Biomarkers at 6 weeks' gestation in the prediction of early miscarriage in pregnancy following assisted reproductive technology.

All authors contributed to the intellectual content of the study. TCL and LCP conceptualized and designed the study. JG, KA were responsible for data collection, JG, QLF and PC were responsible for data management and analysis, JG and QLF were responsible for drafting the manuscript, DSS, BWL and JPC were responsible for interpretating the data. LCP were responsible for revising the manuscript. All authors approved of the final version to be submitted.

Direct Grant for Research (grant no. 4054411) and The Hong Kong Obstetrical and Gynecological Trust Fund 2018/2019 (grant no. 6904963).

A total of 183 pregnancies were recruited in this study. We excluded 14 cases (7.7%) because they had twin pregnancy ( n  = 8), termination for chromosomal or structural abnormalities ( n  = 2), stillbirth ( n  = 2), incomplete data ( n  = 1) and vaginal bleeding presenting at 6 weeks' gestation ( n  = 1). In the remaining 169 pregnancies included in the final analysis, 145 (85.8%) pregnancies progressed to beyond 13 weeks' gestation and had live births at ≥24 weeks' gestation, whereas 24 (14.2%) pregnancies resulted in miscarriage within the first trimester (Figure  1 ). We collected miscarriage data, including the gestational age at diagnosis and type. All miscarriages were spontaneous, with 20 cases (20/169, 11.8%) being complete miscarriage and four cases (4/169, 2.4%) being incomplete miscarriage. Summary of the study population. Maternal demographics and clinical characteristics of the participants are shown in Table  1 . Women with a clinical pregnancy that resulted in subsequent miscarriage had significantly higher maternal age ( p  < 0.001), bodyweight ( p  = 0.003) and body mass index ( p  = 0.028) at 6 weeks' gestation in comparison to those with live births at ≥24 weeks' gestation. In addition, serum hCG concentrations at 4, 5, and 6 weeks' gestation in the miscarriage group were significantly lower than that in the live birth group ( p  = 0.011, <0.001, and 0.002, respectively). Demographic and clinical characteristics between outcome groups. Note : Data are presented as median (interquartile range) or n (%). Abbreviations: BMI, body mass index; FSH, follicle stimulating hormone; hCG, ICSI, intracytoplasmic sperm injection; IVF, in vitro fertilization; y, years. p  < 0.05. The assessment of biomarkers was carried out in 119 cases with live births without placenta‐related complications. As shown in Table  2 , multivariate regression analysis demonstrated that log 10 MAP was associated with maternal weight ( R 2  = 0.048), PlGF was associated with the type of ET (fresh or frozen) ( R 2  = 0.043), and sqrt kisspeptin was affected by the indication of IVF treatment (if unexplained) ( R 2  = 0.108), whereas other biomarkers including mGSD, CRL, mUTPI, sFlt‐1, and glycodelin‐A were not associated with maternal characteristics, type of infertility, indication and categories of IVF treatment, including fertilization method (IVF or ICSI), type of ET (fresh or frozen) and stage of embryo transfer (cleavage stage or blastocyst). Multivariate linear regression analysis for the prediction of biomarkers at 6 weeks' gestation. Abbreviations: ICSI, intracytoplasmic sperm injection; IVF, in vitro fertilization; MAP, mean arterial pressure; PlGF, placental growth factor; Sqrt, square root. p  < 0.05. In the miscarriage group, median MAP was significantly increased, median mGSD, CRL, mUTPI, glycodelin A and sFlt‐1 were significantly decreased and there was a lower rate of the presence of fetal heart activity, compared to the live birth group (Table  3 ). Serum concentrations of PlGF and kisspeptin in the miscarriage group were not significantly different from the concentrations in the live birth group (Table  3 ). Comparison of biomarkers between pregnancies resulted in miscarriages and ongoing pregnancies that resulted in live birth. Note : Data are presented by median (interquartile range) or mean (standard deviation). Comparison of data between two outcome groups was carried out by Mann–Whitney U test. Abbreviation: MoM, multiple of the median. p  < 0.05. Multivariate logistic regression analysis demonstrated that in the prediction of miscarriage before 13 weeks' gestation, there were significant independent contributions from maternal age, fetal heart activity (yes or no), mUTPI MoM and glycodelin‐A MoM (Table  4 ; Y  = −4.606935 (SE, 4.333037) + 0.255347 (SE, 0.107744) * maternal age (year) − 2.469670 (SE, 0.593154) (1 if presence of fetal heart activity; 0 if no fetal cardiac activity) − 4.676286 (SE, 1.806947) * mUTPI MoM − 1.047539 (SE, 0.464865) * glycodelin‐A MoM; R 2  = 0.527; p  < 0.01). Multivariate regression model with maternal age, presence of fetal heart activity, mean uterine artery pulsatility index and serum glycodelin‐A at 6 weeks' gestation for the prediction of miscarriage before 13 weeks' gestation. Abbreviations: CI, confidence interval; MoM, multiple of median; mUTPI, mean uterine artery pulsatility index; OR, odds ratio; SE, standard error. The combination of maternal age, ultrasound (fetal heart activity and mUTPI) and biochemical (glycodelin‐A) markers achieved the highest AUC (0.918, 95% CI: 0.866–0.955, p  = 0.001 when compared to maternal age alone), with estimated detection rates of 54.2% (95% CI :29.2–75.0), 70.8% (95% CI: 50.0–87.5) and 70.8% (95% CI: 48.9–87.4) for miscarriage before 13 weeks' gestation, at fixed false positive rates of 5, 10, and 15%, respectively (Table  5 ; Figure  2 ). The respective positive predictive values were 65.0, 54.8, and 44.7% and negative predictive values were 92.6, 94.9, and 94.6%. Predictive values of screening for miscarriage before 13 weeks' gestation. Abbreviations: AUC, area under the receiver‐operating characteristics curves; CI, confidence interval; DR, detection rate; FHA, fetal heart activity; FPR, false positive rate; mUTPI, mean uterine artery pulsatility index; NPV, negative predictive value; PPV, positive predictive value. All the p ‐values are the comparison of AUC against maternal age. Receiver‐operating characteristics (ROC) curves of maternal age (‐‐‐‐‐‐), maternal age with glycodelin‐A (…….), maternal age with presence of fetal heart activity and mUTPI ( ) and maternal age with glycodelin‐A, presence of fetal heart activity and mUTPI ( ).

The present study has shown that (1) among IVF/ET singleton pregnancies that miscarry prior to 13 weeks' gestation, median maternal age, body mass index and MAP are increased while median CRL, mGSD, mUTPI, sFlt‐1, glycodelin‐A and the rate of presence of fetal heart activity are decreased and (2) the combination of maternal age, fetal heart activity, mUTPI and glycodelin‐A at 6 weeks' gestation can effectively identify IVF/ET singleton pregnancies at risk of miscarriage prior to 13 weeks' gestation. In this study, the findings that women in the miscarriage group were older and with a higher body mass index than those in the live birth group are consistent with existing literature. 24 , 25 , 26 , 27 Furthermore, we have shown that the absence or delayed presence of unequivocal fetal heart activity is associated with a higher chance of miscarriage, which has been reported in earlier studies. 28 , 29 In a retrospective cohort study consisting of 68 IVF pregnancies, 30 Seungdamrong et al. observed positive fetal heart activity at 6 weeks' gestation in 78.3% (36/46) of singleton pregnancies and the authors demonstrated that the presence of fetal heart activity at 6 weeks' gestation achieved a sensitivity of 100%, at a specificity of 100%, with a positive predictive values of 100% and a negative predictive value of 100% in predicting ongoing pregnancy. 30 These results demonstrated higher performance than our results, which may be explained by their relatively small sample size and the differences in the IVF stimulation and embryo transfer protocol. In addition, women with a history of recurrent miscarriage, which is known to be associated with increased risk of fetal loss, were excluded from their study. Similarly, our finding that the maternal circulating levels of glycodelin‐A are reduced in pregnancies that subsequently miscarry (336.9 vs. 718.6 ng/mL, p  = 0.009) is in keeping with other studies. 31 , 32 , 33 Glycodelin‐A is a glycoprotein abundant in the decidua, which rises from implantation and peaks at around 6–12 weeks' gestation. It plays an important role in placental development and fetomaternal tolerance during early pregnancy by regulating T‐helper type 1 (Th1)/ T‐helper type 2 (Th2) lymphocytes balance in the decidua, promoting natural killer cells apoptosis as well as decreasing the production of proinflammatory cytokines from decidua immune cells. 34 Jakubowicz et al. demonstrated that serum concentrations of glycodelin‐A at 3–5 and 6–8 weeks' gestation in pregnancies with subsequent miscarriage were 48% ( p  = 0.02) and 78% lower ( p  = 0.01) than that in ongoing pregnancies. 31 Interestingly, in uterine flushings obtained on days 10 and 12 after luteinizing hormone surge, Dalton et al. found significantly lower concentrations of glycodelin‐A in women who went on to miscarry than those who went on to have a live birth once they got pregnant (1000 vs. 1440 ng/mL, p  < 0.05), 32 which indicated that the decrease of glycodelin‐A might occur during the menstrual cycle in women at risk of miscarriage. As the expression of glycodelin can be upregulated by progestogens, 35 women with low level of glycodelin‐A may benefit from progesterone in reducing the risk of early miscarriage. In addition, we have demonstrated that there is a reduction in resistance to uterine blood flow at 6 weeks' gestation in IVF pregnancies that result in a miscarriage, which is in broad agreement with a previous study reported by Taylor et al. 7 mUTPI was prospectively measured in 913 unselected viable pregnancies at 5–8 weeks and reported that the mean Z score of mUTPI in women who miscarried before 12 weeks' gestation (−0.48, 95% CI: −0.82 to −0.14, n  = 50) was significantly ( p  = 0.001) lower than that of ongoing pregnancies (0.00, 95% CI: −0.07 to 0.07, n  = 863). 7 The investigators developed a multivariate model in predicting miscarriage within the first trimester with a combination of ultrasound parameters including mGSD, yolk sac diameter, fetal heart rate, mUTPI and trophoblast volume, which achieved an AUC of 0.81 (95% CI: 0.74–0.87). 7 At a false positive rate of 10%, the combined test could identify 46% (95% CI: 32–61) of miscarriages with a positive predictive value of 2.9% and a negative predictive value of 75.5%. 7 In our study, we did not measure yolk sac diameter and fetal heart rate due to the early gestation, as these features may not be present at 6 weeks' gestation. However, a combination of maternal age and ultrasound markers of fetal heart activity and mUTPI in the present study achieved a better AUC of 0.889 (95% CI 0.831–0.932) than in the study by Taylor et al., with a detection rate of 66.7%, at a false positive rate of 10.0%, a positive predictive value of 30.6% and a negative predictive value of 97.9%. In another study consisting of 2601 pregnancies with confirmed fetal heart activity following IVF/ET, 36 Yi et al. developed a model, including maternal age, gestational age, mGSD, CRL, yolk sac diameter, fetal heart activity and fluid collection around the gestational sac on days 27–29 after ET, for the prediction of ongoing pregnancy after 12 weeks that could correctly classify 94.44% of all cases with an AUC of 0.909. 36 These results are comparable to the performance of our final model combining maternal age, fetal heart activity, mUTPI, and glycodelin‐A. Evidence suggests that pregnant women with uncertain viability at initial scan could benefit psychologically from a combined test, which includes maternal age, mGSD, and progesterone level, that informs them of the likelihood of an ongoing pregnancy as it can significantly lower the anxiety level irrespective of the pregnancy outcome. 4 The ability to effectively predict the possibility of miscarriage or ongoing pregnancy among women who have undergone IVF/ET might facilitate antenatal counseling and management in clinical practice, allowing close monitoring in targeted high‐risk women and further investigation of potential therapeutic interventions for the prevention of miscarriage among pregnancies following IVF/ET. A major strength of this study was that all biomarkers were measured at a precise timepoint immediately after the clinical pregnancy was confirmed by rising hCG and ultrasonography, so that the observed differences between the two outcome groups were less likely to represent secondary vascular changes after the miscarriage had occurred. Additionally, all women with ongoing pregnancy beyond 13 week's gestation had live births at ≥24 weeks' gestation, thus the high negative predictive value of the combined model at 6 weeks' gestation (92.6%–94.9%) implies excellent predictive value for subsequent live birth in the index pregnancy. Such prediction would be useful for counseling to help relieve the anxiety of the pregnant women. However, there were several limitations. It is well known that serum hCG is a widely used biomarker in diagnosing and predicting pregnancy outcomes prior to 6 weeks' gestation; however, in this study, we did not include hCG as a predictor in the combined model as we used serum hCG to help identify miscarriage. We did not perform placental biopsy for karyotype in all cases of miscarriage, and therefore miscarriages due to aneuploidy could not excluded. In order to determine whether the observed differences between the two outcome groups were less likely to represent secondary vascular changes after the miscarriage had occurred, a relatively early ultrasound was performed for the detection of fetal heart activity at 6 weeks ±1 day. As a result, the performance of the developed model could have lower sensitivity. Lastly, a total of four independent predictors were included in the final model while there were only 24 cases of miscarriage; however, the models with three markers achieved good AUCs of 0.855–0.892.

A combination of maternal age, fetal heart activity, mUTPI and serum glycodelin‐A at 6 weeks' gestation could effectively identify IVF/ET pregnancies at risk of first trimester miscarriage. Further research and external validation are needed to determine the clinical utility of the combined test in predicting miscarriage following IVF/ET.

Miscarriage is the most common complication of early pregnancy. It has been reported that up to 20% of clinical pregnancies result in a miscarriage, with the vast majority occurring within the first trimester of pregnancy. 1 Women undergoing assisted reproductive treatments experience substantial psychological stress related to the fear of early miscarriage, while high level of anxiety during pregnancy has been reported to increase the risk of pregnancy complications such as preterm birth 2 and birth defects. 3 The ability to effectively predict the possibility of miscarriage or ongoing pregnancy beyond the first trimester has the potential to lower anxiety level irrespective of the pregnancy outcome. 4 Among women undergoing in vitro fertilization (IVF)/embryo transfer (ET), a routine blood or urine sample is obtained at 4 weeks' gestation to confirm if pregnancy has occurred and a clinical pregnancy is confirmed by ultrasound at 6 weeks' gestation. A number of ultrasound markers have been demonstrated to be potential predictors for early miscarriage, such as fetal crown‐rump length (CRL), 5 mean gestational sac diameter (mGSD), 6 and mean uterine artery pulsatility index (mUTPI), 7 while the predictive value of these markers varies in different studies. Several promising biochemical markers related to placental function have also been proposed for the prediction of subsequent early miscarriage, including glycodelin‐A which is involved in implantation, placentation and regulation of maternal‐fetal immunotolerance, 8 kisspeptin which reflects the time of trophoblast invasion, 9 and angiogenesis‐related factors such as placental growth factor (PlGF) and soluble fms‐like tyrosine kinase‐1 (sFlt‐1). 10 However, the potential value of these biochemical markers is inconclusive and combined models including ultrasound and biochemical markers are still under exploration. The aims of this study were to: (1) examine potential miscarriage‐related biophysical and biochemical markers at 6 weeks' gestation among women with confirmed clinical pregnancy following IVF/ET and (2) evaluate the performance of a model combining maternal factors, biophysical and biochemical markers at 6 weeks' gestation in the prediction of first trimester miscarriage among singleton pregnancies following IVF/ET.

The authors declare no conflicts of interest.

The data were derived from a prospective observational study for adverse pregnancy outcomes among women who conceived through IVF/ET attending their routine visit at 6 weeks' gestation in the IVF unit at the Prince of Wales Hospital, the Chinese University of Hong Kong, Hong Kong SAR, between December 2017 and January 2020. The inclusion criteria were: (1) conceived by IVF/ET; (2) a single ET; (3) a single intrauterine gestational sac of at least 5 mm in diameter confirmed by transvaginal ultrasound examination at 6 weeks' gestation. Exclusion criteria were: (1) multifetal gestations; (2) with known uterine anomaly and uterine pathology including fibroid, adenomyosis or endometrium polyp; (3) with vaginal bleeding and/or abdominal pain at the time of assessment. Gestational age was calculated based on the date of ET. On day 3 of embryo transfer, the gestational age was counted as 17 days, while on day 5 of embryo transfer, the gestational age was counted as 19 days. Women undergoing IVF/ET routinely returned to the clinic for serum human chorionic gonadotropin (hCG) measurement at gestational age of 28 days and ultrasound examination at gestational age of 43 ± 1 days. Written informed consent was obtained from women agreeing to participate in the study. The protocol of ovarian stimulation has been previously described. 11 , 12 This utilized either recombinant follicle stimulating hormone (Gonad‐F; Serono; or Puregon, MSD) or human menopausal gonadotrophins (Pergonal, Merck Serono) with doses ranging from 150 to 450 IU per day. When three or more follicles of ≥16 mm in diameter were present, 5000 IU of hCG (Profasi, Serono) was administered, followed by transvaginal oocyte retrieval 36 h later and subsequent IVF or intracytoplasmic sperm injection (ICSI). In fresh ET cycles, luteal support was provided by the administration of progesterone (Endometrin 100 mg three times daily, Ferring) immediately from the day after oocyte retrieval (OR) whereas in frozen ET cycles, luteal support was omitted in natural cycles. In hormone replacement cycles, estradiol valerate at a dose of 6 mg/day was administered from day 2 or 3 of the cycle and progesterone in the form of Endometrin (100 mg three times daily, Merck Serono) was added 14 days after the initiation of estrogen therapy. Maternal demographic characteristics including age, weight, height, body mass index, racial origin, smoking habit, and parity; infertility information including infertility type, basal follicle‐stimulating hormone, length and cause(s) of infertility; assisted reproductive treatment information including fertilization method, number, type and stage of embryo transfer, were recorded in a secured electronic database. During the routine clinical visit at 6 weeks' gestation, mean arterial pressure (MAP) was measured by validated automated devices (FHA‐MW‐ BP3AQ1, Microlife). The MAP measurements were made by research staff who had received appropriate training on the use of the blood pressure (BP) devices. Women were in a sitting position with their arms supported at the level of the heart and appropriate size cuffs were used. 13 After resting for 5 min, the BP was measured in both arms simultaneously and two recordings were made at 1‐min intervals. The MAP was calculated from systolic (sBP) and diastolic BP (dBP) with the equation: MAP = (2 * dBP + sBP)/3. The final MAP was the average of four measurements. 13 Transvaginal ultrasound scan was performed with the use of a Voluson E8 Expert series and a RIC6‐12‐D 3D transvaginal transducer (General Electric Healthcare) in a lithotomy position by a single operator (JG). All women had an empty bladder for the transvaginal ultrasound scan. The gestational sac was measured in both longitudinal and transverse views. The mGSD was calculated as the average of three perpendicular diameters measured with the calipers placed at the inner edges of the trophoblast. 14 A sagittal view of the embryo was identified, and the CRL was measured along the embryo's longest axis from the tip of the cephalic pole to the tip of the caudal pole. The presence or absence of fetal heart activity was recorded. With regard to the measurement of mUTPI, a sagittal view of the cervix was visualized to identify the endocervical canal and internal cervical os. 15 By moving the transducer laterally, each uterine artery located laterally to the cervix and uterus at the level of the internal os was identified. Pulsed wave Doppler was used with a sampling gate set at 2 mm to cover the whole vessel to capture the waveforms 15 and care was taken to ensure the angle of insonation was less than 30° 16 and pulse repetition frequency was the same for each measurement. When three similar consecutive waveforms were obtained the UTPI was measured and mUTPI of the left and right arteries was calculated. All measurements were obtained by auto tracing of the waveform. Blood samples were collected using CAT Serum Sep Clot Activator tubes, which were then centrifuged at 2000 g for 10 min at 4°C. Collected serum samples were divided into 1 mL aliquots in new eppendorf tubes, which were labeled with unique patient identifier and stored at −80°C until subsequent analysis. The serum concentrations of glycodelin‐A and kisspeptin were evaluated using quantitative enzyme‐linked immunosorbent assay from Bioserv Diagnostics (BS‐30‐20) and USCN Life Science Inc. (SEC559Hu), respectively. All tests were performed in duplicate according to the manufacturer's instructions. Results were determined as the optical density at 450 nm using microplate reader and concentrations of these proteins were determined from the standard curve with the known concentrations. Serum PlGF and sFlt‐1 concentrations were determined by the Kryptor Compact Plus analyzer (ThermoFisher Scientific). Obstetric and delivery outcomes including the mode of delivery, maternal and neonatal outcomes were obtained from the maternity computerized records or the general medical practitioners of the women and recorded in our secured database. The primary outcome of this study was early miscarriage, which is defined as a nonviable intrauterine pregnancy with either an empty gestational sac or a gestational sac with an embryo or a fetus without fetal heart activity before 13 weeks' gestation. 17 Normal live birth was defined as the live birth at ≥24 weeks' gestation without placenta‐related complications, including preeclampsia, 18 placenta abruption and small for gestational age. 19 Sample size calculation was based on the screening performance of the combined test in predicting miscarriage with an area under the receiver operating characteristic (ROC) curve (AUC) of 0.887, reported by Stamatopoulos et al. 20 At a power of 80% and 95% confidence interval (CI), a minimum of 19 cases of miscarriage would be required. The incidence of early pregnancy loss following IVF/ET was 16%, 21 therefore, a minimum of 119 women was needed. Normality of the data distribution was tested by Kolmogorov–Smirnov test. Log 10 or square root (sqrt) transformation was used to make the data distribution Gaussian prior to statistical analysis if necessary. Quantitative data are presented as median and interquartile range and categorical data are presented as numbers and percentages. Comparison of data between two outcome groups was carried out by Mann–Whitney U test for continuous data and by chi‐square test or Fisher's exact test for categorical variables. Multivariate backward stepwise linear regression analysis was performed to determine which of the factors among maternal characteristics and obstetric history were significant predictors of each biomarker in the normal live birth group. The biomarkers were then expressed as multiple of the median (MoM) after adjustment of significant predictors, where necessary. CRL was converted to the gestational age‐specific standard Z ‐score based on a locally derived nomogram. 22 Multivariate backward stepwise logistic regression analysis was performed to determine significant predictors of miscarriage. The linearity in the logit for continuous variables was tested through Box–Tidwell test. The performance of screening was estimated by ROC curve analysis. Differences between AUCs of different models were assessed by DeLong test. 23 A probability value ( p ‐value) of less than 0.05 was considered statistically significant. Data were analyzed with SPSS statistical software version 23.0 (IBM SPSS Statistics for Windows) and Medcalc (Medcalc Software). This study was approved by the Institutional Review Board on October 20, 2017 (Joint Chinese University of Hong Kong—New Territories East Cluster Clinical Research Ethics Committee, reference no.: CRE‐2017.472).