4.1 Key Findings and Model Optimization in SVBT cycles with APA
The primary objective of this study was to develop and optimize machine-learning models for predicting live birth outcomes following SVBT cycles in infertile couples with APA. By applying ten machine-learning algorithms and comprehensive data processing methods, we obtained several significant findings.
First, the top three models were Extra Trees Classifier, Stacking Classifier, and Random Forest Classifier, which demonstrated superior performance based on several critical metrics for predicting live birth outcomes following single vitrified blastocyst transfer (SVBT) cycles in infertile couples with advanced paternal age (APA).
The Area Under the Curve (AUC) is a crucial metric for evaluating model performance. It indicates the model's ability to distinguish between positive and negative samples, with values closer to 1 representing better discriminatory power. Both the Extra Trees Classifier and Stacking Classifier achieved an AUC of 0.813, the highest among all models, indicating their superior ability to differentiate between live birth and non-live birth outcomes. The Random Forest Classifier, with an AUC of 0.810, also exhibited excellent performance, closely following the top two models.
These models showed strong comprehensive performance across various metrics. Although accuracy maybe misleading, the accuracy of these top three models closely matched their AUC values, signifying good overall classification performance. Sensitivity and specificity together provided a comprehensive view of the model's effectiveness in identifying live birth and non-live birth outcomes. The top three models exhibited high and balanced sensitivity and specificity, indicating no significant bias in classification. Precision and Negative Predictive Value (NPV) further reflected the models' reliability in predicting positive and negative outcomes, respectively, with all three models performing well on these metrics.
The stability of these models was evidenced by their high F1 scores, which balance precision and sensitivity. This indicates that the models maintain reliable performance even.
The model complexity and generalization ability of these classifiers contributed to their superior performance. The extra trees classifier and random forest classifier, both ensemble models based on decision trees, reduce overfitting risks by randomly selecting features and samples to train multiple trees. This enhances their generalization capabilities. The stacking classifier, which combines multiple base models, leverages the strengths of each to improve overall performance. This approach results in more stable and accurate predictions.
The Extra Trees Classifier and Stacking Classifier are considered the best models due to their outstanding performance across various metrics. The Random Forest Classifier closely follows with similarly excellent performance. These findings suggest that the Extra Trees Classifier, Stacking Classifier, and Random Forest Classifier are well-suited for predicting live birth outcomes following SVBT cycles in infertile couples with APA, providing valuable tools for clinical decision-making in reproductive medicine.
Third, when assessing feature importance using the extra trees method, the mean decrease accuracy method highlight the importance of variables like trophectoderm, inner cell mass, maternal age at oocyte retrieval, blastocyst derived from 8-blastomere at the cleavage stage and total gonadotropin dose.
4.2 Interpretation of the important features influencing live birth outcomes
The results of our study highlight several critical factors influencing live birth outcomes following Single Vitrified-Warmed Blastocyst Transfers (SVBT) in infertile couples with advanced paternal age (APA). Among the 1,044 SVBT cycles analyzed, 29.5% resulted in live births. This success rate underscores the need for meticulous evaluation and optimization of various clinical parameters to improve reproductive outcomes in this demographic.
(1) Blastocyst Quality and Developmental Markers
The analysis revealed that the quality of the trophectoderm is the most significant predictor of live birth outcomes. This finding aligns with existing literature25, indicating that a well-developed trophectoderm, which eventually forms the placenta, is crucial for implantation and live birth26–28. Similarly, the inner cell mass, which contributes to the formation of the fetus, was identified as the second most important factor. High-quality inner cell mass is essential for the embryo's viability and successful development. Prior research has demonstrated that the grade of inner cell mass serves as the most significant indicator of live birth following a single embryo transfer. Specifically, blastocysts at stages 4–5 with an inner cell mass grade of A and trophectoderm grades of A or B are deemed optimal for achieving successful pregnancy outcomes29. Additionally, the presence of a blastocyst derived from 8-blastomere at the cleavage stage was another significant predictor. This marker of early embryonic development indicates a healthy and timely progression through critical stages, which is vital for successful implantation and pregnancy continuation. The number of cells on cleavage-stage can reflect the rate of embryo development and is associated with aneuploidy30. The Istanbul consensus states that the optimal developmental speed for a cleavage stage embryo is the achievement of the cleavage stage with 8 blastomeres on the third day after insemination31. An embryo with 8-blastomeres is considered to have a higher implantation potential32 and a greater probability of achieving a live birth33.This result suggests that, when selecting blastocysts, it is crucial to consider not only the Gardner scoring system34 but also the number of cells in the cleavage-stage embryo. This aspect is vital for achieving a live birth.
(2) Maternal Age at Oocyte Retrieval
Maternal age at the time of oocyte retrieval is one of the most critical predictors of live birth success35,36. This finding aligns with existing literature that emphasizes the dominant role of maternal age in reproductive outcomes37. As maternal age increases, oocyte quality declines, leading to reduced implantation and live birth rates38. Despite the influence of advanced paternal age, maternal age holds critical importance. Our findings indicate that an older male partnered with a younger female may have a higher probability of achieving a live birth compared to couples where both partners are of advanced age. This observation may be due to the higher quality of younger oocytes, which can help counterbalance the lower reproductive potential of older semen39.
(3) Total Gonadotropin Dose
The total gonadotropin dose was found to have a negative correlation with both embryo quality and live birth rates.40. Currently, three potential mechanisms have been identified by which gonadotropin may influence oocyte quality and embryo development: i. High doses of exogenous gonadotropin result in superphysiological estrogen levels in vivo, impairing oocytes and subsequently arresting embryo development41; ii. Ovulation induction with simultaneous development of multiple follicles disrupts the natural selection process of dominant follicles, leading to the recruitment of secondary follicles and thereby compromising embryo quality42; iii. High levels of exogenous gonadotropin interfere with oocyte meiosis and chromosome segregation, resulting in an increased rate of embryo aneuploidy43.
(4) Endometrial Thickness
Endometrial thickness is a reliable and noninvasive measure for evaluating endometrial receptivity44,, serving as a positive independent variable predictive of clinical pregnancy and live birth45. Research indicates that an optimal endometrial thickness is crucial for successful implantation and achieving live birth. Typically, an endometrial thickness of 8.7–14.5 mm is considered ideal46. Deviations from this range, whether too thin or too thick, are associated with reduced live birth rates46. Within the optimal range, the endometrium is more likely to be receptive, providing the necessary environment for the embryo to attach and develop47.
(5) The Number of Oocytes Retrieved and Maternal BMI
The number of oocytes retrieved and maternal BMI were also significant factors, though less critical compared to others. A higher number of retrieved oocytes increases the likelihood of obtaining high-quality embryos for transfer. Data indicate that with an increasing number of oocytes retrieved, clinical pregnancy rates and live birth rates also increase, but this trend plateaus beyond a certain point48,49.
Meanwhile, maternal BMI may affect hormonal imbalances, altered endometrial receptivity and metabolic disturbances associated with obesity, influencing pregnancy outcomes. A study involving data from 42,724 FET cycles indicates that women with a higher BMI (≥ 23 kg/m²) have a significantly lower likelihood of live birth compared to those with a normal BMI (18.5–22.9 kg/m²). Obesity negatively impacts the live birth rate in ART cycles50. Another comprehensive meta-analysis on ART outcomes shows that overweight women (BMI 25-29.9 kg/m²) and obese women (BMI ≥ 30 kg/m²) have significantly lower clinical pregnancy rates compared to women with a normal BMI. Additionally, the study notes that women with a high BMI require higher doses of gonadotropins during treatment51, which negatively affects live birth rates, consistent with our findings.
These findings suggest that a comprehensive approach considering these key predictors can enhance the success rates of SVBT cycles in couples with APA. Clinicians should focus on optimizing embryo quality, particularly the trophectoderm and inner cell mass, as well as carefully managing maternal factors such as age and BMI. Additionally, ensuring appropriate gonadotropin dosing and maintaining optimal endometrial conditions are crucial for improving live birth outcomes. Thus, our study not only aligns with existing literature but also provides additional insights into the complex interplay of factors affecting live birth outcomes in infertile couples with APA undergoing SVBT. By identifying and understanding these predictors, clinicians can better tailor ART protocols to individual patients, thereby improving success rates and providing more accurate counseling to couples.