In-vitro fertilization failure risk prediction model and construction method and prediction method thereof

Abstract

The invention discloses an in-vitro fertilization failure risk prediction model and a construction method and a prediction method thereof, based on data of an ART treatment cycle, multi-dimensional indexes such as woman age, BMI, male seminal fluid parameters and infertility years are incorporated, logistic regression, a random forest and an XGBoost model are constructed, and performance is evaluated by adopting a nested cross validation framework (including an SMOTE technology). Results show that the logistic regression model is optimal in performance and is remarkably superior to a random forest model and an XGBoost model, key predictive factors include male age, female BMI, the total number of forward motional sperms and a sperm DNA fragmentation index, the total number of forward motional sperms and the male age are protection factors, the BMI and the sperm DNA fragmentation index are risk factors, and the risk factors are the risk factors. The C-index of the model column diagram is up to 0.722 through internal verification, and the model column diagram has the clinical distinguishing capability of medium or above. The logistic regression model is verified to have high efficiency and stability in IVF fertilization failure prediction, and can provide decision support for clinical early intervention.

Description

Technical Field

[0001] The present invention relates to the technical field of human assisted reproductive technology evaluation, and specifically to an in vitro fertilization failure risk prediction model and its construction method and prediction method. Background Art

[0002] In vitro fertilization (IVF) is the most widely used assisted reproductive technology (ART) procedure. It is primarily suitable for patients seeking treatment due to female factors (such as ovulatory disorders, fallopian tube abnormalities, pelvic endometriosis, and cervical factors), male factors (limited to mild to moderate oligoasthenoteratozoospermia), and unexplained causes. Conventional IVF boasts a fertilization rate of approximately 60%-80%. However, clinically, approximately 20% of cycles still experience low fertilization rates (LFR, <25%), and even 5%-15% experience complete fertilization failure (TFF). To avoid these issues, the best solution is to perform a short-term insemination (SIF) procedure, which involves performing degranulation 4-6 hours after IVF insemination to monitor the release of the second polar body. Early rescue ICSI (Rescue ICSI) is performed for cycles with a fertilization rate below 30%. However, even when an IVF regimen is selected based on the patient's clinical parameters at the outset of treatment, some cycles still experience low fertilization rates, which may be related to a variety of factors.

[0003] To more accurately predict the risk of fertilization failure during IVF cycles, this study retrospectively analyzed clinical parameters from our center's IVF and Rescue ICSI cycles. Based on this data, we developed multiple prediction models (including logistic regression, random forest, and XGBoost algorithms). By integrating multiple clinical indicators, we aim to develop an efficient prediction tool for identifying patients with low fertilization rates who may require early intervention. The application of this prediction model will help optimize patient enrollment strategies, improve IVF success rates, and reduce unnecessary waste of medical resources. Summary of the Invention

[0004] The present invention aims to provide a model for predicting the risk of in vitro fertilization (IVF) failure, a method for its construction, and a prediction method. Based on data from ART cycles, the model incorporates multiple indicators, including female age, BMI, male semen parameters (sperm concentration, total progressively motile sperm count, DNA fragmentation index, etc.), and years of infertility. Logistic regression, random forest, and XGBoost models were constructed, and performance was evaluated using a nested cross-validation framework (including SMOTE). Results showed that the logistic regression model performed optimally, significantly outperforming both the random forest and XGBoost models. Key predictors included male age, female BMI, total progressively motile sperm count (TPMC), and sperm DNA fragmentation index (DFI). TPMC and male age were protective factors, while BMI and DFI were risk factors. The model nomogram achieved an internal validation C-index of 0.722, demonstrating above-average clinical discriminatory power. The logistic regression model demonstrated both high efficiency and stability in predicting IVF fertilization failure, providing decision support for early clinical intervention. The model's generalization ability can be further optimized using multicenter data.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions: The present invention provides a method for constructing an in vitro fertilization failure risk prediction model, comprising: S1. Based on the data of ART treatment cycles, extract the basic information of male and female patients, the hormone status of female patients, the semen status of male patients, and the embryo status; S2. Perform normality and variance analysis on the extracted data to obtain a data set; S3. Screen the indicators with statistical differences in the data set to select the core prediction indicators; S4. Build logistic regression models, random forest models, and XGBoost algorithm models, and use a nested cross-validation framework to evaluate model performance based on the dataset. S5. Based on the selected core prediction indicators, evaluate the ability of the logistic regression model, random forest model, and XGBoost algorithm model to predict the risk of in vitro fertilization failure. Comprehensively determine the area under the receiver operating characteristic curve, confusion matrix, and the accuracy, regression coefficient, and F1 score of the test set prediction to select the optimal prediction model. S6. Output indicator parameters for the model with the best performance to verify its prediction performance.

[0006] Furthermore, in step S3, the indicators with statistical differences in the data set are screened out, the indicators with P < 0.05 are summarized and tested for collinearity, and then the indicators for inclusion in the logistic regression model are screened out through the results of backward stepwise analysis; then, based on the cumulative Gini importance of the indicators being greater than 90%, the indicators for inclusion in the random forest model are screened out; and then, based on the cumulative gain value feature importance evaluation of the XGBoost algorithm model, the core prediction indicators are screened out.

[0007] Furthermore, nine core predictive indicators were finally screened out, namely male age, female BMI index, sperm concentration, progressive motility sperm rate (PR), total progressive motile sperm count (TPMC), very slow motile sperm rate (IM), sperm DNA fragmentation index (DFI), infertility classification, and infertility years.

[0008] Furthermore, in step S4, a nested cross-validation framework is used to evaluate the model performance. In the outer validation stage, an independent test set without any processing is used for the final performance evaluation. The outer validation divides the data set into a training set and a test set through 5-fold stratified cross-validation. Each outer training set is further optimized for model hyperparameters through an inner 5-fold stratified cross-validation.

[0009] Furthermore, in step S5, the indicator parameters of the model with the best performance are output, and all variables pass the significance test; a nomogram is constructed for each variable, and the regression coefficient is mapped into a risk score of 0-150 points. The total score is the sum of the scores of each variable, and the total score is calculated; and then the prediction efficiency is verified through internal Bootstrap verification.

[0010] Furthermore, the data of ART treatment cycles were divided into two categories, namely, IVF group and Rescue ICSI group. Among them, the Rescue ICSI group refers to early rescue when the number of eggs with second polar bodies was less than 30% 6 hours after IVF fertilization.

[0011] The present invention also provides an in vitro fertilization failure risk prediction model, which is constructed using the above method, and the optimal prediction model is a logistic regression model.

[0012] Furthermore, the output indicator parameters of the logistic regression model are male age, female BMI index, total progressive motile sperm count (TPMC), and sperm DNA fragmentation index (DFI).

[0013] The present invention also provides a method for predicting the risk of in vitro fertilization failure using the above-mentioned prediction model, comprising: obtaining four indicator parameters, namely, the male's age, the female's BMI index, the total number of forward-motile sperm, and the sperm DNA fragmentation index; inputting the above-mentioned indicator parameters into a logistic regression model; calculating the contribution score of each indicator parameter through the logistic regression model and summing them to obtain a comprehensive risk score; the comprehensive score ranges from 100 to 150 points; the higher the score, the greater the probability of fertilization failure.

[0014] Furthermore, the risk threshold is set to 112 points. When the calculated comprehensive risk score is greater than 112 points, it is judged as high risk.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention identified a number of statistically significant factors by comparing the patient's basic clinical condition, ovulation induction status, and male semen condition, including female factors (female age, female body mass index), male factors (male age, sperm concentration, total number of progressively motile sperm, sperm acrosome enzyme activity, progressively motile sperm rate, very slow-motile sperm rate, sperm DFI index), and unexplained factors (infertility classification, infertility years). Based on the above research, three prediction models, namely logistic regression, random forest, and XGBoost, were constructed, and the model performance was systematically evaluated using inner and outer five-fold cross-validation. Empirical results showed that logistic regression (average AUC = 0.734 ± 0.049) significantly outperformed random forest (average AUC = 0.714 ± 0.034) and XGBoost (average AUC = 0.697 ± 0.038) in predictive performance. Classification results also revealed that the logistic regression model outperformed the other two models in terms of precision, recall, and F1 score. Based on linear separability of the data and low inter-feature complexity, logistic regression was considered the optimal predictive model. Regression analysis revealed that male age, female BMI, total progressively motile sperm count (TPMC), and sperm DNA fragmentation index (DFI) were the most relevant factors for IVF fertilization, with TPMC being particularly effective in predicting normal or negative fertilization rates. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0017] Figure 1 A flowchart of the modeling algorithm of the present invention; Figure 2 This is a heat map of the backward stepwise regression analysis in the present invention; Figure 3 This is a graphical representation of the top 9 indicators selected in the present invention with cumulative Gini importance greater than 90%, with the X-axis representing Gini importance and the Y-axis representing the indicators arranged in descending order of importance; Figure 4 This is a graphic representation of the top 9 indicators with cumulative gain values greater than 90% selected in the present invention. The X-axis represents the gain value of each indicator, and the Y-axis represents the indicators arranged in descending order of importance. Figure 5 is the nested cross-validation ROC curve of each model in the present invention; Figure 6 This is the IVF fertilization failure risk assessment nomogram of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] The present invention aims to construct a prediction model for in vitro fertilization (IVF) fertilization failure based on multidimensional clinical indicators to identify patients at high risk of fertilization failure in in vitro fertilization (IVF) cycles, optimize treatment decisions and improve success rates.

[0020] 1. Research Objects and Methods 1.1 Research subjects Based on data from the Assisted Reproduction Center of Wenzhou People's Hospital from January 2019 to August 2024, the initial inclusion criteria were all ART cycles. Exclusion criteria were cycles with outcomes of TFF (total fertilization failure), ICSI (intracytoplasmic sperm injection), and HALF-ICSI (half-intracytoplasmic sperm injection). A total of 691 cycles were included. Rescue ICSI was performed when the number of oocytes with a second polar body (PB2) less than 30% 6 hours after IVF insemination was considered a cutoff. The final stratification was as follows: 594 patients in the IVF group and 97 patients in the Rescue ICSI group.

[0021] 1.2 Research Methods The following data were extracted from the above 691 case database: (1) Basic information of enrolled patients The woman's age, the man's age, the woman's BMI index, occupation, education level, years of infertility, infertility factors, infertility type, ovulation induction plan, gonadotropin (Gn) usage, and number of days of Gn use.

[0022] (2) Hormone status Basal hormone levels (estradiol, luteinizing hormone, and follicle-stimulating hormone (FSH)) and trigger-day estradiol and AMH were measured using the Beckman Coulter DxI 800 Immunoassay Analyzer using chemiluminescence. Hepatitis B surface antigen (HBsAg) was measured using the Shenzhen Yhlo Biotech iFlash 3000 chemiluminescent immunoassay analyzer.

[0023] (3) Semen condition Abstinence for 3-5 days, fresh semen was collected on the day of egg retrieval, and standardized testing procedures were carried out in strict accordance with the "World Health Organization Laboratory Manual for the Examination and Processing of Human Semen" (Fifth Edition). The amount of semen was measured by weighing, and key parameters such as sperm concentration, progressive motility sperm rate (PR), non-progressive motility sperm rate (NP) and very slow motility sperm rate (IM) were quantitatively analyzed using a Makler chamber. The total number of progressively motile sperm (TPMC) was calculated by semen volume * sperm concentration * PR. Sperm selection was performed using the Isolate® double-layer density gradient centrifugation method. The separation medium included an upper layer of semen separation medium and a lower layer of semen separation medium. The SOP was strictly followed, and the concentration of the selected sperm was finally adjusted to 3*10 6 / mL is used for IVF fertilization.

[0024] (4) Determination of sperm acrosome enzyme activity Obtain the results of the patient's most recent semen examination (within 3 months). This project uses the modified Kennedy enzyme method. The specific process is as follows: take fresh liquefied semen and mix it, determine the sperm density through preliminary experiments, and accurately measure the sperm containing 7.5×10 6 The volume of semen from each sperm was added to the assay tube (experimental group) and the blank control tube (control group), and the supernatant was removed after centrifugation at 3000×g for 10 minutes. The precipitate was then processed according to the reagent instructions and the enzyme reaction system was constructed. Finally, the absorbance value was measured at a wavelength of 405 nm using an enzyme-labeled analytical system. The experimental data were calculated and analyzed by the absorbance difference between the assay tube and the blank control tube.

[0025] (5) Sperm DNA fragmentation index (DFI) test Obtain the results of the patient's most recent semen analysis (within 3 months). This test utilizes the SCSA method (Sperm Chromatin Structure Analysis). The principle is that after standard acid denaturation, acridine orange dye specifically binds to different DNA structures through its metachromatic properties. Normal sperm double-stranded DNA exhibits green fluorescence (excitation wavelength 530 nm), while abnormal sperm single-stranded DNA emits red fluorescence (excitation wavelength 640 nm). A flow cytometer collects 5,000 sperm events per sample, and the intensity ratio of the green fluorescence signal (double-stranded DNA) to the red fluorescence signal (single-stranded DNA) is used to calculate the quantitative DNA fragmentation index (DFI).

[0026] (6) Embryo status Number of eggs retrieved: The total number of oocyte complexes obtained by transvaginal ultrasound-guided puncture, with the complete oocyte structure observed under a microscope as the counting standard.

[0027] Number of mature oocytes: The cumulus-ovarian complex was mechanically degranulated 4 hours after IVF insemination, and oocyte morphology was observed under an inverted microscope (×200). Mature oocytes were defined as those with homogeneous cytoplasm, an intact zona pellucida, and completion of the first meiotic division (MII). Specific criteria for determining maturity were the presence of the first polar body (PB1), confirming that the oocyte had expelled the first polar body (indicating completion of the first meiotic division and progression to MII). The presence of the second polar body (PB2) was optional: the PB2 was expelled only after fertilization. Its presence or absence was used to determine fertilization status, not oocyte maturity.

[0028] Fertilization number: Observe 18±1 hours after IVF fertilization, and the appearance of clear pronuclear structure (PN) is used as the fertilization criterion.

[0029] Normal fertilization number: The number of eggs that show two pronuclei (2PN) during observation. The two pronuclei must be of equal size and the nucleoli must be clearly visible, which is consistent with the normal fertilization characteristics of humans.

[0030] Abnormal fertilization number: including eggs with one pronucleus (1PN), three pronuclei (3PN) or multiple pronuclei (≥4PN).

[0031] All the above embryo evaluations were completed collaboratively by two senior embryologists.

[0032] 1.3 Statistical analysis All data were analyzed for normality and homogeneity of variance. Data that met the normal distribution were expressed as (x ± s) and the t test was used. If the data did not meet the normal distribution, nonparametric statistical analysis was applied. The data were expressed as median (quartile) [M (P25, P75)] and the two groups were compared using the Mann-Whitney U test. Enumeration data were expressed as percentages (%) and the intergroup comparisons were performed using the χ test.2 test.

[0033] During the model development phase, three prediction models were constructed: a conventional logistic regression model, a random forest model, and an XGBoost algorithm model. A nested cross-validation framework was used to evaluate model performance, in which the outer validation used 5-fold stratified cross-validation to divide the dataset into a training set (80%) and a test set (20%). Each outer 5-fold training set was further optimized for model hyperparameters through an inner 5-fold stratified cross-validation. To address the sample imbalance problem, the synthetic minority oversampling technique (SMOTE) was integrated with the model training process to ensure that only samples were synthesized for the training folds during each inner validation, while the test folds maintained their original distribution. In the outer validation phase, an independent test set without any processing was used for the final performance evaluation. This double validation mechanism effectively controlled the risk of information leakage, while reducing the random bias of data division through repeated sampling (see for details). Figure 1 ).

[0034] Statistically significant differences in various clinical data between the IVF and Rescue ICSI groups were incorporated into the model for analysis. First, significant variables were identified using backward stepwise analysis combined with the chi-square test. These variables were then used in a traditional logistic regression model. Next, variables with an entry probability less than 0.05 were selected using stepwise selection. Secondly, when constructing the random forest model, indicators were selected based on cumulative Gini (Gain) importance. Third, an XGBoost model was used to analyze the Gain value of each variable in predicting fertilization failure, and SHAP analysis was used to select indicators. Hyperparameter optimization was performed for all three models to maximize model performance. Model comparison was performed using the area under the receiver operating characteristic (ROC) curve, confusion matrix, test set prediction accuracy, regression coefficient, and F1 score. All analyses were performed using Python 3.8 software, and data were standardized. A P value < 0.05 was considered statistically significant.

[0035] 2. Results 2.1 Comparison of basic conditions of patients in the two groups (IVF group and Rescue ICSI group) There were statistically significant differences in the female partner's age, male partner's age, female BMI index, infertility type, and infertility duration (P < 0.05), as shown in Table 1.

[0036] Table 1 project IVF group R-ICSI group <![CDATA[χ 2 / U value]]> P-value Number of cases 594 97 Woman's age [years old, M (P25, P75)] 34.0 (30.0, 38.0) 32.0 (28.0, 35.0) 34017.0 <![CDATA[0.004 * ]]> Man’s age [years old, M (P25, P75)] 35.0 (31.0, 40.0) 34.0 (31.0, 37.0) 33667.0 <![CDATA[0.008 * ]]> Female's BMI [kg / m2, M (P25, P75)] 22.2 (20.0, 24.2) 22.9 (20.8, 25.4) 24518.0 <![CDATA[0.019 * ]]> Infertility type (%) 5.802 <![CDATA[0.016 * ]]> Primary infertility 31.31%（186 / 594） 44.33%（43 / 97） Secondary infertility 68.69%（408 / 594） 55.67%（54 / 97） Infertility factors (%) 7.613 0.107 Ovulation factors 17.85%（106 / 594） 23.71%（23 / 97） Male factors 8.59%（51 / 594） 15.46%（15 / 97） Pelvic factors 58.42%（347 / 594） 47.42%（46 / 97） Factors on both sides 4.55%（27 / 594） 4.12%（4 / 97） Other factors 10.61%（63 / 594） 9.28%（9 / 97） HBsAg (%) 0.898 0.343 Positive 6.60%（40 / 594） 4.12%（4 / 97） Negative 93.40%（566 / 594） 95.88%（93 / 97） Infertility years 2（1，3） 2（2，3） 25162.0 <![CDATA[0.037 * ]]> Profession(%) 1.019 0.961 white collar 16.33%（97 / 594） 15.46%（15 / 97） Services and Commerce 4.04%（24 / 594） 3.09%（3 / 97） Blue-collar workers and manufacturing 8.92%（53 / 594） 11.34%（11 / 97） Other and unemployed 5.39%（32 / 594） 4.12%（4 / 97） Professional technology and knowledge 13.13%（78 / 594） 13.40%（13 / 97） Freelancing and Entrepreneurship 52.19%（310 / 594） 52.58%（51 / 97） Educational background (%) 2.615 0.455 Junior high school and below 46.30%（275 / 594） 42.27%（41 / 97） College, undergraduate 35.86%（213 / 594） 32.99%（32 / 97） Technical secondary school, vocational high school, high school 17.00%（101 / 594） 23.71%（23 / 97） Master's degree or above 0.84%（5 / 594） 1.03%（1 / 97） AMH [ng / ml, M (P25, P75)] 2.69 (1.3425, 4.38) 2.9 (1.82, 4.67) 26309.5 0.170 Basal FSH [mIU / ml, M (P25, P75)] 6.87 (5.56, 8.44) 6.59 (5.58, 8.23) 29402.5 0.745 Basal LH [mIU / ml, M (P25, P75)] 4.555 (3.21, 6.79) 4.86 (3.75, 6.38) 26272.0 0.164 Basal E2 [pmol / L, M (P25, P75)] 132.25 (93.1, 199.0) 134.2 (102.3, 191.7) 28003.0 0.658 *P less than 0.05.

[0037] 2.2 Comparison of ovulation stimulation between the two groups (IVF group and Rescue ICSI group) The ovulation stimulation regimens used by the two groups included agonist regimens, antagonist regimens, ovarian stimulation regimens under high progesterone levels in the follicular phase, minimal stimulation regimens, luteal phase ovulation stimulation regimens, and natural cycle regimens. No statistically significant differences were found between the two groups (P>0.05). There were no statistically significant differences in Gn dosage, Gn days, or E2 on the trigger day (P>0.05). See Table 2 for details.

[0038] Table 2 project IVF group R-ICSI group <![CDATA[χ 2 / U value]]> P-value Number of cases 594 97 Ovarian stimulation plan (%) 2.474 0.780 Agonist regimens (super-long regimen, short-acting long regimen, modified long follicular phase regimen, long luteal phase regimen, long follicular phase regimen) 55.56%（330 / 594） 58.76%（57 / 97） Antagonist regimen 23.23%（138 / 594） 22.68%（22 / 97） Ovarian stimulation protocol under high progesterone condition during follicular phase 10.77%（64 / 594） 12.37%（12 / 97） Minimal stimulation program 9.09%（54 / 594） 6.19%（6 / 97） Luteal phase ovulation induction program 1.01%（6 / 594） 0.00%（0 / 97） Natural Cycles Program 0.34%（2 / 594） 0.00%（0 / 97） Gn dosage (U) 2025.0 (1575.0,2584.0) 2175.0 (1775.0,2850.0) 25276.5 0.053 Gn days (days) 11.0 (9.0, 12.0) 11.0 (9.0, 13.0) 27231.5 0.384 E2 on trigger day [pmol / L, M (P25, P75)] 7952.7 (4753.75,15436.3) 8508.0 (5420.0,15231.0) 27643.0 0.523 2.3 Comparison of male semen status between the two groups There were no statistically significant differences in semen volume and NP between the two groups (P>0.05). However, there were statistically significant differences in sperm concentration, number of progressively motile sperm, PR, IM, sperm DFI index, and acrosome enzyme activity (P<0.05), as shown in Table 3.

[0039] Table 3 project IVF group R-ICSI group U-value P-value Semen volume [ml, M (P25, P75)] 2.5 (2.0, 3.0) 2.0 (2.0, 2.5) 31257.0 0.174 <![CDATA[Sperm concentration [10 6 / ml, M(P25, P75)]]> 70.0 (50.0, 97.5) 40.0 (30.0, 70.0) 40360.0 <![CDATA[<0.001 * ]]> <![CDATA[Total number of spermatozoa with forward motility [10 6 , M (P25, P75)]]> 60.0 (35.25, 100.0) 35.0 (18.0, 60.0) 40635.5 <![CDATA[<0.001 * ]]> PR[%,M(P25,P75)] 40.0 (30.0, 50.0) 30.0 (30.0, 40.0) 36311.0 <![CDATA[<0.001 * ]]> NP[%,M(P25,P75)] 20.0 (10.0, 20.0) 20.0 (10.0, 20.0) 30345.5 0.338 IM[%,M(P25,P75)] 40.0 (40.0, 50.0) 50.0 (40.0, 60.0) 20503.5 <![CDATA[<0.001 * ]]> Sperm DFI index [%, M (P25, P75)] 10.5 (6.3, 16.4475) 14.3 (9.49, 20.58) 21946.5 <![CDATA[<0.001 * ]]> <![CDATA[Acrosin activity [uIU / 10 6 spermatozoa, M (P25, P75)]]> 76.1 (53.4, 106.9) 73.3 (40.4, 94.3) 32766.5 <![CDATA[0.030 * ]]> *P less than 0.05.

[0040] 2.4 Comparison of oocyte retrieval and fertilization between the two groups There was no statistically significant difference in the number of mature and immature oocytes between the two groups (P>0.05). Before R-ICSI, there was a statistically significant difference in the fertilization status (normal fertilization, abnormal fertilization, and non-fertilization) between the two groups (P<0.05). After R-ICSI, the normal fertilization rate increased from 17.16% (150 / 874) to 61.78% (540 / 874), the abnormal fertilization rate increased from 2.97% (26 / 874) to 9.95% (87 / 874), and the non-fertilization rate decreased from 79.86% (698 / 874) to 28.26% (247 / 874). The difference between the two groups was still statistically significant (P<0.05). See Table 4 for details.

[0041] Table 4 project IVF group R-ICSI group <![CDATA[χ 2 Value]]> P-value Number of eggs retrieved (%) 1.714 0.190 Number of mature eggs 88.46%（5517 / 6237） 86.97%（874 / 1005） Number of immature eggs 11.54%（720 / 6237） 13.03%（131 / 1005） Fertilization number (%) without Ricsi 1882.518 <![CDATA[<0.001 * ]]> Normal fertilization 67.34%（3715 / 5517） 17.16%（150 / 874） Abnormal fertilization 18.98%（1047 / 5517） 2.97%（26 / 874） Unfertilized 13.68%（755 / 5517） 79.86%（698 / 874） Fertilization number (%) after Ricsi 140.355 <![CDATA[<0.001 * ]]> Normal fertilization 67.34%（3715 / 5517） 61.78%（540 / 874） Abnormal fertilization 18.97%（1047 / 5517） 9.95%（87 / 874） Unfertilized 13.69%（755 / 5517） 28.26%（247 / 874） *P less than 0.05.

[0042] 2.5 Features screened in the model All indicators with P values less than 0.05 in Tables 1, 2, and 3 were summarized, including female age, male age, female BMI, sperm acrosinactivity, sperm concentration, progressive motility (PR), total motile sperm concentration (TPMC), percentage of sperm with normal morphology (IM), DNA fragmentation index (DFI), type of infertility, and years of infertility. After performing a collinearity test (VIF) on these indicators, backward stepwise analysis was first used to select the indicators for inclusion in the logistic regression model: male age, female BMI, total motile sperm concentration (TPMC), and DNA fragmentation index (DFI) (see Figure 2 Then, based on the criterion that the cumulative Gini importance of the indicators is greater than 90%, nine indicators, including female age, male age, sperm acrosome enzyme activity, sperm concentration, progressive motility rate (PR), total motile sperm concentration (TPMC), percentage of sperm with normal morphology (IM), DNA fragmentation index (DFI) and years of infertility, were selected and included in the random forest model (see Figure 3 Finally, based on the cumulative Gain value feature importance evaluation of the XGBoost algorithm model (feature screening threshold greater than 90%), the final 9 core prediction indicators selected include: male age, female BMI index, sperm concentration, forward motility rate (PR), total motile sperm concentration (TPMC), percentage of sperm with normal morphology (IM), DNA fragmentation index (DFI), infertility type and infertility years (see Figure 4 ).

[0043] 2.6 Model Comparison During the development and validation of the model, after nested 5-fold cross-validation, the three models (logistic regression model, random forest model, and XGBoost algorithm model) all showed good predictive ability, with the average AUC of the outer 5 folds being 0.734 (95% CI 0.668-0.802), 0.714 (95% CI 0.667-0.762), and 0.697 (95% CI 0.645-0.748), respectively. Figure 5Due to the limited sample size of the test set, all models demonstrated low precision and recall in the R-ICSI group (Table 5). The logistic regression model performed best in the IVF group (precision 0.91, recall 0.87, F1 score 0.88, and overall accuracy 81%). Although the random forest model's recall (0.88) and F1 score (0.89) in the IVF group were slightly higher than those of the logistic regression model, its precision (0.89) and overall accuracy (81%) were similar to those of the logistic regression model (0.91 and 81%), showing no significant differences. The XGBoost model's performance in the IVF group (precision 0.88, recall 0.85, F1 score 0.87, and accuracy 78%) showed slight fluctuations compared to the random forest model, but overall performance remained comparable. In the RICSI group, the logistic regression model's recall (0.44) remained the highest among the three models, although the absolute value was still relatively low. The random forest model's recall rate (0.34) for this group was slightly higher than that of the XGBoost model (0.32), and its precision rate (0.34) was significantly better than that of the XGBoost model (0.26), suggesting that the XGBoost model had a higher risk of false positives in the R-ICSI group. Overall, the classification performance of all models for the RICSI group was significantly weaker than that for the IVF group (e.g., the F1-score of the logistic regression model dropped from 0.88 to 0.38, and the F1-score of the XGBoost model dropped from 0.87 to 0.29). This suggests that the feature distribution or data volume of this group may pose a significant challenge to model learning. Future efforts will require targeted feature engineering optimization or the exploration of algorithm adaptation strategies more suitable for small sample size / high noise scenarios.

[0044] Table 5 Model Grouping Accuracy (%) Recall rate (%) F1 score (%) Accuracy (%) Logistic regression IVF group 0.91 0.87 0.88 0.81 Logistic regression RICCSI group 0.35 0.44 0.38 0.81 Random Forest IVF group 0.89 0.88 0.89 0.81 Random Forest RICCSI group 0.34 0.34 0.33 0.81 XGBoost IVF group 0.88 0.85 0.87 0.78 XGBoost RICCSI group 0.26 0.32 0.29 0.78 Note: The data in this table are average statistics from five outer cross-validation runs. Precision represents the proportion of samples predicted as positive that are actually positive; recall represents the proportion of samples that are correctly predicted to be positive; the F1 score is the harmonic mean of precision and recall; and accuracy represents the proportion of correct predictions across all samples.

[0045] 2.7 Optimal Model Analysis The logistic regression analysis model with the best performance output index parameters (Table 6) showed that total motile sperm concentration (TPMC) (OR = 0.428; 95% CI: 0.392-0.466; p < 0.001) and male age (OR = 0.642; 95% CI: 0.598-0.689; p < 0.001) were protective factors for fertilization failure, and increased values of these TPMCs were associated with a reduced risk. In contrast, female BMI (OR = 1.268; 95% CI: 1.191-1.351; p = 0.005) and DNA fragmentation index (DFI) (OR = 1.362; 95% CI: 1.274-1.455; p = 0.005) were independent risk factors, and increased values significantly increased the risk of fertilization failure.

[0046] All variables passed the significance test (p < 0.05), supporting their predictive efficacy. 6 ) and DNA fragmentation index (DFI) (0.39%-62.07%). Figure 6 ) shows that the total score is the sum of the scores of each variable. Bootstrap internal validation (1000 replicates) yielded a C-index of 0.722 (95% CI: 0.667-0.773), indicating that the model has above-moderate discriminatory power for the risk of IVF fertilization failure and significantly outperforms random guesswork (C-index = 0.5, p < 0.001). Future external cohort validation will further optimize the model's clinical applicability.

[0047] Depend on Figure 6 IVF Fertilization Failure Risk Assessment Nomogram: The contribution of each variable is calculated using a logistic regression model. The overall score ranges from 100 to 150, with higher overall scores indicating a greater probability of fertilization failure. The bottom of the chart (positive risk axis) displays the corresponding risk probability (0-1) for each overall score. The horizontal dashed line represents the threshold of 0.64, which is the risk threshold. When the overall score exceeds 112 (corresponding to a positive risk of >0.64), the risk probability exceeds the threshold and rises rapidly.

[0048] There are many factors that affect oocyte fertilization, such as the quality of the egg itself (maturity, integrity of genetic material, etc.), the quality of sperm (motility, DNA fragmentation, acrosomal enzymes, etc.), and the in vitro culture environment (culture medium, culture conditions, personnel operation, etc.). To maximize the fertilization rate and embryo quality of in vitro fertilization patients, personalized ovulation stimulation plans are tailored to the patient's follicular development, sperm is optimized, and strict quality control is implemented for the culture medium, culture environment, and personnel operations in the embryo laboratory. However, some patients still experience low IVF fertilization rates. This study found that the incidence of low IVF fertilization rates in the selected database was 14.04% (97 / 691). After Rescue ICSI, the normal fertilization rate increased significantly from 17.16% to 61.78%. However, the overall abnormal fertilization rate and non-fertilization rate still showed statistical differences compared with the IVF group (χ²=140.355, p<0.001). However, the key clinical indicator of normal fertilization rate was only 5.56% between the rescue ICSI group and the IVF group (67.34% vs 61.78%), with a 95% confidence interval of 2.11% to 9.01%. Although this interval does not include 0, the lower limit (2.11%) is close to the clinically acceptable minimum difference (5%). We believe that the rescue ICSI technology can effectively maintain normal fertilization ability comparable to conventional IVF.

[0049] A number of statistically significant factors were identified in the comparison of the patients' basic clinical conditions, ovulation induction conditions, and male semen conditions, including female factors (female age, female BMI index), male factors (male age, sperm concentration, total number of progressively motile sperm, sperm acrosome enzyme activity, progressively motile sperm rate, very slow-motile sperm rate, sperm DNA fragmentation index), and unexplained factors (infertility classification, infertility years). Based on the above analysis, this study constructed three prediction models: logistic regression, random forest, and XGBoost, and used inner and outer five-fold cross-validation to systematically evaluate the model performance. The empirical results show that the prediction performance of logistic regression (average AUC = 0.734 ± 0.049) is significantly better than that of random forest (average AUC = 0.714 ± 0.034) and XGBoost (average AUC = 0.697 ± 0.038). Moreover, according to the classification report results, the precision, recall, F1 score and accuracy of the logistic regression model are all better than the other two models. Based on the linear separability of the data or the low complexity between features, we believe that logistic regression is the best prediction model.

[0050] Regression analysis revealed that male age, female BMI, TPMC, and DFI were the most relevant factors for IVF fertilization, with TPMC showing a stronger predictive power for fertilization. Increasing male age is significantly associated with increased sperm DNA damage and decreased chromatin maturation, negatively impacting sperm motility and thus fertility. However, in this study, both males and females in the Rescue ICSI group were younger than those in the IVF group. This contrasts with related research findings. This may be due to the single-center nature of this study and the limited sample size. It may also be due to the fact that older patients with poor sperm quality were initially offered ICSI due to the anticipated low fertilization rate, leading to a "healthy survivor effect" in older patients (i.e., only older patients in good health undergo IVF). Studies have found that women with a BMI exceeding 25 kg / m² often exhibit decreased ovarian function and ovarian reserve, which can lead to poor ART outcomes. Higher BMI increases the need for ovulation induction medications and is associated with decreased embryo quality, increased miscarriage rates, and lower live birth rates. Optimized TPMC can be used as a predictor of fertilization failure. The sperm DFI index reflects the integrity of sperm DNA. Studies have shown that the sperm DFI index has no effect on fertilization rate, but only reduces the number of transplantable embryos, high-quality embryos, implantation rate, clinical pregnancy rate, delivery rate, and live birth rate after IVF-ET, and increases the risk of early miscarriage in each transfer cycle. Because the sperm DFI index was not assessed on the day of egg retrieval, the sperm DFI index collected in this study was data from the most recent semen analysis (within three months). However, the impact of the sperm DFI index on fertilization was still visible in the logistic regression analysis.

[0051] Previous studies have explored predictive indicators for IVF fertilization failure in a trend of gradual refinement. The results of this study differ from the predictive indicators of current related technologies, which may be due to the following reasons: First, this study innovatively incorporated sperm DFI as a core predictor, which can more sensitively reflect the impact of sperm genetic material integrity on fertilization; second, differences in population characteristics, laboratory quality control standards, and variable screening strategies across different research centers may lead to changes in the weights of key predictive factors; finally, this study adopted a multi-model comparative verification method, optimizing feature selection through random forest and XGBoost algorithms, which may more accurately identify the impact of the interaction between metabolic factors such as BMI and male age on fertilization outcomes.

[0052] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for constructing an in vitro fertilization failure risk prediction model, characterized in that: include: S1. Based on the data of ART treatment cycles, extract the basic information of male and female patients, the hormone status of female patients, the semen status of male patients, and the embryo status; S2. Perform normality and variance analysis on the extracted data to obtain a data set; S3. Screen the indicators with statistical differences in the data set to select the core prediction indicators; S4. Build logistic regression models, random forest models, and XGBoost algorithm models, and use a nested cross-validation framework to evaluate model performance based on the dataset. S5. Based on the selected core prediction indicators, evaluate the ability of the logistic regression model, random forest model, and XGBoost algorithm model to predict the risk of IVF failure. Comprehensively determine the optimal prediction model through the area under the receiver operating characteristic curve, confusion matrix, and the accuracy, regression coefficient, and F1 score of the test set predictions. S6. Output indicator parameters for the model with the best performance to verify its prediction performance.

2. The method for constructing an in vitro fertilization failure risk prediction model according to claim 1, characterized in that: In step S3, the indicators with statistical differences in the data set are screened out, the indicators with P < 0.05 are summarized and tested for collinearity, and then the indicators for inclusion in the logistic regression model are screened out through the results of backward stepwise analysis; then, the indicators for inclusion in the random forest model are screened out based on the cumulative Gini importance of the indicators being greater than 90% as a boundary; and the core prediction indicators are screened out based on the cumulative gain value feature importance evaluation of the XGBoost algorithm model.

3. The method for constructing an in vitro fertilization failure risk prediction model according to claim 2, wherein: Finally, nine core predictive indicators were screened out, namely male age, female BMI index, sperm concentration, progressive motility sperm rate, total progressive motility sperm count, very slow motility sperm rate, sperm DNA fragmentation index, infertility classification, and infertility years.

4. The method for constructing an in vitro fertilization failure risk prediction model according to claim 3, wherein: In step S4, a nested cross-validation framework is used to evaluate the model performance. In the outer validation stage, an independent test set without any processing is used for the final performance evaluation. The outer validation divides the data set into a training set and a test set through 5-fold stratified cross-validation. Each outer training set is further optimized for model hyperparameters through an inner 5-fold stratified cross-validation.

5. The method for constructing an in vitro fertilization failure risk prediction model according to claim 4, characterized in that: In step S5, the indicator parameters of the model with the best performance are output, and all variables pass the significance test; a nomogram is constructed for each variable, and the regression coefficient is mapped into a risk score of 0-150 points. The total score is calculated as the sum of the scores of each variable; and then the prediction efficiency is verified through bootstrap internal verification.

6. The method for constructing an in vitro fertilization failure risk prediction model according to claim 1, wherein: The data of ART treatment cycles were divided into two categories: IVF group and Rescue ICSI group.

7. A model for predicting the risk of in vitro fertilization failure, characterized in that: The method according to any one of claims 1 to 6 is used to construct the optimal prediction model, and the optimal prediction model is a logistic regression model.

8. The in vitro fertilization failure risk prediction model according to claim 7, characterized in that: The output indicator parameters of the logistic regression model are male age, female BMI index, total number of progressively motile sperm, and sperm DNA fragmentation index.

9. A method for predicting the risk of in vitro fertilization failure using the prediction model according to claim 7, characterized in that: include: Four indicator parameters, including the man's age, the woman's BMI index, the total number of forward-motile sperm, and the sperm DNA fragmentation index, were obtained and input into the logistic regression model. The contribution score of each indicator parameter was calculated and summed up through the logistic regression model to obtain a comprehensive risk score. The comprehensive score range is 100-150 points. The higher the score, the greater the probability of fertilization failure.

10. The prediction method according to claim 9, characterized in that The risk threshold is set to 112 points. When the calculated comprehensive risk score is greater than 112 points, it is judged as high risk.

Images