Genome‐Wide Cell‐Free DNA Analysis for Aneuploidy Detection in Miscarriages: Test Performance Meta‐Analysis

ABSTRACT

Objective

To conduct a systematic review and meta‐analysis of published series examining the efficacy of genome‐wide cell‐free DNA (cfDNA) testing in identifying aneuploidy in pregnancies ending in miscarriage.

Methods

A systematic review was conducted encompassing observational studies evaluating aneuploidy detection by genome‐wide cfDNA testing in pregnancy losses before 22 weeks of gestation. A hierarchical summary receiver operating curve was employed to assess pooled sensitivity, specificity, and area under the curve (AUC) of genome‐wide cfDNA versus genetic diagnostic studies in the detection of aneuploidy. Pooled aneuploidy rate, rate of no‐calls, and concordance between cfDNA and diagnostic studies were analyzed using a single proportion meta‐analysis based on the inverse of the variance.

Results

Out of 25 eligible series, eight studies were included for analysis, comprising 552 miscarriages with informative results for both cfDNA and diagnostic testing. Pooled sensitivity, specificity, and AUC were 78% (95% CI: 71%–83%), 91% (95% CI: 86%–95%), and 92%, respectively. Pooled aneuploidy rate, the proportion of no‐calls, and concordance were 61% (95% CI: 53%–69%), 4% (95% CI: 0%–12%), and 84% (95% CI: 81%–87%), respectively. In cases of positive cfDNA results, the risk of aneuploidy increased to 93%, whereas negative results yielded a 28% risk of aneuploidy.

Conclusion

cfDNA testing demonstrates acceptable accuracy in predicting fetal aneuploidy when employed as a screening test in miscarriages. The main advantage of cfDNA testing is that it does not require the availability of products of conception or prior chorionic villi sampling.

Summary.

• What's already known about this topic?

  • ◦Early pregnancy loss is the clinical situation associated with the highest risk of fetal aneuploidy.
  • ◦Cell‐free DNA is not used in pregnancy losses except for vanishing twins where aneuploidy may be unintentionally detected.

• What does this study add?

  • ◦This meta‐analysis assesses the performance of cfDNA in detecting fetal aneuploidy in miscarriages. It includes eight studies encompassing 552 miscarriage cases, revealing a pooled sensitivity of 78% and a pooled specificity of 91%.
  • ◦The primary advantage of cfDNA testing in the context of miscarriages is that it does not require the presence of products of conception or prior chorionic villi sampling.

1. Introduction

Pregnancy loss is the most common adverse event in pregnancy, given that it occurs in up to 20% of gestations [1, 2]. In early pregnancy loss, chromosomal anomalies are recognized to be involved in 50%–70% of cases [3, 4]. Since karyotyping the products of conception (POC) was prone to high risks of both culture failure and maternal cell contamination, this practice was discontinued in many centers [4, 5, 6]. However, either the application of chromosomal microarray analysis (CMA) to POC or the practice of chorionic villi sampling before uterine evacuation can overcome these drawbacks and deliver accurate results [7, 8, 9, 10]. In recent years, medical uterine evacuation has gained acceptance among patients and doctors; therefore, POCs are not available if a medical device for home collection is not used correctly [11, 12, 13].

When an early pregnancy loss happens, women often blame themselves because they feel guilty and experience sadness, anger, and bewilderment [14, 15]. In 2018, a meta‐analysis reviewed 3 decades of published evidence on the profound psychological sequelae in a significant proportion of women experiencing early pregnancy loss. It found evidence of significant depression and anxiety together with post‐traumatic stress symptoms in the first month following the loss [15]. Furthermore, the provision of information as to the cause of early pregnancy loss has been shown to have some beneficial psychological effects in decreasing the feelings of self‐blame [16] and depression [17]. In our experience, when miscarriage is diagnosed, many women express a preference for understanding the reason behind it, even if the loss occurs in early pregnancy and is the first event [10]. Recent guidelines on recurrent pregnancy loss recommend embryo genetic studies through CMA in POC [18, 19]. This is to investigate whether there is a genetic cause of the miscarriage and also to guide recommendations on further investigation, if required, for example, parental peripheral blood karyotyping of both partners where there may be a risk of recurrence in future pregnancies.

cfDNA studies in maternal plasma have been explored to provide an etiologic approach, rather than a definitive genetic diagnosis, and to guide any further investigations that may be required. Unlike ongoing pregnancies, where cfDNA testing can be performed anytime after 9–10 weeks gestation, cfDNA in miscarriages has been assayed as early as five weeks based on ultrasound measurements, due to the expectation of high cell‐free fetal DNA levels [20].

This systematic review and meta‐analysis aims to assess the performance of genome‐wide cfDNA testing in the detection of fetal aneuploidy in miscarriages.

2. Methods

2.1. Protocol and Registration

The protocol was originally registered and published in the International Prospective Register of Systematic Reviews (PROSPERO: CRD42023476110). Neither ethics approval nor patient consent was required for this study, as it did not involve human participants, identifiable data, or clinical interventions. There is no requirement for institutional approval in our hospital for systematic reviews and meta‐analyses.

2.2. Eligibility Criteria

Criteria for inclusion in this systematic review were observational studies of pregnant women with the following: (a) miscarriage defined as a pregnancy loss before 22 weeks; (b) analysis of genome‐wide cfDNA for aneuploidy detection comprising all chromosomes; and (c) confirmation of the aneuploidy by diagnostic genetic methods in POC or chorionic villi samples (karyotype, CMA, or quantitative fluorescent polymerase chain reaction [QF‐PCR]). The following studies were excluded: (a) case reports; (b) opinion articles or letters; (c) cfDNA studies that did not perform genome‐wide analysis comprising all chromosomes; (d) studies applying multigene panels; and (e) studies for which data could not be extracted and, when requested, the corresponding author did not provide additional information. Triploid pregnancies were excluded from the analysis since the vast majority of cfDNA methods are not able to detect them and they were not reported in some of the studies included in this meta‐analysis.

2.3. Information Sources and Search

A systematic search was conducted using PubMed, SCOPUS, and Web of Knowledge to identify relevant manuscripts published without time or language restrictions. References of relevant publications were manually searched for any additional potentially relevant published studies. The final search was run on October 23, 2024. We reviewed the following MeSH terms with word variations of (“cfDNA, non‐invasive prenatal testing, genome‐wide cfDNA”) and “Abortion, Spontaneous,” which comprises “abortions, spontaneous abortions, early pregnancy loss, early pregnancy losses, pregnancy loss, miscarriage, miscarriages, abortion.” The full search is shown in Supporting Information S1.

This meta‐analysis adhered to the Preferred Reporting Items for Systematic Reviews and Meta‐Analyses (PRISMA) guidelines for randomized controlled trials [21]. The study protocol was agreed upon among the authors before running the analysis, and one author (R.M) acted as a reviewer. Ethics approval and patient consent were not required for this study as it did not involve human participants, identifiable data, or clinical interventions.

2.4. Study Selection

Abstracts identified as relevant by one of the researchers (V.A.R.) were assessed by two independent evaluators (M.P. and A.C.J.E.), both blinded to the authorship, authors' institutions, and study results. If studies met the inclusion criteria, full‐text articles were reviewed. A third investigator (R.M.) independently resolved any disagreement between the evaluators. The Supporting Information S1, contains the details of the search strategy and query syntaxes.

2.5. Data Collection Process and Data Items

The following data were extracted on a datasheet based on the Cochrane Consumers and Communication Review Group data extraction template: cities and countries where the study was carried out, study period, study inclusion criteria, sample size, gestational age at analysis, proportion of no calls, aneuploidy rate, concordance, and platform used for the molecular analysis. For diagnostic accuracy analyses, true positive, true negative, false positive, and false negative results were recorded.

2.6. Outcome Measures

The primary outcome was the diagnostic performance of genome‐wide cfDNA for aneuploidy detection in miscarriages. The Phenomena Identification and Ranking Table (PIRT) strategy was as follows. Population: women with a miscarriage; index test: genome‐wide cfDNA; reference standard: diagnostic genetic testing for aneuploidy; treatment effect: Sensitivity, specificity, positive and negative likelihood ratios (LRs).

The secondary outcomes of the study were (a) the concordance between cfDNA and diagnostic results; (b) the proportion of no calls defined as cases where the lab could not perform the test or where the test was conducted but did not produce a result; and (c) the aneuploidy rate, defined as the proportion of aneuploidies detected by a diagnostic genetic method.

2.7. Assessment of Risk of Bias

Two reviewers (J.C. and R.M.) independently assessed the quality of the selected studies. Quality assessment was carried out using the Quality Assessment of Diagnostic Accuracy Studies‐2 (QUADAS‐2) tool [22], which assesses the quality of the included studies in terms of biases affecting their applicability in four domains: patient selection, index test, reference standard, and flow and timing. Answers concerning bias were categorized as low, high, or unclear risk according to the reviewer's judgment about each domain. Results from these questions were plotted and assessed using the Review Manager (RevMan) computer program (Version 5.3. Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration, 2014).

2.8. Strategy for Data Synthesis and Statistical Analysis

Extracted results were pooled in a meta‐analysis. For the diagnostic accuracy analysis, extracted information was used to produce a 2 × 2 table for the calculation of sensitivity and specificity. The obtained results were pooled in a meta‐analysis, and hierarchical summary receiver–operating characteristic (HSROC) curves were constructed. Models were fitted using a bivariate Reitsma model [23] equivalent to the HSROC proposed by Rutter & Gatsonis [24, 25]. Quantitative data synthesis was performed using a random‐effects model.

For the secondary outcomes, proportion was used as the pooled effect size by single‐proportion analysis [26] using random‐effects modeling (weighted by the inverse of the variance), along with the Clopper–Pearson exact method for calculating confidence intervals [27]. Between‐study heterogeneity/variability was assessed using Τau [2], Χ ² (Cochrane Q), and I ² statistics. Results were assessed using forest plots and presented as proportions. Publication bias was visually assessed by funnel plots [28], quantified by the Egger method (weighted linear regression of the treatment effect on its standard error) [29], and adjusted using the Copas model for selection bias [30, 31, 32].

Statistical analyses were conducted using R studio v1.0.136 (The R Foundation for Statistical Computing; package “meta v4.2” [33]).

3. Results

3.1. Study Selection

A total of 288 studies were identified via a database search. Among those, 25 manuscripts were eligible for full‐text review. After review, 17 articles were excluded due to a lack of information on cfDNA and/or diagnostic testing. A total of eight articles were included in this systematic review and meta‐analysis. Figure 1 shows the PRISMA flow diagram of the selected studies. The characteristics of the included articles are described in Table 1.

FIGURE 1.

PRISMA flow diagram of included studies.

TABLE 1.

Characteristics of the included studies.

First author	Year	City, country	Cases (n)	Excluded triploidies	Cases with both cfDNA and diagnostic testing (n)	Mean gestational age (range) (weeks)	Platform for genome‐wide cfDNA	Method of genetic diagnostic testing	Number of no calls cfDNA (n/n)	Aneuploidy rate (n/n)	Concordance (n/n)
Clark	2015	Washington DC, USA	37	0	10	10.1 (5–22)	Illumina v3	Karyotype	10/37	4/10	9/10
Yaron	2020	Barcelona, Catalonia, Spain	109	2	84	6.4 (5–14)	Illumina Verifi Plus and adjusted log‐LR	QFPCR and karyotype	2/109	53/84	73/84
Colley	2020	Birmingham, England, UK	102	7	57	7.1 (5–12)	Ilumina VeriSeq NIPT v2	QF‐PCR and CMA	0/57	27/57	43/57
Hartwig	2023	Copenhagen, Danmark	1000	NR	222	7.7 (6–22)	Illumina Wisecondor Defrac	STR analysis and direct sequencing	31/333	131/222	189/222
Balaguer	2023	Valencia, Spain	120	3	87	8.0 (5–13)	Ilumina VeriSeq NIPT v2	STR analysis and low‐pass sequencing	20/120	63/87	74/87
D'Ippolito	2023	Rome, Italy	14	0	10	10.1 (9–11)	Ilumina VeriSeq NIPT v2	CGH+SNP CMA	0/10	6/10	9/10
Kutteh	2024	Memphis, TE, USA	78	NR	65	6.7 (5–21)	Ilumina VeriSeq NIPT v2	Oligo‐SNP CMA	2/78	49/65	52/65
Antolin	2024	Barcelona, Catalonia, Spain	68	1	17	8.6 (5–13)	Ilumina VeriSeq NIPT v2	QF‐PCR and karyotype	0/17	9/17	16/17

Abbreviation: NR = Not reported.

3.2. Risk of Bias of Included Studies

Among the eight articles included, the risk of bias using the QUADAS‐2 scale revealed an unclear risk for patient selection in two studies due to a lack of information regarding consecutive sampling enrollment [34, 35], two had a high risk of bias [36, 37], and three showed an unclear risk of bias for the reference standard due to a lack of information on whether the diagnostic genetic analysis used was sufficient to diagnose the genetic condition [38, 39, 40]. Lastly, two studies presented an unclear risk of bias in flow and timing due to a lack of information regarding whether all participants were analyzed using the same reference standard [34, 35]. Supporting Information S1: Figure 1 shows the risk of bias and applicability concerns in summary with the review author's judgments regarding each domain for each study included.

3.3. Synthesis of Results

Among the 1528 miscarriages included, only 552 cases had a result for both cfDNA and diagnostic testing. The mean ultrasound gestational age for the entire study population was 7.5 weeks, with only a small number of cases enrolled at 5 weeks and after 12 weeks. All articles used various Illumina‐based genome‐wide cfDNA assays (detailed in Table 1) as the preferred platform for analysis; however, one study adjusted the log‐likelihood ratios according to the prevalence of each aneuploidy in miscarriages, and another used WISECONDOR (Within Sample Copy Number Aberration Detector) as the algorithm to detect aneuploidy using low‐coverage next‐generation sequencing. Thirteen triploidy cases mentioned in four of the studies were excluded from the analysis [34, 36, 38, 40]. Regarding the no‐call results, three studies did not mention them [34, 35, 36]; in two studies, there were two no‐call cases, both due to inadequate samples [37, 40]; and in the remaining three studies, 10–31 samples were excluded because they did not meet the established fetal fraction and/or quality control parameters. All studies were published after 2014.

3.4. Aneuploidy Rate

Genetic diagnostic testing was performed in 552 patients, of which the weighted aneuploidy rate by random effects modeling was 61% (95% CI: 53%–69%). Also, heterogeneity was due to true effect rather than random sampling, where an I ² of 61% shows the proportion of heterogeneity among studies. Figure 2 shows the forest plot of the rate of aneuploidies among selected studies. The sample was too small to conduct a publication bias analysis.

FIGURE 2.

Forest plot of the rate of aneuploidies among selected studies.

3.5. Prognostic Accuracy of cfDNA Testing for the Diagnosis of Fetal Aneuploidies

Figure 3 shows the sensitivity, specificity, and positive and negative LRs for each study. The pooled sensitivity from the eight included studies was 78% (95% CI: 71%–83%). If the 13 triploid cases were not excluded, sensitivity would decrease to 72% (95% CI: 56%–84%). The pooled specificity was 91% (95% CI: 86%–95%), while the positive and negative LRs were 9.19 (95% CI: 5.91–13.60) and 0.25 (95% CI: 0.20–0.31), respectively. The area under the curve (AUC) was 92%. The prognostic accuracy of genome‐wide cfDNA testing for the detection of aneuploidies in pregnancies ending in miscarriage was calculated using an HSROC, as shown in Supporting Information S1: Figure 2. Using a Fagan plot for the pre‐test and post‐test probability of aneuploidy in the case of positive or negative results with genome‐wide cfDNA, considering a baseline aneuploidy rate of 61% (Figure 4), a positive result increases the probability of true aneuploidy by diagnostic methods to 93%, while a negative result reduces the post‐test probability to 28%.

FIGURE 3.

Diagnostic accuracy of individual studies.

FIGURE 4.

Fagan plot of the pre‐ and post‐test after a positive or negative genome‐wide cfDNA in miscarriage.

3.6. Concordance of Aneuploidy Between Genome‐Wide cfDNA and Diagnostic Genetic Testing in Miscarriage

Among the 552 cases with complete information on cfDNA results and diagnostic genetic testing, the weighted proportion of the concordance rate between both tests, using a single proportion analysis by the inverse of the variance under random‐effects modeling, was 84% (95% CI: 81%–87%). There was no significant heterogeneity among studies (p = 0.14), while an I ² of 0% means that there was no variability among studies. Figure 5 shows the forest plot of the concordance between cfDNA and diagnostic testing. The sample was too small to conduct a publication bias analysis.

FIGURE 5.

Forest plot of the concordance between cfDNA and diagnostic testing results.

3.7. No‐Call Result for Genome‐Wide cfDNA

There was information on cfDNA results from 761 cases. The weighted proportion of cfDNA no calls by random effects modeling was 4% (95% CI: 0%–12%). Heterogeneity was due to true‐effect rather than random sampling with up to 88% heterogeneity by I ². Figure 6 shows the forest plot of the proportion of cfDNA no calls among the included studies. The sample was too small to conduct a publication bias analysis.

FIGURE 6.

Forest plot of the proportion of cfDNA no calls among the included studies.

4. Discussion

4.1. Main Results

The main results from this study are the following: (a) The performance of genome‐wide cfDNA testing for the detection of fetal aneuploidy in pregnancies ending in miscarriage demonstrated a 78% detection rate with a 9% false‐positive rate; (b) The probability of aneuploidy was 61%, and a positive cfDNA result increased this probability up to 93%, while a negative result did not exclude aneuploidy, with a post‐test probability of 28%; (c) The rate of concordance between genome‐wide cfDNA and diagnostic genetic testing was 84% when polyploidies are excluded; and (d) The observed proportion of no calls was 4%.

4.2. Comparison of Literature Studies

A preliminary small study by Clark et al. [39] in 2015 explored the use of cfDNA applied to pregnancy losses, reporting a good concordance (90%) between results obtained by cfDNA and diagnostic studies. However, the high no‐call rate (27%) observed, particularly in miscarriages under 8 weeks, discouraged other groups from further exploring this approach. In 2020, our group [40], based on the previous description of high levels of cfDNA in missed miscarriages [20], presumably due to increased apoptosis of chorionic cells, assumed that this effect would result in acceptable fetal fractions for cfDNA analysis, even under 10 weeks of pregnancy. That study demonstrated a good concordance rate (87%) with an extremely low no‐call rate of 2%, even though 5‐week pregnancy losses (pre‐embryonic losses) were also included in the study, suggesting that cfDNA could be used as a screening test when POC are not available. The most extensive study included in our meta‐analysis was the Copenhagen Pregnancy Loss Study (CPOS) [41], which obtained complete results for both cfDNA and genetic diagnostic studies in 222 pregnancy losses up to 22 weeks. The fact that cases were successfully included in the study up to 6 hours post‐uterine evacuation widens the application of cfDNA not only to missed miscarriages but also to recent complete miscarriages.

4.3. Clinical Applicability

Parental karyotyping has been traditionally recommended after the second or third miscarriage, according to the local definition of recurrent miscarriage, to detect balanced parental rearrangements that can cause recurrent loss. However, while the percentage of cases in which the cause of miscarriage can be identified via parental karyotyping is only about 4% [42], that of karyotyping POC is about 50%–60% [9]. There is a general trend in scientific society guidelines to focus genetic testing in recurrent miscarriage on POC rather than on the parents to detect genetic causes of miscarriage due to sporadic aneuploidies, which may also impact clinical care, rather than just those with a recurrence risk inherited as a result of a parental balanced translocation. This can be observed in the latest guidelines of the European Society of Human Reproduction and Embryology (ESHRE) [18] and the Royal College of Obstetricians and Gynaecologists (RCOG) [19].

Regarding the threshold number of miscarriages for offering diagnostic testing, our group, along with others, contends that sporadic miscarriages cannot be dismissed as trivial events. This is not just because of their high prevalence, occurring in up to 20% of clinical pregnancies, but also because several studies have demonstrated their high psychological impact [16, 43]. In the United Kingdom, the Tommy's Group provides support to women and couples who suffer pregnancy loss and promotes changes in the National Health Service regulation to include a progressive work‐up for miscarriage, instead of waiting for a third event to occur [43]. The grief and psychological impact of miscarriage are often not comprehended, both by society and healthcare professionals. Miscarriage can often be minimized by the view that a pregnancy before 12 weeks “isn't a real baby” or that “it just wasn't meant to be” [44]. Interestingly enough, only the smallest study included in the present meta‐analysis was focused on recurrent miscarriage, and the results did not differ from the studies performed in sporadic miscarriages, suggesting that disclosing the cause of the loss is not an issue that should apply exclusively to recurrent miscarriage.

In recent years, genetic studies in POC have evolved from conventional karyotyping to CMA, because karyotyping on POC is limited by a high culture failure (32%) and a remarkable incidence of maternal cell contamination (15%) [7, 45], achieving an accurate result in only about half of the cases. Before the introduction of CMA, chorionic villi sampling was suggested by our group as a method to achieve an accurate karyotype, but specific expertise in transcervical chorionic villi sampling was required to replicate our results [9, 11]. Finally, the availability of POC is increasingly threatened by the growing acceptance of medical evacuation of uterine contents. In our center, the recommendation for surgical removal of POC has been confined to 10–11‐week miscarriages. The use of a container to retrieve POC at home was used in almost half of the cases in the COPL study; however, this approach, similar to transcervical chorionic villi sampling, cannot be easily applied in routine practice. On the contrary, cfDNA‐based screening can be easily adopted into the healthcare workflow and is viewed positively by consumers, given the great acceptance of cfDNA testing prenatally, despite being recognized as a screening rather than a diagnostic test.

A main limitation of the studies included in our meta‐analysis is that polyploidies, accounting for up to 10% of early pregnancy losses, cannot be detected by cfDNA sequencing methods, which are used in all the studies. Regarding polyploidies, the eight studies included adopted discordant policies; they were included and undetected in two [37, 40], no triploid cases were found in two [35, 39], they were excluded in two others [34, 38], and not mentioned in the remaining two studies [35, 37]. Furthermore, the expected false negatives of this method are those related to uniparental disomy and copy number variants, as well as low fetal fraction and placental mosaicisms, which account for a small proportion of miscarriages. False positives with cfDNA testing may also occur as a consequence of a vanished twin, maternal mosaicism, or maternal malignancy, similar to that occurring in ongoing pregnancies. As circulating cfDNA mainly originates from trophoblast cells, the cfDNA result will likewise be a false positive in high‐grade confined placental mosaicism.

4.4. Limitations and Strengths

The main limitation of our review is the reduced sample size and number of studies that fulfilled the final criteria, given that among 1528 cases, only 552 (36%) had an informative result for both POC and cfDNA analysis. The main strength of our review is that a pattern of high specificity has been exhibited approaching 100%; therefore, it is important to minimize false positive results in recurrent miscarriage as this would result in failure to investigate other non‐genetic causes.

5. Conclusions and Future Directions

In conclusion, cfDNA analysis for aneuploidy detection in miscarriage has a lower accuracy than in ongoing pregnancies; nevertheless, it can be widely applied as a screening test with the advantage of an earlier application window than in ongoing pregnancies: from 5‐week miscarriages (instead of 9–10 weeks) to 6 hours after uterine evacuation (instead of delivery). The results of this meta‐analysis support the use of cfDNA testing when a POC specimen is unavailable, such as in medical evacuations, when a previous CVS or home collection device has not been used, or when maternal cell contamination is detected in the POC sample. However, this is an unconventional screening test as the results cannot be definitively confirmed. Larger prospective studies and method refinements to detect triploidy would provide significant benefits.

Acknowledgements

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1

Data Availability Statement

Data sharing is not applicable to this article as no datasets were generated or analyzed during the current study.