrence in a participant, which did not, necessarily, have to have a causal relationship with the intervention. Owing to the high incidence of AEs routinely expected in this patient population (e.g. abnormal laboratory findings, new symptoms), only those AEs identified as serious were recorded for the trial.

Serious adverse events

A SAE was any untoward medical occurrence that:

• resulted in death

• was life-threatening

• required participant hospitalisation or prolongation of existing hospitalisation

• resulted in persistent or significant disability/incapacity

• was a congenital anomaly/birth defect

• was an important medical event.

The term ‘severe’ was often used to describe the intensity (i.e. severity) of a specific event; however, the event itself may have been of relatively minor medical significance. This was not the same as ‘serious’, which was based on participant/event outcome or action criteria, usually associated with events that posed a threat to a participant’s life or functioning.

The term ‘life-threatening’ in the definition of serious refers to an event in which the participant was at risk of death at the time of the event; it does not refer to an event that hypothetically might have caused death had it been more severe. Medical and scientific judgement was exercised in deciding whether or not an AE was serious in other situations.

Foreseeable serious adverse events

Foreseeable SAEs were events that were expected in the patient population or as a result of the routine care/treatment of a patient. The events were foreseeable in women or couples undergoing IVF treatment and, therefore, did not need to be reported as SAEs. Data on foreseeable SAEs were collected on the eCRF as part of routine data collection.

The foreseeable events relating to the female partner or couple were:

• OHSS

• miscarriage

• hypertensive disorders of pregnancy

• antepartum haemorrhage

• GDM

• multiple pregnancy

• no embryos surviving thawing.

The foreseeable events relating to the baby when born were:

• low birthweight

• very low birthweight

• low weight for gestational age

• high weight for gestational age

• preterm delivery

• very preterm delivery.

Unforeseeable serious adverse events

An unforeseeable SAE was any event that met the definition of a SAE and was not detailed as foreseeable. The following unforeseeable SAEs were reported:

• maternal death

• stillbirth

• congenital anomaly detected antenatally or postnatally

• neonatal death.

Unforeseeable SAEs were reported up to 6 weeks post delivery. They were reported to the NPEU CTU as soon as possible after staff at the site became aware of the event. The SAEs were reported in one of the following ways:

1. Using the clinical database OpenClinica – only staff with access to OpenClinica could report SAEs in this way; site staff were required to print the OpenClinica SAE form and obtain the information and signature of the study clinician carrying out the causality assessment. The completed and signed SAE form was e-mailed or faxed to the NPEU CTU. NPEU CTU staff were automatically informed by e-mail of any SAEs reported electronically.

2. By completing a SAE form that was e-mailed or faxed to the NPEU CTU. Paper copies were available with the trial documentation to enable anyone to report a SAE. Guidance for the research site was provided on the paper SAE reporting form.

3. If it was not possible to report a SAE using the methods detailed in points 1 and 2, the unforeseeable SAE could be reported by telephone and the SAE form was completed by staff at the NPEU CTU.

If any additional information regarding the SAEs became available, this was detailed on a new SAE form and e-mailed or faxed to the NPEU CTU or reported electronically using OpenClinica. The SAE forms were sent to the sponsor by the NPEU CTU as soon as possible after they were received. The chief investigator assessed whether or not a SAE was ‘related’ (i.e. resulting from administration of any of the research procedures) and ‘unforeseeable’ in relation to those procedures. Any reports of related and unforeseeable SAEs were submitted to the following places, in line with the protocols of the trial unit: the North of Scotland Research Ethics Committee (which gave a favourable opinion of the study), the sponsor and the centre at which the SAE occurred within 15 working days of the chief investigator becoming aware of the event. All recorded SAEs were reviewed by the DMC at regular intervals. The chief investigator informed all principal investigators (PIs) concerned of relevant information that adversely affected the safety of the participants.

Trial Steering Committee

The role of the TSC was to provide overall supervision of the study. The TSC monitored the progress of the study and conduct, and advised on its scientific credibility. The TSC considered and acted, as appropriate, on the recommendations of the DMC and, ultimately, carried the responsibility for deciding whether or not the trial needed to be stopped on grounds of safety or efficacy.

The TSC consisted of an independent chairperson and two other independent members. The committee members were deemed to be independent if they were not involved in study recruitment and were not employed by any organisation directly involved in the study’s conduct. Representatives from Fertility Network (London, UK) (patient/public involvement groups), the chief investigator and other investigators/co-applicants were joined by observers from the NPEU CTU. A TSC charter was prepared in advance of and agreed on at the first TSC meeting to document how the committee operated.

Data Monitoring Committee

A DMC that was independent of the applicants and the TSC reviewed the progress of the trial at frequent intervals and provided advice on the conduct of the trial to the TSC, which reported to the HTA programme manager. The committee periodically reviewed study progress and outcomes. The timing and content of the DMC reviews were detailed in a DMC charter, which was agreed on at its first meeting.

Project Management Group

The study was supervised on a day-to-day basis by the PMG. This group reported to the TSC, which had overall responsibility for the conduct of the study. The core PMG met regularly (i.e. at least monthly). The Co-Investigators Group (CIG) met at regular intervals during the trial, and comprised all co-applicants and the members of the core PMG. The full membership of the committees is listed in Appendix 5.

Trial management

The trial co-ordinating centre was the NPEU CTU, University of Oxford, where the trial manager was based. The NPEU CTU was responsible for trial oversight; information technology (IT) system/functions, such as randomisation, clinical and administrative databases; all programming and statistical analyses; servicing both the DMC and the TSC; and, in collaboration with the chief investigator and the local research nurse, the general day-to-day running of the study, including the recruitment of sites and training of staff. A 24/7 (365 days per year) emergency helpline was available for out-of-hours queries related to the trial. The economic analysis was conducted at the University of Aberdeen (Aberdeen, UK).

Risk assessment and monitoring

A study risk assessment and monitoring plan was completed as part of the development of this study by the NPEU CTU. This risk assessment and monitoring plan was reviewed at regular intervals during the study to ensure that appropriate and proportionate monitoring activity was performed.

Recruitment by sites

Eighteen clinics across the UK signed up for the trial. Only 13 clinics randomised any participants, of which four randomised > 50 participants. The number of recruited participants from each clinic is presented in Table 2.

Baseline comparability of randomised arms

The demographic and clinical characteristics at trial entry are described for all couples in the ITT population by trial arm.

Clinical characteristics post randomisation

The clinical characteristics of the embryo and the endometrium, which were collected post randomisation, are described for the ITT population by trial arm in Table 5.

Of those randomised, 298 out of 307 participants underwent embryo transfer in the freeze-all arm and 303 out of 309 participants underwent embryo transfer in the fresh-embryo transfer arm. Most transfers (93.8%) were at the blastocyst stage. Overall, > 80% of participants in both arms underwent single embryo transfer; the remaining participants had two embryos, except for one participant, who had three embryos.

The endometrial appearance was recorded in only half of the cases in both arms; in 96.3% of these cases, it was triple layer, with a mean thickness of > 9.3 mm (SD 1.9 mm) in the freeze-all arm and 10.2 mm (SD 2.3 mm) in the fresh-embryo transfer arm.

The number of embryos remaining frozen (i.e. spare embryos) after the first embryo transfer was similar in both arms (median 2, IQR 1–4); 78.5% of couples had at least one remaining embryo frozen after embryo transfer.

The clinical characteristics post randomisation of those randomised to the freeze-all arm were similar whether or not they complied with the allocated intervention.

Post-randomisation characteristics of those who received frozen-embryo transfer

As shown in Table 6, 202 couples in the freeze-all arm and 21 couples in the fresh-embryo transfer arm received frozen-embryo transfer. Most couples had embryos frozen by vitrification (88.1% in the freeze-all arm and 95.2% in the fresh-embryo transfer arm). Most embryos were frozen at the blastocyst stage. The median number of embryos frozen among those who underwent frozen-embryo transfer was 4 (IQR 3–6). The median number of embryos thawed was one. One-fifth (19.8%) of those randomised to the freeze-all arm and 14.3% of those randomised to the fresh-embryo transfer arm who underwent frozen transfer had more than one embryo thawed. Very few embryos were thawed and discarded (i.e. this occurred in a total of 19 couples). The most common method used for endometrial preparation was artificial cycle with oestrogen and progesterone, or downregulation with gonadotropin-releasing hormone (GnRH) agonist.

Intention-to-treat analysis

The ITT analysis (Table 7) showed that the healthy baby rate (i.e. term singleton live birth with appropriate weight for gestation) was 20.3% in the freeze-all arm and 24.4% in the fresh-embryo transfer arm (p = 0.28). There was no statistical difference with/without adjustment of confounding factors (adjusted RR 0.84, 95% CI 0.62 to 1.15). The proportion of singletons born was 27.7% in the freeze-all arm and 34.0% in the fresh-embryo transfer arm. The proportion of babies born at term was 25.4% in the freeze-all arm and 30.2% in fresh-embryo transfer arm. Similarly, the proportion with an appropriate weight for gestation was 22.5% in the freeze-all arm and 26.9% in the fresh-embryo transfer arm.

Sensitivity analysis by compliance status of the freeze-all arm

When the analysis for the primary outcome was undertaken by compliance status of those randomised to the freeze-all arm, there was no difference in the outcome healthy baby rate between those who complied and those who did not comply with the allocated intervention (21.3% vs. 20.0%, respectively) (Table 8).

Sensitivity analysis: complier-average causal effect

The CACE RR was 0.77 (0.204/0.264). This was calculated for participants with no missing primary outcome data.

The values in Table 9 are calculated for the would-be non-compliers and would-be compliers of the fresh-embryo transfer arm, assuming the same non-compliance rate and event rate as the non-compliers of the freeze-all arm. 20 The CACE RR suggests that there is no difference in the healthy baby rate between the two arms.

Exploratory analysis on primary outcome

In both arms, the healthy baby rate did not differ depending on whether the analysis was restricted to those who had received the allocated intervention (p = 0.45) or the analysis was undertaken as treated (p = 0.59) (Table 10).

Subgroup analysis

The prespecified subgroup analysis for the primary outcome of healthy baby rate was undertaken based on the age of the female partner, number of previous embryo transfers, stage and number of embryos, and fertility clinic. As shown in Figure 10, there was no statistical difference in the healthy baby rate in different age groups (< 35, 35 to < 40 and > 40 years), number of previous embryo transfers (0 or ≥ 1), whether the transfer was undertaken at the cleavage or blastocyst stage, or whether one or two embryos were transferred (Table 11). There was a difference in the healthy baby rate between clinics; however, this is unlikely to be meaningful because of the very small numbers recruited by most clinics. Exploratory analyses were undertaken for method of endometrial preparation and type of freezing. Most frozen-embryo transfers were hormonally mediated and most embryos were frozen by vitrification.

FIGURE 10.

Subgroup analysis of the primary outcome. Adjusted for minimisation factors at randomisation. p-values from test of heterogeneity. Reproduced from Maheshwari et al. ,23 Elective freezing of embryos versus fresh embryo transfer in IVF: a multicentre randomized controlled trial in the UK (E-Freeze), Human Reproduction, 2022, deab279, by permission of European Society of Human Reproduction and Embryology.

Reproduced from Maheshwari et al. ,23 Elective freezing of embryos versus fresh embryo transfer in IVF: a multicentre randomized controlled trial in the UK (E-Freeze), Human Reproduction, 2022, deab279, by permission of European Society of Human Reproduction and Embryology.

Maternal safety: ovarian hyperstimulation syndrome

The risk of OHSS was lower in the freeze-all arm (3.6%) than in the fresh-embryo transfer arm (8.1%); however, this difference did not reach statistical significance (RR 0.44, 99% CI 0.15 to 1.30). The severity of OHSS was mild to moderate in the freeze-all arm, whereas six (1.9%) women had severe OHSS in the fresh-embryo transfer arm (Table 12).

Measures of clinical effectiveness

Table 13 presents the measures of clinical effectiveness.

There was no significant difference in the live birth rate between the freeze-all arm and the fresh-embryo transfer arm (28.3% vs. 34.3%; adjusted RR 0.83, 99% CI 0.65 to 1.06).

The singleton baby rate was 27.7% in the freeze-all arm and 34.0% in the fresh-embryo transfer arm, but the difference was not statistically significant (adjusted RR 0.82, 99% CI 0.64 to 1.06).

The rate of singleton babies born at term was 25.4% in the freeze-all arm and 30.2% in the fresh-embryo transfer arm. There was no statistically significant difference (adjusted RR 0.85, 99% CI 0.67 to 1.08). The details of gestational age were missing for one baby in the fresh-embryo transfer arm.

The rate of singleton babies with an appropriate weight for their gestational age was 22.2% in the freeze-all arm and 26.9% in the fresh-embryo transfer arm. There was no statistically significant difference (adjusted RR 0.83, 99% CI 0.55 to 1.26). Details were missing for one baby in each arm.

The rate of positive pregnancy tests was 45.3% in the freeze-all arm and 49.8% in the fresh-embryo transfer arm. There was no statistically significant difference (adjusted RR 0.91, 99% CI 0.77 to 1.08).

The clinical pregnancy rate was 33.9% in the freeze-all arm and 40.1% in the fresh-embryo transfer arm. There was no statistically significant difference (adjusted RR 0.85, 99% CI 0.65 to 1.11).

Complications of pregnancy and delivery

Tables 14 (ITT analysis) and 15 (clinically relevant denominators) present the complications in pregnancy and delivery. All complications in pregnancy and delivery are described in text with the clinically relevant denominator.

Measures of effectiveness of the process of freezing embryos

The following measures were used to determine the effectiveness of the freezing process:

• the total number of embryos frozen, thawed and transferred for all randomised couples

• the proportion of thawed embryos that were then transferred for all randomised couples

• no embryos surviving after thawing, leading to no embryo transfer.

To transfer 248 embryos, 280 embryos had to be thawed (i.e. 88.6% of embryos were suitable for transfer after being thawed) (Table 16).

Three couples in the freeze-all arm did not have any embryos to transfer because their embryos did not survive the freezing–thawing process (Table 17).

Evaluation of emotional status

The couples’ emotional state was assessed using the STAI questionnaire. 16 Both partners completed the questionnaire at two time points (at consent and at embryo transfer).

Table 18 reports the data from the STAI questionnaire.

Study design and participants

Details of the trial design are provided in the published protocol1 and in Chapter 2. The economic analysis was based on all women randomised, except for three post-randomisation exclusions, and follows the same ITT principle as the statistical analysis in Chapter 3.

Cost and outcome assessment

Costs and outcomes were assessed from post randomisation up to and including delivery, using the trial eCRFs completed by the research staff post embryo transfer, during early pregnancy (i.e. 6–8 weeks’ gestation), at the 12- and 28-week follow-ups and 6 weeks post delivery, and the paper-based economic questionnaire completed by couples at embryo transfer. Health-care utilisation data collected post embryo transfer, during early pregnancy, at 28 weeks’ gestation and at 6 weeks post delivery were used in the economic analysis. Questions related to participant time and travel costs incurred during treatment were included in the embryo transfer economic questionnaire (see Report Supplementary Material 2). All costs are reported in 2018/19 Great British pounds. Adjustments for inflation were applied to unit costs, where necessary, using the NHS Cost Inflation Index (NHSCII). 25

Assessment of health service costs

Given that the economic evaluation seeks to inform the efficient allocation of scarce health-care resources, the base-case analysis adopted the NHS and Personal Social Services perspective. The effect on patient time and travel costs was considered separately as a secondary analysis.

Cost of the primary intervention

The cost of the embryo transfer procedure was estimated based on the health service resource use observed for each arm. The additional cost of freezing all embryos, followed by frozen-embryo transfer, was estimated from resource use data recorded in the eCRF post embryo transfer for each participant who received frozen-embryo transfer. The eCRF captured the number of monitoring visits, blood tests and ultrasound scans prior to embryo transfer, and the method of endometrial preparation. The cost of preparing frozen embryos was also included for the participants who received frozen-embryo transfer. The cost of embryo freezing was applied to all participants who received frozen-embryo transfer, as well as to the participants who had their remaining embryos frozen following a fresh-embryo transfer. The unit costs that were used to value the resource use associated with the intervention are reported in Appendix 5.

The costing approach assigned costs to each component of resource use to capture patient-level variation in costs. Each resource use item was mapped to an appropriate Healthcare Resource Group (HRG), where available, and costed using the relevant NHS reference cost. 26 The monitoring visit prior to transfer was assumed to be a 30-minute session led by a nurse. 25 The reported methods of endometrial preparation were valued based on routine regimens obtained from clinical advice and unit costs from the British National Formulary (BNF). 27 For embryo freezing and the preparation of frozen embryos, the time spent by an embryologist for each procedure was assumed to be 1 hour based on clinical advice. 25

Costs of ovarian hyperstimulation syndrome

The clinical management costs associated with OHSS and pregnancy were valued using the NHS Reference Costs 2018/19. 26 The resource use for managing OHSS was obtained from the eCRFs post embryo transfer and during early pregnancy. The non-elective inpatient stay cost for a 1-day stay was valued using the inpatient short stay cost, whereas inpatient stays of > 1 day were valued using the inpatient long stay cost, adjusted for length of stay using the excess bed-day cost if the stay was longer than the average length of stay. As some of the relevant data needed for the adjustment of inpatient long-stay costs were not available in the NHS Reference Costs 2018/19, the average length of stay for non-elective long stays and the non-elective excess bed-day cost were obtained from the NHS Reference Costs 2017/1828 and inflated. This approach was undertaken for all costs associated with inpatient stays in the analysis.

Costs of pregnancy outcomes

Data on pregnancy outcomes (e.g. miscarriage, biochemical pregnancy, ectopic pregnancy, pregnancy of unknown location and termination) were obtained from the eCRF during early pregnancy, at the 12- and 28-week follow-ups. Miscarriage and ectopic pregnancy were costed by applying the average reference cost per case, whereas termination cost varied by gestation. It was assumed that, on average, a minimum of one ultrasound scan and three beta-human chorionic gonadotropin (βHCG) blood tests would be required for biochemical pregnancy and pregnancy of unknown location. 29

Costs of antenatal care

The antenatal care (ANC) costs were based on the NHS Reference Costs 2018/1926 and varied according to the maternal complications reported. Primary and secondary care contacts during pregnancy and any complications were captured in the 28-week follow-up CRF and post-delivery eCRF. It was assumed that participants had a community midwife visit following an early pregnancy scan (EPS). The complications were grouped using the Code to Group: HRG4+ 2019/20 Local Payment Grouper30 workbook to determine the corresponding HRG. Episodes of care were costed by applying either the day-case reference cost for an inpatient day care visit or the non-elective inpatient cost (stay of ≥ 1 day), adjusted for length of stay using the excess bed-day cost.

For antenatal ultrasound scans, the resource use was obtained from the eCRFs during early pregnancy, at the 28-week follow-up and post delivery. The costs of ultrasound were determined using the NHS reference cost26 and were varied in accordance with the standard recommendation for antenatal ultrasound scanning of one scan throughout the pregnancy. Any additional ultrasound scans were costed as non-routine.

Costs of delivery

The delivery method was obtained from the post-delivery eCRF and valued based on the NHS reference cost. 26 The costs varied according to the delivery mode, onset method (with or without induction) and length of inpatient stay for delivery.

Participant travel and time costs

Participant costs associated with travelling to and from appointments and the time spent for a clinic visit during treatment were estimated from post randomisation to embryo transfer. Travel costs were estimated based on the number of visits to the clinic and travel expenses per visit or distance travelled by car, obtained from the economic questionnaire. Travel costs were estimated based on the expenses reported for participants who travelled using public transport and taxis, whereas the costs for travel by car were calculated using the mileage reported and the private car rate per mile of 45p per mile published by Her Majesty’s Revenue and Customs (HMRC). 31 Time costs, which account for time lost from productive activities, were estimated from the economic questionnaire based on the time taken to visit the clinic. Time taken away from normal productive activities was estimated in hours, and appropriate unit costs were used to estimate the opportunity cost of time. Gross age- and sex-specific wage rates obtained from the Annual Survey of Hours and Earnings (ASHE),32 published by the Office for National Statistics, were used to cost the time lost from paid employment. To estimate the cost associated with time lost for the accompanying male partner, the partner’s age was assumed to be the same as that of the participant. The cost of time lost from unpaid work was estimated using the value of unpaid work published by the Office for National Statistics. 33 Forgone leisure time was valued using the current value of non-working time, available from the Department for Transport. 34 The unit costs that were used to value the time lost are reported in Appendix 5.

Outcome measures

Effectiveness for the economic evaluation was measured in terms of the number of healthy babies born and the secondary clinical outcomes of the trial. Secondary outcomes included live births, maternal safety outcomes (i.e. OHSS), pregnancy outcomes, complications of pregnancy and delivery and adverse birth outcomes.

Aggregating costs and effects

Resource use, costs and health outcome data were summarised by trial arm, based on the participants who had the event of interest and by ITT. The IVF costs were broken down into the following categories: freezing of embryo, endometrial preparation, embryo transfer, monitoring visits prior to frozen-embryo transfer, blood tests prior to frozen-embryo transfer, transvaginal ultrasound scans prior to frozen-embryo transfer and preparation of frozen embryos. All cost elements were summed over the follow-up period (up to and including the cost of delivery) to estimate a total NHS cost per patient.

Cost data were fully present with respect to resource use associated with embryo transfers and OHSS; however, given that no further resource use data were collected for 15 participants who did not undergo embryo transfer, these participants were conservatively assigned no further treatment costs other than embryo freezing and thawing, where applicable. This assumption favours the freeze-all approach because three patients in the freeze-all arm may have incurred further work-up costs prior to cancellation of their embryo transfer owing to the embryos failing to survive the thawing process. Therefore, we also conducted a sensitivity analysis in which multiple imputation was used to impute the pre-embryo-transfer monitoring visit, blood test and scan costs for these three participants.

Elements of resource use data were also missing for a small number (n = 13) of resultant pregnancies. Two of these participants who had uncomplicated pregnancies were missing entries for the number of antenatal midwifery, outpatient and inpatient attendances between 12 and 28 weeks’ gestation only. These two participants were included in the complete-case analysis by assigning a zero cost to these elements; based on a comparison with similar participants with no missing data, it was considered plausible that these elements were missing because the costs were zero and they were not missing at random. Thus, for the complete-case analysis, we retained 616 participants for the cost-effectiveness analysis using treatment plus OHSS costs, and 605 participants for the analysis of total NHS costs (inclusive of ANC and delivery care).

Recognising the uncertainty in the above approach, we also conducted a sensitivity analysis using multiple imputation to impute plausible values for all 13 participants with missing ANC and delivery care cost elements, and the work-up costs for the three cancelled frozen-embryo transfer cycles where missingness could not be ascertained. We did not, however, impute missing values for the primary clinical effectiveness outcome, allowing 614 participants to be included in the cost-per-healthy-live birth multiple-imputation analysis and 616 participants to be included in the cost-per-live birth multiple-imputation analysis.

Within-trial cost-effectiveness and cost-consequence analysis

The within-trial economic analysis was performed on an ITT basis using individual participant-level data from the trial. The cost differences were summarised by ITT against the primary and secondary clinical outcomes using a cost–consequence balance sheet. All analyses were performed using Stata^®, version 15 (StataCorp LLC, College Station, TX, USA).

For the within-trial cost-effectiveness analysis, two cost categories were used: treatment costs (including post-randomisation preparation and embryo transfer costs); and OHSS costs and full NHS costs, which, in addition to treatment and OHSS costs, included pregnancy and delivery costs. Cost-effectiveness was expressed in terms of the incremental cost per healthy baby and per live birth of freezing all embryos, followed by frozen-embryo transfer, compared with fresh-embryo transfer. The incremental treatment cost (inclusive of OHSS costs) per additional healthy baby born was estimated as the primary measure of cost-effectiveness, as the management and incidence rate of complications per pregnancy were similar between arms, with no statistically significant differences. Generalised linear models (GLMs), with adjustment for minimisation factors, were used to estimate MDs in costs and effects between the trial arms. For cost outcomes, a gamma family with log-link was selected using the modified Park test, Pearson’s correlation, Pregibon link and modified Hosmer–Lemeshow tests. 35 For the effectiveness outcomes (i.e. healthy babies and live births), a Poisson family with log-link was used, as per the statistical analysis. Recycled predictions were used to recover adjusted mean values by trial arm, as well as the incremental differences between trial arms. 35 The incremental cost-effectiveness ratio (ICER) for frozen-embryo transfer compared with fresh-embryo transfer was calculated as the difference in mean cost divided by the difference in mean effect. The variance surrounding the joint incremental costs and effects was characterised using non-parametric bootstrapping (i.e. 1000 iterations), with simulated output summarised graphically using cost-effectiveness plane and cost-effectiveness acceptability curves. For the sensitivity analysis using multiple imputation, the imputation model used chained equations with predictive mean matching (k = 5) to generate five imputed data sets (m = 5) nested within each bootstrapped resample (n = 1000). The imputation model included all of the missing cost elements, treatment allocation, women’s age and indicators for pregnancy and live birth as auxiliary variables.

Sensitivity analysis

The sensitivity analysis focused on the costing methodology for the initial interventions and missing data. A sensitivity analysis using multiple imputation chained equation was performed to assess the impact of missing data (including participants with partial missing data) on the robustness of the cost-effectiveness findings of the incremental NHS cost per baby born. The trial-based cost-effectiveness analysis was also conducted using pre-defined subgroups for women’s age at ovarian stimulation (< 35 years vs. ≥ 35 years).

Modelling of subsequent frozen-embryo transfers

Although the within-trial analysis is useful for informing cost-effectiveness in the short term, a longer time horizon is required to determine the relative cost-effectiveness of the alternative embryo transfer strategies in the context of routine IVF practice, whereby the subsequent transfer of the remaining frozen embryos can take place if the initial fresh-embryo transfer or frozen-embryo transfer fails to achieve a live birth. Therefore, a Markov model was developed to simulate progression through the subsequent transfer of the remaining frozen embryos for those failing to achieve live birth in both trial arms. This compares the policy of offering a full cycle of IVF using the freeze-all approach and a full cycle of IVF using the routine approach, in which a fresh embryo is used for the index transfer and the remaining frozen embryos are replaced in subsequent associated transfers. Although NICE3,36 recommends up to three full cycles of IVF treatment, with remaining good-quality embryos frozen and, subsequently, transferred as part of the same cycle, implementation of this guidance is low in England. 37 Furthermore, the most efficient approach to the first full IVF cycle is also likely to offer the best value for money if repeated full IVF cycles are permitted. Therefore, the economic model considered one full IVF cycle. The index transfer in the model was parametrised using event rates and costs derived from the ITT analysis of the E-Freeze trial data, including post-randomisation costs for the initial embryo transfer, OHSS costs and ongoing costs associated with pregnancies. The costs and outcomes associated with subsequent frozen-embryo transfers were extrapolated using several assumptions (see Figure 11 for details). The state transition diagram for the model is provided in Figure 11. The details of the derived model parameter inputs are provided in Results.

FIGURE 11.

State transition diagram for the Markov model. a, Tunnel states consisting of nine 4-week temporary states to capture progression through pregnancy. b, States were multiplied by 6 to allow for up to six subsequent frozen-embryo transfers.

The model structure is replicated for the fresh-embryo transfer and freeze-all arms of the E-Freeze trial and runs on a fixed 4-week Markov cycle. Couples start the model in the ‘index fresh embryo transfer’ or the ‘index frozen-embryo transfer’ state based on the observed distribution in the respective arms of the E-Freeze trial. Following this, couples proceed to embryo transfer and either achieve a positive pregnancy test result or fail to become pregnant. A small proportion also receive no transfer, as observed in both arms of the trial. Those who fail to become pregnant move to either the ‘not pregnant – frozen embryos remaining’ or the ‘cycle complete – no child’ state, depending on the availability of remaining frozen embryos. Those who become pregnant move to the ‘pregnant’ state, which is a tunnel state consisting of nine temporary states that can be occupied for one 4-week cycle only. This captures progression and outcomes through the stages of pregnancy. During pregnancy, women incur ANC costs relevant to their stage of pregnancy, as observed by trial arm in the E-Freeze trial. Women can also lose their pregnancy; in these situations, they incur the cost of this event and transition to the ‘not pregnant – frozen embryos remaining’ or ‘cycle complete – no child’ state.

For those who carry their pregnancy to ≥ 24 weeks’ gestation, birth outcomes are modelled as observed by trial arm in the E-Freeze trial, and the relevant costs of delivery care are applied. To fit with the model structure and cycle length, preterm deliveries were categorised into three mutually exclusive categories as follows: < 28 weeks’, 28–32 weeks’ and 33–36 weeks’ gestation. For those transitioning to the ‘not pregnant – frozen embryos remaining’ state, the observed numbers of remaining embryos by treatment allocation arm in E-Freeze were used, in conjunction with several assumptions informed by external data,38 to inform transitions through subsequent frozen transfers of remaining embryos (see Figure 11). The model states representing subsequent embryo transfers, pregnancy following subsequent transfer, and failure to achieve pregnancy following subsequent transfer (see Figure 11), are multiplied by six to allow for up to six frozen transfers for those failing to achieve a live birth. Once a live birth has been achieved, no further transfers are modelled.

For the key outcomes that drive differences in the live birth rate and healthy baby rate between the trial arms, the model utilises relative risks (and 95% CIs) for frozen-embryo transfer and fresh-embryo transfer (ITT) applied to the baseline probabilities of events observed in the fresh-embryo transfer arm of E-Freeze [positive pregnancy following embryo transfer, RR 0.91 (95% CI 0.77 to 1.08); any pregnancy loss following a positive pregnancy test, RR 1.18 (95% CI 0.76 to 1.84); delivery prior to 33 weeks’ gestation for ongoing pregnancies at 24 weeks’ gestation, RR 0.60 (95% CI 0.15 to 2.34); and delivery between 33 and < 37 weeks’ gestation for ongoing pregnancies at end of week 32 of gestation, RR 1.17 (95% CI 0.39 to 3.52)]. The relative risk for OHSS is also applied (0.44 95% CI, 0.15 to 1.3). For other parameters, trial arm-specific event counts and costs are derived from the observed E-Freeze outcome and cost data. The details of these derived model parameter inputs are provided in Results.

Following internal validation of the model output for the index embryo transfer against the trial-based cost-effectiveness findings, the model was used to extrapolate the costs and consequences of transferring the remaining frozen embryos. For this, the distribution of remaining embryos by trial arm in the E-Freeze trial for those not achieving live birth was used to estimate the proportion of women expected to have remaining embryos available for a second, third, fourth, fifth, sixth and seventh embryo transfer attempt. This required two key assumptions:

1. The remaining frozen embryos are thawed and replaced one at a time.

2. The transfer of each remaining frozen embryo has an equal chance of resulting in pregnancy and live birth, regardless of the approach to the first transfer and the total number of embryos remaining frozen.

For all subsequent frozen-embryo transfers, the event rates and costs were based on those observed for the index transfer in the frozen arm of the E-Freeze trial. The following adjustments were also made:

1. It was assumed that not all couples with remaining frozen embryos available after each failed transfer would return to use them (i.e. a discontinuation rate was applied after each failed transfer).

2. The survival rate for thawed embryos was set to 88%, in line with data from the E-Freeze trial.

3. The chance of pregnancy with subsequent frozen-embryo transfers was assumed to be lower, relative to the chance of pregnancy in the index transfer in the freeze-all arm of the E-Freeze trial.

4. A duration of three model cycles (i.e. 12 weeks) was assumed between each embryo transfer attempt.

The adjustments described in 1 and 3 above were based on an Australian population-based study reporting on the cumulative live birth rate (CLBR) following a freeze-all strategy compared with a fresh-embryo transfer strategy,38 including a good-prognosis group with a similar number of embryos available and a similar live birth rate as those of the E-Freeze cohort. Following each failed embryo transfer and using the reported data, a discontinuation rate was calculated as the proportion of women with remaining frozen embryos who did not return for a further transfer. This was taken as the average across the fresh-embryo transfer and freeze-all groups in the best prognosis subgroup of the Australian study. The relative adjustment to the pregnancy rate in subsequent frozen-embryo transfers compared with index frozen-embryo transfer cycles was derived using data reported for the freeze-all group of the best-prognosis subgroup of the Australian study. No more than six subsequent frozen-embryo transfers were incorporated in the model, as < 5% of the E-Freeze cohort had more than six embryos remaining frozen following failure of the index transfer, and the increase in the expected live birth rate from allowing a sixth transfer was < 0.1% compared with that from allowing up to five subsequent frozen-embryo transfers.

The model also included final birth outcome states, as it was originally intended to extrapolate longer-term costs and health outcomes per child born following treatment. This was in anticipation of a potential trade-off between maximising the live birth rate and maximising the health baby rate. However, the E-Freeze trial data indicated that, as a proportion of live births, the healthy baby rate was the same (71%) in both arms. Therefore, a decision was made not to include these in the current model. Although there were differences in the percentages of live births falling into the categories of preterm delivery and abnormal birthweight, these differences were based on very small numbers and, therefore, are highly uncertain. This would consequently translate into a high degree of uncertainty around any modelled difference in expected child health service costs or health outcomes.

Model-based analysis

The model was run probabilistically over a time horizon of 5 years, using 1000 random draws from distributions assigned to each clinical and cost input parameter. The results of the model were assessed in terms of the incremental cost per healthy baby and the incremental cost per live birth. Costs incurred beyond year 1 in the model were discounted using a rate of 3.5%, in line with the NICE reference case,39 but birth outcomes were not discounted because there is no guidance on the discount rate to apply to these outcomes in the context of fertility treatment. The probabilistic model output was summarised using cost-effectiveness scatterplots and cost-effectiveness acceptability curves, indicating the probability of each strategy being preferred on grounds of cost-effectiveness by increasing levels of societal willingness to pay per additional healthy baby or additional live birth. A further deterministic sensitivity analysis was undertaken to assess the robustness of the model findings to key structural uncertainties and costing assumptions.

Health service resource use and costs

Table 19 summarises NHS resource use by trial arm and Table 20 summarises costs by trial arm. A total of 601 participants received an embryo transfer, with 117 participants not receiving their allocated treatment. As a result, 223 participants received frozen-embryo transfer, including 21 who were allocated to fresh-embryo transfer. Prior to frozen-embryo transfer, participants had additional monitoring visits, blood tests and transvaginal ultrasound scans. The crossover participants had slightly fewer monitoring visits and transvaginal scans, but more blood tests. The resource use associated with IVF translated to an average post-randomisation treatment cost of £1538 and £1216 in the freeze-all and fresh-embryo transfer arms, respectively, giving an unadjusted cost difference of £322. Compared with frozen-embryo transfer, a larger number of participants developed OHSS in the fresh-embryo transfer arm (8% vs. 4%, respectively), and participants in the fresh-embryo transfer arm had more outpatient visits and inpatient day case visits and a longer inpatient length of stay than participants in the freeze-all arm. The mean OHSS costs by treatment allocation were £17 for frozen-embryo transfer and £201 for fresh-embryo transfer.

Of the 293 participants with a positive pregnancy test result at 2 weeks post embryo transfer, 100 suffered pregnancy loss. The number with pregnancy loss was slightly larger in the freeze-all arm, at 52 (37%), than in the fresh-embryo transfer arm, at 48 (31%). The mean cost of pregnancy loss was £89 and £75 for frozen-embryo transfer and fresh-embryo transfer, respectively. As a result, 193 participants had a live birth delivery, with more babies delivered in the fresh-embryo transfer arm than in the freeze-all arm. No obvious, notable differences were observed in the resource use associated with ANC in participants with ongoing pregnancy or in the delivery costs for those achieving live birth. The mean cost of ANC and delivery was higher in the fresh-embryo transfer arm (see Table 22) owing to the higher pregnancy rate and the smaller proportion of participants experiencing pregnancy loss.

The resource use from randomisation to delivery translated to a total, average, unadjusted NHS cost of £3431 in the freeze-all arm and £3574 in the fresh-embryo transfer arm, resulting in an unadjusted difference of £143. A breakdown of direct medical costs per participant experiencing each type of resource use event is presented in Appendix 6.

Cost-effectiveness analysis results

The cost-effectiveness analysis was conducted using the complete-cases data set, and the incremental cost per baby born is presented in Table 22. The adjusted mean treatment cost (including OHSS) per participant was £1402 for the fresh-embryo transfer arm and £1573 for the freeze-all arm, resulting in an adjusted MD of £171. The mean treatment cost was significantly higher in the freeze-all arm than in the fresh-embryo transfer arm. The mean adjusted healthy baby rate was 0.242 in the fresh-embryo transfer arm and 0.204 in the freeze-all arm, producing an adjusted MD of –0.039 (95% CI –0.104 to 0.023) in favour of the fresh-embryo transfer arm. The cost-effectiveness scatterplot, using 1000 bootstrapped iterations, in Figure 12 shows that frozen-embryo transfer was more costly than fresh-embryo transfer in the majority of iterations (≈ 99%). In addition, the healthy baby rate was lower for frozen-embryo transfer than fresh-embryo transfer in 89% of iterations, in line with the MD, in effect favouring fresh-embryo transfer over frozen-embryo transfer. Thus, frozen-embryo transfer was dominated by fresh-embryo transfer. Based on the cost-effectiveness acceptability curves in Figure 13, frozen-embryo transfer had a low chance of being cost-effective at all willingness-to-pay thresholds. Similar findings were noted for the cost-effectiveness analysis using live birth as the measure of effect. (see Table 24, and Figures 12 and 13).

FIGURE 12.

Trial-based incremental cost-effectiveness scatterplots for frozen-embryo transfer vs. fresh-embryo transfer. (a) Treatment costs: healthy baby; (b) treatment costs: live birth; (c) NHS costs: healthy baby; and (d) NHS costs: live birth.

FIGURE 13.

Trial-based cost-effectiveness acceptability curves for frozen-embryo transfer vs. fresh-embryo transfer. (a) Treatment costs: healthy baby; (b) treatment costs: live birth; (c) NHS costs: healthy baby; and (d) NHS costs: live birth.

When the total NHS costs were used in the analysis, the adjusted mean NHS cost per participant was £3551 for the fresh-embryo transfer arm and £3454 for the freeze-all arm, resulting in an adjusted MD of £97 (fresh-embryo transfer vs. freeze all). The mean treatment costs for fresh-embryo transfer were slightly higher than those for frozen-embryo transfer because of higher ANC and delivery costs, driven by the higher pregnancy and delivery rates. The ICER for frozen-embryo transfer compared with fresh-embryo transfer, representing cost savings per unit reduction in effect, came to £2425 and £1742 for one less healthy baby and one less live birth, respectively.

Sensitivity and subgroup analyses results

As the cost of transvaginal scans was the main driver of the increased treatment costs for the freeze-all arm compared with the fresh-embryo transfer arm, several sensitivity analyses were conducted using alternative costing methodologies for the scan (Table 23). In addition, multiple imputation was conducted to assess the impact of the base-case assumptions around missing cost data (Table 24). Prespecified subgroup analyses based on age and the number of previous embryo transfers were also conducted (Table 25). The cost-effectiveness scatterplots and cost-effectiveness acceptability curves are shown in Appendix 8.

Subgroup analyses

Table 25 reports the results of subgroup analyses by age. In women aged ≥ 35 years, the healthy baby rate was slightly higher in the freeze-all arm, although the difference was not statistically significant, leading to a positive ICER of £24,308 per additional healthy baby. Based on the results of the non-parametric bootstrap, frozen-embryo transfer was found to have a 46.5% chance of being the preferred intervention at the willingness-to-pay threshold of £20,000 per healthy baby (see Appendix 8). The cost-effectiveness findings for other subgroups remained less favourable to frozen-embryo transfer than to fresh-embryo transfer.

Model-based sensitivity analysis

Table 28 presents the results of several key sensitivity analyses around the model-based estimates of cost-effectiveness, using incremental treatment cost (including OHSS) per additional healthy baby as the measure of cost-effectiveness. Assuming no discontinuation among those eligible for subsequent frozen-embryo transfers and applying more conservative costs for transvaginal scans and monitoring prior to frozen-embryo transfer, fresh-embryo transfer remains, on average, more effective and less costly than thawed frozen-embryo transfer. Cost-effectiveness acceptability curves for the model-based sensitivity analysis are provided in Appendix 10.

Live birth rate

The live birth rate in our trial was 28.3% in the frozen-embryo transfer arm and 34.3% in the fresh-embryo transfer arm. Although there were no statistically significant differences between the two arms, these figures were similar to the live birth rates reported by the two trials from Europe45,46 and the trial from Viet Nam. 42 However, the rates are much lower than those of both of the trials reported from China. 43,44 This could be because the trials reporting from China had an upper age limit of 35 years and, therefore, included patients with a better prognosis.

The combined data from these five trials showed no difference in the outcome of live birth between the two arms, which is similar to the results of the E-Freeze trial (Figure 16).

FIGURE 16.

Meta-analysis of existing trials,42–46 with and without the E-Freeze trial, for live birth rate.

Based on nearly 5246 women randomised to fresh-embryo transfer compared with frozen-embryo transfer, an aggregate-data meta-analysis did not seem to favour either fresh-embryo transfer or a strategy of freezeing all embryos, followed by frozen-embryo transfer, with a combined RR of 1.01 (99% CI 0.78 to 1.31). An updated meta-analysis, including randomised data from an additional 616 women from the E-Freeze trial, does not appear to result in a convincing change in the direction, size or precision of the overall effect, with a combined RR of 0.97 (95% CI 0.77 to 1.24).

Ovarian hyperstimulation syndrome

The risk of OHSS reported in the E-Freeze trial was 3.6% in the freeze-all arm and 8.1% in the fresh-embryo transfer arm. Most cases of OHSS were mild, with moderate to severe OHSS in only 1.6% in the freeze-all arm and 5.8% in the fresh-embryo transfer arm. This is higher than the rate quoted for moderate to severe OHSS in national statistics. 6 This difference is because of better ascertainment in the trial setting; it is well known that cases of OHSS are not reported in clinical practice. 49

The total OHSS figures were higher than those reported by other trials. This could be because we also reported mild OHSS, whereas other trials reported only moderate and severe OHSS. Combined data from the five trials reporting on OHSS42–46 showed a statistically significant reduction in OHSS when all embryos were frozen compared with fresh-embryo transfer. The addition of data from the E-Freeze trial did not change the direction or magnitude of these figures, but increased the precision (Figure 17).

FIGURE 17.

Meta-analysis of existing trials,42–46 with and without the E-Freeze trial, for OHSS.

Miscarriage

The risk of miscarriage per couple randomised was 14.3% in the frozen-embryo transfer arm and 12.9% in the fresh-embryo transfer arm. The corresponding figures for risk of miscarriage per pregnancy were 31.7% and 26.0%, respectively. Miscarriage rates were slightly higher in the E-Freeze trial than in other studies (ranging from 9.9% to just over 25% per pregnancy42–45). We were unable to explain this higher rate of miscarriage. To explore the higher risk of miscarriage, we undertook a post hoc analysis of the miscarriage rate per centre; there were no significant differences between the five clinics that contributed most data. The combined data from other trials showed no difference in miscarriage rates between trial arms, which is similar to the results reported in the E-Freeze trial. 42–45 Wong et al. 46 did not report on miscarriage per pregnancy; hence, their data are not included in the meta-analysis graph. The addition of E-Freeze to existing data does not change the direction, magnitude or precision of effect (Figure 18).

FIGURE 18.

Meta-analysis of existing trials,42–45 with and without the E-Freeze trial, for miscarriage rate.

Obstetric and perinatal complications

There was no statistical difference in obstetric and perinatal complication between arms in the E-Freeze trial. Observational studies and meta-analysis of published data50–52 suggested that the risk of pre-eclampsia and large for gestational age (LGA) babies is increased in pregnancies that are a result of frozen-embryo transfer. The numbers of each individual complication were too small to draw any definite conclusions from this trial alone, or from any of the other existing trials individually. 42–46 In comparison with the risk in the general population (see Table 30), not all risks were higher in IVF pregnancies, irrespective of whether they were the result of fresh-embryo transfer or frozen-embryo transfer. Although this is reassuring, it is in contrast to the previous findings from systematic review of observational data7 and could be owing to small numbers of each complication in our trial.

Natural compared with hormone replacement therapy frozen-embryo transfer

Most of the frozen-embryo transfers in the E-Freeze trial were undertaken in hormonally mediated cycles. By contrast, worldwide,53 45% of cycles are natural cycles. This can be explained by the fact that participants in the E-Freeze trial were eager to receive treatment and wanted a planned date for their frozen-embryo transfer, after having their treatment delayed by participation in the trial. By contrast, the worldwide data are based on observational, non-randomised data and are more likely to be from patients undergoing transfer of surplus frozen embryos after fresh-embryo transfer has failed, rather than frozen-embryo transfer as their first treatment.

The two Chinese trials43,44 had a large proportion of participants who underwent endometrial preparation for frozen-embryo transfer using natural cycles. The two European trials45,46 and the Vietnamese trial42 had the largest number of patients undergoing hormonally mediated hormone replacement therapy (HRT) cycles.

Hormonally mediated frozen-embryo transfer is more convenient for the clinic and patients, as the date of thaw and transfer can be planned in accordance with the workload of the clinic and at the convenience of the staff and patients. Recently, some concerns have been raised that pregnancies following HRT-mediated frozen-embryo transfer may be at a higher risk of complications than those following a natural cycle. 54

Non-adherence

Non-adherence in the freeze-all arm was very high in this trial. Other trials have also shown non-adherence to the allocated intervention, but at much lower levels than those seen in the E-Freeze trial. The highest level was reported by Shi et al. 43 (freeze-all, 18%; fresh-embryo transfer, 15%). However, non-adherence was reported in both arms, whereas in the E-Freeze trial non-adherence was predominantly seen in the freeze-all arm (31.3%). This level of non-adherence occurred despite the fact that the trial was specifically designed to reduce non-adherence, with consent being reconfirmed just before randomisation and information being provided throughout the process. There could be two reasons for this non-adherence. The funding of IVF treatment is limited across the UK, especially in England, where most of the participating centres were based, and > 60% of treatments in England are funded by the patients themselves. Where funding is available, only one cycle of treatment is funded by most CCGs. Hence, there will be apprehension about freezing all embryos among clinicians, scientists and patients, especially when there are only one or two embryos. There is always a fear that the embryos may not survive the freezing–thawing process, leading to loss of funding for the only funded treatment. Our data showed that, on average, 86% of embryos survived the freezing–thawing process across the participating clinics; hence, the survival rate is far below 100%.

Although there was an intention and acceptance from patients to be randomised to either arm at the time of consent, this did not translate to real practice, with a noticeable preference seen for fresh-embryo transfer. Studies from two different parts of the world55,56 have shown that patients would accept the intervention of freeze-all, followed by frozen-embryo transfer, if freeze-all reduces the side effects and has at least equal, if not better, success rates. Both conditions must be fulfilled to accept the delay.

Difficulties in recruiting to in vitro fertilisation studies in the UK

Most IVF/ICSI treatment across the UK is funded by patients. Where it is funded by the NHS, the full entitlement reported in the NICE guidance3 (i.e. three full cycles of IVF/ICSI) is not available for most patients. To randomise patients to a clinical IVF trial in such a set-up is always going to be difficult, especially when the intervention causes a delay and extra costs, and is the only funded treatment that patients may receive. This is, possibly, one of the reasons for the large proportion of patients who were non-compliant. RCTs are the gold standard; for successful recruitment in any future clinical trials in the UK in the field of IVF/ICSI, NHS funding must improve to the level recommended by NICE. 3

Further analyses of the E-Freeze trial data

Further research questions raised