Prediction of blastocyst development using cleavage-stage embryo metrics and maternal age

Abstract

In assisted reproductive technology (ART), predictive models are becoming increasingly important for improving pregnancy rates and reducing the risks associated with multiple pregnancies. The current standards for selective single-embryo transfer, especially the use of day 5 (D5) blastocysts, are known to enhance pregnancy outcomes. However, variability in embryonic development poses ongoing challenges, necessitating more accurate predictive tools for embryo viability. This study developed a novel predictive model to forecast D5 blastocyst viability using early cleavage stage embryo indicators. Based on a retrospective analysis of 764 in vitro fertilization-embryo transfer cycles, 13 key factors influencing blastocyst formation were identified. Multivariate analysis revealed four independent predictors, including the number of cleavage-stage embryos with > 10 cells, high-quality embryos, 2-pronuclei cleavages, and a menopause-related age metric. The predictive model was formulated and demonstrated high predictive accuracy with an area under the curve of 0.929. Diagnostic testing and internal validation using an independent cohort of 318 blastocyst culture cycles demonstrated that the combined predictor L performed better than the empirical prediction. This study highlights the importance of ART prediction, improving clinical decision-making, and reducing multiple pregnancy risks. This approach empowers patients and enhances the overall effectiveness of reproductive treatments.

Introduction

Integrating patient-reported outcomes (PROs) into medical research is gaining traction, illustrated by their inclusion in cancer symptom management clinical trials supported by National Cancer Institute-sponsored networks, and is increasingly applied in the field of assisted reproductive technology (ART), where PROs play a crucial role in enhancing predictive capabilities¹. This trend aligns with the principles of predictive, preventive, personalized, and participatory medicine, collectively known as “4P medicine”^2,3. Among these four principles, prediction has emerged as a key focus, driving innovations aimed at improving pregnancy rates while minimizing the risks associated with multiple pregnancies.

In the context of in vitro fertilization-embryo transfer (IVF-ET), advancements have led to higher success rates; however, the prevalence of multiple pregnancies remains a significant concern due to associated maternal and infant complications. To address this, selective single-embryo transfer (SET) has become the standard practice in assisted reproduction institutions. Research has demonstrated that single blastocyst transfers yield higher pregnancy rates than single-cleavage embryo transfers^4,5. This reinforces the need for improved predictive models that can accurately forecast embryo viability, thereby guiding clinicians and patients toward safer and more effective treatment decisions⁶.

Despite these advancements, variability in embryo quality and developmental speed following superovulation remains a significant challenge in the IVF-ET process^7,8. Studies have shown that blastocysts transferred on day 5 (D5) are generally associated with higher pregnancy success rates compared to those transferred on day 6 (D6) or day 7 (D7)^9–13. To improve outcome prediction, many ART centers assess day 3 (D3) embryo morphology and the degree of blastomere compaction observed on day 4 (D4) as indicators of a given embryo’s potential to reach the blastocyst stage by D5¹⁴. However, these estimations can often be inaccurate and influenced by the conditions of embryo culture and the expertise of embryologists. When D5 blastocysts do not meet embryo transfer standards, transfers may be canceled, which can extend the treatment cycles and cause emotional and financial distress for patients.

To address these challenges, we aimed to construct a predictive model, named combined predictor L, using early cleavage-stage embryo development indicators from D3 to provide a reliable forecast of the likelihood of forming viable D5 blastocysts. The flow diagram of the study is shown in Fig. 1. By focusing on predictions within the framework of 4P medicine, this tool was designed to enhance clinical outcomes and empower patients through active involvement in their care. Ultimately, by improving our ability to predict successful embryo transfers, pregnancy rates can be increased while effectively reducing the incidence of multiple pregnancies, thereby reaffirming the vital role of patient participation and informed decision-making in reproductive medicine.

Fig. 1.

Detailed workflow for research.

Results

Univariate analysis of factors affecting transferable D5 blastocysts

Univariate analysis was performed on the 19 factors to evaluate their association with the formation of transferable D5 blastocysts. Of these, 13 were significantly different between the non-transferable and transferable D5 blastocyst groups (P < 0.05). These factors included the insemination method, number of 2-pronuclei cleavage (2PN) fertilization, number of 2PN cleavages, high-quality embryos, total number of embryos for blastocyst culture, number of cleavage-stage embryos with > 10 cells, female age, male age, duration of infertility, estradiol levels on the day of human chorionic gonadotropin injection, egg maturation rate, female basal follicle-stimulating hormone levels, and sperm DNA fragmentation index (Table 1).

Table 1.

Univariate analysis of factors affecting the formation of transferable blastocysts on D5.

		Blastocysts formed on D5		Formation rate (%)	χ2/Z price	P-value
		Non-formation group (n = 105)	Formation group (n = 659)
Maternal age, years		32 (30–37)	30(28–33)		-5.616	< 0.001
Male age, years		33.5 (31–37)	32 (29–34)		-4.882	< 0.001
Maternal BMI, kg/cm2		21.25 (19.83–23.05)	21.3 (19.60–23.30)		-0.09	0.928
Infertility duration, years		3 (2–5)	3 (1–4)		-1.971	0.049
Estradiol levels on the day of human chorionic gonadotropin injection (pg/mL)		2268.5 (1,663.25–3,479.25)	3,536 (2,415–4,859)		-6.323	< 0.001
The number of days gonadotropin was injected, days		10 (9–11)	10 (9–11)		-0.278	0.781
Basal follicle-stimulating hormone levels (mIU/mL)		8.02 (6.49–9.21)	7.12 (6.09–8.54)		-2.898	0.004
Male BMI		23.43 (21.88–25.87)	23.84 (21.63–25.95)		-0.134	0.893
Normal sperm morphology rate		7 (2–12)	8 (2–12)		-0.122	0.903
Sperm DNA fragmentation index		13.03 (7.24–20.69)	8.41 (6.25–16.58)		-3.316	0.001
Insemination	IVF	80	570	87.69	7.575	0.006
	ICSI	25	89	78.07
Infertility type	Primary	59	397	87.06	0.618	0.432
	Secondary	46	262	85.06
Egg retrieval rate		86.19% (75–100%)	88.89% (77.78–100%)		-1.458	0.145
Egg maturation rate		86.5% (67.75–100%)	91% (83–100%)		-2.593	0.01
Number of 2PN fertilization		4 (3–6)	9 (7–12)		-12.066	< 0.001
Number of 2PN cleavages		4 (3–5)	9 (6–12)		-12.18	< 0.001
Number of high-quality embryos		0.5 (0–1)	3 (1–4)		-10.863	< 0.001
Total number of embryos for blastocyst cultures		3 (2–4)	8 (5–12)		-12.315	< 0.001
Number of cleavage-stage embryos with > 10 cells		0 (0–0)	1 (0–3)		-9.299	< 0.001

2PN, 2-pronuclei cleavage; BMI, body mass index; D5, day 5; ICSI, intracytoplasmic sperm injection; IVF, in vitro fertilization.

Multivariate analysis of 13 factors influencing outcomes

The 13 significant factors identified in univariate analysis were further evaluated using binary logistic regression. The results revealed that the number of 2PN cleavages, high-quality embryos, cleavage-stage embryos with > 10 cells, and female age were independent predictors of transferable D5 blastocyst formation (P < 0.001) (Table 2). Subsequently, female age was replaced with the menopause-related variable “M,” which was included in the multivariate analysis along with the other variables (Table 3).

Table 2.

Multivariate analysis of factors influencing the formation of transferable D5 blastocysts.

	Regression coefficient B	Standard error	Wald value	Free degree	Conspicuousness	OR (B)	95% confidence interval for the OR (B)
							Lower limit	Upper limit
Number of 2PN cleavages	0.322	0.062	26.835	1	< 0.001	1.380	1.222	1.559
Number of high-quality embryos	0.683	0.132	26.776	1	< 0.001	1.979	1.528	2.563
Number of cleavage-stage embryos with > 10 cells	1.196	0.262	20.869	1	< 0.001	3.307	1.980	5.525
Female age	-0.150	0.034	19.182	1	< 0.001	0.861	0.805	0.920
Constant	2.888	1.128	6.561	1	0.010	17.965

2PN, 2-pronuclei cleavage; D5, day 5; OR, odds ratio.

Table 3.

Multivariate analysis of factors affecting the formation of transferable D5 blastocysts (after female age was converted to M^†).

	Regression coefficient B	Standard error	Wald value	Free degree	Conspicuousness	OR (B)	95% confidence interval for the OR (B)
							Lower limit	Upper limit
Number of 2PN cleavages	0.325	0.062	27.692	1	< 0.001	1.384	1.226	1.563
Number of high-quality embryos	0.659	0.129	26.213	1	< 0.001	1.934	1.502	2.489
Number of cleavage-stage embryos with > 10 cells	1.108	0.246	20.248	1	< 0.001	3.028	1.869	4.906
M†	0.148	0.034	19.468	1	< 0.001	1.160	1.086	1.238
Constant	-4.405	0.689	40.833	1	< 0.001	0.012

^† M = 49 − actual age.

2PN, 2-pronuclei cleavage; D5, day 5; OR, odds ratio.

Construction of the combined predictor L

Based on the data in Table 3, the regression coefficient for 2PN cleavage was set to 1, and the regression coefficients of the other factors were adjusted accordingly by dividing them by the coefficient for 2PN cleavage. This resulted in the following formula for L:

L = 2PN cleavage + () × number of high-quality embryos + () × number of the cleavage-stage embryos with > 10 cells + () × M

Simplifying, the formula becomes

L = 2PN cleavage number + 2.027 × the number of high-quality embryos + the number of cleavage-stage embryos with > 10 cells +  M

Efficacy assessment of the combined predictor L

Receiver operating characteristic (ROC) curves were generated for each independent predictor and combined predictor L. The results indicated that the area under the curve (AUCs) for 2PN cleavage, high-quality embryos, and cleavage-stage embryos with > 10 cells were 0.866, 0.769, and 0.670, respectively. Combined predictor L achieved the highest AUC of 0.929, with a sensitivity of 0.876, specificity of 0.848, and Youden index of 0.724. The optimal cut-off value for L was determined to be 17.7525 (Fig. 2). The Hosmer–Lemeshow goodness-of-fit test yielded a χ² value of 4.009 (P = 0.856).

Fig. 2.

ROC curves of the four independent factors affecting transferable D5 blastocyst formation and of the combined predictor L. D5, day 5; ROC, receiver operating characteristic.

For internal data validation, the formula for the combined predictor, L, was applied to 318 cases with blastocyst culture cycles (Table 4). The L value was calculated using the number of 2PN cleavages, high-quality embryos, cleavage-stage embryos with > 10 cells, and a menopause-related age. Among these 318 cycles, 257 had L values > 17.7525, indicating their potential to yield transferable blastocysts on D5, whereas 61 cycles fell below this threshold. The actual outcomes showed that 239 cycles produced transferable blastocysts, and 79 did not. The sensitivity and specificity of combined predictor L in this internal data validation were 92.9% and 55.7%, respectively. The positive and negative predictive values were 86.381% and 72.131%, respectively, with an AUC of 0.743 (Table 5).

Table 4.

Clinical data related to the internal validation dataset (318 cycles).

	Forming transferable D5 blastocysts (positive)	Forming non-transferable D5 blastocysts (negative)
Female age, years	31.29 ± 3.81	32.13 ± 3.94
Number of 2PN cleavage	9 (6–12)	6 (3–8)
Number of high-quality embryos	4 (2–5)	1 (0–2)
Number of cleavage-stage embryos with > 10 cells	1 (0–3)	0 (0–1)

2PN, 2-pronuclei cleavage; D5, day 5.

Table 5.

Diagnostic test for using the combined factor L to predict blastocyst formation.

	Formation	Non-formation	Row total	Predictive values
Positive prediction	222	35	257	PPV: 86.381% (83.168–89.062%)
Negative prediction	17	44	61	NPV: 72.131% (61.121–80.993%)
Row total	239	79
	SN: 92.887% (88.856–95.802%)	SP: 55.696% (44.077–66.875%)

NPV, negative predictive value; PPV, positive predictive value; SN, sensitivity; SP, specificity.

Before using the combined predictor L, we empirically predicted whether the embryos in this internal validation set would form transferable blastocysts on D5, and the results are shown in Table 6. The sensitivity was 77.406%, the specificity was 62.025%, the positive predictive value was 86.047%, the negative predictive value was 47.573%, and the area under AUC was 0.697.

Table 6.

Diagnostic test for empirically predicting blastocyst formation.

	Formation	Non-formation	Row total	Predictive values
Positive prediction	185	30	215	PPV: 86.047% (82.188–89.179%)
Negative prediction	54	49	103	NPV: 47.573% (40.409–54.838%)
Row total	239	79
	SN: 77.406% (71.571–82.547%)	SP: 62.025% (50.405–72.715%)

NPV, negative predictive value; PPV, positive predictive value; SN, sensitivity; SP, specificity.

Discussion

This study represents a significant effort to enhance predictive capabilities for embryo development and implantation potential with the aim of optimizing clinical outcomes in ART. Limited clinical data suggests that the formation of transferable D5 blastocysts is crucial for achieving smooth clinical outcomes. Although numerous studies have explored artificial intelligence (AI) applications for predicting embryo development^15–18, many rely on time-lapse imaging systems, which are often prohibitively expensive in most domestic centers^19,20. Consequently, the applicability of AI-based prediction models is limited, particularly in facilities that do not have access to such technologies. Thus, the assessment of embryo development in the present study primarily depended on the clinical experience of the medical staff and retrospective analyses, which were aligned with the principles of 4P medicine. Our study provides a novel approach for predicting embryo development potential using clinical data, specifically focusing on early embryo development within the first three days post-insemination. By combining clinical practice, diagnostic insights, and treatment experiences, our findings hold significant application value and can serve as a reference for reproductive centers.

Experience can complement logic when selecting embryos with the highest potential for transfer and minimizing the risk of multiple pregnancies. Experience provides physicians with insights and intuition from practical operations, helping them to understand the complexities of embryonic development and the specific circumstances of each patient. By integrating experience with logic and data analysis, a more comprehensive evaluation of the embryo potential can be achieved. For instance, experienced physicians can identify subtle and potentially unquantifiable factors that enhance logical reasoning. This combination improves the decision-making accuracy, ultimately leading to better clinical outcomes. Therefore, synergy between experience and logic is a crucial pathway for achieving precision medicine.

Previous studies have highlighted that various factors, including the embryo culture environment and patients’ underlying conditions, affect blastocyst formation, although not all directly correlate with pregnancy outcomes^21–23. Among these, egg maturation and the number of mature eggs are critical determinants of embryo development^24,25. Specific morphological characteristics and kinetic parameters have been identified; however, some of these have minimal data before blastocyst formation. For example, Zhang et al. noted that the presence of granules in the central cytoplasm of uninseminated oocytes could negatively affect fertilization, embryonic development, and subsequent blastocyst formation²⁶. Additionally, the number of blastomeres at the D2 stage serves as an important predictor of blastocyst formation^27,28, with partial compaction at D4 being associated with lower rates of successful blastocyst development^29–33.

In our analysis, we compared 19 factors between the transferable and non-transferable D5 blastocyst groups, identifying 13 factors that significantly influenced the outcomes. Four independent factors emerged, including 2PN cleavage, number of high-quality embryos, number of cleavage-stage embryos with > 10 cells, and female age. Notably, when we substituted female age with the menopause-related variable M, the regression coefficient became positive, indicating improved predictive capability for the combined predictor L. The overall predictive value of L was notably high, suggesting its effectiveness compared to individual factors. Furthermore, the Hosmer–Lemeshow test indicated a good fit, and the model was validated using an independent internal dataset. In diagnostic tests, specificity and sensitivity are two indices with opposing meanings. In this study, sensitivity indicated whether our model could correctly predict the successful formation of D5 transferable blastocysts, whereas specificity indicated whether the model could correctly predict the failure of D5 transferable blastocysts. In the same internal validation dataset, empirically predicted results were compared with those calculated using the combined factor L, demonstrating that the combined factor L improved the sensitivity of prediction and AUC. Although the AUC was slightly lower in this validation phase, the model demonstrated robust clinical predictive ability without the need for advanced imaging techniques. Overall, our findings align with existing literature on the influence of common embryonic parameters on development, highlighting the predictive potential of cleavage-stage embryos.

Notably, the number of cleavage-stage embryos with > 10 cells, which reflects the embryo’s developmental speed, emerged as the most highly weighted factor in our combined predictor, L, with a coefficient of 3.409. This underscores the importance of considering developmental speed when predicting D5 blastocyst formation³⁴. Studies have indicated that embryos exhibiting faster developmental rates are more likely to result in successful pregnancies^35,36. Our results not only reinforce the relationship between embryo developmental speed and pregnancy outcomes but also suggest that patients with more embryos exhibiting ≥ 10 cells on D3 should continue to culture to the blastocyst stage^37–39. However, the importance of age cannot be ignored. Factors related to both male and female age significantly influence gamete quality and, subsequently, embryonic development potential^40–42. Poor sperm quality, often due to suboptimal semen parameters or surgical extraction, can hinder embryo quality, particularly in intracytoplasmic sperm injection (ICSI)-derived embryos when compared with those from conventional IVF^43–45. Key metrics, such as the number of 2PN fertilization, cleavage-stage embryos, high-quality embryos, and total number of embryos available for blastocyst culture, also significantly impact the efficiency of blastocyst formation⁴⁶. This strategy may enhance the chances of achieving transferable blastocysts by D5, ultimately supporting the success of single-blastocyst transfer.

Although promising, this study is not without its limitations. First, the sample size was modest, and the majority of enrolled patients had relatively unremarkable baseline characteristics. Furthermore, the variability in embryo assessment criteria and treatment protocols across assisted reproductive centers⁴⁷ may compromise the broader applicability of the model, potentially favoring patient populations with inherently better prognoses. Lastly, the retrospective design inherently carries a risk of bias, highlighting the need for future prospective studies to rigorously validate and refine predictive models within the complexities of real-world clinical practice.

In conclusion, this study underscores the importance of predicting transferable D5 blastocysts by constructing a concise model, the combined predictor L. This model considers female age combined with three parameters from D3 of embryo development to facilitate personalized embryo transfer strategies. A calculated L value exceeding the critical threshold of 17.7525 indicated a higher likelihood of developing transferable blastocysts by D5, allowing timely endometrial preparation. Conversely, lower L values provide a scientific basis for advising patients to cancel blastocyst transfers and to prevent unnecessary costs and unsuccessful procedures. Even if a usable blastocyst forms on D6 and is cryopreserved, the opportunity for transfer in future cycles remains, potentially leading to more satisfactory outcomes and enhanced communication between healthcare providers and patients.

Methods

Ethical approval

This retrospective study was approved by the Ethics Council of Human Research in Xiamen Maternal and Child Health Care Hospital, Fujian, China (Approval Number: KY-2023-135-K01), and the requirement for informed consent was waived due to the retrospective nature of the study. All experiments were conducted in accordance with relevant ethical guidelines and regulations.

Data collection

We conducted a comprehensive analysis of 764 IVF-ET cycles from January 1, 2020, to December 31, 2022, using the Clinical Reproductive Medicine Management System (version 15.5). To ensure the integrity of our study, we excluded cycles that involved ICSI, puncture surgery, rescue ICSI, in vitro maturation, and use of donor eggs or sperm. In accordance with our center’s established protocols, most embryos were cultured in conventional incubators and assessed through routine static observations. However, each day, embryos from one randomly selected patient were placed in a time-lapse incubator, allowing for dynamic imaging and continuous developmental monitoring using time-lapse technology. Evaluations of gametes and embryos, as well as sequential culture procedures, are performed at standardized time points recommended by authoritative sources, such as the Istanbul Consensus, to ensure consistency and adherence to best practices⁴⁸.

Based on a thorough literature review and clinical expertise^24,49,50, we focused on the following 19 key factors influencing blastocyst formation: infertility type, male age, female age, body mass index, duration of infertility, basal follicle-stimulating hormone levels, insemination method, gonadotropin dosage, stimulation duration, estradiol levels on the day of human chorionic gonadotropin injection, egg retrieval rate, egg maturation rate, number of 2PN fertilizations, number of 2PN cleavages, number of high-quality embryos, total number of embryos for blastocyst culture, number of cleavage-stage embryos with > 10 cells, normal sperm morphology rate, and sperm DNA fragmentation index.

Construction of the combined predictor L

To derive our predictive model, we began with univariate analyses to assess the impact of the 19 factors on the likelihood of forming a viable blastocyst on D5, defined as the “outcome.” Factors exhibiting a significance level of < 0.05 were selected for further investigation in a multivariate analysis. Using forward-stepwise binary logistic regression, we identified 13 independent factors that significantly influenced the outcome.

Recognizing that menopause typically signifies the natural limit of female fertility, with an average age of 49 years in China⁵¹, we reformulated the variable “female age” into a new metric, M (where M = 49 − actual age). This transformation has allowed for a more nuanced understanding of declining fertility potential as women approach menopause. We subsequently adjusted the regression coefficients for each factor and constructed a preliminary formula for the combined predictor, L.

Efficacy assessment of the combined predictor L

To evaluate the effectiveness of our combined predictor, we plotted ROC curves for both independent predictors and combined predictor L. The AUCs were calculated to assess predictive accuracy, and the optimal cut-off point was determined using the Youden index, which identifies the threshold that maximizes sensitivity and specificity. A Hosmer–Lemeshow goodness-of-fit test was conducted, and data from 318 blastocyst culture cycles collected between January and April 2023 were analyzed following the same criteria (Fig. 1). This validation dataset did not overlap with the dataset used to construct the formula for L. Diagnostic testing was used to compare empirical predictions with the combined predictor L. Sensitivity, specificity, negative predictive value, and positive predictive value were used to evaluate the predicted effects separately.

Statistical analysis

Statistical analyses were performed using SPSS (version 22.0; IBM Corp., Armonk, NY, USA) and GraphPad Prism 8.0.1.244 (La Jolla, CA, USA), with significance set at P < 0.05. The diagnostic test was performed using MedCalc Software, version 23.0.1 (Washington, DC, USA). Categorical data were analyzed using the χ² test. Normally distributed continuous variables are expressed as mean ± standard deviation and were compared using the t-test. Non-normally distributed data were reported as medians and quartiles, with comparisons made using the rank-sum test.

Binary logistic regression analyses were employed to confirm the significant variables identified in the univariate analysis (P < 0.05), with the forward stepwise regression methodology utilized for model refinement. The risk prediction model was formulated based on the final regression coefficients of the significant predictors¹⁵.

The multiple logistic regression model can be expressed as follows:

In this equation, π(x) represents the probability of transferable D5 blastocyst formation based on the given predictor variables X₁, X₂,., X_p. The coefficients (β₀, β₁, β₂, …) in the logistic regression model represent the strength and direction of the relationship between each predictor variable and log odds of the outcome¹⁶.

Author contributions

Hong Ji: study design, article writing, statistical analysis, and drawing; Qiaomei Bai: mapping of statistical maps; Lu Ding: data compilation, article writing; Lizhi Jiang: mapping of workflow; Yingying Shi: research guidance; Longmei Wang: statistical analysis; Li Meng: review and proofreading; Ping Li: ethical approval.

Data availability

Data are available from the corresponding author upon reasonable request.

Competing interests

The authors declare no competing interests.

.