A prediction model for influencing factors of pregnancy risk in assisted reproductive technology and a construction method thereof

Abstract

This invention relates to the field of biomedical technology, specifically to a predictive model and its construction method for pregnancy risk factors in assisted reproductive technology (ART). The method constructs the predictive model based on clinical data of women of childbearing age undergoing ART. Univariate analysis is performed sequentially using Matlab (non-parametric test, chi-square test). Based on the univariate analysis results, a generalized linear logistic regression predictive model and an XGBOOST predictive model are constructed to conduct multivariate analysis of the correlation between pregnancy risk and risk factors, predicting pregnancy risk factors. This model can accurately predict positive factors influencing pregnancy risk, effectively reducing the psychological burden on some patients who conceive through ART, thus making clinical intervention more targeted, more conducive to the rational allocation of clinical resources, and suitable for large-scale application.

Description

Technical Field

[0001] This invention relates to the field of biomedical technology, specifically to a predictive model for pregnancy risk factors in assisted reproductive technology and its construction method. Background Technology

[0002] Assisted reproductive technology (ART) is a commonly used technique that provides more options for conception. Currently, with the increasing trend of older pregnancies and births, the trend of using ART instead of natural conception is also gradually increasing. However, as both men and women age, their body mass index (BMI) also increases, leading to a gradual decrease in ART pregnancy rates and an increased risk of miscarriage. Furthermore, based on the presence of secondary diseases, women can be classified as primary or secondary infertility; similarly, men can be classified as primary or secondary infertility based on whether their infertility has lasted for more than one year. Different types of infertility require different ART interventions, resulting in significant differences in outcomes. Therefore, exploring the influence mechanisms and predictive studies of multidimensional factors on ART outcomes is of great significance.

[0003] Although studies have found that clinical pregnancy rate and live birth rate may be related to factors such as the age of the couple, BMI, and secondary disease status, there is still a lack of effective predictive models that can predict the factors affecting pregnancy risk based on different types of infertility. This would allow for accurate prediction of positive and negative factors affecting pregnancy risk, reduce the psychological burden on patients undergoing ART, and make clinical interventions more targeted and allocate clinical resources more rationally. Summary of the Invention

[0004] The purpose of this invention is to provide a predictive model for pregnancy risk factors in assisted reproductive technology and a method for constructing the model.

[0005] To achieve the above-mentioned objective, this invention provides a method for constructing a predictive model of pregnancy risk factors for assisted reproductive technology, comprising the following steps:

[0006] S1. Obtain clinical data of the sample, perform statistical analysis on the clinical data of the sample to obtain statistical data of the sample; use 60-80% of the sample as training set and the remainder as test set; the statistical analysis includes: continuous data are expressed as mean (standard deviation), and categorical variables are expressed as frequency (percentage);

[0007] S2. Based on the statistical data of the sample, perform univariate analysis on the correlation between each factor and pregnancy risk, and screen out the factors whose correlation with pregnancy risk reaches the preset value as the univariate analysis results (the main factors affecting pregnancy risk); among them, nonparametric tests (Mann-Whitney U test) are used to analyze the correlation of continuous data factors; chi-square test is used to analyze the correlation of categorical variables factors.

[0008] S3. Based on the results of the univariate analysis, construct the generalized linear logistic regression model and the XGBOOST model respectively, and obtain the generalized linear logistic regression prediction model and the XGBOOST prediction model respectively.

[0009] S4. Evaluate the performance of the generalized linear logistic regression prediction model and the XGBOOST prediction model based on the test set. Based on the evaluation results, optimize the generalized linear logistic regression prediction model and the XGBOOST prediction model to obtain prediction models for pregnancy risk factors in assisted reproductive technology (including the optimized generalized linear logistic regression prediction model and the XGBOOST prediction model).

[0010] This invention discloses a method for constructing a predictive model of pregnancy risk factors for assisted reproductive technology (ART). Based on clinical data of women of childbearing age undergoing ART, univariate analysis is performed sequentially using Matlab (non-parametric test and chi-square test). Based on the results of the univariate analysis, a generalized linear logistic regression prediction model and an XGBOOST prediction model are constructed to conduct multivariate analysis of the correlation between pregnancy risk and risk factors, predicting pregnancy risk factors. This model can accurately predict positive factors affecting pregnancy risk, effectively reducing the psychological burden on some patients who conceive through ART, thereby making clinical intervention more targeted, more conducive to the rational allocation of clinical resources, and suitable for large-scale promotion and application.

[0011] In step S1, preferably, the clinical data includes: age, BMI, education level, egg retrieval cycle (ORC), infertility duration (DI), female endocrine anti-Müllerian hormone (AMH), female endocrine follicle-stimulating hormone (FSH), female endocrine luteinizing hormone (LH), female endocrine prolactin (PRL), female BMI, number of follicles (AFC), male endocrine luteinizing hormone (LH), semen analysis results (SM), semen analysis A (SA), first progressive motility sperm rate (PR), total number of oocytes (TOC), number of usable embryos (UE), blastocyst formation rate (BFR) of this cycle, number of high-quality blastocysts (GB), number of excellent embryos (GQE), high-quality embryo rate (GQER), total number of fertilized oocytes (TFO), total number of mature oocytes (TMO), total number of normally fertilized oocytes (TNFO), total number of cleaved oocytes (TCO), number of high-quality embryos transferred (TGQE), gestational age (GW), implanted embryos (IE), and female endocrine index. The parameters include estradiol (E2), female endocrine luteinizing hormone (LH), female endocrine progesterone (P), female endocrine testosterone (T), male endocrine follicle-stimulating hormone (FSH), male endocrine testosterone (T), as well as routine semen volume (SV), routine semen analysis B (SB), routine semen concentration (SC), male BMI, total fertilization rate (TFR), total normal fertilization rate (TNFR), total number of embryos transferred (TTE), total normal cleavage rate (TCNCR), total cleavage rate (TCR), number of embryos transferred (TQE), gestational age (GW), and implanted embryos (IE).

[0012] Preferably, the exclusion criteria for the samples include: samples for which male and female infertility cannot be determined, samples with abnormal values ​​exceeding the human physiological range (such as: weight, height, BMI, AFC, number of usable embryos, rate of high-quality embryos, blastocyst formation rate, LH-positive women, P-positive women, routine semen motility, routine semen volume, routine semen concentration, and first PR, etc., exceeding the human physiological range), or samples with missing clinical data.

[0013] Preferably, R software version 4.4.1 is used to perform statistical analysis on the clinical data of the sample.

[0014] Preferably, the Student's test is used for continuous data; for categorical variables expressed as frequencies (percentages), the χ² test or Fisher's exact test is used to compare differences between groups; error data are excluded by test.

[0015] Preferably, in the samples, female samples are divided into primary infertility and secondary infertility according to the infertility status; male samples are divided into primary infertility and secondary infertility according to the infertility status (uncontraception and no pregnancy for more than one year).

[0016] Preferably, based on the type of infertility, the samples are divided into: primary infertility group, secondary infertility group, primary infertility group, secondary infertility group, and secondary infertility group; then, for each group, a predictive model for pregnancy risk factors for assisted reproductive technology is constructed using the above method.

[0017] Preferably, the number of samples in each group is not less than 400.

[0018] In step S2, the preset value is: H=1 and P<0.05.

[0019] In step S4, preferably, the evaluation method includes: (1) evaluating the odds ratio (OR), confidence interval, and p-value of the generalized linear logistic regression prediction model, and evaluating the discrimination, calibration, clinical effectiveness, and net benefit of the prediction model through the receiver operating characteristic curve, locally weighted calibration curve, and decision curve; (2) evaluating the XGBOOST prediction model through the discrimination, calibration, clinical effectiveness, and net benefit of the prediction model through the receiver operating characteristic curve, locally weighted calibration curve, and decision curve.

[0020] Preferably, the performance of the generalized linear logistic regression prediction model and the XGBOOST prediction model is evaluated using ROC curves.

[0021] Preferably, the criteria for evaluating the predictive model are: the AUC of the ROC curve of the generalized linear logistic regression predictive model is greater than or equal to 0.6; the AUC of the ROC curve of the XGBOOST predictive model is greater than or equal to 0.6.

[0022] Preferably, the optimization method for the prediction model includes, but is not limited to, increasing the number of samples.

[0023] To achieve the above-mentioned objectives, the present invention further provides a predictive model for pregnancy risk factors in assisted reproductive technology, which is constructed using the above-described method.

[0024] To further achieve the above-mentioned objectives, the present invention also provides an application of a predictive model for pregnancy risk factors in assisted reproductive technology, used to predict factors affecting pregnancy risk in assisted reproductive technology.

[0025] Preferably, the application method includes performing SHAP analysis on the XGBOOST prediction model. SHAP (SHapley Additive exPlanations) is used to explain the feature importance of the XGBOOST model and the impact of each feature on the prediction results. Features are sorted from top to bottom according to the average absolute value of their impact on the model output. The horizontal axis represents the impact of a feature on the model prediction; a positive value indicates that the feature pushes the final predicted value higher, and a negative value indicates that the feature pushes the final predicted value lower.

[0026] A predictive device for pregnancy risk factors in assisted reproductive technology, the predictive device being capable of running the above-mentioned predictive model; the predictive device can be used to predict factors affecting pregnancy risk in assisted reproductive technology.

[0027] Beneficial effects of the invention

[0028] 1. The present invention provides a method for constructing a predictive model for pregnancy risk influencing factors in assisted reproductive technology. This method uses Matlab (non-parametric test, chi-square test) to perform univariate analysis in sequence. Based on the results of the univariate analysis, a generalized linear logistic regression prediction model and an XGBOOST prediction model are constructed to perform multivariate analysis of the correlation between pregnancy risk and the influencing factors, thereby predicting pregnancy risk influencing factors.

[0029] 2. The predictive model for pregnancy risk factors in this invention can accurately predict positive factors affecting pregnancy risk, effectively reducing the psychological burden on some patients who conceive with ART, thus making clinical intervention more targeted and more conducive to the rational allocation of clinical resources, and suitable for large-scale promotion and application. Attached Figure Description

[0030] Figure 1 This is a bar chart showing the results of the U-test and chi-square test for continuous variables in Embodiment 1 of the present invention.

[0031] Figure 2 This is a heatmap showing the correlation between different variables and the significant correlation between them and pregnancy outcomes in Embodiment 1 of the present invention.

[0032] Figure 3 The results are multifactor regression analysis of the four infertility groups in Example 1 of this invention.

[0033] Figure 4The results of using XGBOOST, Logistic Regression, and Random Forest to evaluate the pregnancy outcome prediction of the primary infertility and secondary infertility group in Example 1 of this invention.

[0034] Figure 5 This is a graph showing the prediction results of the prediction model for pregnancy risk influencing factors in Embodiment 2 of the present invention. Detailed Implementation

[0035] The technical solutions in some embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments provided in this disclosure are within the scope of protection of this disclosure.

[0036] Example 1

[0037] This example uses couples who received treatment at the Third Affiliated Hospital of Guangzhou Medical University from January 2012 to December 2024, with an initial sample size of 156,380.

[0038] Given the presence of outliers, missing values, and samples inconsistent with the research objectives in the current data, the data screening and cleaning process is as follows:

[0039] (1) Based on the pregnancy outcome data of miscarriage or delivery, an initial sample of 47,842 cases was selected.

[0040] (2) To ensure the accuracy of the research results, 1229 samples that could not determine male and female infertility were deleted; samples with abnormal values ​​exceeding the human physiological range, such as female weight, height and BMI (166 samples), male height (8799 samples) and weight (36 samples), AFC (1 sample), number of available embryos (24825 samples), high-quality embryo rate (152 samples), blastocyst formation rate since this week (9 samples), endocrine LH_female (2 samples), endocrine P_female (1 sample), routine semen viability (3 samples), routine semen volume (2 samples), routine semen concentration (6 samples), and first PR (1 sample), were also deleted.

[0041] (3) Remove missing values ​​for categorical variables: male education level (16) and female education level (2).

[0042] (4) For variables that have both numerical and categorical aspects: Semen routine liquefaction time 1.x, the categorical aspect is converted into a numerical value according to the minimum value in the range, such as ">60" is converted to "60" and "<30" is converted to "30".

[0043] (5) This study included 12,586 valid samples after screening and cleaning as the research subjects. To ensure the integrity of the data and the robustness of the analysis, the median imputation method was used to process the small number of missing data.

[0044] The type of infertility in both men and women is a significant factor affecting pregnancy outcomes. To facilitate the differentiation of different types of infertility in clinical practice, this example 1 divides the included samples into four groups: primary infertility (5,966 cases), secondary infertility (4,657 cases), primary infertility with secondary infertility (414 cases), and secondary infertility with primary infertility (1,579 cases). This grouping method better reflects different types of infertility symptoms, facilitating subsequent targeted analysis and treatment plan development.

[0045] Based on the above sample, a method for constructing a predictive model of pregnancy risk factors for assisted reproductive technology includes the following steps:

[0046] S1. Obtain clinical data from four groups of samples, and perform statistical analysis on the clinical data of the four groups of samples to obtain statistical data for each group; use 60-80% of the sample size in each group as the training set and the remainder as the test set; the statistical analysis includes: continuous data are expressed as mean (standard deviation), and categorical variables are expressed as frequency (percentage).

[0047] Clinical data include: age, BMI, education level, egg retrieval cycle (ORC), duration of infertility (DI), female endocrine anti-Müllerian hormone (AMH), female endocrine follicle-stimulating hormone (FSH), female endocrine luteinizing hormone (LH), female endocrine prolactin (PRL), female BMI, number of follicles (AFC), male endocrine luteinizing hormone (LH), semen analysis results (SM), semen analysis A (SA), first progressive motility sperm rate (PR), total number of oocytes (TOC), number of usable embryos (UE), blastocyst formation rate (BFR) in this cycle, number of high-quality blastocysts (GB), number of good-quality embryos (GQE), high-quality embryo rate (GQER), total number of fertilized oocytes (TFO), total number of mature oocytes (TMO), total number of normally fertilized oocytes (TNFO), total number of cleaved oocytes (TCO), number of high-quality embryos transferred (TGQE), gestational age (GW), number of implanted embryos (IE), and female endocrine indicators such as estradiol (E). 2) Female endocrine luteinizing hormone (LH), female endocrine progesterone (P), female endocrine testosterone (T), male endocrine follicle-stimulating hormone (FSH), male endocrine testosterone (T), and semen analysis (SV), semen analysis (SB), semen concentration (SC), male BMI, total fertilization rate (TFR), total normal fertilization rate (TNFR), total number of embryos transferred (TTE), total normal cleavage rate (TCNCR), total cleavage rate (TCR), number of embryos transferred (TQE), gestational age (GW) and implanted embryos (IE), female endocrine estradiol (E2), female endocrine testosterone (T), male endocrine follicle-stimulating hormone (FSH), male endocrine testosterone (T), semen analysis (SV), semen analysis (SB), total fertilization rate (TFR), total normal fertilization rate (TNFR), total number of embryos transferred (TTE), total normal cleavage rate (TCNCR) and number of embryos transferred (TQE).

[0048] The clinical data of the samples were statistically analyzed using R software version 4.4.1.

[0049] For continuous data, use the Student's test; for categorical variables expressed as frequencies (percentages), use the χ² test or Fisher's exact test to compare differences between groups; exclude error data using tests.

[0050] S2. Based on the statistical data of each group of samples, perform univariate analysis on the correlation of each factor with pregnancy risk, and screen out the factors whose correlation with pregnancy risk reaches the preset value (preset value: H=1 and P<0.05), as the univariate analysis results (main factors affecting pregnancy risk) (e.g. Figure 1 , Figure 2(As shown in the figure). For continuous data, the factors were analyzed using a non-parametric test (Mann-Whitney U test); for categorical variables, the factors were analyzed using a chi-square test.

[0051] S3. Based on the univariate analysis results for each group, construct a generalized linear logistic regression model and an XGBOOST model respectively to obtain the generalized linear logistic regression prediction model and the XGBOOST prediction model (e.g., Figure 3 (as shown)

[0052] S4. Evaluate the performance of the generalized linear logistic regression prediction model and the XGBOOST prediction model based on each test set. Based on the evaluation results, optimize the generalized linear logistic regression prediction model and the XGBOOST prediction model to obtain prediction models for pregnancy risk factors in assisted reproductive technology (including the optimized generalized linear logistic regression prediction model and the XGBOOST prediction model).

[0053] The assessment methods included: using XGBOOST, Logistic Regression, and Random Forest to assess the predictive value of pregnancy outcomes in the primary infertility and secondary infertility groups (e.g., Figure 4 As shown in the figure); the evaluation results show that Logistic Regression (AUC=0.74) has relatively good prediction results, while XGBOOST (AUC=0.68) and Random Forest (AUC=0.69) have similar prediction results.

[0054] Example 2

[0055] A predictive model for pregnancy risk factors in assisted reproductive technology is provided, which is constructed using the method described in Example 1.

[0056] An application of a predictive model for pregnancy risk factors in assisted reproductive technology, used to predict factors affecting pregnancy risk in assisted reproductive technology.

[0057] Application method: Perform SHAP analysis on the XGBOOST prediction model (e.g., Figure 5 (As shown). SHAP (SHapleyAdditive exPlanations) is used to explain the feature importance of the xgboost model and the impact of each feature on the prediction results. Features are sorted from top to bottom according to the average absolute value of their impact on the model output. The horizontal axis represents the impact of a feature on the model prediction. A positive value indicates that the feature pushes the final prediction value higher, while a negative value indicates that the feature pushes the final prediction value lower.

[0058] Model results indicate that P_Female, DI, and female age class are the most important characteristics for primary and secondary infertility. Meanwhile, female endocrine P, DI, male education level (secondary), implanted embryos, and male age class (25-34 years) have a higher predictive probability of impacting pregnancy outcome and delivery. However, the duration of infertility, female age class (35-44), male age over 35, and male education level (higher education) have a lower predictive probability of impacting pregnancy outcome and delivery.

[0059] A predictive device for pregnancy risk factors in assisted reproductive technology, the predictive device being able to run the predictive model constructed in Example 1; the predictive device can be used to predict factors affecting pregnancy risk in assisted reproductive technology.

[0060] In the description of this specification, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for constructing a prediction model of pregnancy risk influencing factors for assisted reproductive technology, characterized in that, Includes the following steps: S1. Obtain clinical data of the sample; perform statistical analysis on the clinical data of the sample to obtain statistical data of the sample; 60-80% of the sample size in each group is used as the training set, and the remainder is used as the test set; the statistical analysis includes: continuous data are expressed as mean (standard deviation), and categorical variables are expressed as frequency (percentage); S2. Based on the statistical data of the sample, perform univariate analysis on the correlation between each factor and pregnancy risk, and screen out the factors whose correlation with pregnancy risk reaches the preset value as the univariate analysis results (the main factors affecting pregnancy risk); among them, nonparametric tests (Mann-Whitney U test) are used to analyze the correlation of continuous data factors; chi-square test is used to analyze the correlation of categorical variables factors. S3. Based on the results of the univariate analysis, construct the generalized linear logistic regression model and the XGBOOST model respectively, and obtain the generalized linear logistic regression prediction model and the XGBOOST prediction model respectively. S4. The performance of the generalized linear logistic regression prediction model and the XGBOOST prediction model are evaluated based on the test set. Based on the evaluation results, the generalized linear logistic regression prediction model and the XGBOOST prediction model are optimized to obtain a prediction model for pregnancy risk factors in assisted reproductive technology.

2. The construction method according to claim 1, characterized in that, In S1, the clinical data includes: age, BMI, education level, egg retrieval cycle, duration of infertility, female endocrine anti-Müllerian hormone, female endocrine follicle-stimulating hormone, female endocrine luteinizing hormone, female endocrine prolactin, female BMI, number of follicles, male endocrine luteinizing hormone, semen analysis motility, semen analysis A, first progressive motility sperm rate, total number of oocytes, number of usable embryos, blastocyst formation rate in this cycle, number of high-quality blastocysts, number of excellent embryos, high-quality embryo rate, total number of fertilized oocytes, total number of mature oocytes, total number of normally fertilized oocytes, total number of cleaved oocytes, number of high-quality embryos transferred, gestational age, number of implanted embryos, and female endocrine... Indicators such as estradiol, female endocrine luteinizing hormone, female endocrine progesterone, female endocrine testosterone, male endocrine follicle-stimulating hormone, male endocrine testosterone, as well as semen volume, semen analysis (B), semen concentration, male BMI, total fertilization rate, total normal fertilization rate, total number of embryos transferred, total normal cleavage rate, total cleavage rate, number of embryos transferred, gestational age and implanted embryos, female endocrine estradiol, female endocrine testosterone, male endocrine follicle-stimulating hormone, male endocrine testosterone, semen volume, semen analysis (B), total fertilization rate, total normal fertilization rate, total number of embryos transferred, total normal cleavage rate and number of embryos transferred.

3. The construction method according to claim 1, characterized in that, In step S1, the samples are divided into primary infertility and secondary infertility based on the infertility status of female samples; and primary infertility and secondary infertility based on the infertility status of male samples. Based on the type of infertility, the samples are divided into: primary infertility group, secondary infertility group, primary infertility group, and secondary infertility group.

4. The construction method according to claim 1, characterized in that, In step S2, the preset value is: H=1 and P<0.

05.

5. The construction method according to claim 1, characterized in that, In S4, the evaluation method includes: (1) calculating the odds ratio (OR), confidence interval, and P-value using the generalized linear logistic regression model, and evaluating the discrimination, calibration, clinical effectiveness, and net benefit of the prediction model using the receiver operating characteristic curve, locally weighted calibration curve, and decision curve; (2) evaluating the XGBOOST model using the discrimination, calibration, clinical effectiveness, and net benefit of the prediction model using the receiver operating characteristic curve, locally weighted calibration curve, and decision curve.

6. The construction method according to claim 1, characterized in that, In S4, ROC curves are used to evaluate the performance of the generalized linear logistic regression prediction model and the XGBOOST prediction model.

7. The construction method according to claim 6, characterized in that, The criteria for the prediction model to meet the requirements are: the AUC of the generalized linear logistic regression model is greater than or equal to 0.6; the AUC of the XGBOOST model is greater than or equal to 0.

6.

8. A predictive model for pregnancy risk factors in assisted reproductive technology, characterized in that, The prediction model is constructed using the construction method described in any one of claims 1-7.

9. The application of the predictive model for pregnancy risk factors in assisted reproductive technology as described in claim 8, characterized in that, Factors used to predict pregnancy risks in assisted reproductive technologies.

10. The application according to claim 9, characterized in that, Application methods include: performing SHAP analysis on the XGBOOST model.

Images