Diagnostic yield of exome sequencing for prenatal diagnosis of fetal structural anomalies: A systematic review and meta‐analysis

1Ultrasound scanning in pregnancy can reveal fetal structural abnormalities from the first trimester onwards, ranging from isolated minor anomalies to severe, multisystem disorders. These abnormalities may indicate an underlying genetic condition but making a definitive diagnosis can be difficult because many conditions have overlapping fetal presentations. A genetic diagnosis can clarify the prognosis for the fetus, informing parental decision making and clinical management during pregnancy and the perinatal period.

2Conventional genetic testing for fetal anomalies comprises quantitative fluorescence PCR (QF‐PCR) for rapid aneuploidy exclusion, and karyotype or chromosomal microarray (CMA) for detection of chromosomal and sub‐chromosomal abnormalities. Together, these can provide a diagnosis in up to 40% of cases,,, but leave the majority undiagnosed. Targeted sequencing of individual genes can be offered in cases where clinical features or family history suggest a specific monogenic disorder but genome‐wide next generation sequencing (NGS) is now being used more frequently to allow interrogation of multiple genes in a single test. Exome sequencing (ES) examines the 1%–2% of the genome that encodes proteins and harbours around 85% of known disease‐causing genetic variants, and is generally less costly and easier to interpret than whole genome sequencing (WGS). ES can be applied in practice as whole ES (sequencing or analysing all exons), medical/clinical ES (sequencing or analysing only the exons of known Mendelian disease genes), or targeted ES (applying ‘virtual’ gene panels to ES data to limit analysis to genes of interest for a specific phenotype).

3In the postnatal setting ES has proven a powerful tool for the diagnosis of children with neurodevelopmental delay but its application in prenatal diagnosis initially lagged behind. This is largely because long turnaround times precluded the return of results within the timeframe of a pregnancy. Additionally, incomplete or unrecognised fetal phenotypes complicate the interpretation of genetic variants from prenatal ES. Over recent years, rapid analytical pipelines have made it possible to return ES results during an ongoing pregnancy and it is now considered for prenatal diagnosis of fetal structural anomalies in the guidance of several professional bodies.,, Two large prospective studies demonstrated the diagnostic utility of ES for prenatal diagnosis of unselected fetal structural abnormalities, reporting diagnostic rates of 8.5%–10%,, while numerous smaller studies report diagnostic rates above 80% in highly‐selected case series.,, This body of literature has been previously reviewed,, but has rapidly expanded over the last 2 years.

4As prenatal ES is increasingly adopted in clinical settings there remain important questions around best practice for implementation. It will be of interest to health professionals and policy makers to define expected diagnostic yields from prenatal ES for different fetal presentations. This information may help to manage parental expectations during pre‐test counselling and will inform development of further guidelines on when and how to offer prenatal ES. In this systematic review and meta‐analysis we aim to address the question ‘what is the diagnostic yield of ES for prenatal diagnosis in pregnancies complicated by fetal structural anomalies where fetal karyotype/microarray is normal?’ We also explore the diagnostic yield in different phenotypic sub‐groups.

5We devised a systematic review protocol in line with PRISMA guidance, and registered it prospectively on the PROSPERO international register of systematic reviews (reference CRD42020200600). Study authors agreed the protocol prior to conducting the searches. Any required small amendments were made with the consensus of all authors.

6Studies were included in this review if they met the following criteria: (i) Retrospective or prospective cohorts of 10 or more pregnancies undergoing ES (whole, clinical or targeted) or WGS for diagnosis of fetal structural anomalies, (ii) CMA/karyotype was negative or non‐diagnostic, (iii) Testing was initiated based on the prenatal phenotype, (iv) Full text report was available in English language.

7The focus of this review is prenatal diagnosis of structural anomalies, therefore studies with mixed pre‐ and postnatal presentations were only included if it was possible to extract the prenatal cases. Similarly, for studies with mixed indications for testing (e.g. some cases initiated because of fetal phenotype and some because of family history) we only included studies where fetal anomaly cases could be extracted. Studies where ES was completed postnatally for some or all cases but had been initiated based on the prenatal phenotype alone were included. Only variants classified as pathogenic/likely pathogenic according to diagnostic‐standard guidelines and determined to be causative of the fetal phenotype were considered in diagnostic yield, therefore studies using ES purely for novel candidate gene discovery were not included. We also excluded case reports, series of fewer than 10 cases, reviews, editorials and commentaries, and studies using gene panel testing rather than ES.

8Electronic searches of four databases (MEDLINE, Embase, Cochrane library, and Web of Science) were conducted for records published between January 2010 and October 2021. This date range was chosen because prenatal ES was not used prior to 2010. The search was initially conducted on 27/08/2020 and repeated on 15/10/2021. Additional potentially relevant studies were identified by manually searching reference lists of relevant studies and published reviews, and conference proceedings of prenatal and genetics conferences of relevant major societies in last 3 years.

9Search terms were variations on the keywords ‘prenatal diagnosis’ and ‘exome sequencing’. Alternative terms for ‘prenatal diagnosis’ included ‘fetal’, ‘fetus’, ‘prenatal’, and ‘antenatal’. Alternative terms for ‘exome sequencing’ included ‘whole exome sequencing’, ‘exome’, ‘whole genome sequencing’, ‘genome, human’ and ‘sequence analysis, DNA’. The full search strategy is detailed in supporting information S1. We chose relatively broad search terms in an effort to capture as many relevant studies as possible; for the same reason no search limits were applied apart from date range.

10After the removal of duplicates, two reviewers (RM and KO) independently screened titles and abstracts. For abstracts identified as potentially relevant, full text articles were retrieved and reviewed against the inclusion and exclusion criteria. Any disagreement between reviewers was resolved by discussion. Where the same data were presented more than once, the most recent study was selected.

11The following data, where available, were extracted by two reviewers (RM and ES) into a datasheet: study setting, sample size, study inclusion criteria, ES approach, phenotypes used for interpretation (prenatal only or prenatal plus postnatal/post‐mortem), number of fetuses with diagnostic variants, variants of uncertain significance (VUS), incidental findings, gestation at testing, test turnaround time, pregnancy outcomes, and impact on management. For studies that performed CMA in parallel with sequencing, the cases with negative CMA were extracted, in order to be comparable with other studies where chromosomal abnormalities were ruled out prior to ES.

12The quality of the studies was assessed using a modified Standards for Reporting of Diagnostic Accuracy (STARD) checklist, comprising 22 items. The items for reporting considered most important for these studies (as they may influence the reported diagnostic yield) were: (i) clearly described selection criteria, (ii) whether diagnostic variants were confirmed by Sanger sequencing, (iii) whether American College of Medical Genetics and Genomics (ACMG) or equivalent standardised classification was used for variant interpretation, (iv) approach to analysing and reporting VUS and incidental findings addressed. For eight studies the assessment was done independently by two reviewers (RM and MH) with discrepancies resolved by discussion. Once consensus was reached on how to apply the criteria and reviewers' scores were concordant on new study assessments, the remaining studies were assessed by RM only.

13The primary outcome of interest was the incremental diagnostic yield of ES (i.e. the proportion of cases in which a diagnostic genetic variant is detected by ES) over CMA/karyotype, expressed as a risk difference. This was calculated for each study with 95% confidence intervals (CI) and pooled for all studies in a meta‐analysis, using a random effects model with inverse variance weighting. A subgroup analysis was performed to investigate the effect on diagnostic yield of pre‐test case selection for higher likelihood of monogenic disease, where this information was available. Where studies reported data on isolated system anomalies classified by different phenotypic groups, a further subgroup analysis was performed to estimate pooled incremental yields by phenotypic group. Only phenotype groups reported in at least three studies were included in this analysis. Findings were displayed using forest plots. Between‐study heterogeneity was assessed visually and using an I2 statistic with a Q‐test. Publication bias was assessed graphically using a funnel plot with Egger's test. All statistical analyses were performed in R studio v1.3.1093 using the ‘metafor’ package.

14Database searches identified 8123 unique records in total. After screening by title and abstract, and retrieving full text reports, 148 articles were assessed, of which 72 reports from 66 studies were deemed eligible and included in the final review (Figure 1, Table 1). Six reports examined specific phenotypic sub‐groups from previously published larger cohorts,,,,,, hence were not separate studies. These were excluded from the overall meta‐analysis of incremental yield and the subgroup analysis by case selection, to avoid double‐counting of cases. However, these reports provide updated information in their respective phenotypic groups so were included in the subgroup analysis by phenotypic group, in place of the previously reported sub‐groups from their parent studies. Across 66 studies, this review represents 4350 probands undergoing prenatal ES.

15The characteristics of the included studies are summarised in Table 1. Of the 72 included reports, 48 (67%) were published in 2020‐21. Three studies used WGS rather than ES, with or without applying phenotype‐specific panels for analysis.,, Eighteen studies used a clinical exome sequencing (CES) approach in some or all cases, which here we define as targeting either the sequencing or the analysis to Online Mendelian Inheritance in Man disease genes.,,,,,,,,,,,,,,,,, Eight studies targeted ES analysis more tightly using phenotype‐specific gene panels,,,,,,,, while the remainder took a whole ES approach.,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,, Where stated (15 studies),,,,,,,,,,,,,,, the median turnaround time for ES was 20 days (range 4–141). Studies included fetuses with a range of structural abnormality phenotypes, with some studies recruiting cases with fetal anomalies in a specific body system (e.g. renal anomalies or cardiac defects) and others including any structural anomaly. In total, 42/66 studies recruited cases with anomalies in a specific body system or broke down fetal abnormalities within the cohort into distinct phenotypic groups (Table 1). The quality of included reports, assessed using modified STARD criteria, was generally high (Figure 2).

16The diagnostic yield of ES for fetal structural anomalies among the included studies ranged from 5% (95% CI −8 to 18%) to 89% (95% CI 72%–99%). The pooled incremental yield of ES over CMA/karyotype from all studies was 31% (95% CI 26%–36%, p < 0.0001), illustrated in the forest plot (Figure 3). Between‐study heterogeneity was high, indicated by an I2 statistic of 94% (Q = 682, p < 0.001). Testing for publication bias showed significant funnel plot asymmetry (z = 5.6, p < 0.0001 by Egger's test; supporting information S1), indicating possible publication bias.

17Some of the heterogeneity between included studies may be due to differing approaches to case selection. Some studies included fetuses with structural anomalies without any prior review of the clinical features to ascertain risk of a monogenic aetiology, whereas others selected only pregnancies where the pattern of fetal anomalies was highly suggestive for a monogenic disorder (as determined by multidisciplinary teams [MDTs] usually including clinical genetics specialists). Seven studies were excluded from this analysis because it was not possible to determine how the cases were selected, or because mixed selection criteria were used. The other 59 studies were classified as selected or unselected cohorts (Table 2) for a subgroup analysis (Figure 4). This demonstrated that the pooled incremental diagnostic yield of ES for selected cohorts was 42% (95% CI 35%–48%), which is significantly higher than that in unselected cohorts (15%, 95% CI 11%–18%; p < 0.0001).

18Further heterogeneity may arise from the varied types of fetal structural anomaly included in different studies. To investigate this, a second subgroup analysis was performed for 42 studies that reported cases disaggregated by phenotypic group, and where it was possible to distinguish isolated system anomalies from multisystem anomalies (Figure 5). Data from at least 3 studies were available for fetuses in each of 13 phenotypic subgroups based on affected body system (Table 3). The number of fetuses analysed in each subgroup varied widely, from seven cases with isolated abdominal wall abnormalities, to 773 cases with isolated cardiac abnormalities. The subgroups for multisystem anomalies, isolated skeletal anomalies, and isolated brain or central nervous system (CNS) anomalies were the next largest, with over 400 cases each. The subgroup analysis found that the incremental diagnostic yield of ES differed significantly between different phenotypic groups (I 2 96.4%, Q = 177, p < 0.0001). Table 3 details the pooled estimated incremental diagnostic yield for each subgroup, from highest to lowest. The highest diagnostic yields occurred in fetuses with skeletal abnormalities (53% [95% CI 42%–63%], p < 0.0001), neuromuscular abnormalities, usually presenting as fetal akinesia deformation sequence (FADS), (37% [20%–54%], p < 0.0001), and abnormalities in multiple systems (29% [22%–35%], p < 0.0001). By contrast, the lowest rates of ES diagnoses occurred in fetuses with isolated increased nuchal translucency (NT; 2% [0%–5%], p = 0.04), and those with gastrointestinal abnormalities (2% [−4 to 8%], p = 0.5). There were no diagnoses from ES in 7 fetuses with abdominal wall defects and 38 fetuses with chest abnormalities.

19Only seven reports from six studies explicitly assessed the impact of prenatal ES on clinical management. These studies reported that ES results influenced at least one of (i) decisions about termination of pregnancy, (ii) pre‐ or perinatal management, (iii) future reproductive planning, or (iv) counselling on recurrence risk in 26%–79% of cases.,,,,,,

20The findings of our review underscore the value of applying ES in the prenatal context. Meta‐analysis results show an incremental diagnostic yield of 31% from ES over CMA/karyotype alone for prenatal diagnosis of fetal structural anomalies. This diagnostic yield could be considered a conservative estimate because we included only definite diagnoses (variants classified as pathogenic/likely pathogenic and deemed to be causing the fetal phenotype) in our definition of diagnostic yield. Variants reported as ‘possibly diagnostic’, ‘probably diagnostic’, or ‘potentially relevant’ were not included in our analysis but may have been re‐classified as diagnostic after further study. For example, one study of 19 fetuses with severe multiple anomalies of suspected genetic aetiology found a diagnostic yield of 31.6% when strictly using ACMG guidelines, but this rose to 63% when including variants not reportable by ACMG criteria but which were plausible based on animal models, in silico prediction tools or functional studies. Other studies have also shown potentially increased diagnostic yields when probable diagnoses are included.,

21In this way, diagnostic yields can increase over time, as has been previously documented in large postnatal ES studies such as the UK Deciphering Developmental Disorders (DDD) study, where iterative reanalysis was performed after several years. This results from improved bioinformatics methodologies, newly discovered gene‐disease associations and, in the case of prenatal ES, evolving knowledge of fetal presentations of known disorders. For example, in one study from our review, Petrovski et al. reported four variants which were re‐classified (from bioinformatic signatures) to diagnostic variants over the course of the study period in light of new published evidence or additional phenotypes emerging postnatally.

22Our second key finding is that the diagnostic yield of ES for fetal anomalies is higher when cases are pre‐selected for likelihood of a monogenic aetiology by MDTs with expertise in genetics. This is intuitive but, as prenatal ES is increasingly implemented in clinical diagnostic settings, it highlights the value of pre‐test review and MDT collaboration to direct testing where it is most likely to yield a diagnosis. Furthermore, our findings demonstrate that diagnostic yield varies significantly between phenotypic groups: the highest yield observed was for skeletal dysplasias (53%) and the lowest yields were for fetuses with isolated increased NT (2%) and isolated gastrointestinal anomalies (2%).

23This understanding of likely diagnostic yield for different fetal presentations may inform clinical guidelines on offering prenatal ES, however our findings should be interpreted with care. Classification of cases by phenotype and comparison across studies is challenging because of different approaches to the classification (e.g. whether ‘soft markers’ are classed as separate abnormalities, and whether cases were categorised strictly based on isolated system anomalies or based on the system that best describes the predominant phenotype in the fetus). Moreover, fetal phenotypes evolve as organ development progresses, so classification of the same case may differ depending on the gestation at assessment. Further research is needed to clarify the diagnostic utility of prenatal ES for isolated anomalies in systems that have been less thoroughly investigated to date. For example, in this review, for fetuses with chest (respiratory) abnormalities, fetal growth restriction, gastrointestinal anomalies and abdominal wall defects, the sample sizes were too small to reach statistical significance in estimates of pooled diagnostic yield.

24Studies examining other aspects of clinical utility such as impact on clinical management and parental decision‐making will be an important complement to this, however we found only six to include in our review.,,,,,, Three of these studies found that negative as well as diagnostic results impacted upon parental and clinical decision‐making,,, leading Deden et al. to conclude that ‘the efficacy of prenatal [ES] should be evaluated by more variables than diagnostic yield alone.’ This is an area where the need for further research is pressing, including more studies exploring the perspectives of parents undergoing prenatal ES to investigate what personal utility they derive from it.

25This systematic review is the most detailed and comprehensive in this area to date. It included 72 reports of 66 studies, many of which were published in the last 12 months. The reporting of research in the included studies was generally of high quality with a standardised approach to reporting which made it easier to compare study characteristics, and we were able to perform meta‐analyses for a range of phenotypic groups as well as overall. There are however signs of possible publication bias, as indicated by funnel plot asymmetry showing a paucity of published studies with small sample size and low diagnostic yield (supporting information S1). This could reflect publication bias or indeed may represent a true relationship between small sample size and higher diagnostic yield: among the studies included in this review, smaller cohorts were often highly selected (usually after review by clinical genetics specialists) for cases where the clinical presentation suggested a monogenic aetiology. The higher a priori likelihood of a monogenic disorder among such cases translates to higher diagnostic yields, whereas the largest studies to date have been unselected cohorts, including fetuses with isolated and minor abnormalities and anomalies associated with multifactorial aetiology, as well as those more likely to have a monogenic disorder.

26The main limitation of our review was the high degree of heterogeneity between included studies, limiting the precision of comparisons. On performing subgroup analyses to investigate the effect on ES diagnostic yield of case selection criteria or fetal phenotypic group, there remained substantial heterogeneity within many of the subgroups as well as between them, which suggests that neither of these factors alone fully explain the observed heterogeneity amongst the studies. Studies also varied in sample size, sequencing and analysis approach (gene agnostic or targeted), approach to variant interpretation, and whether or not interpretation considered postnatal/post‐mortem phenotype information in addition to prenatal phenotypes. All of these factors are likely to influence the diagnostic yield. For example, even when adhering to standardised ACMG variant interpretation guidelines, there may be some subjectivity in how the guidelines are applied and how it is judged whether a variant explains the fetal phenotype. Studies done in a purely research setting without reporting results back to families may have interpreted variants with candidate causality more generously, leading to higher reported diagnostic yields than in studies that used more conservative reporting standards. In addition, for the phenotype subgroup analysis we only included studies that disaggregated cases by phenotypic group and distinguished cases with a single affected body system from those with multiple system anomalies. This resulted in the exclusion of 24/66 studies from this subgroup analysis. Manual inspection of the raw phenotype data from these studies could enable extraction and classification of more cases in their phenotypic groups and allow a more granular analysis of the incremental yield of ES in different phenotypic groups, as has been reported recently for congenital heart disease, fetal hydrops, and isolated increased NT respectively.,,

27Prenatal ES is a powerful tool to improve the diagnosis of monogenic conditions in fetuses with structural abnormalities. However, it also presents challenges for interpreting variants in the face of incomplete and non‐specific fetal phenotypes while also aiming to minimise identification of VUS and incidental findings. In ongoing pregnancies, results must be returned quickly if they are to inform decisions on termination of pregnancy or prenatal management. Prenatal ES is rapidly being adopted in many clinical settings and its successful implementation will require robust laboratory infrastructure and bioinformatics pipelines, expert parental counselling, multidisciplinary collaboration between fetal medicine and clinical and laboratory genetics specialists to offer and interpret tests appropriately, and clear guidelines to ensure equity of access. The need to sequence fetus‐parent trios in parallel to minimise turn‐around times means that ES remains relatively costly, so in many health systems it may be necessary to triage cases to allocate finite resources. Further research is needed on the clinical impact of prenatal ES, to better understand which pregnancies benefit most and how to prioritise potential cases in a way that maximises benefit to families, and is efficient for healthcare systems.

28None of the authors have any conflict of interests to declare.

29This work is a systematic review of literature and no ethical committee approval was required.