Introduction

Pregnancy loss is defined as the spontaneous termination of a pregnancy prior to fetal viability1. Recurrent pregnancy loss (RPL) is defined as the occurrence of two or more pregnancy losses before 24 weeks of gestation1, encompassing biochemical pregnancy losses, resolved and treated pregnancies with unknown location. The reported incidence of RPL ranges from 1 to 3%, depending on the definitions1. According to previous research, the number of previous pregnancy losses has been identified as an independent risk factor for recurrent pregnancy loss, irrespective of maternal age2. The number of spontaneous pregnancy losses was significantly associated with the subsequent pregnancy success rate1. For the majority of women and their partners, pregnancy loss signifies the loss of a child along with the shattered aspirations and plans associated with that baby1. Feelings of loss and grief, commonly experienced following a single pregnancy loss, can be further intensified with subsequent losses1. Therefore, in addition to imposing significant financial burdens on affected families, RPL also exerts a detrimental impact on their mental well-being3. As a result, many individuals actively seek medical assistance to mitigate the risk of future pregnancy losses and frequently undergo prenatal examinations to ensure optimal health conditions for their infants, especially during the first trimester.

Ultrasound examination is routinely performed in the first trimester following recurrent pregnancy loss to assess viability and confirm pregnancy location. Currently, established ultrasound criteria for initial pregnancies include mGSD (mean gestational sac diameter) of 25 mm without an embryo, CRL (crown-rump length) of 7 mm without detectable fetal heartbeat two weeks after identifying a gestational sac or eleven days after identifying both gestational sac and yolk sac; all indicative of nonviable gestation4. However, there is still a requirement for a simplified and precise predictive model for pregnancy outcomes in cases of recurrent pregnancy loss. Patients with RPL seek clear information from their healthcare providers regarding the condition and its progression as it represents a significant life event5. While various models utilizing ultrasound scans have been developed with good predictive performance6,7,8,– 9, there remains a lack of prediction models for RPL using ultrasound scans during each week of pregnancy despite the urgent need. A predictive model specific to different gestation weeks will provide more accurate information to patients and clinicians while also saving time and unnecessary costs for revisits.

In this study, we will utilize ultrasound indices in the first trimester of pregnancy along with demographic features and commonly used serum markers such as human chorionic gonadotropin (hCG), progesterone (P), and estradiol (E2). Our aim is to develop a predictive model that integrates widely used clinical indicators on a weekly basis during the first trimester; provide cut-points of predictors based on clinical data for multiple sonographic characteristics for healthcare professionals and patients with RPL to improve clinical utility.

Materials and methods

Study design

We conducted a retrospective analysis on patients seeking treatment for recurrent pregnancy loss at our reproductive center of The Second Hospital of Lanzhou University between September 2019 and December 2021, with a minimum follow-up period of 18 months. The data were extracted from the reproductive center database located in Lanzhou, China. This study was approved by the institutional research ethics review board of The Second Hospital of Lanzhou University (2024A-322), and the informed consent was obtained from all subjects. All methods were carried out in accordance with relevant guidelines and regulations.

Patients

All these patients underwent a comprehensive assessment prior to their initial visit for preconception planning or at their first visit in our clinic during the early stages of pregnancy, including transvaginal ultrasound to exclude uterine structural abnormalities, thyroid function test, antiphospholipid syndrome related antibodies, and other RPL related tests according to European Guideline for RPL. Our hospital also implemented appropriate therapeutic interventions for these patients in cases of pregnancy loss. The pregnancy outcomes of all these patients were monitored through telephone follow-up or by accessing the databases in our hospital until July 2023. We included women with RPL aged between 18 and 45 years old. We excluded patients who were lost to follow up, had no plan to conceive any more, and those who were pregnant in the first trimester at the last follow-up (Fig. 1). Additionally, we excluded patients with multiple pregnancies and those who conceived a second time. Finally, 603 RPL women were included in this analysis.

Figure 1
figure 1

Flow chart of the included population in our study.

Full size image

Transvaginal ultrasonography (TVS) was performed using a 5 MHz transducer for B mode imaging (ARIETTA 60) by experienced obstetricians due to its high sensitivity and accuracy for earlier recognition of pregnancy. All of the ultrasound measurements were measured following Chinese practice guidelines for obstetric ultrasound. The presence of a yolk sac, embryo, and heartbeat were all recorded during the ultrasound examination. mGSD was obtained by the average of measurements taken in three dimensions of gestational sac.

Outcome measures

Pregnancy loss is defined as the spontaneous termination of a pregnancy prior to fetal viability, encompassing all losses occurring from conception until 24 weeks gestation1. First trimester pregnancy loss was defined the pregnancy loss before 13 weeks of gestation1. Ongoing pregnancy was defined as the continuation of pregnancy at 12 weeks of gestation. Gestation time was calculated by weeks.

Sample size calculation

Sample size was calculated by an online interactive computing tool (https://mvansmeden.shinyapps.io/BeyondEPV/) developed by van Smeden et al10.This tool makes sure that when a new prediction model is estimating the outcome probability, it has a minimum prediction error which is evaluated by mean absolute prediction error (MAPE). Based on the calculation, with a pregnancy loss rate of 20%, an available number of pregnant women of 391 at least in each gestation week and a MAPE of 0.05, a maximum of five predictors are required to develop the pregnancy loss prediction model.

Data preprocessing

In the fifth week, the CRL can’t be seen in most cases, hence we exclude it in the analysis. In our study, data with age, previous pregnancy losses, mGSD and pregnancy outcomes were complete, while BMI, P, hCG, E2, YSD and CRL data has a missing less than 20% in every gestational weeks. Therefore, multiple imputation (based on the pattern of missingness and incorporating all variables in the imputation model, with 10 iterations for each imputation cycle) was performed in the logistic regression analysis to mitigate the impact of the missing values.

Statistical analysis

Variables were tested for normality by the Shapiro–Wilk test. The data were presented as mean ± standard deviation (SD) or median with interquartile range for continuous data, and n (%) for dichotomous data. Descriptive statistics were compared using appropriate statistical tests, including Student’s T test, Mann–Whitney U test, Chi-squared test, or Fisher’s exact test as deemed suitable for the analysis. Logistic regression analysis was performed to construct a predictive model for the fifth to ninth week of pregnancy. The area under the curve (AUC) was calculated using receiver operating characteristic (ROC) curve analysis. Restricted cubic spline (RCS) analysis was also employed to identify non-linear relationships between continuous predictors and pregnancy outcomes. To explore the threshold effect of predictors in the first trimester on the risk of pregnancy loss and to find the inflection point, we used smooth curve fitting and generalized additive models. R packages “rms” was used to identify the threshold of predictors in early stages of pregnancy. Gestation period was depicted by week. Sensitivity analysis was conducted to mitigate potential bias from possible known factors on pregnancy outcomes. In this study, sensitivity analysis was performed with the exclusion of abnormal thyroid stimulating hormone (TSH), positive antibody of anticardiolipin antibody (ACA), and antiβ2glyco protein1 antibody (β2GP1). Statistical analyses were conducted using SPSS 27.0 and R studio software packages with a significance level set at p < 0.05.

Ethical approval and consent to participate

This study was approved by the institutional research ethics review board of The Second Hospital of Lanzhou University (2024A-322). All methods were carried out in accordance with relevant guidelines and regulations.

Results

Among the 691 RPL patients, a total of 14 participants with second pregnancies and 3 participants with multiple pregnancies were excluded. Subsequently, 12 were excluded due to lack of pregnancy outcomes (lost to follow-up or being pregnant in the first trimester). Additionally, ultrasound scans were unavailable for 58 patients between 5 and 11th gestation weeks. Finally, one participant was excluded due to termination in the 12th week for fetal anomaly. Consequently, a total of 603 patients were included in this study. Among these participants, 103 experienced pregnancy loss in the first trimester, while 500 had ongoing pregnancies. The distribution across gestational weeks was as follows: fifth week (426), sixth week (495), seventh week (462), eighth week (420), ninth week (391), tenth week (372) and eleventh week (264). Demographic data were presented in Table 1. The mean age of the entire population was found to be 31 years old with a range from 20 to 44 years and an interquartile range(IQR) of 31(29, 33). The mean BMI was recorded as 22.3 kg/m2. Among the participants, there were 358(59.4%) individuals who experienced two previous pregnancy losses; 159(26.4%) had encountered three previous pregnancy losses; and 86(14.2%) reported four or more previous pregnancy losses. The mean values of hCG, P, E2, mGSD, YSD, and CRL can be found in Table 2.

Table 1 Characteristics of included patients in our study.
Full size table
Table 2 Basic characteristics of the population in different weeks.
Full size table

In the pregnancy loss group, 71, 84, 60, 31, 17, 1 and 0 cases were reported in each week from the fifth to the eleventh week. Correspondingly, in the ongoing pregnancy group, a total of 355, 411, 402, 389, 374, 371 and 264 participants were observed from the fifth to the eleventh week (Table 3). In the pregnancy loss group, there were only 1 in the tenth week and 0 in the eleventh week. Therefore, predictive model was only conducted from the fifth week to ninth week. The BMI and previous pregnancy losses were comparable between the ongoing pregnancy group and the first trimester pregnancy loss group in 5th–9th gestation weeks; however, age, hCG, P, E2, mGSD, YSD and CRL exhibited statistically significant differences as shown in Table 3. Age was higher in first trimester pregnancy loss group from fifth to ninth weeks except for eighth week. Higher hCG levels were observed from sixth to ninth weeks except for fifth week in the ongoing pregnancy group. Similarly, higher E2 levels were identified in ongoing pregnancy group during pregnancy from fifth to ninth weeks. In addition, no statistical differences in mGSD, YSD, and CRL were observed between the two groups in the fifth week. However, in the sixth and seventh weeks, all three parameters (mGSD, YSD, and CRL) were higher in the ongoing pregnancy group. YSD did not show any significant difference in the eighth week whereas both mGSD and CRL exhibited significantly higher values in the ongoing pregnancy group. By the ninth week, CRL remained significantly higher in the ongoing pregnancy group while mGSD and YSD showed no significant differences in univariate analysis.

Table 3 Comparisons between different groups based on pregnancy outcomes.
Full size table

After conducting multivariate analysis (incorporating all variables in the logistic model), as shown in Table 4, it was found that age (OR: 1.129, 95% CI: 1.055–1.208) and P (OR: 0.961, 95% CI: 0.935–0.988) were independent risk factors for pregnancy loss in the first trimester in the fifth week, with a corresponding AUC of 0.671 (95% CI: 0.601–0.740). The RCS analysis revealed a positive linear correlation between age and pregnancy loss, whereas the relationship between serum progesterone and pregnancy loss exhibited a non-linear pattern. The cut-off point for P level, adjusted by age, BMI and previous pregnancy losses, was determined to be 23.9 ng/ml (Fig. 2B), indicating a higher risk of pregnancy loss when the P level falls below this threshold. In the sixth week, age (OR: 1.102, 95% CI: 1.030–1.179), mGSD (OR: 0.887, 95% CI: 0.809–0.973) and CRL (OR: 0.761 95% CI: 0.662–0.875) showed statistically significant associations after multivariate logistic regression analysis with a corresponding AUC of 0.796 (95% CI: 0.734–0.857). Furthermore, RCS analysis indicated non-linear relationships between mGSD and first trimester pregnancy loss (p < 0.05), as well as CRL and pregnancy loss (p < 0.05). Adjusted by age, BMI and previous pregnancy losses, mGSD < 18.3 mm and CRL < 2.4 mm in the sixth week were associated with an increased risk of first trimester pregnancy loss (Fig. 2C, D). In the seventh week, age (OR: 1.111, 95% CI: 1.013–1.218), hCG (OR: 0.998, 95% CI: 0.996–0.999) and CRL (OR: 0.659, 95% CI: 0.574–0.756) were identified as independent risk factors after logistic regression with an AUC of 0.872 (0.814, 0.930). The RCS analysis indicated that CRL < 9.9 mm in the seventh week adjusted by age, BMI, and previous pregnancy losses is associated with an increased risk of first trimester pregnancy loss (Fig. 2E). Additionally, hCG < 69636.6 mIU/ml is also associated with a higher risk of pregnancy loss in the seventh gestation week (Fig. 2F). Although mGSD and YSD in ongoing pregnancy group were larger than those in first trimester pregnancy loss group in the seventh week, they no longer showed statistical significance after multivariate regression analysis. In the eighth week, CRL was identified as the only independent risk factor for first trimester pregnancy loss (OR: 0.670, 95% CI: 0.610–0.816), with a corresponding AUC of 0.871(0.789–0.953). The RCS analysis indicated that there is a linear relationship between CRL and pregnancy loss; CRL < 16.9 mm in eighth weeks is associated with an increased risk of pregnancy loss in the first trimester (Fig. 2G). The independent risk factors identified after logistic regression analysis in the ninth week were CRL (OR: 0.659, 95% CI: 0.574–0.756) and mGSD (OR: 1.221, 95% CI: 1.056–1.412), with an AUC of 0.813 (95% CI: 0.679–0.947). The RCS analysis revealed that a CRL < 18.6 mm in the ninth week is associated with an increased risk of pregnancy loss in the first trimester (Fig. 2H), while mGSD < 33.3 mm and > 48.3 mm are also associated with a higher risk of pregnancy loss; however, caution should be treated due to the small sample size of only 17 women in the pregnancy loss group at this gestational week.

Table 4 Logistic model for predicting first trimester pregnancy loss and the corresponding AUC (95% CI).
Full size table
Figure 2
figure 2

The relationship between predictors and pregnancy loss in the restricted cubic spline analysis. (a) There exists a linear relationship between age and pregnancy loss. (b) Non-linear association exists between P level and pregnancy loss during the fifth week of gestation, the threshold is 23.9 ng/ml. (c) Non-linear association exists between mGSD and pregnancy loss during the sixth week of gestation, the threshold is 18.3 mm. (d) Non-linear association exists between CRL and pregnancy loss during the sixth week of gestation, the threshold is 2.4 mm. (e) Non-linear association exists between CRL and pregnancy loss during the seventh week of gestation, the threshold is 9.9 mm. (f) Non-linear association exists between hCG level and pregnancy loss during the seventh week of gestation, the threshold is 69,636.6 mIU/ml. (g) Linear association exists between CRL level and pregnancy loss during the eighth week of gestation, the threshold is 16.9 mm. (h) Non-linear association exists between mGSD and pregnancy loss during the ninth week of gestation. The mGSD between 33.3 mm and 48.3 mm is in a lower risk of pregnancy loss (HR < 1), while mGSD below 33.3 mm and above 48.3 mm is in a higher risk of pregnancy loss. (i) Linear association exists between CRL and pregnancy loss during the ninth week of gestation. The threshold is 18.6 mm. All the infection points above were determined by adjusting for age, BMI and previous pregnancy losses. P: progesterone; mGSD:mean of the gestational sac diameter; CRL:Crown-rump length; HR: hazard ratio. BMI:body mass index.

Full size image

The sensitivity analysis was conducted by excluding pregnant women with abnormal TSH levels and those who tested positive for antibodies of ACA and β2GP1 during pregnancy during pregnancy (we didn’t exclude those with abnormal TSH level and positive antibody of ACA and β2GP1 before pregnancy for the missing information). A total of 46, 51, 48, 43, and 41 women were excluded from the sensitive analysis at the 5th, 6th, 7th, 8th, and 9th gestation weeks respectively. The remaining numbers included in the sensitive analysis were 380, 444, 414, 377, and 350 at these respective gestation weeks. It is worth noting that after conducting logistic regression analysis (Tables S1, S2, S3, S4 and S5), the independent risk factors identified were consistent with our previous findings (Table S6 and Fig. S1). Therefore, our findings may be applicable to most RPL women.

Discussion

Recurrent pregnancy loss is a distressing disorder, prompting couples to proactively schedule regular visits and scans during subsequent pregnancies, including ultrasound scans as well as measurements of hCG, P, and E2 levels in the first trimester. In the present study, we propose a predictive model utilizing commonly employed markers from the fifth to the ninth week for RPL patients during the first trimester.

In the present study, we found that age and P level were predictive factors in the fifth week for first trimester pregnancy loss. In contrast to previous studies11, our investigation revealed a linear association between age and first trimester pregnancy loss. We support that age is an independent predictor for pregnancy loss1. Additionally, serum progesterone level is also a predictive marker for RPL patients in fifth week, it is worth noting that all of these patients were administered various forms of progesterone once they became pregnant12. Progesterone is released in pulses under the influence of luteinizing hormone, but its release from the corpus luteum is determined by the increase in levels of human chorionic gonadotropin following implantation13. The role of serum progesterone level has been a topic of debate. Coomarasamy et al14 found that progesterone therapy did not significantly increase the incidence of live births compared with placebo in a large cohort study. However, current recommendations support the use of progesterone for women with a history of pregnancy loss and presenting with bleeding during early pregnancy15. Consistent with the previous study16, our study also supports its utility as a predictive marker during early stages of pregnancy for recurrent pregnancy loss women. Notably, in our study, cases of RPL in subsequent pregnancies with serum progesterone levels below 23.9 ng/ml are at a higher risk of experiencing another first trimester pregnancy loss. However, the progesterone cutoff values were lower in other studies17,18. This may be due to the recurrent pregnancy loss participants in our study, leading us to administer individualized progesterone for our RPL patients. In summary, our findings suggest that a higher level of progesterone in the fifth gestation week is beneficial to pregnancy outcomes19. Human chorionic gonadotropin, a placenta-derived glycoprotein, plays a crucial role in ensuring the well-being of pregnancy20. Lower hCG concentration in the seventh gestation week was independently related to first trimester pregnancy loss. hCG facilitates embryonic implantation process, promotes blood vessel growth, regulates trophoblast cell development and plays a vital role in regulating immune responses at the maternal-embryonic or fetal interface throughout pregnancy21. Liu et al. reported that a threshold level of 88,000 IU/L in the peak for hCG may serve as a predictive indicator for early pregnancy outcomes in women who have experienced recurrent pregnancy loss22. In this study, a cutoff value of 69,636.6 mIU/ml is the inflection point for higher risk of pregnancy loss in the seventh gestation week. Therefore, we support that hCG plays an essential role in ensuring the well-being of pregnancy.

The yolk sac plays a crucial role in maternal–fetal exchange before the placental circulation is established. In the sixth and seventh weeks of pregnancy, YSD was smaller in the group that experienced first trimester pregnancy loss. However, after conducting regression analysis, it no longer showed significant differences, which is not consistent previous research23. This discrepancy may be attributed to the inclusion of different variables in our regression model. Our model incorporated important demographic factors and commonly used serum hormones (hCG, P and E2). It is worth noting that larger YSD has been identified as an adverse marker in other studies24,25, our study found a smaller YSD in the pregnancy loss group, consistent with some previous studies7. Our findings indicate a positive correlation between YSD and gestational week, suggesting that YSD increases as gestation progresses as reported by previous studies26. The gestational sac, a fluid-filled structure surrounding the embryo in the early stages of embryonic development, can be visualized via ultrasound as early as 4.5–5 weeks of gestational age27. Its size increases with gestational age. mGSD serves as an early indicator for pregnancy loss23, which is consistent with our research. Based on our RCS analysis, we have determined that an mGSD < 18.3 mm in the sixth week of pregnancy indicates a higher risk for first trimester pregnancy loss. Additionally, RCS analysis conducted in our study showed that by the ninth week, an mGSD < 33.3 mGSD > 48.3 mm is associated with a higher risk of first-trimester pregnancy loss; however, given the small sample size at nine weeks, this result should be interpreted cautiously. Furthermore, CRL has been identified as a reliable predictor for pregnancy loss in the present study. Specifically, a small CRL measurement (less than 2.4 mm, 9.9 mm, 16.9 mm and 18.6 mm) in the sixth, seventh, eighth and ninth week was found to be independently associated with first trimester pregnancy loss. As reported by a previous study28, they determined a different cut-points of 6.0, 8.5, and 10.9 mm for gestational weeks 6, 7, and 8 respectively. The differences might be attributed to the different inclusion of participants and the different methods for identifying the inflection point. In the present study, we included RPL women and used restricted cubic spline to find the inflection point. Simultaneously, the cut-off points were determined by adjusting for age, BMI and previous pregnancy losses, thereby enhancing accuracy. Additionally, our results demonstrated the strong predictive ability of CRL in the eighth week with an AUC value of 0.871. It has been reported that low first-trimester CRL is significantly associated with an increased risk of chromosomal anomalies29. However, another study29,30 reported that short CRL were not associated with an increased risk of fetal chromosomal abnormality. In this study, the majority of patients experienced pregnancy loss before the tenth week, and further investigation is needed to determine specific underlying causes.

The etiology of RPL is intricate and multifactorial, underscoring the necessity for a robust predictive model for pregnancy outcomes. According to Cecilia et al.’s findings6, the final scoring system incorporating maternal age, bleeding score, mGSD, mYSD, and presence of a fetal heart beat achieved an AUC of 0.901. In contrast, utilizing only demographic variables (maternal age and amount of bleeding) resulted in an AUC of 0.724 for predicting ongoing pregnancy. Similarly, using only ultrasound variables (mGSD, mYSD and presence of fetal heart beat) yielded an AUC of 0.873. In our model, logistic regression was employed to incorporate more crucial factors for each gestation week, resulting in a predictive model with only one to three variables. Despite the intricate nature of RPL examined in this study, the discrimination power demonstrates good performance. Furthermore, when compared with another study17 on serum markers for predicting miscarriage, dual markers (estradiol and progesterone; AUC = 0.871) or three markers (hCG, estradiol and progesterone; AUC = 0.869) showed similar performance with the seventh and eighth gestation week of our findings. However, their lack of comparisons across each gestation week may not be suitable for RPL women who require precise evaluation at every stage. In our study, the model in seventh(AUC = 0.872, 95% CI: 0.814–0.930) and eighth(AUC = 0.871, 95% CI: 0.789–0.953) gestation week achieved better performance than fifth (AUC = 0.671, 95% CI: 0.601–0.740), sixth week(AU = 0.796, 95% CI: 0.734–0.857), and ninth week( AUC = 0.813, 95% CI: 0.679–0.947), this is inconsistent with a previous study9, which reported a better model in sixth week and they supposed that it may be related to the morphologic changes of the gestational sac during embryonic development. In the present study, the variability of the different performance in various gestation weeks may lie in that most pregnancy loss occurs in seventh and eighth gestation weeks.

As a special population, women with RPL deserve more attention. RPL brings significant confusion and psychological burden to affected families, particularly for the women involved1. A simplified and practical predictive model could alleviate their stress and anxiety in subsequent pregnancies, ultimately improving pregnancy outcomes. Maternal stress during pregnancy is potentially linked to an increased risk of various adverse birth outcomes1. Given the frequent visits by RPL patients, our study provided a specific predictive marker for different gestational weeks, which may be more practical and simplified in clinical settings. Furthermore, we used RCS to identify the inflection point for higher risk of pregnancy loss. However, it is important to acknowledge that our study has limitations as it was retrospective and conducted at a single center. The findings are limited by potential selection and confounding bias; therefore, further validation through multi-center studies is necessary. Furthermore, the inclusion of RPL women is not limited to unexplained participants in this study, however, we conducted sensitive analysis to further mitigate the potential selection bias, and the results indicate similar predictors and similar AUC in every gestational weeks, therefore, these models might be robust and generalized to most RPL women. Thirdly, despite administering appropriate individualized treatment for patients with recurrent pregnancy loss in our hospital, there may be underlying factors influencing the reproductive outcomes. Preventive measures for subsequent pregnancy loss into the RPL predictive model should also be considered in the future. Additionally, despite their collective experience in ultrasound, inherent variability may exist among the different ultrasound operators included in our study, however, this variation could potentially reflect the true clinical environment. Therefore, future research should consider a well-designed prospective study with a consistent and experienced ultrasound operator. Finally, women in the present study were Chinese, which limited the generalisability of current findings to other populations.

Conclusion

These models and cut-points may help clinicians and patients to make more informed decisions together. More studies are needed to reassure the findings.