Background

Infertility is a serious human reproductive health issue, which affects about fifteen percent of married couples [1]. Nowadays, In vitro fertilization and embryo transfer (IVF-ET) is applied more and more widely worldwide. In some circumstances, when the patients have a tendency to suffer ovarian hyperstimulation syndrome or their endometrium is too thin in fresh IVF cycle, or they have a plan to conduct preimplantation genetic testing, frozen-thawed embryo transfer (FET) become an effective alternative. The proportion of FET is increasing gradually and now accounts for about 30–40% of all the transfer cycles. Therefore, more and more patients are receiving FET treatment.

Although the live birth rate (LBR) of FET improved recently in the world, the average success rate remains low. It is hard to predict clinical pregnancy rate and live birth rate objectively because of the heterogeneity in etiology and treatment protocols et cl. How to identify the risk factors and predict live birth after embryo transfer has received widespread attention both from the infertility population and clinical researchers.

As we all knows, pathophysiology and pregnancy outcomes between ovulatory infertility and anovulatory infertility are different to some extent. For example, the live-birth rate of FET is higher than that of fresh-embryo transfer in anovulatory patients who suffered polycystic ovary syndrome (PCOS) [2]. Frozen single blastocyst transfer resulted in a higher singleton livebirth rate than did fresh single blastocyst transfer in ovulatory women [3]. However, there is no difference in ovulatory women between FET and fresh-embryo transfer of cleavage-stage embryos [4]. Recently, there are studies trying to figure out risk factors and construct predictive models in specific infertility population [5]. For example, obesity and metabolism factors were used to develop model to predict the pregnancy outcomes in PCOS patients who received IVF/ICSI or FET [6, 7]. In ovulatory women, there are some studies trying to figure out the best FET protocols [8, 9]. However, until now, studies trying to predict clinical pregnancy outcome in ovulatory women remain limited.

Many mathematical models such as the nomogram have been used in the field of assisted reproduction technology [10,11,12]. Nomogram converts complex regression analyses into visual figures, making predictive models more readable and convenient. The purpose of this study was to identify factors that affect ovulatory women's live birth rate. By logistic regression and nomogram, we tried to create a convenient tool for predicting the live birth in ovulatory women undergoing FET.

Methods

Study design and population

We enrolled patients younger than 42 years old with regular length of menstrual cycles between 24 and 38 days. They received their first or second FET cycles from June 2015 to December 2021 in the reproductive center of Qilu hospital, Shandong University. Exclusion criteria include recurrent spontaneous abortion (defined as three or more spontaneous abortions prior to 20–28 weeks gestation), abnormal anatomy of uterine, intrauterine adhesions, endometriosis, polycystic ovary syndrome, endocrine disorders and chromosomal diseases (exclude chromosome polymorphism). A total of 2884 patients were enrolled and assigned randomly into training (n = 2307) and validation cohort (n = 577). The former was used for the construction of the model and the latter was used for internal validation. Indicators for IVF/ICSI treatments were tubal factor (64.9%), male factor (12.7%), unexplained (4.5%), mixed factor (13.6%) and diminished ovarian reserve [13] (4.3%).

Ethics Committee approval for this study has been granted by Qilu Hospital, Shandong University (registration number: KYLL-202306–019). As a retrospective study, we were exempt from informed consent. Our protocol of research was performed according to the Declarations of Helsinki.

Data collection

We collected clinical data including male and female age, duration and type of infertile history, gravidity and labor history, body mass index (BMI), reproductive endocrine hormones and anti-Müllerian hormone (AMH). The protocols of FET, detailed information of embryos transferred and endometrial thickness on transplant day were recorded. The outcome was live birth. We defined live birth as live newborn ≥ 28th week of pregnancy.

Embryo score

Morphological score of embryos was assessed before embryo transfer. The qualities of the Cleavage-stage embryo were evaluated into four grades according to their speed of development, degree of fragmentation and evenness of cleavage sphere. We scored blastocysts according to Gardner morphological criteria [14].

Protocol of FET

Natural cycle (NC) or artificial cycle (AC) were selected based upon patients’ individual prefer and the experience of physicians in charge. For NC, clinicians examined the follicle size and endometrial thickness in order to get the ovulation day by transvaginal ultrasound examination. HCG or GnRH-a was administrated for ovulation induction in some patients. FET was scheduled based on ovulation day. For AC, estrogens are administered for approximately 14 days from cycle day 3. Endometrial thickness and the serum estrogen level are monitored from cycle day 13. Patients were given progesterone when their endometrial thickness were at least 7 mm and the serum estrogen level of over 100 pg/ml. Thawing and transferring of embryo was scheduled based on the supplementation of progesterone. On the day of ovulation, intramuscular injection of progesterone 40–80 mg/d or oral dydrogesterone 20-40 mg/d was administered for luteal support of NC. Intramuscular injection of progesterone 40–100 mg/d, or progesterone vaginal sustained-release gel 90 mg/d or micronized progesterone 300–800 mg/d was administered for luteal support of AC. The luteal support was maintained until intrauterine pregnancy is seen on ultrasound, and then the dose was reduced until 8 ~ 10 weeks.

Data analysis

We performed statistical analyses with statistical packages R and SPSS 26.0. For continuous variables, mean ± SD (for normally distributed variables) or median and IQR (for skewed distributed variables) was used and normality test was detected by the Kolmogorov–Smirnov test. Student t test or Mann–Whitney U tests were applied to compare continuous variables. Frequencies (percentages) were presented and Chi-square test was applied to compare categorical variables. An analysis of the relationship between variables and live birth was conducted by using logistic regression. We first selected variables (P < 0.1) in univariate analyses to avoid missing important risk factors [15, 16]. Then we screened out final variables by multivariable logistic regression analysis (MLR) (p < 0.05) and enrolled them into the nomogram model. Variance inflation factor was used to detect the multicollinearity with the criterion of less than 10.

Nomogram was constructed by the data in training cohort to predict the probability of live birth in ovulatory women receiving FET. The variables enrolled in nomogram are female age, AMH, number of high-quality embryos, FET protocol and stage of embryos. Each variable was mapped on its scale to get a score, then the total score of all the covariates was calculated and the probability of live birth was finally obtained. The area under the receiver operating characteristic (ROC) curve (AUC) was used to detect the discriminatory ability. Calibration curve was conducted with a bootstrapping technique (1,000 repetitions) to detect the calibration. Decision curve analyses (DCA) was applied to calculate the net benefit of making a decision [17].

Results

Baseline characteristics

Patient characteristics and parameters related to FET cycle were shown in Table 1. Clinical pregnancy rate was 55.61% and 57.71%, separately. Live birth rate was 45.60% and 48.01%, separately. The characteristics observed between the two cohorts were not significant.

Table 1 Characteristics of the training and validation cohorts (N = 2884)
Full size table

As shown in Table 2, the characteristics of women with and without live birth in the training cohort were displayed. Women with live birth had younger age, lower BMI and baseline FSH, higher AMH, thicker endometrium on the day of ET. They transferred more high-quality embryos and higher percentage of blastocyst. The FET protocol is also different. The percentage of patients with live birth and their physician who chose natural cycle is 72.24%, while the percentage of patients without live birth and their physician who chose natural cycle is 68.13%.

Table 2 Baseline characteristics among women with or without live birth in the training cohort (N = 2307)
Full size table

ROC analyses were conducted to analyze the proper cutoff values of variables. The cutoff point of the female age related to live birth was 32.5 and the optimal cutoff point of AMH was 2.265 (Supplementary Table 1). For the cutoff point of the female age, the sensitivity and specificity for live birth is 0.525 and 0.641. For the cutoff point of AMH, the sensitivity and specificity for live birth is 0.717 and 0.440.

Logistic analysis of the risk factors of live birth

The univariable and multivariable logistic analyses in the training cohort revealed that female age, AMH, protocol of FET, stage of embryos and number of high-quality embryos were significantly correlated with live birth. The MLR showed that the live birth was associated with age > 32 (OR: 0.629; 95% CI:0.511–0.773, P = 0.000), AMH > 2.26 (OR: 1.406; 95% CI: 1.143–1.730, P = 0.001), number of high-quality embryos (OR:1.896; 95% CI: 1.602–2.243, P = 0.000), artificial FET cycle (OR: 0.734; 95% CI:0.600–0.897, P = 0.003) and blastocyst embryos transferred (OR, 3.829; 95% CI, 2.850–5.145, P = 0.000) (Table 3).

Table 3 The univariate and multivariate logistic regression analysis for factors associated with live birth in the training cohort (N = 2307)
Full size table

Construction of nomogram to predict live birth

Based on the logistic regression analyses in the training cohort, five factors including age, AMH, cleavage embryo/blastocyst, number of high-quality embryos and protocol of FET were selected. we incorporated these factors into nomogram which could predict live birth (Fig. 1A).

Fig. 1
figure 1

Nomogram to predict the probability of live birth for ovulatory women undergoing FET. A To calculate a patient’s live birth rate, first label her value on each variable axis and get the corresponding points through the top Points scale. Then, add the points of each variable value together and label it on the Total Points scale. Finally, obtain the personalized probability of live birth by the bottom axis. B The ROC curves of the nomogram for probability of live birth in the training cohort. C The ROC curves of the nomogram for probability of live birth in the validation cohort. D Calibration curve of the nomogram in the training cohort. E Calibration curve of the nomogram in the validation cohort. After 1000 repetitions of bootstrap, the calibration curves showed a good correlation between the predicted probability and actual probability. ROC: receiver operating characteristics curves

Full size image

Validation of the nomogram

ROC curves were shown. In the training dataset, AUC for live birth (Fig. 1B) was 0.666. In the validation dataset, the AUC was 0.669 (Fig. 1C). The sensitivity of the training cohort was 73.5% and specificity was 50.1%. The sensitivity of the validation cohort was 56.3% and specificity was 68.3%.The calibration curves for live birth were shown in Fig. 1D and E. Decision curve analyses were used to calculate the net benefit which were shown in Fig. 2. The black line means the net benefit of no patients got live birth. The gray one means the net benefit of all patients got live birth. The closer the curve to these two reference lines, the less valuable the model is. Our curve is above the reference line in a relative wide threshold interval.

Fig. 2
figure 2

The decision curve analysis of the nomogram for probability of live birth in the training cohort and validation cohort. A The decision curve analysis of the nomogram for probability of live birth in the training cohort. B The decision curve analysis of the nomogram for probability of live birth in the validation cohort

Full size image

Discussion

Live birth is the most important issue concerned by patients who received FET and their physician. Based on 2884 patients with regular menstrual cycles, we constructed an easy-used nomogram to predict the live birth probability in this study. The indicators enrolled in the nomogram include female age, AMH, embryo transfer protocols, cleavage embryos/blastocysts and number of high-quality embryos. Moreover, we evaluated our nomogram by calibration curve, discrimination curve and decision curve analysis. Our model showed moderate discriminatory ability and well-calibrated ability overall. Decision curve analyses showed that our model could get net benefit in a relative wide threshold interval.

In recent years, there are many predictive models to predict the outcome of infertility population or those who received assisted reproductive technology. For example, Scientists constructed different predictive models to calculate cumulative live birth rate in patients who accepted fresh IVF treatment [18, 19]. However, only a few predictive models were constructed to predict pregnancy outcomes in patients who received FET treatment. Yu et al. tried to establish a predictive nomogram for early clinical pregnancy but not live birth in FET cycles [20]. In another study, a data-driven predictive model using machine learning was constructed to calculate LBR in FET, with a satisfactory performance [21]. However, the infertile population is highly heterogeneous. The pathophysiology, treatment protocol and clinical outcome of ovulatory people and anovulatory people is different. It is necessary to construct predictive model for ovulatory people who received FET. Our model aims to do this and it helps those patients to have a reasonable expectation of their LBR. It also helps clinicians to choose proper transfer protocol and embryos.

As we all know, female fecundity declines with age [22]. Advanced maternal age reduces ovarian reserve, decreases oocyte/embryo competence and negatively associates with pregnancy outcomes [23]. No remedies could effectively antagonize the aging-related fertility decline [24]. It was reported that age is the only variable predicting live birth in patients undergoing IVF treatment [25]. Our result was in accordance with those previously published researches. In this study, we calculated the ideal cutoff values of age for live birth and found that the cut-off value was 32.5 years old.

AMH is a proper metric for ovarian reserve function or ovarian response. Research suggests that AMH could predict live birth independently of age in ART cycles, although the predictive ability is poor [26]. AMH was also proven to be related to implantation and clinical pregnancy [27]. We incorporated AMH in our model according to results of MLR. In our study, the OR for AMH as an indicator of live birth was 1.406 and the optimal cutoff value for AMH was 2.26 ng/ml. This shows a high degree of consistency with previous study. In another study predicting early pregnancy rate in FET, the cutoff value of AMH was 2.23 ng/ml [20].

The studies about the relationship between basal FSH and pregnancy outcome was inconsistent in previous studies. It was reported that basal FSH was independent predictive factor of live birth in poor ovarian response patients [28] and women elder than forty years [29]. However, in donor/recipient pairs undergoing oocyte donation IVF, donor basal FSH was not associated with pregnancy outcomes [30]. In our study, after adjusting the confounding factors, multivariable logistic analysis revealed that basal FSH level was not related with live birth. This difference may be due to the different populations enrolled in different studies.

Artificial cycle (AC) and natural cycle (NC) are the main endometrial preparation regimen in FET. However, no consensus was reached as which protocol is optimal. Although previous researches were controversial, it seemed that NC was better in ovulatory women. For example, a study found that modified NC was linked to a higher LBR in patients with regular menstruation than AC [31]. A similar study found that AC was linked to higher ectopic pregnancy rate and lower LBR in ovulatory women [32]. A recent study found that AC was related with higher risk of large for gestational age, hypertensive disorders and higher birthweight in 3,639 patients who got live-born singletons after FET [33]. In our study, we found NC regimen had a higher probability of live birth in ovulatory women.

In our nomogram, cleavage embryo/blastocysts and number of high-quality embryos are two dominant factors. The OR of “stage of embryo” and “number of high-quality embryos” is 3.829 and 1.896, separately. These two indicators were also the common indicators in other models. For example, Zhang. et al. incorporated the “stage and number of transferred embryos” into the nomogram predicting the early pregnancy in women who received FET [20]. Zhou et al. found that “number of high-quality embryos” are related to live birth in artificial FET [34].

As we all know, the factors influencing live birth are complex. The discriminatory ability and calibration ability of the models predicting live birth in patients receiving IVF treatment are usually not so high. For example, Dhillon et al. constructed a predictive model of live birth probabilities with AUC of 0.62 in the validation cohort [35]. Qui et al. constructed a personalized prediction of live birth prior to the first IVF cycle with AUC of 0.73 by using complex machine learning method [36]. In our study, we analyzed risk factors of live birth in ovulatory women with regular menstruation in our center. To our knowledge, it is the first predictive model of live birth for ovulatory women receiving FET. Secondly, our nomogram is a visual and simple model. The indicators we incorporated in our model are all easily got clinical measurements, which made our model be more applicable than other complicated models.

However, there are also some disadvantages in our study. Firstly, our study did not analyze the female age at oocyte retrieval, which may neglect the potential relationship between female age at oocyte retrieval and live birth. Secondly, it is single-center research, and we still need outer validation from other center’s datasets. Thirdly, it is retrospective research, in which bias is unavoidable to some extent. Prospective study is needed to validate the sensitivity and specificity of our nomogram. Finally, our predictive model has a limited accuracy partly because the indicators we enrolled in the nomogram are all the simple and easy-got clinical data. The heterogeneity of the infertile population also makes it difficult to obtain an accurate predictive model. Further studies focusing on those limitations need to be conducted.

Conclusions

To sum up, we constructed a practical predictive nomogram for live birth in ovulatory women undergoing FET. We used five variables including age, AMH, protocol of FET, types of embryos and number of high-quality embryos, which were screened out by logistic regression analyses. Our model provided a reference for patients with regular ovulation and help them establish a reasonable expectation of their live birth. It aims to help the couples minimize their anxiety and confusion, which will further enhance their confidence in the long course of ART treatment. However, our data were from a single reproductive center and the accuracy of our predictive model is limited, outer validation and further exploration to improve the performance of our model are needed in the future.