MRI in the diagnosis of fetal developmental brain abnormalities: the MERIDIAN diagnostic accuracy study

Ultrasonography and fetal screening

For many years, fetal imaging with ultrasonography has been the mainstay of antenatal screening programmes in the UK. However, no imaging method is perfect and various technical factors and physical limitations of ultrasonography may result in suboptimal images of the fetus being obtained. Suboptimal images may lead to incorrect diagnoses of structural abnormalities and inaccurate counselling and prognostic information being given to parents. The fetal brain is a particular area of concern because of the relatively high frequency of developmental abnormalities (approximately 3/1000 pregnancies). 1 A wide spectrum of neuropathologies can be shown, many of which are associated with serious clinical morbidities.

Development of magnetic resonance imaging

Magnetic resonance imaging (MRI) technology has advanced over the years and because of significant improvements in spatial and contrast resolution, highly reliable and accurate diagnoses of comparable pathology can now be made for children. In the 1990s, MRI became a realistic clinical possibility because of advances in hardware and software. The Academic Unit of Radiology at the University of Sheffield was a pioneer in this field. 2 Clinical magnetic resonance sequences have been developed that do not require maternal sedation or fetal neuromuscular blockade, allowing several groups, including our own, to believe that in utero magnetic resonance imaging (iuMRI) may be a powerful adjunct to ultrasonography for detecting fetal brain abnormalities from 18 weeks’ gestational age.

Magnetic resonance imaging in the literature: diagnostic accuracy

Much of the early literature focused on describing the techniques required to perform iuMRI and relied on anecdotal cases to compare the additional information provided with ultrasonography alone. 3–8 A key limitation of the majority of the literature was the lack of an outcome reference diagnosis (ORD), with discrepancies between iuMRI and ultrasonography generally assumed as errors on the part of the latter. Doing so inevitably leads to bias in favour of iuMRI, and fetal ultrasonography experts have criticised these studies for the artificially high detection rates for iuMRI that resulted from biased patient selection,9,10 although the extent of this bias is unknown. One such study was published by our group;11 it was a study of 100 singleton pregnancies in which a fetal brain abnormality was suspected on ultrasonography but for which optimal diagnostic information had not been obtained. When iuMRI was used in conjunction with ultrasonography, an increase in diagnostic accuracy of 48% was observed. Although these findings are clearly prone to the aforementioned bias, they nevertheless demonstrate the potential for a significant improvement in diagnoses based on ultrasonography and provide an upper limit on the improvement attributable to the introduction of iuMRI.

Ventriculomegaly (VM) has been the focus of much research in the field of fetal neuroimaging with iuMRI because of the high prevalence of the finding (1 or 2 out of 1000 pregnancies) following a screening ultrasonography. In our more recent study, 147 fetuses were diagnosed with isolated VM on ultrasonography, with high confidence and no technical limitations. 12 When iuMRI was performed in these cases, additional clinically important brain findings were detected in 17% of cases. In a second study of 61 fetuses with cerebral VM, iuMRI was found to provide more information than ultrasonography in 33% of cases and was able to identify the cause of the VM in 21% of cases. 13 A further study of 185 fetuses with isolated VM in the third trimester found that 11 out of 185 (5.94%) fetuses had other brain abnormalities. 14 The use of iuMRI as an adjunct to ultrasonography for detecting fetal brain abnormalities has been further supported by recent studies and systematic reviews;15–18 however, the extent of diagnostic improvement remains unclear.

Magnetic resonance imaging in the literature: clinical management

The literature on the clinical impact of iuMRI on the counselling and management of pregnancies complicated by fetal brain abnormalities is limited. One prospective study of fetuses with a suspected central nervous system abnormality observed that 24 out of 52 (46%) pregnancies were managed differently after iuMRI,19 and a further prospective study found that the counselling and management of pregnancies were changed after iuMRI in 49.7% and 13.5% of cases, respectively. 20 Our study of fetuses with isolated VM found significant changes in clinical management after iuMRI in the majority of cases when additional brain abnormalities were identified. Changes in management more frequently happened between the gestational ages of 20 and 24 weeks. 12 Our group conducted a systematic review and meta-analysis18 including 34 studies, of which only 11 reported on the changes in counselling or management of pregnancies brought about by the iuMRI. These 11 studies11,21–30 comprised 186 fetuses and changes were observed in 78 (41.9%) cases. Our review concluded that the clinical impact of iuMRI through changes to the counselling and management of pregnancies remains unclear owing to being reported in only a small proportion of studies.

It is important to recognise that iuMRI is unlikely to be incorporated as a routinely used diagnostic tool for pregnancies. Constraints on resources, both financial and human, mean that iuMRI is likely to be used selectively, in combination with ultrasonography and other tests, for fetuses that are considered to be at a high risk of severe developmental abnormalities. The term ‘high risk’ can be interpreted in many different ways and may legitimately incorporate genetic features and/or a previous pregnancy with a fetal abnormality, but the most obvious risk factor is when the fetus has an abnormality diagnosed by the ultrasonography. Therefore, the project set out to determine whether or not adding iuMRI to the usual diagnostic pathway helps to improve the diagnosis and appropriate management of fetuses with a suspected brain abnormality based on previous ultrasonography.

Rationale for the MERIDIAN study

In the light of the uncertainties surrounding the extent of improvement in diagnostic accuracy associated with iuMRI as an adjunct to ultrasonography and its impact on clinical management and counselling, a large prospective study with a representative population of fetal brain abnormalities was required to inform clinical practice in the UK.

For a full evaluation of the use of iuMRI in clinical practice, it was essential that patient acceptability was assessed. New technologies in fetal medicine can raise ethical and social dilemmas for the patients involved and, therefore, the views of patients and their partners are important when considering the implementation of iuMRI. 31 A survey of 227 women who were not pregnant but underwent MRI suggests that it is perceived positively in terms of comfort and impact on care;32 however, other studies have shown a potential association between MRI and anxiety33,34 and that patients may find it less acceptable than ultrasonography. 35 A study published in 2007 found that overall satisfaction with the prenatal diagnosis (PND) is high in women carrying a fetus with a congenital abnormality. 36 Satisfaction with prenatal care has been found to be associated with the attitude of the medical staff,37,38 the amount of information provided38 and the patient’s involvement in decision-making. 37 Data on the use of MRI in fetomaternal medicine are limited but do suggest that MRI may cause additional distress, especially in women for whom the prognosis for the fetus is poor,39,40 and can lead to more anxiety than that of undergoing an ultrasonography. 41

The research described in this report was a logical extension of our previous work and essential in providing definitive data on the diagnostic accuracy, clinical impact and acceptability of iuMRI in clinical practice in the UK.

Participants and eligibility criteria

Participants were pregnant women aged ≥ 16 years and carrying a fetus with a suspected brain abnormality on detailed ultrasonography as diagnosed by a fetal medicine consultant. The potential participants were screened against the following criteria.

Participant identification and recruitment

Potentially eligible participants were identified from 16 participating fetal medicine units that acted as research sites. Following a detailed ultrasonography examination that confirmed the suspicion of a fetal brain abnormality, consecutive patients were screened by the referring clinician or local research midwife according to the inclusion and exclusion criteria listed above. Written informed consent was obtained after the confirmation of eligibility and after the patient had been given the opportunity to consider participation and ask any questions. Data collection and management provides details of the referral process once consent had been obtained.

Test methods

Data collection and management

After the detailed ultrasonography performed by a fetal medicine consultant, participants were recruited into the study and referred for an iuMRI scan at the Academic Unit of Radiology at the University of Sheffield or at one of the five collaborating sites (Belfast, Birmingham, Leeds, Newcastle or Nottingham).

The referral form was completed by the referring clinician and a copy was sent to the MRI site through the standard referral pathway. The referral form collected details of the type of brain abnormalities suspected, confidence of diagnosis and the prognosis for the outcome of the pregnancy. It also collected data on whether or not termination of pregnancy (TOP) was discussed or offered because the abnormalities on ultrasonography were sufficient to consider that option under Ground E of the Abortion Act 1967 [section 1(1)(d) – substantial risk of serious mental or physical disability]. 45

The reporting radiologist completed a similar iuMRI feedback form and was required to comment on all suspected diagnoses along with their diagnostic confidence. Additional diagnoses could be added or excluded when required, as detailed previously.

Following the iuMRI examination, the participant attended a follow-up appointment with the referring fetal medicine consultant. The iuMRI report was available to the clinician at this appointment and the clinical feedback form was completed. The purpose of the clinical feedback form was to assess the changes in the clinical management and prognosis of the pregnancy after the iuMRI. In addition, further details of diagnostic and prognostic information were collected, along with the management plan for the pregnancy. From a diagnostic perspective, participants were asked if iuMRI had provided additional information and, if so, if the ‘new’ pathology was visible when the clinician performed a follow-up ultrasonography. To assess the prognosis, the clinicians were asked if iuMRI had changed the prognostic information given to the patient and to regrade the prognosis in the light of the iuMRI information using the same five categories used at referral. The clinicians also recorded whether or not TOP had been discussed or offered. Finally, the clinicians were requested to state if the iuMRI had altered the counselling and management of the pregnancy and, if so, what the degree of change was.

The pregnancy was then followed up for up to 6 months post delivery and data were collected on the outcome of the pregnancy, as described in Statistical analysis. The outcome data form was used to collect relevant information.

Sample size

A recruitment target of 750 pregnant women was set, from whom we anticipated 504 fetuses would have a complete ORD. It was assumed that 336 of these fetuses would be 18 to < 24 weeks’ gestation at the time of iuMRI. A diagnostic accuracy of 70% in ultrasonography46–54 and 80% in iuMRI was assumed, and it was further assumed that iuMRI would correctly contradict the ultrasonography in 20% of cases and erroneously contradict the ultrasonography in 10% of cases. A sample size of 336 participants allowed the confidence interval (CI) for the difference to be estimated to within a standard error of 3%, and ensured that any improvement in diagnostic accuracy of < 5% with 90% power and at 85% confidence could be ruled out. A change of 5% was deemed to be clinically important as it would lead to changes in fetal prognostic information and management intent in ≈5% of all cases. A total recruitment target of 504 fetuses with complete ORDs was set, assuming that one woman with a fetus of > 24 weeks’ gestation would be scanned for every two fetuses in the 18 to < 24 weeks’ gestation age group.

Statistical analysis

The diagnostic strategies are referred to as ultrasonography (diagnostic ultrasonography alone) and iuMRI (iuMRI following diagnostic ultrasonography).

Statistical analysis was conducted using R version 3.3.3 (The R Foundation for Statistical Computing, Vienna, Austria) and Stata^® version 14 (StataCorp LP, College Station, TX, USA).

Approach/consent

The number of mothers who were approached and consented is presented in the STARD flow diagram (Figure 1). When given, the reasons for not offering iuMRI and reasons for not consenting to join the study have been presented.

FIGURE 1.

Flow diagram to show the pathway and recruitment numbers in the MERIDIAN study. a, One participant carried two fetuses, only one of which had an ORD. Therefore, the participant is counted in both the incomplete and complete participant groups.

Background

Introducing iuMRI to diagnose fetal brain abnormalities during pregnancy will introduce additional direct costs to the NHS. However, iuMRI may also lead to cost savings in other areas and may improve outcomes. Therefore, there is a need for health economic analysis to assess the cost-effectiveness of iuMRI in the diagnosis of fetal brain abnormalities.

Overview

This section considers an economic evaluation based on data collected as part of the MERIDIAN study. Information on resource use and outcomes is analysed for participants in the MERIDIAN study, to estimate the total additional costs and the change in outcomes associated with iuMRI in the diagnosis of fetal brain abnormalities.

A health economic analysis involves the comparison of two or more alternative courses of action. One of the comparators should be usual care, so that it can be determined whether or not the new intervention represents good value for money compared with current practice. The MERIDIAN study is not a randomised controlled trial with an intervention and a comparator, but instead it reports the diagnostic accuracy of iuMRI. In the base case, the economic analysis compares the costs and outcomes associated with iuMRI with those of single ultrasonography. In scenario analyses, iuMRI is compared with repeat ultrasonography.

Costs

The base-case analysis considers the costs incurred until the end of pregnancy. This includes costs of delivery for continued pregnancies and costs of termination for terminated pregnancies. Scenario analyses consider the exclusion of costs of delivery and termination.

In the base-case analysis, resource use is calculated from the analysis of MERIDIAN study data. Costs in the analyses include costs of ultrasonography scans, iuMRI scans, consultations, further investigations and TOP. Resource use is multiplied by unit costs to calculate total costs. Unit costs considered in the base case are shown in Appendix 1. All costs are reported for 2016; when necessary, costs were inflated using the Unit Costs of Health and Social Care 2016 by Curtis and Burns. 60

There are several NHS costs available for TOP. 61 In almost all cases within the study (96.1%), gestational age at termination was > 20 weeks, so ‘Medical or surgical termination of pregnancy, over 20 weeks’ gestation’ (MA20Z) appears to be the most appropriate classification. Clinicians advised that terminations would be planned and so elective inpatient costs are appropriate. Clinicians further advised that the majority of women would stay in hospital for 2 days, with 10% of women staying in hospital for longer and none staying for < 2 days. The elective inpatient cost of £1232 per patient corresponds to an average length of stay of 1.26 days, so additional excess bed-days were included (costed at £277 per day), such that the total average stay was 2.1 days, costing £1581 per patient. From the NHS costs it is unclear whether or not the total costs include feticide, which clinicians advised would be required for gestational ages of > 21⁺⁶ weeks. 61 Clinicians advised that the cost of feticide would be similar to the cost of a specialised fetal diagnostic procedure, which is £366 on average per procedure (NZ72Z). The NHS reference cost for an elective inpatient medical termination at 14–20 weeks’ gestation (MA18D) would not include feticide and is £322 less than the equivalent cost category for > 20 weeks’ gestation. 61 This suggests that the cost for termination at > 20 weeks’ gestation already includes a feticide cost (at least for a proportion of admissions). Therefore, we used our cost for termination at 20 weeks’ gestation for a 2.1-day stay for terminations that involved a feticide. For validation, clinicians advised that this cost should be similar to the cost of a normal delivery, which costs an average of £1643 per procedure (NZ30C0). 61

For terminations that did not require feticide, we assumed that a 2.1-day stay cost for 14–20 weeks’ gestation was appropriate, which was calculated as £1136.

Scenario analysis additionally considers the inclusion of costs borne by participants, such as travel (valued at £0.45 per mile as per UK Government expenses62) and time (valued using earnings by age taken from the 2016 Annual Survey of Hours and Earnings63). A further scenario analysis includes the long-term costs associated with caring for children born with disabilities.

Costs for the pathway with iuMRI and with ultrasonography alone are reported separately for participants who had and for those who did not have TOP. Resource use differs for participants who have TOP, there is the cost of TOP itself, and further imaging or consultation costs may differ. The proportion of participants having a TOP is required to calculate the total cost of iuMRI and ultrasonography. The proportion of participants having a TOP following iuMRI was observed, but because the MERIDIAN study was not a comparative trial, the proportion of TOP for ultrasonography alone was not directly available and had to be estimated (see Analyses).

Outcomes

Health economic analysis from an NHS perspective typically uses the quality-adjusted life-year (QALY) as the measure of benefit. However, the relevance of the QALY in iuMRI is questionable as decision-makers may be uncomfortable with the use of QALYs to reflect the value of an unborn child. Furthermore, incorporating QALYs for unborn children may lead to the perverse situation whereby the intervention with the lowest rate of true-negative detections maximises QALY gain.

In utero magnetic resonance imaging is intended to provide information, additional to ultrasonography, that is of value to parents and treating clinicians. This additional information is of value only if it is both accurate and leads to changes in the management of pregnancy. Therefore, the primary outcome selected for the economic analysis is ‘appropriate management decisions’. The decision for an individual mother is classified as appropriate if the revised decision is consistent with the presence or otherwise of neuro-developmental abnormalities at birth or postmortem. Therefore, appropriate management decisions are:

• TOP advised following ultrasonography, continuation of pregnancy advised following iuMRI. Continuation of pregnancy chosen and infant exhibits no abnormalities at birth.

• TOP advised following ultrasonography, continuation of pregnancy advised following iuMRI. TOP chosen and the fetus exhibits no abnormalities at postmortem.

• Continuation of pregnancy advised following ultrasonography, TOP advised following iuMRI. Continuation of pregnancy chosen and infant exhibits abnormalities at birth.

• Continuation of pregnancy advised following ultrasonography, TOP advised following iuMRI. TOP chosen and the fetus exhibits abnormalities at postmortem.

In the base-case and scenario analyses, the reported cost-effectiveness estimate is the cost per management decision appropriately revised. The proportion of management decisions appropriately revised is calculated by cross-tabulating the number of cases in which iuMRI was correct with the number of cases in which iuMRI changed the diagnosis. The proportion of management decisions appropriately revised is equal to the number of cases in which iuMRI was correct and iuMRI changed the diagnosis, divided by the total number of cases for which both outcomes were available.

Analyses

All analyses were conducted using Stata to calculate resource use and outcomes, and data were analysed in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA) to calculate total costs and cost-effectiveness. All analyses used completed cases with the exception of the imputed missing data, which used bootstrapping in Stata. The analyses conducted are summarised in Table 1 and explained further in the sections that follow.

Study governance

The study conduct was governed by a number of oversight committees: the Trial Management Group (TMG), the TSC and the Data Monitoring and Ethics Committee (DMEC). The trial was conducted in accordance with Sheffield CTRU standard operating procedures (SOPs) and the committees convened at appropriate intervals as dictated by both study requirements and SOPs.

The TMG was composed of the chief investigator (chairperson), study collaborators and key staff within the CTRU and Newcastle University. The TMG met monthly or bimonthly via teleconference during trial recruitment and follow-up.

The TSC members were approved and appointed by the funder. The TSC was composed of a professor of fetal and maternal medicine as the independent chairperson, an independent consultant in neuroradiology, experts in antenatal screening and supporting families who provided expert advice and acted as PPI representatives. The DMEC consisted of an independent statistician, a professor of health-care research (chairperson) and a fetal medicine consultant. The TSC received formal recommendations from the DMEC.

Ethics arrangements and regulatory approvals

The study and subsequent amendments were approved by the Yorkshire and The Humber – South Yorkshire Research Ethics Committee (REC) (reference number 11/YH/0006). Each participating site gave UK NHS Research and Development approval. The study was conducted in accordance with the principles of good clinical practice.

Protocol amendments after study initiation

Details of substantial amendments submitted to the REC, which were important changes to trial methodology, are listed below.

Overall

The overall diagnostic accuracies of ultrasonography and iuMRI were 68.1% and 93.0%, respectively, with a difference of 24.9% (95% CI 21.1% to 28.7%; Table 5). The difference between ultrasonography and iuMRI increased with gestational age: in the 18–23 week group, the diagnostic accuracies were 69.9% for ultrasonography and 92.4% for iuMRI (difference of 22.5%, 95% CI 17.8% to 27.2%); in the ≥ 24 week group, the diagnostic accuracies were 64.7% for ultrasonography and 94.0% for iuMRI (difference of 29.3%, 95% CI 22.6% to 36.1%).

In 386 out of 570 cases (67.7%), both the ultrasonography and iuMRI reports were correct, and in 144 out of 570 cases (25.3%), the ultrasonography report was incorrect but the iuMRI report was correct. There were two fetuses (0.4%) for whom the ultrasonography was correct and the iuMRI was incorrect and 38 (6.7%) fetuses for which both the ultrasonography and the iuMRI were incorrect.

Missing data

Of the 175 fetuses with no ORD, ultrasonography and iuMRI agreed in 86 cases and disagreed in 89 cases. In the most conservative assumption possible, ultrasonography would be correct in all 175 cases, with iuMRI incorrect in all cases of disagreement, resulting in a diagnostic accuracy for ultrasonography of 76.6% and for iuMRI of 82.7%. The difference of 7.1% (95% CI 3.0% to 11.2%), although vastly reduced, remains highly statistically significant (p = 0.0007). The improved diagnostic accuracy of iuMRI over ultrasonography was, therefore, not attributable to missing data.

A less extreme method of handling a missing ORD was to assume that it was missing at random and then to predict what the outcome would have been based on other, similar fetuses for which ORDs were available. The prediction model took into account the relationship between diagnostic accuracy, the probability of missing ORD data and the following characteristics: maternal age, fetal age (weeks) at time of iuMRI, single/multiple pregnancy, outcome (TOP/no TOP) and prognosis.

A missing ORD was more commonly encountered among individuals with a normal, poor or unknown prognosis, but was similar in relation to whether or not iuMRI changed the prognosis. The association of missing data with a prognosis was similar regardless of whether the prognosis was based on ultrasonography or iuMRI. As there were no missing data for the ultrasonography prognosis, it was used in the prediction model. Missing data were also more commonly observed among individuals who underwent TOP, although this was highly associated with the prognosis. Missing data were slightly more common among younger fetuses but showed little association with maternal age or single/multiple pregnancy. Because gestational age was associated with diagnostic accuracy (the diagnostic accuracy of ultrasonography was reduced in those of older gestational age), this was retained in the prediction model along with the prognosis based on ultrasonography.

Using the prediction model to predict missing outcomes (under the assumption of missing at random), the diagnostic accuracy of ultrasonography was 66.9% and iuMRI was 92.8%, resulting in a slightly greater difference of 25.9% (95% CI 22.3% to 29.7%).

All participants with outcome reference diagnosis

The primary analysis was repeated on all participants who had scans and an ORD (Table 6). Similar results to the primary analysis were observed and the overall diagnostic accuracies of ultrasonography and iuMRI, at 68.8% and 92.3%, respectively, resulted in a difference of 23.5% (95% CI 20.0% to 27.1%; p < 0.0001).

Repeat in utero magnetic resonance imaging

There were 65 fetuses for which repeat ultrasonography and iuMRI were undertaken and an ORD was obtained. As with the primary analysis, the majority (46/65) underwent iuMRI at the host institution. The median gestational age was 31 weeks with just five fetuses under 24 weeks. The overall diagnostic accuracies of ultrasonography and iuMRI were 67.7% and 87.7%, respectively, indicating a difference of 20.0% (95% CI 7.8% to 32.2%; p < 0.0001; Table 7).

A summary of the analysis repeated on first, single and repeat pregnancies can be found in Appendix 5.

Isolated ventriculomegaly

Ventriculomegaly was the most common brain abnormality for referral to the MERIDIAN study. 67 In 421 out of 570 (74%) fetuses, VM formed part of the ultrasonography diagnosis; in 306 out of 570 (54%) fetuses, VM was the only brain abnormality diagnosed on ultrasonography. This analysis is based on the 306 cases with isolated VM on ultrasonography.

A total of 199 out of 306 (65%) cases were in the 18–23 weeks group at the time of iuMRI and 107 out of 306 (35%) cases were in the ≥ 24 weeks group. The category of VM was determined by the largest trigone measurement and were divided into mild (10–12 mm), moderate (13–15 mm) and severe (≥ 16 mm) categories. Ultrasonography diagnosed mild VM in 244 out of 306 (80%) fetuses, moderate VM in 36 out of 306 (12%) fetuses and severe VM in 26 out of 306 (8%) fetuses.

The ORD found additional brain abnormalities other than VM in 31 out of 306 fetuses (10.1%); 17 out of 199 (8.5%) of those were in the 18–23 weeks group and 14 out of 107 (13.1%) of those were in the ≥ 24 weeks group. Of those fetuses, 5.7% (14/244) of those referred with mild VM had other brain abnormalities on ORD, 19.4% (7/36) of those referred with moderate VM had other brain abnormalities and 38.5% (10/26) of those referred with severe VM on ultrasonography showed other brain abnormalities on ORD (Table 8). The other brain abnormalities identified were agenesis of the CC (ACC) or hypogenesis of the CC in 17 out of 31 (55%) fetuses, encephalomalacia and/or intracranial haemorrhage in 8 out of 31 (26%) fetuses, cortical formation abnormalities in 4 out of 31 (13%) fetuses and absent cavum septum pellucidum (CSP) caused by septo-optic dysplasia in 2 out of 31 (6%) fetuses.

The iuMRI was correct in 302 out of 306 cases when compared with the ORD. Of these 302 cases, 275 were correctly diagnosed with isolated VM and in 27 cases the iuMRI correctly identified additional brain abnormalities. The iuMRI provided an incorrect diagnosis in 4 out of 306 fetuses (three cases of hypogenesis of the CC and one case of absent CSP).

There were complete clinical impact data available for 295 out of 306 (96%) fetuses; of these, 29 fetuses had additional brain abnormalities. Fetal medicine clinicians reported that iuMRI provided additional information compared with ultrasonography in 106 out of 295 (35.9%) cases. Overall, there was a change in the prognosis category after iuMRI in 69 out of 295 (23.4%) cases. This was a change to a worse prognosis in 22 cases and to an improved prognosis in 47 cases.

Forty-two (14.2%) women were offered TOP based on the ultrasonography diagnosis alone and in eight of those cases the offer of TOP was reversed after iuMRI. After iuMRI, an additional 14 women were offered TOP based on the diagnosis of brain abnormalities other than VM. The decision to offer TOP, therefore, was changed in 22 out of 295 (7.5%) fetuses after iuMRI and TOP was performed in 14 out of 295 (4.7%) cases. The contribution of iuMRI to the final choice of management was rated by the fetal medicine expert to be ‘of no value’ in 41 out of 295 (14%) cases, minor in 177 out of 295 (60%) cases, significant in 64 out of 295 (22%) cases, major in 12 out of 295 (4%) and decisive in one (0.3%) case.

For the 29 fetuses that had brain pathology other than VM confirmed on ORD, the prognosis was considered worse in 13 out of 29 (44.8%) cases after the iuMRI and the effect of iuMRI on overall clinical management was rated as ‘significant’, ‘major’ or ‘decisive’ in 15 out of 29 (51.7%) cases.

Failed commissuration

Ultrasonography reported some form of ‘failed commissuration’ abnormality in 92 out of 570 (16.1%) referrals into the MERIDIAN study. 68 Thirteen of those referrals also had additional brain pathology other than VM identified on ultrasonography and were, therefore, not included in this subgroup analysis. The remaining 79 cases form the basis of this analysis, 54 of which had associated VM. The iuMRI was performed at 18–23 weeks of gestational age in 51 out of 79 (65%) fetuses and at ≥ 24 weeks of gestational age in 28 out of 79 (35%) fetuses. In total, 24 out of 79 cases were diagnosed with hypogenesis of the CC on ultrasonography and 55 out of 79 cases were diagnosed with ACC.

The diagnostic accuracy for ‘isolated failed commissuration’ as a whole group was 34.2% for ultrasonography, compared with 94.9% for iuMRI (difference of 60.7%, 95% CI 47.6% to 73.9%; p < 0.0001) (Table 9). Of the 52 fetuses whose diagnosis was incorrect on ultrasonography, the ORD recorded a normal CC in 46 fetuses and additional diagnoses not detected on ultrasonography in 12 fetuses (five fetuses had a normal CC but additional diagnoses).

The iuMRI had an error rate of 4 out of 79 fetuses; all four of these errors were in relation to the CC diagnosis. All 12 other brain abnormalities shown on ORD were diagnosed correctly on iuMRI.

Further analysis looked at the diagnostic accuracy of ultrasonography and iuMRI in specifically diagnosing hypogenesis of the CC or ACC as discrete abnormalities, irrespective of the presence of other brain abnormalities (see Table 9).

Ultrasonography was shown to have a diagnostic accuracy of 8.3% for detecting hypogenesis of the CC. In comparison, iuMRI demonstrated 87.5% accuracy (difference of 79.2%, 95% CI 55.1% to 100%; p < 0.0001). The diagnostic accuracy for detecting ACC specifically was 40.0% for ultrasonography and 92.7% for iuMRI (difference of 52.7%, 95% CI 37.7% to 67.7%; p < 0.0001).

Ultrasonography correctly diagnosed hypogenesis of the CC in 2 out of 24 fetuses (8.3%). The majority of ultrasonography errors (20/24) for hypogenesis of the CC occurred when the ORD showed a normal CC, and in 2 out of 24 cases the ORD reported ACC.

Ultrasonography correctly diagnosed ACC in 22 out of 55 cases (40.0%). Of the remaining 33 cases, which ultrasonography diagnosed with ACC, the ORD reported hypogenesis of the CC in seven cases and a normal CC in 26 cases.

The iuMRI reported a normal CC in 45 out of 79 fetuses, 43 (95.5%) of which were subsequently confirmed as normal on ORD; the other two fetuses were incorrectly diagnosed by iuMRI and were found to have hypogenesis of the CC on ORD. In 8 out of 79 cases, the iuMRI reported hypogenesis of the CC, of which 6 (75%) were found to be correct. Finally, iuMRI diagnosed 26 cases of ACC, of which 23 (88.5%) were subsequently found to be correct.

Posterior fossa abnormalities

Ultrasonography reported some form of abnormality in the posterior fossa in 81 out of 570 (14.2%) fetuses. 69 The iuMRI was performed at 18–23 weeks of gestational age in 57 out of 81 (70%) fetuses and at ≥ 24 weeks of gestational age in 24 out of 81 (30%) fetuses. A total of 67 out of 81 cases were diagnosed with a parenchymal abnormality confined to the posterior fossa on ultrasonography and 14 out of 81 cases were diagnosed with a CSF-containing abnormality. Associated VM was found on ultrasonography in 25 out of 81 (31%) cases overall and in 3 out of 15 (20%) fetuses with cerebellar hypoplasia, 6 out of 21 (29%) fetuses with Dandy–Walker spectrum abnormalities and 11 out of 21 (52%) fetuses with Chiari type II malformations.

The overall diagnostic accuracy of ultrasonography was 65.4%. A normal brain was reported on ORD in 18 out of 28 of the ultrasonography errors and other supratentorial brain abnormalities were reported on in 10 out of 28 of the ultrasonography errors (the other brain abnormalities consisted of six ACC or hypogenesis of the CC cases, two cortical formation abnormalities and two acquired pathologies). The iuMRI gave a diagnostic accuracy of 87.7%, showing an improvement in diagnostic accuracy of 22.3% (95% CI 14.0% to 30.5%; p < 0.0001) over ultrasonography.

In the 10 iuMRI errors, eight were cases in which the ORD was normal and two were cases in which iuMRI failed to detect an abnormality of the CC shown on the ORD.

The specific diagnoses of parenchymal abnormalities on ultrasonography were:

• cerebellar hypoplasia in 25 out of 67 (38%) fetuses – vermian hypoplasia in 15 fetuses and hypoplasia of the cerebellar hemispheres in 10 fetuses

• Dandy–Walker spectrum abnormalities in 21 out of 67 (31%) fetuses

• Chiari type II malformation in 21 out of 67 (31%) fetuses.

The label ‘parenchymal abnormality’ in Table 10 requires the exact pathological diagnosis of the posterior fossa abnormality to be correct when compared with the ORD and ignores the presence or absence of other brain abnormalities. The correct specific diagnosis was made by ultrasonography in 54 out of 67 fetuses, demonstrating a diagnostic accuracy of 80.6%. The iuMRI correctly diagnosed 64 out of 67 fetuses, demonstrating a diagnostic accuracy of 95.5%. The difference in diagnostic accuracy was 14.9% (95% CI 5.7% to 24.1%; p = 0.002) in favour of iuMRI.

The CSF-containing abnormalities made up 14 out of 81 of the diagnoses on ultrasonography. Eight fetuses had an enlarged CM and six fetuses had abnormal cystic structures (four arachnoid cysts and two unspecified). In total, 3 out of 14 cases with an ultrasonography diagnosis of a CSF abnormality were correct (diagnostic accuracy of 21.4%). In the 11 out of 14 cases of incorrect diagnoses on ultrasonography, the ORD consisted of ‘no brain abnormality’ in eight fetuses, failed commissuration in two fetuses and durovenous thrombosis with ectasia70 in one fetus. The iuMRI was correct in 8 out of 14 of those cases (diagnostic accuracy of 57.1%), with a difference of 35.7% (95% CI –1.6% to 73.0%; p = 0.0625) in favour of iuMRI. All of the iuMRI errors were cases in which iuMRI reported a CSF-containing abnormality but this was not confirmed on ORD, although in one case a CC abnormality was also missed.

Prognosis

Clinical feedback was available for 783 out of 823 pregnancies. Of the remaining 40 pregnancies, there were 26 participants for whom no follow-up took place, 13 participants for whom the clinical feedback form was not fully completed and one participant who withdrew from the study after the iuMRI but before the clinical feedback visit. The ORD was available for 545 pregnancies but was not a requisite for this analysis, meaning that the analysis was based on 783 mothers. A full summary of prognosis categories on ultrasonography and iuMRI can be found in Appendices 6 and 7.

Fetal medicine clinicians reported that the iuMRI provided additional diagnostic information in 387 out of 783 (49%) cases. In 201 out of 387 (52%) cases, the ‘new’ fetal brain abnormalities identified by iuMRI were visible on follow-up ultrasonography.

In 189 out of 783 (24%) cases, referring clinicians answered ‘yes’ to the question ‘Did iuMR change your prognosis?’. In 138 out of 189 cases, additional non-neuroimaging investigations, such as karyotyping, fetal cardiac echo and infection screening, were performed in conjunction with the iuMRI examination. The results of such tests were considered to have a major influence on the prognosis in 32 of those cases. Therefore, we estimate that iuMRI per se changed the prognostic information in at least 157 out of 783 cases (20%). Although 24% of clinicians reported that iuMRI changed their prognosis, the actual prognostic information was modified in 342 out of 783 (44%) participants. The changes were primarily shifting away from ‘intermediate’ and ‘unknown’ to ‘normal’, ‘favourable’ or ‘poor’. The most frequent change in prognosis was from ‘unknown’ on ultrasonography to a quotable risk category after iuMRI (113/783, 14% of cases). Conversely, on 33 occasions (4%) after iuMRI the prognosis was changed from a quotable known risk to ‘unknown’. A full summary of these changes can be found in Appendix 8.

The prognosis was quotable after both scanning methods in 77% of cases; of these, the prognosis was improved after iuMRI in 102 out of 783 (13%) cases and worsened in 94 (12%) cases (Table 14). Most participants showed no change in prognosis (n = 409, 68%).

Counselling and management

Termination of pregnancy was discussed in 51% of consultations before iuMRI and in 49% of consultations after iuMRI. However, the rate of TOP being offered increased from 25% before iuMRI to 36% after iuMRI, with an additional 84 (11%) cases in which TOP was offered after iuMRI (Table 15). A higher proportion of cases had TOP offered based on substantial risk of disability following iuMRI than ultrasonography (33% following iuMRI and 23% following ultrasonography alone). There were also fewer cases when TOP was discussed but not offered following iuMRI (12% following iuMRI and 25% following ultrasonography alone).

In response to the question ‘Did iuMR change your counselling in this case?’ the referring clinician reported no influence on counselling in 172 out of 783 (22%) cases, minor influence in 496 out of 783 (63%) cases and major influence in 115 out of 783 (15%) cases. The contribution of iuMRI to the final choice of management was felt to be of ‘no value’ in 95 out of 783 (12%) cases, a ‘minor influence’ in 419 out of 783 (53%) cases, ‘significant’ in 201 out of 783 (26%) cases, a ‘major influence’ in 49 out of 783 (6%) cases and ‘decisive’ in 19 out of 783 (3%) cases (Table 16). 44 The contribution of iuMRI was rated particularly highly among those participants whose chosen management plan was TOP (Table 17). A total of 31 out of 103 (30%) participants said that it was ‘significant’, 9 out of 103 (8.8%) participants said that it was ‘major’ and 12 out of 103 (12%) participants said that it was ‘decisive’.

Outcomes

A total of 570 women were part of the primary analysis set, had outcomes recorded and had iuMRI within 2 weeks of the ultrasonography. Information on whether or not iuMRI changed the prognosis was missing for 19 women. Table 23 shows whether or not iuMRI was correct and if it changed the prognosis for the remaining 551 women.

Of the 570 women in the primary analysis set, information on whether or not TOP was offered after ultrasonography was missing in 23 women and information on whether or not TOP was offered after iuMRI was missing in 33 women.

Of the 570 women with information on whether or not TOP happened, 68 (11.93%) women had a TOP.

Of the 547 women with information on whether or not TOP was offered after ultrasonography, 135 (24.68%) women were offered TOP after ultrasonography.

Of the 537 women with information on whether or not TOP was offered after iuMRI, 189 (35.20%) women were offered TOP. A total of 58 of the 189 (30.69%) women offered TOP after iuMRI had a TOP (Table 24).

Costs

Disaggregated and total costs for iuMRI resulting in TOP, iuMRI not resulting in TOP, ultrasonography resulting in TOP and ultrasonography not resulting in TOP for complete cases are shown in Table 25.

Base case, scenario 1 and scenario 2

Results for the base case, scenario 1 and scenario 2 analyses are shown in Table 26.

Cost-effectiveness

In the base case, 21.78% (120/551) of decisions were appropriately revised following iuMRI. The incremental cost per management decision appropriately revised is therefore £2203.

For scenario 1, if we assumed that each woman offered TOP after iuMRI had a TOP, then the average cost to the NHS of iuMRI would be £3268. The cost of ultrasonography is unchanged from the base case. The incremental cost is therefore £396 and the incremental cost per management decision appropriately revised is £394.

For scenario 2, if we assumed that 30.69% of women offered TOP after ultrasonography had a TOP, then 7.57% of women who underwent ultrasonography would have had a TOP. The average cost to the NHS of ultrasonography would be £3472. The cost of iuMRI is unchanged from the base case. The incremental cost is therefore £301 and the incremental cost per management decision appropriately revised is £873.

The cost of a delivery is higher than the cost of a termination. In the base case, iuMRI leads to fewer TOPs than ultrasonography; together with the additional cost for iuMRI, this means that the total cost is higher for iuMRI than for ultrasonography. In scenarios 1 and 2, the proportion of TOPs for iuMRI and ultrasonography is more similar than in the base case. In scenarios 1 and 2, iuMRI leads to more TOPs than ultrasonography, so the costs are more comparable, although the cost of iuMRI means that the total cost is higher for iuMRI than for ultrasonography but the incremental cost is lower than the base case.

Imputed missing data

When costs were analysed to impute missing data, the total cost to the NHS for iuMRI was £3848 without TOP and £2143 with TOP. The total cost for ultrasonography was £3584 without TOP and £1879 with TOP. Results for the base case, scenario 1 and scenario 2 using imputed missing data for costs are shown in Table 26.

Probabilistic sensitivity analysis

Mean results from 1000 probabilistic simulations are shown in Table 26.

Cost-effectiveness acceptability curves are presented for the base case, scenario 1 and scenario 2 in Figure 2. For the base-case analysis, if we are willing to pay ≥ £4000 for an appropriate management decision, iuMRI has a probability of 100% of being the most cost-effective option. For scenarios 2 and 3, we can be certain that iuMRI imaging is the most cost-effective option.

FIGURE 2.

Cost-effectiveness acceptability curve.

Graphs for the per-person EVPI at varying willingness-to-pay thresholds are presented in Figure 3. The EVPI peaks at the willingness-to-pay threshold at which there is most uncertainty about whether iuMRI or ultrasonography is the most cost-effective technique. The maximum EVPI is relatively low (£14–30), indicating that there is little value in conducting further research on the parameters in this short-term analysis.

FIGURE 3.

The EVPI.

Scenario analyses

Subgroup analyses

Results are presented for gestational ages of < 22 weeks, 22–24 weeks and > 24 weeks at iuMRI scan in Table 28.

Comparison with repeat ultrasonography

A comparison of iuMRI with repeat ultrasonography is shown in Table 29. In this analysis, the incremental cost is negative, meaning that iuMRI is cost-saving compared with repeat ultrasonography. The incremental effectiveness of iuMRI over repeat ultrasonography is reduced.

Different costing scenarios

Results excluding delivery costs only, excluding delivery and termination costs, and increasing termination costs are shown in Table 30.

When only delivery costs are excluded, TOP becomes more costly than continuation of pregnancy. In the base case, there are more TOPs for ultrasonography than for iuMRI, but these costs are offset somewhat by the cost of iuMRI itself. However, the incremental cost is much less than when delivery costs are included (deterministic results in Table 26) and, therefore, the cost per management decision appropriately revised decreases. In scenarios 1 and 2, the incremental costs have increased (compared with deterministic results in Table 26) because there are more TOPs for iuMRI than for ultrasonography, which lead the cost per management decision appropriately revised to increase. The estimates of cost-effectiveness for the base case, scenario 1 and scenario 2 are closer together, because the incremental cost of iuMRI itself, which is the same between the scenarios, is a bigger driver of costs.

When delivery and termination costs are excluded, the incremental costs for the base case, scenario 1 and scenario 2 are very similar to each other, as the only differences in the costs are the cost of the iuMRI and the differences in investigations and further imaging.

When the cost of termination increases, it becomes closer to the cost of delivery. Therefore, there is less difference in the costs for people with and without TOP. The incremental costs for the base case, scenario 1 and scenario 2 become closer to each other than they were in the deterministic results in Table 26, but the difference between them is still larger than when termination and delivery costs are excluded.

Longer-term horizon

In this scenario (Table 31), the cost of a continued pregnancy becomes much higher than the cost of a termination, and iuMRI becomes much more costly than ultrasonography when it results in fewer TOPs (as in the base case) and becomes cost-saving when it results in more TOPs (as in scenarios 1 and 2). This analysis demonstrates the uncertainty in the longer-term economic evaluation. Longer-term follow-up studies of children diagnosed with fetal brain abnormalities would be required to address this uncertainty.

Questionnaire surveys

Women who were approached about the sociological study (after recruitment into the study) were provided with a site-specific information pack. Survey 1 was administered after the fetal medicine consultation during which the iuMRI findings were discussed and a provisional management plan was made. Survey 2 was administered 3–6 months post pregnancy outcome (either birth or TOP). Both surveys included questions to evaluate (1) the overall satisfaction with care, (2) the practical utility of results from iuMRI to inform understanding and decision-making and (3) to what extent participants’ experiences were influenced by their situation (e.g. psychological state) at the time of iuMRI. This last objective utilised the HADS questionnaire data at the end of each survey.

Survey 1 differed slightly from survey 2. Both surveys comprised questions on age, marital status, education and questions on the impact of practical issues on the acceptability of the care package related to referral pathways (e.g. travel times). Survey 2 included (1) an open-text question that allowed participants to raise issues they felt were important to evaluating their health-care experience and (2) a filter question with an expression of interest (EOI) form for an in-depth interview. Follow-up letters and survey forms were sent to participants who had not responded after a certain time (Figure 4), improving returns by 4% (survey 1) and 6% (survey 2).

FIGURE 4.

Substudy survey flow chart. Black, MERIDIAN database; green, site fetal medicine unit clinicians; blue, substudy researchers.

The primary focus of the survey was the quantitative data collected, but there were also a small number of open-text responses in survey 2 and marginalia. Content analysis using NVivo 10 (QSR International, Warrington, UK) was employed to code these textual data that were then used, when appropriate, to inform the analytic process.

Qualitative interviews with women

Survey respondents who completed an EOI form in survey 2 formed the sampling frame for the qualitative interviews with women. This approach facilitated the use of background characteristics, diagnostic category and pregnancy outcome alongside fetal medicine unit and iuMRI centre in the purposive sampling for maximum variation. 78 Women were contacted using information provided in EOI forms. Following the verification of the agreement to be interviewed, they were sent a study pack with information about the interview phase (including a reply slip and information on how a partner/relative/friend could participate in the woman’s interview if she desired).

Following written consent, interviews were conducted in participants’ homes and recorded on an encrypted digital voice recorder. A narrative approach to interviewing allowed participants to develop their own account of their ‘story’. The topic guide prompted participants to describe their feelings on learning about the abnormality, undergoing the iuMRI and decision-making about pregnancy management. The topic guide also explored perceived differences between ultrasonography and iuMRI, information-seeking behaviour and the formal and informal support received.

Qualitative interviews with health professionals

Health professional participants were recruited from an identified pool of 42 health professionals associated with the clinical study, including 10 sites that were recruited in a suitable time frame for inclusion. Purposive sampling ensured that the participant group was sufficiently diverse (1) to cover a range of clinical sites, (2) to cover a range of professionals (e.g. fetal medicine specialists, radiology specialists), (3) in gender and (4) in length of service in the field. Potential participants were contacted by post with a study information pack, including an information sheet and a reply slip. On receipt of a reply agreeing to take part, the interviewer made contact using the details provided by the potential participant.

Interviews were carried out at two time points. The first round of interviews took place during the first year of clinical recruitment at the participant’s site, and the second round of interviews took place during the third year of recruitment at that site (originally the final year of recruitment, prior to the extension of the clinical study). Whenever possible, participants interviewed in the first round were reinterviewed in in the second round, but this was not possible in all cases. In the second round of interviews, participants were offered a telephone or face-to-face interview and additional participants were included. Face-to-face interviews were held at the participants’ place of work. The topic guide included prompts on key issues, including experiences of the clinical project, perceived value of iuMRI in the PND and potential barriers to and facilitators of future policy implementation.

Qualitative data management

Interviews continued until data saturation was reached (i.e. we achieved a sufficiently diverse sample with no new themes emerging in the ongoing thematic analysis, leading to a stable descriptive coding framework). All digital sound files were stored on a password-secure server and deleted from the digital recorders. The interviews were transcribed, edited for accuracy, anonymised and entered into a qualitative data-management software for indexing and retrieval, ATLAS.ti version 7 (ATLAS.ti Scientific Software Development GmbH, Berlin, Germany).

Data analysis approach for survey data

Responses to the four key questions were investigated using multiple logistic regression. For the purpose of these analyses, responses were recategorised into ‘positive’ and ‘negative’ binary variables as follows:

• For Likert scale questions – 1 if the response was ‘strongly agree’ or ‘agree’, 0 for everything else.

• For rating scale questions – 1 if the response was ‘excellent’ or ‘very good’, 0 for everything else.

A variable selection approach was used with variables that were selected for use in the final model if they had a p-value of < 0.1. The following models were calculated for each of the four outcomes in questionnaires 1 and 2:

• background – maternal age, gestational age, level of education

• service – Sheffield versus non-Sheffield MRI centres, journey length

• prognosis on iuMRI

• diagnosis on iuMRI

• change from ultrasonography after iuMRI – change in counselling, change in prognosis.

A further model was produced for questionnaire 2:

• outcome – outcome for pregnancy, agreement between ultrasonography and iuMRI.

Odds ratios (ORs) with CIs and corresponding p-values were calculated. Only complete cases were included (i.e. those mothers who had a value for the outcome and each of the covariates were included in the model). Model fit was assessed for each of the models by the Hosmer–Lemeshow goodness-of-fit test. All 49 questionnaire 2 responders who underwent TOP strongly agreed or agreed, meaning that the model could not be estimated owing to separation; therefore, TOP was removed.

Data analysis for qualitative interview data

For the service user and the health professional interview data analyses, anonymised interview transcripts (primary documents) constituted the formal data. Service users’ transcripts were coded separately from health professionals’ transcripts. The health professionals’ transcripts were initially subdivided by round. When a high degree of similarity in the data from the two data-collection periods became clear, the health professionals’ transcripts were analysed as one set of primary documents. In both the service user and the health professional interview data analyses, a generative thematic approach was used, incorporating a multistage process similar to that described by Braun and Clarke79,80 (Table 32). Our approach to generative thematic analysis is informed by Silverman81,82 and shares some principles with grounded theory approaches,83 for example the identification of themes that emerge from the data. 84 However, we did not follow any specific grounded theory protocol. Initial coding frames were developed by each lead analyst and with coding samples reviewed by the co-analyst, providing a qualitative equivalent to inter-rater reliability, enhancing consistency and theme refinement. The final theme content was then condensed and further refined by the lead analyst, with input from the other two members of the substudy team.

The sensitive topic area meant that interviews were sometimes emotional for service user participants, but still generated rich, in-depth narratives. In most cases, the presence of a partner or relative was supportive and their participation gave an added dimension to the interviews. Field notes were written for unusual cases (e.g. when family circumstances were difficult).

To guard against the fragmentation of data through the ‘code and retrieve’ method of analysis, synopses of each woman’s interview were composed by the research associate, providing an overview of all interview cases and read by the qualitative lead. A composite case study was produced to highlight the difficulties faced by patients (see Appendix 10).

Interviews with health professionals were, on occasion, challenging for pragmatic reasons. Participants were usually interviewed at work, where they were busy and often interrupted. However, even the most fragmented interviews retained a coherent narrative structure and detailed discussion, producing a rich, complex data set. The interviews included transparent discussion of the context leading to the interview and field notes were produced only as needed (e.g. when disruption had been experienced).

Service user substudy participants

The questionnaire survey was conducted between September 2011 and August 2014 and included the first 681 participants recruited to the MERIDIAN study (questionnaire 1) and 630 participants who were eligible for the follow-up questionnaire (questionnaire 2). The response rate was higher for questionnaire 1 (428/681, 62.8%) than for questionnaire 2 (284/630, 45.1%) (p = 0.005). A total of 206 participants completed both questionnaires. Respondents’ characteristics are shown in Table 33. The proportion of respondents to questionnaire 1 who opted for TOP was 18%, in contrast to the 11% of questionnaire 2 respondents who opted for TOP.

In total, 104 EOI forms were returned with survey 2. Of these, 10 respondents were not contactable and 49 respondents were withdrawn because their forms were returned after recruitment was completed. Those eligible for the qualitative interview phase were contacted by telephone. Three respondents subsequently declined and data from one interview was prematurely deleted, leaving data from 41 participants available for analysis (39% of EOIs returned, 14% of surveys for questionnaire 2 completed).

Patient survey findings

A total of 97% of questionnaire 1 respondents and 95% of questionnaire 2 respondents would choose to have iuMRI as part of their care if they were ever in a similar situation again. In response to the question about whether or not iuMRI was undertaken at an acceptable time and place, 73% of questionnaire 1 respondents and 68% of questionnaire 2 respondents strongly agreed or agreed. Only 3% of respondents in both groups strongly disagreed and 9% of questionnaire 1 respondents and 11% of questionnaire 2 respondents disagreed. The following open-text responses in the survey are illustrative:

We didn’t mind travelling a long distance there, but the return journey was extremely difficult after being told such heart-breaking news.

ID239

. . . we had to stay overnight in a hotel with a 7-year-old, hours away from home. It would be very helpful if this test was more widely available.

ID125

Responses to questions about the impact of iuMRI on understanding and decision-making are shown in Figure 5. Overall, 83% of respondents to questionnaire 1 and 80% of respondents to questionnaire 2 gave positive responses to the statement ‘The information from the iuMRI scan helped me to understand the baby’s problem’. Positive responses were more likely if the iuMRI was performed at Sheffield compared with non-Sheffield sites [questionnaire 1 (OR 2.22, 95% CI 1.28 to 3.86; p = 0.005) and questionnaire 2 (OR 3.41, 95% CI 1.82 to 6.48; p < 0.001)] and less likely if the iuMRI had been performed after 24 weeks of pregnancy [questionnaire 1 (OR 0.52, 95% CI 0.29 to 0.91; p = 0.022) and questionnaire 2 (OR 0.39, 95% CI 0.20 to 0.750; p = 0.005)].

FIGURE 5.

Responses to questions about the impact of iuMRI on understanding and decision-making in questionnaire 1 (black) and questionnaire 2 (green). (a) Responses to the question ‘The information from the MRI helped me to understand the baby’s problem’; (b) responses to the question ‘The information from the MRI helped me to understand how the baby’s problem could affect his/her future quality of life’; and (c) responses to the question ‘The information on the MRI scan was useful to me when I made the decision about whether or not to continue with the pregnancy’. Of the 206 women who responded to both surveys, the proportion of those who either strongly agreed or agreed to (a) were 73% (Q1) and 68% (Q2), to (b) were 73% (Q1) and 68% (Q2), and to (c) were 76% (Q1) and 72% (Q2).

A total of 69% of respondents to questionnaire 1 and 71% of respondents to questionnaire 2 gave positive responses to the statement ‘The information from the iuMRI scan helped me to understand how the baby’s problem could affect his/her future quality of life’. Positive responses to questionnaire 1 were more likely when the prognosis was graded as poor (i.e. < 50% chance of a normal outcome) (OR 4.20, 95% CI 1.64 to 11.10; p = 0.003). Positive responses to questionnaire 2 were more likely if iuMRI was performed at the Sheffield site (OR 1.96, 95% CI 1.15 to 3.35; p = 0.013).

Of the questionnaire 1 respondents who answered either one of the previous two questions (see Figure 5a and 5b), 37% of them did not respond to the statement ‘The information from the iuMRI scan was useful to me when I made the decision about whether or not to continue with the pregnancy’ (see Figure 5c). In total, 71% of the remaining respondents gave positive responses to the statement. None of the factors included in the model influenced the response to questionnaire 1. Of the questionnaire 2 respondents who answered either one of the two previous questions, 27% did not respond to the same statement but 74% of the remaining questionnaire 2 respondents responded positively.

Women who were not faced with the dilemma of whether to have a termination because the condition was ‘mild’ or were going to continue regardless of what the iuMRI scan revealed might have opted to leave the question blank because it was not ‘relevant or appropriate’. The only factor that influenced the response to the statement about whether or not the iuMRI scan was useful in decision-making was gestational age at iuMRI; fewer mothers who were scanned at ≥ 34 weeks of pregnancy gave positive responses (OR 0.11, 95% CI 0.02 to 0.47; p = 0.007).

Overall, 90% of questionnaire 1 respondents rated their care as ‘excellent’ or ‘very good’, 9% of respondents rated their care as ‘good’ and only 1% of respondents rated their care as ‘fair’. Responses to questionnaire 1 did not appear to be influenced by any of the factors examined and responses were consistently positive across groups. Overall, 88% of questionnaire 2 respondents rated their care as ‘excellent’ or ‘very good’, 9% of respondents rated their care as ‘good’ and 3% of respondents rated their care as ‘fair’. Only one respondent to questionnaire 2 rated their care as ‘poor’. Responses to questionnaire 2 were only influenced by journey length; positive responses were less likely if the journey length was 2–4 hours (OR 0.38, 95% CI 0.13 to 0.99; p = 0.05).

Patient qualitative interview findings

The sample of 41 women interviewed included experiences from three of the six iuMRI sites. A total of 18 family members were included: 16 family members were in joint interviews and two were individually interviewed after their partners. Four of the women’s relatives were their mothers. Interviews were carried out between July 2012 and December 2013, 4–12 months post pregnancy outcome and with the majority > 7 months post outcome. Interviews lasted between 21 minutes (with a family member only) and 146 minutes (with the participant). The average length of the interviews was 69 minutes. Four key themes were identified from the data.

Health professional participant characteristics

From 30 health professional recruitment packs, 23 (77%) returned a reply slip, of which 21 (70%) health professionals were willing to be interviewed (50% of the originally identified sample). In round 1, 11 were interviewed between March and August 2013. In round 2, 19 were interviewed [six face to face, 12 over the telephone and one via Skype™ (Microsoft Corporation, Redmond, WA, USA)] between July 2014 and March 2015. In round 1, interviews lasted between 41 and 89 minutes (average length of 63 minutes). In round 2, interviews lasted between 26 and 62 minutes (average length of 44 minutes).

A total of 21 health professionals were recruited and interviewed at least once. Two participants were interviewed in round 1 only, seven participants in round 2 only and nine participants were interviewed in both. Participants’ characteristics are described in Table 34. Sites have not been included to protect participant anonymity in this close-knit professional field. However, participant recruitment covered 10 clinical sites and included sites with diverse provision structure (tertiary and non-tertiary care, those providing iuMRI on site and those referring elsewhere).

Most participants fell within the fetal medicine and radiology specialties, but a number of participants were in other categories (e.g. midwifery, radiography) or did not fit neatly into one category. To ensure the confidentiality of participants in smaller subgroups, some specific categories are not identified.

Health professional qualitative interview findings

Background

Although the main study demonstrated improved diagnostic accuracy and clinical impact of iuMRI, further considerations must be addressed to fully answer whether or not iuMRI should be embedded in routine clinical practice. First, the number of cases without an ORD was noted; a likely consequence of this is that some diagnoses will not have been picked up and will only become clinically apparent as the child develops. It is also possible that some incorrect diagnoses have been made. Thus, the 6-month follow-up used in the original study may include diagnoses that differ from the confirmed diagnosis later in life.

Perhaps more importantly, the main study focused purely on the diagnosis, whereas to parents the functional significance (i.e. motor, cognitive and behavioural) of the brain abnormality is of a greater interest. The increased diagnostic accuracy only partially addresses the question of whether or not MRI improves outcomes for the child and the family; outcomes in this context are very wide-ranging and include more timely medical interventions, such as appropriate specialist referrals or the early involvement of therapists, as well as improved understanding/knowledge, impacts on quality of life and parental anxiety. Following the diagnosis of a fetal abnormality, parents often want to hear all the possible outcomes85 in order to build a picture of what those difficulties would be like to live with. Aspects such as the emotional impact of caring for a child with potentially serious disabilities and the implications for family life (financial and logistical) affect their perceived ability to be able to provide for the child’s possible needs. 86 The follow-up study was intended to address these concerns by examining the predictive validity of antenatal diagnosis of a brain abnormality for longer-term functional outcomes over the course of life. Specifically, we evaluated whether or not a prognosis based on iuMRI offers an improvement on those prognoses made by ultrasonography alone through a longer-term follow-up of our cohort.

Study aims and objectives

There were three distinct projects within the follow-up study, the aims and objectives of which are detailed below.

Participants and eligibility criteria

Study participants had undergone ultrasonography and iuMRI as part of the original MERIDIAN study during their pregnancy. During the original consent process, women were asked whether or not they were happy to be contacted by the research team for future studies about their child’s development. Participants were eligible if they:

• participated in the MERIDIAN study and had a surviving child aged ≥ 2 years

• underwent both ultrasonography and iuMRI during pregnancy as part of the MERIDIAN study

• consented to be contacted by the research team about future studies related to their child’s development.

For children who were no longer alive, the date of death and the cause of death were obtained when possible. No contact was made with the family.

Children who were aged > 42 months (term corrected) were not eligible for an assessment with the Bayley Scales of Infant and Toddler Development, Third Edition (BSID3), because there is a ceiling for the upper age at which a child can be tested using this tool, but they did complete an age-appropriate Ages and Stages Questionnaire-3 (ASQ-3); the medical case notes were reviewed for additional diagnoses up until the child was 42 months of age.

Participants were excluded from the study if any of the following criteria were met:

• if the child born to a participant in the MERIDIAN study was no longer alive

• if the child was no longer in the care of the biological mother who consented to participate in the original MERIDIAN study

• the parent or guardian was unable to give informed consent

• the parent or guardian was unable to understand English (except in situations when another parent/guardian of the child could translate and provide consent)

• if the participant was withdrawn at any stage of the MERIDIAN study.

Participant tracking and recruitment

To assess eligibility prior to contacting the participant for the follow-up study, the research team completed a consent form audit to identify those participants who had consented to be approached about future studies regarding their child’s development (optional question 7 in the original consent form). Of those who had consented, medical notes and available NHS systems were checked to assess eligibility and suitability for the study.

Section 251 Confidentiality Advisory Group approval (reference 15/CAG/0155) was obtained to allow the central research team to check that the child was still alive using, when applicable, the Health and Social Care Information Centre (now NHS Digital) Patient Tracking system.

Once participants were screened and deemed eligible, they were sent a letter of invitation and a consent form/reply form to the most recent known address. The ASQ-3 was posted out along with the consent form.

Outcome measures and assessments

Sample size

The sample size was constrained by the number of MERIDIAN study participants consenting to further study. This did not affect project 1 for which the 6-month gestational age-corrected diagnosis was used if no follow-up was available, meaning that the original sample size calculation remained applicable. The sample size for project 2 was informed by the fact that there were almost 200 instances when the prognosis changed as a result of iuMRI, of which 38 are now classified as the poorest prognosis. It was also assumed that there would be approximately 400 surviving infants. Scaling these prevalences down, 400 cases would have a 90% power to detect a 20% increase in sensitivity and a 10% increase in specificity using the tests outlined above at a two-sided significance level of 5%.

Statistical methods

In project 1, the diagnostic accuracy was analysed as per the main MERIDIAN study, with the most recent diagnosis (original MERIDIAN study diagnosis or subsequent) used as the ORD.

For project 2, the sensitivity, specificity, PPVs and negative predictive values (NPVs) of iuMRI and of ultrasonography were reported. The sensitivity of each test was calculated for surviving children and also for non-surviving children who had died because of reasons other than TOP. For surviving children, the sensitivity was defined as the:

• number of participants with abnormal development and a prognosis of ‘poor’

• number of participants with abnormal development and a non-missing prognosis.

The ability of ultrasonography and iuMRI to detect abnormalities associated with non-survival was defined as the:

• number of non-surviving infants and a prognosis of ‘poor’

• number of non-surviving infants and a non-missing prognosis.

In order to provide an overall sensitivity averaged across both surviving and non-surviving infants, an adjustment was required to ensure that surviving and deceased infants were represented proportionately to their incidence. By definition, all deceased infants may be considered to have an ‘abnormal’ outcome and, therefore, a response rate of 1. Defining the overall uptake of project 2 among surviving infants by p, the sensitivity was therefore:

in which N_s infants were classified as having an abnormal prognosis and of whom n_s had received a poor prognosis; the total number of deceased infants at 2–3 years was N_d, of whom n_d were classified as poor prognosis.

The specificity was calculated for surviving children and was defined as the:

• number of infants with normal development and a prognosis of either ‘favourable’ or ‘normal’

• number of infants with normal development and a non-missing prognosis.

The relative prognostic accuracy of ultrasonography and iuMRI was compared by calculating the difference in their respective sensitivities and specificities using McNemar’s paired-sample chi-squared test. CIs were calculated using the Wilson method for simple proportions. 88

The analyses were repeated with children classified as ‘abnormal’ and ‘at-risk’ neurodevelopment considered as poor outcomes. Developmental scores (BSID3, GMFCS, SDQ and ASQ-3) were summarised using box plots.

To further understand the characteristics of those recruited to the study compared with the original cohort, specifically in relation to the prognosis on ultrasonography and iuMRI, simple bar charts were produced. The number of infants who were not approached, who had died within the original study, who were approached but did not consent and those who consented were presented by ultrasonography and iuMRI prognosis category. Waffle charts comparing the proportion within each prognosis category in the original cohort with those recruited into each project were produced. The number and type of deaths were also presented by prognosis in a bar chart.

Owing to the small numbers who had completed developmental assessment, no further subgroup analysis was conducted.

Protocol amendments after study initiation

Details of substantial amendments submitted to the REC, which were important changes to trial methodology, are listed below.

Participant flow

Of the 829 participants from the MERIDIAN study who had both ultrasonography and iuMRI, 214 could not be approached (Figure 6). A total of 615 participants were screened and 39 were found to be ineligible at screening owing to the following: not wanting to be contacted (n = 12), being unable to confirm if the child was alive (n = 3), the child was no longer in the care of the mother (n = 12) and it was deemed to be inappropriate to contact (n = 12). A further 22 participants were excluded because a correct address could not be found to which an invitation letter could be sent. A total of 554 invitation letters were sent; 196 participants did not respond and 41 declined to enter the study. A total of 317 participants agreed to initial consent. Of these, 79 participants did not agree to formal consent.

FIGURE 6.

Participant flow in the MERIDIAN follow-on study.

Eight participants were screened in error as they did not complete iuMRI; two of these participants completed consent and a developmental assessment despite being ineligible. For simplicity, these cases have been excluded from the flow diagrams.

A total of 238 participants formally consented to project 1 (Figure 7); of these participants, medical case note review could not be undertaken for 28 children. Medical case note review was undertaken in 210 children and updated imaging was found in 81 cases. Of those with updated imaging, five cases had imaging that changed the original ORD and five had an ORD when they had not had one previously. When combined with the original MERIDIAN study data, 574 cases were included in the primary analysis and 643 cases were included in the secondary analysis.

FIGURE 7.

Participant flow in the project 1 component of the follow-on study.

A total of 132 participants formally consented to project 2 (Figure 8); 127 of these completed at least one type of developmental assessment. A further 52 participants completed the ASQ-3 as part of project 1. When combined, a total of 179 participants completed at least one developmental assessment. Eleven participants had relevant information in their medical case note review as part of project 1 that could be used by clinicians to provide an overall assessment of development. A total of 190 participants completed the clinicians’ review. The primary analysis was carried out on these 190 cases and on 78 children who had died. Fifty-three children had died during the original MERIDIAN study and a further 25 children were known to have died since the end of the MERIDIAN study.

FIGURE 8.

Participant flow in the project 2 component of the follow-on study.

Participant characteristics

Participants in projects 1 and 2 appear to be fairly representative of the overall MERIDIAN study cohort (Table 35). Compared with the original cohort, a lower proportion of participants with a poor prognosis on ultrasonography or iuMRI were recruited into project 1 or 2. A high proportion (55% on ultrasonography and 62% on iuMRI) of participants with a poor prognosis died within the original MERIDIAN study follow-up period (Figure 9). This can also be seen in Figure 10, demonstrating that a high proportion of non-surviving infants had a poor or intermediate prognosis. Additional tables and figures summarising the participants recruited in the study in comparison with the original cohort can be found in Appendix 11.

FIGURE 9.

Bar chart showing the MERIDIAN study cohort (n = 829), by prognosis on (a) ultrasonography and (b) iuMRI for project 2. a, Death within original MERIDIAN include terminations, stillbirths, fetal deaths and deaths within 6 months of birth.

FIGURE 10.

Bar chart showing the non-surviving infants in the MERIDIAN study cohort, by prognosis on (a) ultrasonography and (b) iuMRI (n = 202).

Project 1

Project 2

Developmental assessment scores and categories

Box plots showing the distribution of BSID3 scores appear to show very little difference between an ultrasonography prognosis and an iuMRI prognosis (Figure 11). Median BSID3 composite scores are fairly similar in those patients with a normal or favourable prognosis on ultrasonography and those with the same prognosis on iuMRI. There is a small difference between ultrasonography and iuMRI in median composite scores in those patients with a poor or an intermediate prognosis. iuMRI demonstrates slightly lower medians in each of the composite subscale scores. Lower scores indicate lower developmental ability. Box plots of ASQ-3 and SDQ can be found in Appendix 12.

FIGURE 11.

Box plots of BSID3 subscale composite scores, by ultrasonography and iuMRI prognosis in the MERIDIAN study. (a) Cognitive index; (b) language index; and (c) motor index.

The majority of participants (106/190) who were followed up were assessed as normal (Table 38). A total of 58 out of 190 (31%) participants were assessed as abnormal. Descriptive statistics tables for the developmental scores are presented in Appendix 13.

TABLE 38 - Number and percentage of participants in each development category overall and according to each individual assessment tool

Development tool (N = 190)	n (%)
Overall development category
Abnormal	58 (31.0)
Borderline	26 (14.0)
Normal	106 (56.0)
BSID3 category
Abnormal (< 70)	21 (24.0)
Borderline (70 to < 85)	12 (14.0)
Normal (≥ 85)	55 (63.0)
ASQ-3 category
Abnormal	49 (29.0)
Borderline	24 (14.0)
Normal	98 (57.0)
SDQ category
Abnormal (≥ 16)	17 (17.0)
Borderline (13–15)	5 (5.0)
Normal (0–12)	79 (78.0)
GMFCS level
I	26 (84.0)
II	2 (6.5)
III	1 (3.2)
IV	2 (6.5)
V	0 (0.0)
Cerebral palsy
Yes	4 (2.1)

Project 3

Background

Recruitment into the MERIDIAN study was restricted to women with fetuses for which a brain abnormality was seen on ultrasonography. This allowed an assessment of the diagnostic accuracy of ultrasonography and iuMRI in terms of the false-positive rate of detecting a fetal brain abnormality; however, the false-negative rate cannot be estimated by the main MERIDIAN study. The add-on study addressed this by recruiting pregnant women carrying a fetus without any form of abnormality detected on detailed ultrasonography. This allowed us to describe the false-negative rate of ultrasonography.

Study aims and objectives

The aim of the add-on study was to answer the question ‘What is the false-negative rate of antenatal ultrasound performed by fetal medicine specialists when trying to detect fetal brain abnormalities?’.

Participants and eligibility criteria

Sample size

Starting from the assumption that no ultrasonography false negatives will be found, the study aimed to recruit 200 fetuses on the basis of the 3/n rule,89 a large sample approximation of the upper 95% CI for very rare events. This allowed the NPV of ultrasonography to be estimated to an upper confidence limit of 1.5% in the absence of any abnormal scans and to within a standard error of ≤ 2% for an incidence of < 10%.

Statistical methods

In this add-on study, the diagnostic accuracy of ultrasonography was defined as the probability that a negative scan was correct; this corresponds to the NPV. For iuMRI, diagnostic accuracy was the percentage of cases correctly diagnosed, and the proportion of normal diagnoses corresponds to the NPV of iuMRI following ultrasonography. No attempt was made to estimate the sensitivity or specificity as the study was not a prospective sample of all pregnancies.

The fetal brain was assumed to be normal if the iuMRI study was normal, but in cases when an abnormality was suspected by iuMRI the pregnancy was followed up to ascertain whether ultrasonography, iuMRI or neither were correct.

Accuracy of diagnosis

Our main finding was that the overall diagnostic accuracies of ultrasonography and iuMRI were 67.9% and 92.8%, respectively, meaning that iuMRI showed an improvement in diagnostic accuracy of 24.9%. The diagnostic accuracy of ultrasonography was lower in fetuses with a gestational age of > 24 weeks (64.2%) than in those with a gestational age of 18–23 weeks (69.9%), whereas iuMRI remained similar at 93.5% for those > 24 weeks’ gestational age and 92.4% for those with a gestational age of 18–23 weeks. This indicates an improvement in diagnostic accuracy of 22.5% in the 18–23 weeks group and of 29.3% in the > 24 weeks group. The reduced accuracy of ultrasonography observed in fetuses at > 24 weeks’ gestation may be a result of several factors, including difficulties caused by the ongoing ossification of the fetal skull, the increased physical size of the woman and the descent of the fetal head into the maternal pelvis.

The accuracy of positive ultrasonography in diagnosing fetal brain abnormalities has been the subject of several previous studies. 15–18 Our finding that ultrasonography was accurate in 69.9% of fetuses at < 24 weeks’ gestation and in 64% of fetuses in later stages of gestation is consistent with those reports. This leads us to suggest that iuMRI significantly increases the accuracy of fetal brain diagnoses compared with ultrasonography alone in all fetuses at ≥ 18 weeks’ gestation.

The three commonest ultrasonography diagnoses were VM as the only intracranial abnormality (54%), an abnormality restricted to the contents of the posterior fossa (15%) and failed commissuration (14%). There were statistically significant improvements in diagnostic accuracy for iuMRI compared with ultrasonography in all three subgroups (VM, 8.5%; posterior fossa, 22%; failed commissuration, 60.7%), indicating that iuMRI has advantages in diagnosing a broad range of developmental abnormalities of the brain and should be included in the diagnostic pathway regardless of the suspected pathology type. Notably, additional brain abnormalities were present in 10.1% (31/306) of cases in which ultrasonography diagnosed isolated VM. iuMRI accurately recognised 27 of these additional brain abnormalities, resulting in a diagnostic accuracy of 98.4%, compared with ultrasonography alone at 89.9%.

Further to this, we were able to recruit women with low-risk pregnancies and normal fetuses on ultrasonography to undergo iuMRI of the fetal brain. To our knowledge, this is the first study of this population and our results confirm the ability of both ultrasonography and iuMRI to correctly confirm when the brain development of the fetus is normal. This highlights the validity of ultrasonography as the primary screening imaging method for pregnancy and further supports the need for additional iuMRI when abnormalities are detected.

Diagnostic confidence

The importance of diagnostic confidence as well as diagnostic accuracy when assessing an imaging technology is often overlooked and less well studied, even though Ng and Palmer58 have explained its relevance. To the authors’ knowledge, there are no published studies that report on the diagnostic confidence of iuMRI. Our study was designed to address this issue and our data show that iuMRI increased the proportion of high-confidence diagnoses by 13%, from 82% on ultrasonography alone to 95% after iuMRI. As highlighted by Ng and Palmer,58 an incorrect diagnosis made with high confidence can result in an inappropriate change in management (in this case an inappropriate TOP). We found that iuMRI gave fewer high-confidence but incorrect diagnoses than ultrasonography (6% vs. 22% of the total number of cases). Alternatively, TOP may not be offered if an imaging diagnosis is made with low confidence. This again can bring about medical errors if the diagnosis is found to be correct, as withholding an intervention may have detrimental effects. The MERIDIAN study data demonstrate that iuMRI resulted in fewer low-confidence diagnoses (5% iuMRI vs. 18% ultrasonography alone) and fewer incorrect low-confidence diagnoses (3% iuMRI vs. 8% ultrasonography alone). 44

Counselling and management

As far as we were aware, this is the first prospective analysis of the clinical impact of iuMRI in fetuses with brain abnormalities; in this paper we have shown that iuMRI brings about changes in counselling and management in a high percentage of cases. 44 Specifically, iuMRI provided additional diagnostic information in 49% of cases, caused a documented change in prognosis in 44% of cases and had major effects on counselling in 15% of cases. The contribution of iuMRI to overall clinical management was judged to be ‘significant’ in 26% of cases and had either a ‘decisive’ or ‘major’ influence in a further 9% of cases. This is a considerably larger effect than the 5% change in management anticipated at the design stage, an increase that can be attributed to the greater than predicted improvement in diagnostic accuracy. 44

Prognostic importance

The importance of the long-term functional significance of the suspected brain abnormality must not be overlooked and is of great importance in preparing parents for the potential outcomes. There are limited data available for many brain abnormalities (or combinations) of fetal abnormalities from which fetal medicine experts can draw to predict outcome66 and to inform their counselling, although data on outcomes for certain abnormalities, such as isolated VM and ACC, are more plentiful. The 2- to 3-year follow-up study aimed to address this gap. Although recruitment figures were lower than anticipated, our findings indicate that, despite the increase in diagnostic accuracy and confidence provided by iuMRI, fetal medicine experts still find it difficult to prognosticate accurately, especially in terms of predicting abnormal development. There are a number of possible explanations for this uncertainty when providing a prognosis, including the experience of the fetal medicine expert in reading and interpreting the iuMRI report. It is possible that a neonatologist or paediatric neurologist, who are more familiar with these types of reports or conditions and may see affected children for long-term follow-up, may be more able to predict a prognosis with greater accuracy. The growing culture of litigation in health care may also be leading to concerns among fetal medicine experts regarding the counselling and management of a PND. 90 The impact of a brain abnormality, even if the abnormality is confirmed, can vary markedly between individual infants. The prognosis of an abnormality can include a great deal of uncertainty and the information and steps required to reach a decision can be complex. 64

Our key finding in relation to long-term prognostic importance is that iuMRI appears to be better at ruling out an abnormal developmental pattern than ultrasonography, most likely by excluding the diagnoses that were suspected on ultrasonography. iuMRI was associated with a statistically significant improvement of 19% in predicting a ‘normal’ and ‘at-risk’ developmental outcome than ultrasonography alone. By contrast, there was a smaller, non-statistically significant difference observed between iuMRI and ultrasonography in predicting an abnormal development. This finding may be a result of the small number of participants from the original cohort with a poor prognosis being included in the follow-up study.

Among the 54 children identified as having abnormal neurodevelopment at 2–3 years and for whom a prognosis was available for both ultrasonography and iuMRI, 11 (19%) and 13 (24%) had received a poor prognosis from ultrasonography and iuMRI, respectively (difference of 5%, 95% CI –7% to 19%; p = 0.505). A total of 78 children died, either in utero or prior to the 2- to 3-year follow-up, of whom 73 had received a prognosis on both modalities. Ultrasonography identified 29 (37%) children as having a poor prognosis compared with 36 (48%) children on iuMRI (difference of 11%, 95% CI –3% to 22%; p = 0.146).

Isolated mild VM is often associated with a favourable outcome;91 however, within our cohort 22% of cases of isolated mild VM diagnosed by ultrasonography had abnormal development at 2–3 years of age compared with 18% of cases on iuMRI. These findings are higher than the 11% identified in a review of 90 articles investigating the counselling and outcome when isolated mild VM is diagnosed on prenatal screening,92 yet it remains unclear whether or not this value is higher than in the general population. Our finding could be as a result of its association with other abnormalities. 93 The majority of children with isolated mild VM on either screening modality were recorded as having a resolved VM and no further diagnoses on ORD, but proportionately more unresolved VMs were found among those children with abnormal development at 2–3 years. Fifteen children had additional (non-brain-related) conditions or significant life events (e.g. cardiac complications, chromosomal abnormalities or prolonged hospitalisation) that are likely to have been associated with delayed development. These were proportionately more common among children assessed as having abnormal neurodevelopment (7/19) than among those assessed as having at-risk (4/12) or normal development (4/54).

Selection bias may play a role in this finding. Parents who have a particular concern about their child may be more likely to enrol in a study such as this, especially if they have already been discharged from local services or feel that their concerns are not listened to by health professionals.

Acceptability to patients

Considerable work has already been undertaken and published confirming the safety of iuMRI. The general risks of MRI apply and include scanning people with contraindications (e.g. pacemakers) and the risk of taking ferromagnetic objects into the scan room, which become a projectile threat. Rigorous screening procedures are conducted prior to the procedure to mitigate these risks. Screening was performed successfully in our study and led to a high completion rate of iuMRI (> 99%). Acceptability of the procedure was also high, with at least 95% of women reporting that they would have an iuMRI study again if a future pregnancy was complicated by a fetal brain abnormality.

Radiologists’ experience

The radiologists had a wide range of iuMRI-reporting experience before the start of the study. The most experienced radiologist was from the host site (Sheffield), where approximately two-thirds of the cases in the study were performed. The overall iuMRI error rate was 7.0% and our analysis has shown that those with more experience were less likely to make errors. It must be noted that even the less experienced radiologists consistently had an error rate of < 10%. The analysis of errors has allowed us to investigate the factors that may influence the diagnostic accuracy of iuMRI, and this work has helped to identify practices that could be implemented into a national screening programme to reduce the risk of diagnostic errors.

Health economics

The health economic analysis estimated the incremental cost and benefits from iuMRI compared with ultrasonography alone. As the MERIDIAN study was not a controlled trial, the outcomes following ultrasonography alone had to be estimated. Three scenarios were considered to estimate the proportion of people who had a TOP after iuMRI or ultrasonography. The base case used the proportion of people who had a TOP after iuMRI for iuMRI, and the proportion of people who were offered TOP after ultrasonography (and before iuMRI) for ultrasonography. Scenario 1 used the proportion of people offered a TOP after iuMRI for iuMRI, and the proportion of people who were offered TOP after ultrasonography (and before iuMRI) for ultrasonography. Scenario 2 used the proportion of people who had a TOP after iuMRI for iuMRI, and adjusted the proportion of people offered a TOP after ultrasonography (and before iuMRI) by the proportion of those who were offered TOP following iuMRI and who had a TOP for ultrasonography. Therefore, the addition of iuMRI to the pathway changed the proportion of people who had a TOP instead of a delivery. In the base case, more people had a TOP after iuMRI than after ultrasonography, but in scenarios 1 and 2 more people had a TOP after ultrasonography than after iuMRI.

The addition of iuMRI to one ultrasonography scan increases costs per person by £264. The resource use for investigations, further imaging and further consultations differed in the MERIDIAN study for participants who had a TOP and for those who did not, with higher costs for those who had a TOP. However, a delivery is much more expensive than a TOP, meaning that the total cost for a participant who had a TOP is less than that for a participant who had a delivery. This means that if iuMRI led to sufficiently more participants having a TOP than a delivery, it would be cost-saving. In the base case, the proportion of participants having TOPs after iuMRI was much smaller than after ultrasonography, and so there was an incremental cost of £480 and a cost per management decision correctly revised of £2203. In scenarios 1 and 2, the proportions of participants having a TOP were much closer after iuMRI and ultrasonography, so the incremental costs were much lower.

In addition, the difference in costs and in the cost per management decision appropriately changed are sensitive to the assumptions around which costs should be included. Inclusion of participant-borne costs increases the incremental cost as participants have to spend additional time travelling to and attending iuMRI scans, but participant-borne costs are not routinely included in the evaluation of health technologies conducted by the National Institute for Health and Care Excellence in England. The exclusion of TOP and delivery costs reduces the uncertainty associated with the approach taken to estimate the number of TOPs with iuMRI and ultrasonography, as the major cost driver is the cost of the iuMRI itself.

In all scenarios, the incremental cost is consistently < £600 and the cost per management decision appropriately changed is always < £3000. In a scenario comparing iuMRI with repeat ultrasonography, iuMRI is cost-saving.

Whether or not iuMRI represents value for money depends on the willingness to pay per management decision appropriately changed. The probabilistic sensitivity analysis demonstrated that there is 100% probability that iuMRI is cost-effective if decision-makers are willing to pay ≥ £4000 per management decision appropriately changed.

Although the parameters for costs and effectiveness are uncertain, the maximum EVPIs for the base case and scenarios 1 and 2 are very low, indicating that there is little value in conducting additional research to determine more precise estimates for the parameters.

Sociological substudy

Prenatal diagnosis technologies attract significant social science attention because they constitute a case study of intrinsic interest for those exploring ethical, legal and social dilemmas, an opportunity to interrogate how social contexts shape understandings of health and medical knowledge. Our findings provide insights from both the service users’ and professionals’ perspectives on how these issues affect a PND of brain abnormality in the context of iuMRI. Service users provide insights into lay perspectives on acceptability of and barriers to care, and professionals provide insights into the possible implementation of NHS health-care policies. We consider both perspectives when exploring the social issues that, as Williams31 notes, are often neglected when rolling out new technologies in fetal medicine. In both service users’ and health professionals’ accounts, there was much similarity in how the understandings of acceptability of care were conceptualised in complex ways, acknowledging trade-offs and weighing up options.

Our findings suggest that iuMRI in a PND for brain abnormality is broadly acceptable to both service users and health professionals, but it is not considered a self-evident good. The acceptability of iuMRI was conditional, with service users’ and professionals’ accounts demonstrating careful consideration of the costs and benefits of iuMRI, and providing suggestions for how service user and professional acceptability could be assured and enhanced in future implementation. Nevertheless, professionals found it easy to recruit women to the trial, and the quantitative data from the patient survey demonstrate the high level of women’s baseline acceptance of iuMRI as part of their care. Interviews with women showed that they were usually ‘information hungry’ and prepared to tolerate significant discomfort to maximise information prior to decision-making about their pregnancy. Apart from specific issues, such as claustrophobia, the main barriers to patient acceptability centred on having to travel (linked to, but not limited to, socioeconomic costs). Professionals also conceptualised iuMRI as useful in the PND of brain abnormality, but with diverse views on the types of brain abnormality for which iuMRI was considered to ‘add value’ to the PND process.

The balancing of the advantages and disadvantages of iuMRI in the PND for brain abnormality reveals three main issues: accessibility, triangulation and quality assurance.

Strengths and weaknesses of the study

The MERIDIAN study was designed to address the weaknesses in the current literature by conducting an appropriately powered, pragmatic, prospective, multicentre study. This allowed us to complete a robust evaluation of the extent to which iuMRI improves the diagnostic accuracy of ultrasonography alone and to assess the clinical impact on the counselling and management of the pregnancy. The pragmatic nature of the study allowed us to assess and demonstrate how iuMRI could be incorporated in the diagnostic pathway in the UK for diagnosing fetal brain abnormalities.

Our primary analysis included only cases for which the iuMRI was performed within 2 weeks of the ultrasonography and an ORD was available, which addressed a weakness of many prior studies that often did not rely on an ORD. To our knowledge, this is the first study to present an analysis of diagnostic confidence and assess its impact on appropriate management decisions.

Within this study, we have not only been able to confirm the improved diagnostic performance of iuMRI when diagnosing fetal brain abnormalities but we have also shown that ultrasonography and iuMRI are both capable of confirming when the development of a fetal brain is normal. We believe that this is the first study within this population to confirm this and will have implications for designing a national antenatal screening programme.

The limitations of the study include the potential for reporting bias from the referring clinicians on the diagnostic and prognostic outcomes, as well as the information about their confidence in the findings. However, changes in the prognosis were under-reported by clinicians, suggesting that they were not systematically going back to review their initial decisions at the time of reviewing the iuMRI results.

Two further limitations were the high proportion of participants without an ORD (24%) and the high proportion of iuMRI scans undertaken at the host institution (67%). The first limitation arose because we were unable to undertake diagnostic investigations solely for research purposes, meaning that our ORD relied on collecting data from investigations completed for clinical purposes, which was not always done. This was particularly an issue following TOP or fetal demise, which accounted for over half of the cases of missing ORDs. The second issue is an important limitation because two-thirds of the scans were undertaken by two radiologists who were experienced in undertaking MRI specifically in fetal abnormalities; other centres will, at present, not have this level of expertise. To address this, we undertook two sensitivity analyses. The first found that the improvement in diagnostic accuracy following iuMRI could not be attributed to missing data, with even extreme imputation of missing data showing iuMRI following ultrasonography to be superior to ultrasonography alone. The second showed only a marginally higher accuracy at the host institution compared with other centres. Both of these analyses appear to confirm the robustness of the difference in favour of iuMRI.

Strengths and weaknesses of the 2- to 3-year follow-up study

The main limitation of the follow-up study was the small number of children included and for whom developmental data could be collected. Owing to delays in study initiation, many of the children were out of the age range to be offered the BSID3 assessment, owing to the ceiling effect. Therefore, we relied on ASQ-3 data for many children. The ASQ-3 is a screening tool and its sensitivity, specificity, PPV and NPV are not perfect. It is designed not to miss true positives at the risk of erroneously diagnosing some children as abnormal who would then receive a correct diagnosis on further assessment by a clinician. Our analysis of ASQ-3 classification compared with the BSID3 assessment confirmed that the PPV was relatively low. Although this affects the generalisability of the data and limits the conclusions that can be drawn, it does highlight the importance of further research into the uncertainty of the prognostic importance of brain abnormalities diagnosed antenatally. There are a number of factors that influenced our recruitment rate; these included a delayed start to the study (including delay with funding approval and issues with site set-up). These delays meant that a number of the children were out of the age range for the BSID3 assessment and the ASQ-3 and, therefore, only a medical case note review could be completed. There was also a high proportion of families who could not be contacted. The approach protocol was updated during the study to increase the number and type of contact options available to the research team (as detailed in Chapter 5, Protocol amendments after study initiation); however, many families had moved and there was no follow-up address available or contact telephone numbers were out of use.

The BSID3 is a well-validated tool for assessing development in early infancy; it is widely used and generates standardised scores that allow corrections for differences in age at measurement. We wish to highlight that despite the widespread use of BSID3 in follow-up studies, and the lack of better alternatives, it remains a tool to assess developmental stage and not a tool to accurately predict the final developmental or cognitive outcome later in life; however, it is currently widely considered to be the reference standard. It is entirely possible for a child who performs well on the BSID3 between 2 and 3 years to manifest developmental or learning difficulties later in life and vice versa. Owing to the complexity of early childhood development, the BSID3 may also fail to detect or predict the presence of important behavioural or psychological problems in later life. It is also important to note that the cognitive measure is affected by the child’s motor ability and, therefore, a lower score may be obtained that may not accurately reflect the child’s ultimate level of cognitive function. The BSID3 is also time-consuming and more than half of the children did not undergo the full BSID3 assessment, with the development instead ascertained by a brief questionnaire-based assessment. Although unavoidable, this is likely to have contributed a degree of measurement error to the final classifications.

Strengths and weaknesses of the economic analysis

The key limitations of this analysis lie in the form that the analysis takes and the outcomes that are used. The outcome of a management decision appropriately revised combines both cases in which a decision to continue the pregnancy was changed to TOP and cases in which a decision to terminate was changed to a continuation of pregnancy. In reality, the willingness to pay per appropriately revised decision may differ between those situations in which the intention to terminate a pregnancy was revised to a continuation of the pregnancy and those situations in which the continuation of the pregnancy was revised to a TOP. The analysis does not consider the implications of inappropriately changing a management decision (e.g. when iuMRI was incorrect and changed the prognosis in 1.81% of cases). The impacts on decision-makers, parents and children are clearly profound, both when a decision is taken to terminate a healthy fetus or when the decision is to continue the pregnancy when a fetal abnormality exists. These limitations should not be overlooked when considering the cost implications. Findings relating to these aspects of the study are reported in other sections of this report and should be considered in conjunction with evidence of cost-effectiveness.

Strengths and weaknesses of the sociological substudy

There are limitations to this study. The use of purposive sampling for maximum variation in the substudy samples allowed the small samples to encompass broadly representative ranges of views. The findings from the analysis of the data are not intended to be generalised in a simplistic way to the relevant population, but they do provide a robust account of the likely range of views within that population. 78 Although all eligible participants were included in the survey, not all participants responded to both surveys. Therefore, the results of the surveys do not include information on non-respondents and we do not know whether or not that subgroup had different views from those expressed by participants. In addition, all of our participants were recruited from a subgroup that actively chose to get involved in the MERIDIAN study (whether as a health professional or as a patient), and so the findings do not encompass the views of those who did not take part in the larger clinical study.

Strengths and weaknesses of the add-on study

The add-on study was a prospective study of fetuses with no brain abnormalities on ultrasonography. Although the sample size of 205 participants is notable, it was small in terms of its ability to estimate the NPV of ultrasonography. The fact that one abnormality was found (a NPV of 99.5%, 95% CI 97.3% to 100.0%) is reassuring in terms of the low error rate but may raise concern in mothers who may not grasp the possibility that ultrasonography is potentially an imperfect rule-out in 2% of pregnancies. Although the false-negative rate is likely to be lower than this, our study was not of a sufficient size to estimate the error rate more precisely.

Although the main focus of this study was on the use of iuMRI in fetuses with a suspected brain abnormality, it is likely that iuMRI will be considered for fetuses that are considered ‘at risk’ but whose brain appears normal on ultrasonography. The incident of a fetus with an abnormality not being detected by ultrasonography (and indeed the second case with mild VM) is a timely reminder that ultrasonography is imperfect. For the foreseeable future, it is not viable for iuMRI to be used routinely in the manner in which ultrasonography currently is, but future research may be useful in assessing in which circumstances the addition of iuMRI to the diagnostic pathway would be of benefit.

Writing group

Paul D Griffiths (Academic Unit of Radiology, University of Sheffield), Michael Bradburn (CTRU, School of Health and Related Research, University of Sheffield), Michael J Campbell (School of Health and Related Research, University of Sheffield), Cindy L Cooper (CTRU, School of Health and Related Research, University of Sheffield), Nicholas Embleton (Newcastle Neonatal Service, Royal Victoria Infirmary), Ruth Graham (School of Geography, Politics and Sociology, Newcastle University), Anthony R Hart (Department of Perinatal and Paediatric Neurology, Sheffield Children’s Hospital NHS Foundation Trust), Deborah Jarvis (Academic Unit of Radiology, University of Sheffield), Mark D Kilby [(1) Centre for Women’s & Newborn Health, Institute of Metabolism & Systems Research, University of Birmingham; (2) Fetal Medicine Centre, Birmingham Women’s Foundation Trust (Birmingham Health Partners)], Mabel Lie (Institute of Cellular Medicine, Newcastle University), Gerald Mason (Leeds Teaching Hospitals NHS Trust), Laura Mandefield (CTRU, School of Health and Related Research, University of Sheffield), Cara Mooney (CTRU, School of Health and Related Research, University of Sheffield), Rebekah Pennington (Health Economics and Decision Science, School of Health and Related Research, University of Sheffield), Stephen C Robson (Institute of Cellular Medicine, Newcastle University) and Allan Wailoo (Health Economics and Decision Science, School of Health and Related Research, University of Sheffield).

Trial Management Group

Paul D Griffiths (Academic Unit of Radiology, University of Sheffield), Michael Bradburn (CTRU, School of Health and Related Research, University of Sheffield), Michael J Campbell (School of Health and Related Research, University of Sheffield), Cindy L Cooper (CTRU, School of Health and Related Research, University of Sheffield), Nicholas Embleton (Newcastle Neonatal Service, Royal Victoria Infirmary), Ruth Graham (Newcastle University), Anthony R Hart (Department of Perinatal and Paediatric Neurology, Sheffield Children’s Hospital NHS Foundation Trust), Gerald Mason (Leeds Teaching Hospitals NHS Trust), Stephen C Robson (Newcastle University) and Allan Wailoo (Health Economics and Decision Science, School of Health and Related Research, University of Sheffield).