Method and system for predicting cumulative live productivity of multiple complete cycles based on in-vitro fertilization

Abstract

The invention discloses an in-vitro fertilization-based method and system for predicting the cumulative live productivity of multiple complete cycles, and the method comprises the steps: obtaining clinical data, embryo information and pregnancy follow-up visit data of a patient, and screening and preprocessing a sample in combination with a collection and discharge standard; feature variables are extracted before treatment, after treatment and at the early stage of pregnancy, an optimal set is determined, and key factors influencing live birth at all stages are accurately captured; after the nonlinear variables are processed by the RCS, a pre-treatment / post-treatment COX regression model and a Logistic regression model are adaptively constructed, and the prediction precision is improved by fitting data characteristics; and meanwhile, by evaluating the iterative optimization model, the variable weight is visually displayed by using a Nomogram graph after the model is qualified. The method can accurately predict the multi-cycle cumulative live productivity, assists a doctor in formulating a personalized scheme, helps a patient to reasonably plan treatment, reduces the time and economic burden of blind treatment, and has clinical practical value and benefits of the patient.

Description

Technical Field

[0001] This invention belongs to the field of assisted reproductive clinical treatment technology, and relates to a method and system for predicting the cumulative live birth rate based on multiple complete cycles of in vitro fertilization. Background Technology

[0002] In vitro fertilization-embryo transfer (IVF-ET, or test-tube baby for short) is a recommended solution for infertility. With the maturity of frozen-thawed embryo transfer (FET) technology, the cumulative live birth rate has become a core indicator for evaluating the effectiveness and safety of assisted reproductive technologies. Currently, various IVF live birth rate prediction models have been developed both domestically and internationally, such as the McLernon model from the UK, the SART model from the US, the POPI-Plus model from Canada, and related domestic models. These models mostly employ algorithms such as Logistic Regression and machine learning, and some have validated their performance through calibration curves and decision curve analysis (DCA). However, significant technical problems remain: First, they fail to dynamically assess the cumulative live birth probability across all stages—before, after, and in early pregnancy—based on the actual IVF workflow, or they focus only on a single stage, lacking comprehensiveness. Second, most Logistic Regression models cannot handle the time-series data and non-linear relationships between variables and live birth rates involved in cumulative live births, and a few high-order machine learning models lack user-friendly interpretation mechanisms, making them difficult to use for patient consultation. Third, some models only predict the outcome of a single stimulation cycle, failing to include multiple complete cycles (including fresh and frozen embryo transfer) and the number of complete cycles as discrete-time variables, and also failing to adequately handle censored data from incomplete cycles or interrupted treatment, thus failing to accurately reflect the dynamic changes in cumulative live birth rates across multiple cycles. Furthermore, the current CLBR (Clear Flow Ratio Prediction) model... The results are mostly general averages (stratified only by female age), making it difficult for clinical medical staff to provide patients with individualized estimates of the probability of live birth. Patients feel that they do not have comprehensive and understandable decision-making information, which is not conducive to patients establishing reasonable treatment expectations. Summary of the Invention

[0003] The purpose of this invention is to address the problems in existing IVF live birth rate prediction models, such as the inability to dynamically assess the cumulative live birth probability at all stages before, after, and in early pregnancy based on the actual IVF workflow, the difficulty of handling time series data and nonlinear relationships between variables, and the lack of patient-friendly interpretation mechanisms in high-order models. This invention provides a method and system for predicting the cumulative live birth rate based on multiple complete IVF cycles.

[0004] To achieve the above objectives, the present invention employs the following technical solution: Compared with the prior art, the present invention has the following beneficial effects: This invention acquires patient clinical data, embryo information, and pregnancy follow-up data. It then uses inclusion and exclusion criteria to screen samples and preprocess them, eliminating data errors, encoding categorical variables, and imputing missing values ​​to ensure data quality and lay a reliable foundation for prediction. Furthermore, it extracts characteristic variables from pre-treatment, post-treatment, and early pregnancy stages and determines the optimal set to accurately capture key factors affecting live birth at each stage. After processing nonlinear variables using RCS, it adapts and constructs pre- / post-treatment Cox regression and Logistic regression models to better match data characteristics and improve prediction accuracy. Simultaneously, it iteratively optimizes the model through evaluation, and once successful, uses a Nomogram to visually display variable weights. This method can accurately predict the cumulative live birth rate over multiple cycles, assisting doctors in developing personalized treatment plans, helping patients rationally plan their treatment, reducing the time and financial burden of blind treatment, and possessing both clinical practical value and patient benefit. Attached Figure Description

[0005] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0006] Figure 1 This is a flowchart illustrating the method for predicting the cumulative live birth rate across multiple complete cycles of in vitro fertilization according to the present invention. Figure 2 This is a schematic diagram of the structure of the prediction system for the cumulative live birth rate of multiple complete cycles based on in vitro fertilization according to the present invention. Figure 3 A schematic diagram of the Nomogram for predicting cumulative live birth rate over four cycles in the pre-treatment model; Figure 4 A schematic diagram of the calibration curve and clinical decision curve (DCA) for the first cycle of the pre-treatment model; Figure 5 A schematic diagram of the Nomogram for predicting the cumulative live birth rate over four cycles after treatment; Figure 6 A schematic diagram of the calibration curve and clinical decision curve (DCA) for the first cycle after treatment in the model; Figure 7 A schematic diagram of a Nomogram for predicting the probability of live birth in early pregnancy. Figure 8 This is a schematic diagram of the calibration curve and clinical decision curve (DCA) for an early pregnancy model. Detailed Implementation

[0007] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0008] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.

[0009] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0010] In the description of the embodiments of the present invention, it should be noted that if terms such as "upper," "lower," "horizontal," or "inner" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product of the invention is in use, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as a limitation of the present invention. Furthermore, terms such as "first" and "second" are only used to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0011] Furthermore, the use of the term "horizontal" does not imply that the component must be absolutely horizontal, but rather that it can be slightly tilted. For example, "horizontal" simply means that its direction is more horizontal than "vertical," and does not mean that the structure must be completely horizontal, but can be slightly tilted.

[0012] In the description of the embodiments of the present invention, it should also be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in the present invention according to the specific circumstances.

[0013] The present invention will now be described in further detail with reference to the accompanying drawings: See Figure 1This invention discloses a method for predicting the cumulative live birth rate across multiple complete in-vitro fertilization cycles, comprising: S101: Obtain clinical data, embryo information and pregnancy follow-up data of IVF patients, establish inclusion and exclusion criteria to screen patient data, and preprocess the screened data. The inclusion criteria are patients who have undergone ovarian stimulation and corresponding frozen embryo transfer data for the first time, excluding patients who did not retrieve eggs, those involved in sperm donation, egg donation, and those who underwent preimplantation genetic testing (PGT). The preprocessing of the filtered data specifically includes: The filtered data underwent cleaning to eliminate errors, redundancy, and inconsistencies. Categorical variables were coded to ensure data consistency and analyzability. Decision tree imputation was then used to imput missing variables, including anti-Müllerian hormone (AMH), progressive sperm motility (PR), and DNA fragmentation index (DFI). Missing AMH, PR, and DFI values ​​were identified by reviewing patients' original medical records and added to the modeling dataset. Imputation was performed using decision tree imputation only when the AMH missing rate was less than 5%, the PR missing rate was less than 8.6%, and the DFI missing rate was less than 22.6%, to avoid data distortion caused by excessive imputation rates.

[0014] S102, Based on the preprocessed data, feature variables of the three stages of IVF treatment (before treatment, after treatment, and early pregnancy) are extracted, and the optimal set of predictive variables for each stage is determined. The specific extraction of characteristic variables for the three stages of IVF treatment—before treatment, after treatment, and early pregnancy—is as follows: Characteristic variables in the pre-treatment phase include: female age, body mass index, antral follicle count (AFC), antral follicle-stimulating hormone (bFSH), AMH, male age, percentage of progressively motile sperm (PR), DNA fragmentation index (DFI), duration of infertility, pregnancy history, and cause of infertility. The characteristic variables in the post-treatment phase include: in addition to the variables before treatment, the number of oocytes retrieved, the fertilization method IVF / ICSI, the number of embryos obtained, the number of usable embryos, the number of frozen embryos, the type of transfer (ET / FET / no transfer), and the stage of the transferred embryos (day 3 cleavage embryo / day 5-6 blastocyst). The characteristic variables for the early stage of pregnancy include: in addition to the characteristic variables for the post-treatment stage, the initial serum β-hCG test value and the number of days for which β-hCG is equivalent to the test value are added. The number of days for which β-hCG is equivalent to the test value is the sum of the number of days since the embryo was transferred and the number of days since the test was performed after the transfer.

[0015] The determination of the optimal set of predictor variables for each stage specifically involves: The optimal predictor variable set for each stage was screened using the whole subset regression method. The optimal variable set for the pre-treatment and post-treatment stages was determined by the minimum Bayesian information criterion (BIC), and the optimal variable set for the early pregnancy stage was determined by the minimum Mallows Cp value. The variables that have a non-linear relationship with the probability of live birth include: female age, antral follicle count (AFC), number of available embryos, and initial β-hCG value.

[0016] S103, Based on the optimal predictor variable set for each stage, the variables that have a non-linear relationship with the live birth probability are subjected to restricted cubic spline RCS processing to obtain the optimal predictor variable set after RCS processing for each stage. For variables that have a nonlinear relationship with the probability of live birth, restricted cubic spline (RCS) processing is performed. Specifically, three nodes are set at the 10%, 50%, and 90% quantiles. Based on the three nodes, the data is divided into four intervals by restricted cubic spline (RCS) processing to obtain the optimal set of predictive variables after RCS processing at each stage, thus avoiding fitting bias caused by nodes being concentrated in a certain area.

[0017] S104, using the number of complete cycles as a discrete-time variable, and taking the optimal predictive variable set after RCS processing for each stage as input, constructs a Cox regression model for the pre-treatment and post-treatment stages, and a Logistic regression model for the early pregnancy stage; the Cox regression model and the Logistic regression model together constitute a cumulative live birth rate prediction model for multiple complete IVF cycles. Cox regression model in the pre-treatment phase: ; in, middle, Indicates the number of live births or discrete cycles. This indicates the status of a live birth, where 1 represents a live birth and 0 represents a non-live birth. The result is a 3-node restrictive cubic spline processing of the woman's age. The results of 3-node restrictive cubic spline processing of the baseline antral follicle count (AFC); Cox regression model in the post-treatment phase: ; in, The number of embryos available for the patient in the current cycle. The results of 3-node restricted cubic spline processing are shown for the number of available embryos. This indicates the type of embryo transfer, including fresh embryo transfer (ET=0) and frozen-thawed embryo transfer (FET=1). Logistic regression model in early pregnancy: ; in, For live birth probability, For the period and number of embryos to be transferred, For the equivalent detection days of β-hCG, The value is the β-hCG detection value. , , , , and These are the model parameters obtained through the maximum likelihood estimation method; The numbers represent the embryo stage and number of embryos transferred: 0 for single cleavage embryo transfer (SCT), 1 for diploid cleavage embryo transfer (DCT), 3 for single blastocyst transfer (SBT), and 4 for other types, including double blastocyst transfer (DBT) or SCT+SBT. It also includes prediction equations for the probability of non-live birth in each cycle. The probability of non-live birth is calculated based on an integral point, which is derived from the woman's age, BMI, and AFC using a polynomial formula. The cumulative probability of live birth is 1 minus the probability of non-live birth in the corresponding cycle. Specifically: The non-live birth probability prediction equation is applicable to both the pre-treatment and post-treatment Cox regression models. The equation uses integration points as the core intermediate variable. After calculating the integration points based on individual patient characteristics, the equation is substituted into the corresponding cycle equation to obtain the non-live birth probability. The final cumulative live birth probability = 1 - the corresponding cycle non-live birth probability. The early pregnancy model is a Logistic regression, which directly outputs the live birth probability and does not have a separate non-live birth probability equation. The non-live birth probability prediction equation in the Cox regression model during the pre-treatment phase includes: ; in, ; First cycle non-live birth probability prediction equation ; The second cycle non-live birth probability prediction equation is: ; The third cycle non-live birth probability prediction equation ; The fourth cycle non-live birth probability prediction equation ; The non-live birth probability prediction equation in the Cox regression model during the post-treatment phase includes: ; in, ; ; ; ; 0018. The prediction equation for the probability of non-live births in the first cycle is: ; The second cycle non-live birth probability prediction equation is: ; The third cycle non-live birth probability prediction equation is: ; The prediction equation for the non-live birth probability in the fourth cycle is as follows: ;

[0018] S105. The constructed cumulative live birth rate prediction model for multiple complete IVF cycles is evaluated. If the evaluation is qualified, it can be used in clinical practice. If the evaluation is unqualified, S102 to S104 are repeated until the constructed cumulative live birth rate prediction model for multiple complete IVF cycles is qualified. A Nomogram is constructed to visually show the contribution weight of each variable to live birth.

[0019] The evaluation of the constructed IVF cumulative live birth rate prediction model across multiple complete cycles is specifically as follows: The feature variables extracted from the three stages of IVF treatment—before treatment, after treatment, and early pregnancy—are input into the corresponding Cox regression models for the pre-treatment and post-treatment stages, and the Logistic regression model for the early pregnancy stage, respectively, to output the cumulative live birth probability. The probability of no live birth in the first N cycles = probability of no live birth in the first cycle × probability of no live birth in the second cycle × … × probability of no live birth in the Nth cycle; The cumulative live birth probability of the first N cycles = 1 - the probability of no live birth in the first N cycles, that is, the probability of achieving a live birth at least once in the first N cycles; Based on the cumulative live birth probability, the AUC value, C-index, Brie score, calibration curve, and decision curve are used as model evaluation indicators. The model is validated using 10-fold cross-validation and Bootstrap sampling. If the cumulative live birth rate prediction model for multiple complete IVF cycles does not exceed the set threshold, it is considered that the current model meets the practical requirements. The Brie score ranges from 0 to 0.25. The closer the Brie score is to 0, the better the model calibration.

[0020] The constructed Nomogram visually displays the contribution weight of each variable to live birth, specifically as follows: By standardizing the regression coefficients of the Cox regression and Logistic regression models, the score ranges of each predictor variable are displayed on a number axis in the form of risk scores. Patients determine the corresponding scores for each variable based on their actual indicators and accumulate them to obtain a total score. The total score directly reads the non-live birth probability or live birth probability for the corresponding cycle, thereby obtaining the cumulative live birth rate. The risk score is obtained by converting different values ​​of each predictor variable in the model into a score on a uniform scale according to their weighted impact on the non-live birth risk or live birth probability. The greater the contribution of a variable value to the non-live birth risk, the higher the corresponding risk score; conversely, the greater the contribution to the live birth probability, the lower the corresponding risk score.

[0021] See Figure 2 This invention discloses a prediction system for the cumulative live birth rate across multiple complete in vitro fertilization cycles, comprising: The preprocessing module acquires clinical data, embryo information, and pregnancy follow-up data of IVF patients, establishes inclusion and exclusion criteria to screen patient data, and preprocesses the screened data. The determination module extracts characteristic variables from three stages of IVF treatment—before treatment, after treatment, and early pregnancy—based on the preprocessed data, and determines the optimal set of predictive variables for each stage. The processing module performs restricted cubic spline (RCS) processing on variables that have a non-linear relationship with the live birth probability based on the optimal predictor variable set for each stage, thereby obtaining the optimal predictor variable set after RCS processing for each stage. The acquisition module uses the number of complete cycles as a discrete-time variable, takes the optimal set of predictive variables after RCS processing for each stage as input, constructs a Cox regression model for the pre-treatment and post-treatment stages, and a Logistic regression model for the early pregnancy stage; the Cox regression model and the Logistic regression model together constitute a cumulative live birth rate prediction model for multiple complete IVF cycles. The module evaluates the cumulative live birth rate prediction model for multiple complete IVF cycles. If the evaluation is satisfactory, the model is put into clinical application and a Nomogram is constructed to visually display the contribution weight of each variable to live birth. Example

[0022] This invention discloses a method for predicting the cumulative live birth rate across multiple complete in-vitro fertilization cycles, comprising: Patient clinical data, embryo information, and pregnancy follow-up data were extracted using the HIS, LIS, and ART specialist management systems. Inclusion criteria were established, and information was collected from 1082 patients who underwent their first ovarian stimulation at a specific hospital between January 2019 and December 2022. Records of all complete IVF cycles (including fresh and frozen) initiated during this period were extracted. Each patient's data, treatment characteristics, embryo information, and follow-up data were matched with a unique ID number. Frozen embryo transfer data related to these cycles were also collected between January 1, 2023, and December 31, 2023. This data selection method ensured at least one year of follow-up for all embryo transfer attempts within a complete cycle. Women who did not obtain oocytes in their first treatment cycle were excluded, as were women involved in donor sperm, oocyte donation, and / or PGT. Each participant followed the same clinical and laboratory procedures and underwent detailed calibration to ensure homogenization of operational practices.

[0023] Data cleaning and verification were performed. Categorical variables were coded to ensure data consistency and analyzability. The model data came from a single center, with no missing follow-up outcomes. A few covariates were missing: anti-Müllerian hormone (AMH), and male semen quality-related variables such as progressively motile sperm percentage (PR) and DNA fragmentation index (DFI). Original records were checked, and any missing entries were added. For the remaining unverifiable proportions, AMH missing percentages were less than 5%, PR missing percentages were less than 8.6%, and DFI missing percentages were less than 22.6%. When AMH missing percentages were less than 5%, female age, bFSH, and AFC variables were used to fit a decision tree to impute AMH missing values. When male semen analysis showed PR missing percentages of less than 8.6% and DFI missing percentages of less than 22.6%, male age and insemination method were used to fit a decision tree to impute missing values.

[0024] Feature extraction: Fully integrate predictive variables included in existing technologies. The primary outcome measure is the cumulative chance of a couple receiving a first live birth within a maximum of four full IVF cycles. CLBR = (Number of patients with first deliveries during fresh + frozen cycles) / Number of patients initiating (or egg retrieval) cycles × 100%.

[0025] The characteristic variables of the pre-treatment model include female age, BMI, ovarian reserve indicators (bFSH, AMH and AFC), male age, semen quality (PR, DFI), duration of infertility (years), pregnancy history (yes or no), and cause of infertility (female factors, male factors, multifactorial or unknown causes for both partners).

[0026] The characteristic variables of the post-treatment model were added as predictive variables after the first transfer, including the number of oocytes retrieved, fertilization method (IVF or ICSI), number of embryos obtained, number of usable embryos, number of frozen embryos, transfer type (ET, FET or no transfer), number of embryos transferred (1 or 2), and embryo stage, i.e., cleavage embryo (day 3) or blastocyst (day 5 or 6).

[0027] The characteristic variable of the early pregnancy model is the initial serum β-hCG test value. Given the time-dependent differences in serum β-hCG secretion after IVF pregnancy, the evaluation criterion for the initial serum β-hCG value is determined by the equivalent number of days of embryonic development, i.e., the number of days since embryo transfer plus the number of days of testing after transfer. For example, if a cleavage embryo is transferred on day 3, and β-hCG is tested 14 days post-transfer, then the equivalent number of days of testing is 3 + 14 = 17 days.

[0028] Considering the interactions between features, full subset regression was used to screen covariates, and the optimal variable set was selected by comprehensively considering clinical and model performance.

[0029] Using the number of complete cycles as a discrete-time variable, a Cox regression model was constructed to calculate the cumulative live birth rate before and after treatment. A logistic regression model was then used to predict the probability of live birth in cycles after β-hCG positivity. In the multi-model adjusted univariate curve fitting analysis, it was found that female age (ovum retrieval OS), AFC, number of available embryos, and initial β-hCG value all exhibited non-linear relationships with the live birth probability; therefore, restricted cubic spline modeling was used.

[0030] AUC and C-index were used to validate the model's discriminative power, while the Brier score and calibration curve were used to evaluate its predictive accuracy. Furthermore, decision curve analysis (DCA) was employed to quantify the net benefit at different threshold probabilities and assess the model's clinical utility. The Brier score, ranging from 0 to 0.25, is an indicator of the consistency between predicted and observed probabilities; a score closer to 0 indicates better model calibration. Nomograms were used to standardize regression coefficients and display them as hazard scores on a number axis. To prevent overestimation of model performance, 10-fold cross-validation and bootstrap sampling were used to assess the reproducibility of model performance.

[0031] Using the selected features as input, and the patient's cumulative live birth status (live birth status) and discrete cycle time (number of first live birth cycles) as output, a Cox regression model is trained. The early pregnancy model focuses on the probability of live birth in the current cycle, constructing a logistic regression model. R4.2.2 is used (…). http: / / www.Rproject.orgThe model was constructed and validated using the R Foundation (Vienna, Austria) and Free Statistics 2.2 software. The model coefficients explain the impact of each feature on the cumulative live birth rate. The nomogram standardizes the regression coefficients, assigns weights to each predictor variable, and displays them as probability scores on a number axis. Patients find the score corresponding to each variable on the nomogram based on their individual circumstances, and these scores are summed to obtain a total score. The COX and the total nomogram score correspond to the non-live birth probability at different cycles, from which the cumulative live birth probability (1 - non-live birth probability) is calculated. The logistic nomogram total score directly corresponds to the live birth probability predicted by the model. Patients can directly read the model-predicted probability through the total score. The score range for each variable is clearly visible.

[0032] Specific application examples The wife of a couple is 31 years old and has a BMI of 25.58 kg / m². 2 With an AFC of 24, the pre-treatment model score was 95+20+75=190, predicting a 35% chance of no live birth and a 65% chance of live birth in the first cycle, an 82% cumulative live birth rate in the second cycle, and a 90% cumulative live birth rate in the third cycle. The patient had a high probability of achieving a live birth in the first complete cycle. After the first cycle of treatment, three usable embryos were formed, and two cleavage embryos were transferred via fresh embryo transfer. The post-treatment model score was 80+10+60+42=192, predicting a 55% live birth rate in the first cycle, an 80% cumulative live birth rate in the second cycle, an 88% cumulative live birth rate in the third cycle, and a 90% cumulative live birth rate in the fourth cycle. The patient underwent fresh embryo transfer of 2 cleavage embryos in the first cycle. The β-hCG equivalent testing period was 17 days (12 days after blastocyst transfer). The initial β-hCG value was 649 IU / L. The total score of the early pregnancy model was 28+10+2+30+70=140 points, the revised live birth probability was 82%, and the patient ultimately achieved her first live birth.

[0033] Another 39-year-old infertile woman, with a BMI of 21.12 kg / m². 2The patient had 4 AFCs (Acute Failures), and a pre-treatment model score of 60+25+10=95. The live birth probabilities for the first four cycles were 13%, 20%, 25%, and 30%, respectively, indicating a high probability of needing multiple cycles of preparation. After the first cycle of treatment, one usable embryo was formed, but embryo transfer was not performed. The post-treatment model score was 55+15+20+0=90. The cumulative live birth probabilities for cycles 2-4 were 10%, 15%, and 17%, respectively (no live birth in cycle 1). In cycle 4, at age 40, the patient underwent frozen embryo transfer of one cleavage-stage embryo and one blastocyst. The β-hCG equivalent detection time was 17 days, with an initial β-hCG value of 688.5 IU / L. The early pregnancy model score was 15+0+13+30+80=138, with a revised live birth probability of 79%, ultimately resulting in the first live birth.

[0034] This invention uses a Cox regression model to process time-dependent and censored data, intuitively predicting multi-cycle cumulative live birth rate. The accuracy and generalization ability of the model were verified through internal cross-validation, and the specific results are shown in Table 1.

[0035] Table 1: Raw data and internal cross-validation results before and after treatment and in early pregnancy. Model Time Type Raw data Cross-validation Pre-treatment model Cycle 1 AUC 0.787 0.784 Brier 0.188 0.191 Cycle 2 AUC 0.853 0.849 Brier 0.141 0.144 Cycle 3 AUC 0.887 0.880 Brier 0.120 0.125 Cycle 4 AUC 0.950 0.950 Brier 0.088 0.090 Post-treatment model Cycle 1 AUC 0.916 0.910 Brier 0.128 0.133 Cycle 2 AUC 0.903 0.896 Brier 0.118 0.122 Cycle 3 AUC 0.880 0.868 Brier 0.120 0.124 Cycle 4 AUC 0.934 0.932 Brier 0.092 0.095 Early pregnancy model AUC 0.870 0.860 Brier 0.121 0.127

[0036] The pooled C-index of the model before treatment was 0.735 (95% CI: 0.706–0.763), and the C-index after calibration was 0.730. The Brier scores of the model in the four complete cycles were 0.188, 0.141, 0.120, and 0.088, respectively. The absolute error continued to decrease, indicating that its predictive stability was good. Figure 3 The cumulative live birth rate prediction status of the pre-treatment model over four cycles was disclosed, along with the calibration and DCA data for the first cycle. Figure 4 As shown; decision curve analysis indicates that when the threshold probability is between 14% and 74%, using nomogram to predict the live birth rate for the first full cycle provides a net benefit greater than the “all treatment” or “no treatment” strategy, demonstrating the clinical efficacy of the pretreatment model.

[0037] The pooled C-index of the model after treatment was 0.826 (95% CI: 0.804–0.848), and the calibrated C-index was 0.824. The Brier scores of the model in the four complete cycles were 0.128, 0.118, 0.120, and 0.092, respectively. Figure 5 The cumulative live birth rate prediction status of the model over four cycles after treatment has been disclosed. Figure 6 The calibration and clinical decision curves for the first complete cycle are shown. The model delivers sustained and broad net benefits within the 11%–86% threshold range of the decision curves.

[0038] The C-index of the early pregnancy model was 0.870 (95% CI: 0.844–0.896), and the calibrated C-index was 0.866. The intercept was close to 0, and the slope was close to 1. (Specific details are as follows...) Figure 7 As shown. Figure 8 The calibration and clinical decision curves are shown, and the model will provide broad benefits when the decision curves are within the threshold range of 2% to 93%.

[0039] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for predicting the cumulative live birth rate based on in vitro fertilization of a plurality of complete cycles, characterized in that, The application relates to an IVF multiple complete cycle cumulative live birth rate prediction model and a construction method thereof. Step 1: obtaining the clinical data, embryo information and pregnancy follow-up data of an in vitro fertilization (IVF) patient, screening the patient data according to the screening standard, and pre-processing the screened data; Step 2: based on the pre-processed data, extracting feature variables of three stages of pre-treatment, post-treatment and early pregnancy of IVF treatment respectively, and determining the optimal prediction variable set of each stage; Step 3: based on the optimal prediction variable set of each stage, performing a restricted cubic spline (RCS) processing on the variables having a nonlinear relationship with the live birth probability, and obtaining the optimal prediction variable set of each stage after the RCS processing; Step 4: taking the number of complete cycles as a discrete time variable, taking the optimal prediction variable set of each stage after the RCS processing as input, constructing a COX regression model for the pre-treatment and post-treatment stages, and constructing a Logistic regression model for the early pregnancy stage; the COX regression model and the Logistic regression model jointly form an IVF multiple complete cycle cumulative live birth rate prediction model; Step 5: evaluating the constructed IVF multiple complete cycle cumulative live birth rate prediction model, if the evaluation is qualified, the model is applied to clinical application, if the evaluation is unqualified, steps 2 to 4 are repeated until the constructed IVF multiple complete cycle cumulative live birth rate prediction model is qualified, and a Nomogram diagram is constructed to intuitively display the contribution weight of each variable to the live birth.

2. The method for predicting the cumulative live birth rate based on a plurality of complete cycles of in vitro fertilization according to claim 1, wherein, The screening standard is a patient who performs ovary stimulation and corresponding frozen embryo transplantation for the first time, and patients who do not obtain eggs, involve sperm donation, ovum donation and pre-implantation genetic testing (PGT) are excluded. The pre-processing of the screened data is as follows: cleaning the screened data to eliminate errors, redundancies and inconsistent information, and encoding the classification variables; meanwhile, a decision tree interpolation method is used to interpolate the missing variables; wherein the missing variables include anti-Mullerian hormone (AMH), percentage of forward moving sperm (PR) and DNA fragmentation index (DFI) variables; when the AMH missing rate is less than 5%, the PR missing rate is less than 8.6% and the DFI missing rate is less than 22.6%, the decision tree interpolation method is used to interpolate the missing variables. The feature variables of the pre-treatment, post-treatment and early pregnancy stages of IVF treatment are extracted respectively as follows:

3. The method for predicting the cumulative live birth rate based on a plurality of complete cycles of in vitro fertilization according to claim 2, characterized in that, The feature variables of the pre-treatment stage include: female age, body mass index, basal antral follicle count (AFC), basal follicle stimulating hormone (bFSH), AMH, male age, percentage of forward moving sperm (PR), DNA fragmentation index (DFI), duration of infertility, pregnancy history and cause of infertility; The feature variables of the post-treatment stage include: on the basis of the pre-treatment variables, the number of obtained eggs, insemination method (IVF / ICSI), the number of obtained embryos, the number of available embryos, the number of frozen embryos, transplantation type (ET / FET / without transplantation) and transplantation embryo stage (3-day blastocyst / 5-6-day blastocyst) are added. ​ The characteristic variables of the early stage of pregnancy include: the initial serum β-hCG detection value and the β-hCG equivalent detection days which are the sum of the embryo transfer days and the detection days after embryo transfer, in addition to the characteristic variables of the post-treatment stage.

4. The method for predicting the cumulative live birth rate based on a plurality of complete cycles of in vitro fertilization according to claim 3, wherein, The optimal prediction variable set of each stage is determined, in particular: The optimal prediction variable set of each stage is screened by using the full subset regression method, the optimal variable set of the pre-treatment stage and the post-treatment stage is determined by the minimum Bayesian information criterion BIC, and the optimal variable set of the early stage of pregnancy is determined by the minimum Mallows Cp value; The variables in a non-linear relationship with the live birth probability include: the female age, the AFC, the available embryo number and the initial β-hCG value; The variables in a non-linear relationship with the live birth probability are processed by the RCS, in particular: 3 nodes are set, the nodes are located at 10%, 50% and 90% quantiles, the RCS is processed based on 3 nodes to divide the data into 4 intervals, the optimal prediction variable set of each stage after RCS processing is obtained, and the fitting deviation caused by the nodes concentrated in a certain area is avoided.

5. The method for predicting the cumulative live birth rate based on a plurality of complete cycles of in vitro fertilization according to claim 4, wherein, The optimal prediction variable set of each stage after RCS processing is taken as the input, the COX regression model is constructed for the pre-treatment stage and the post-treatment stage, and the Logistic regression model is constructed for the early stage of pregnancy, in particular: The COX regression model of the pre-treatment stage: ； wherein, in which, represents the number of live births or discrete periods, represents the live birth status, wherein 1 represents live birth and 0 represents no live birth; is the result of a 3-node restricted cubic spline treatment of the female age, is the result of a 3-node restricted cubic spline treatment of the antral follicle count AFC; The COX regression model of the post-treatment stage: ； wherein, is the number of available embryos for the current cycle of the patient, is the result of a 3-node restricted cubic spline treatment of the number of available embryos, denotes the type of embryo transfer, including fresh embryo transfer ET = 0 and frozen-thawed embryo transfer FET = 1; The Logistic regression model of the early stage of pregnancy: ； wherein, is the live birth probability, is the implantation embryo period and number, is the equivalent days of β-hCG detection, is the β-hCG detection value, , , , , and are the model parameters obtained by the maximum likelihood estimation method; embryo transfer (DCT), 3 for single blastocyst transfer (SBT), 4 for other types including double blastocyst transfer (DBT) or SCT+SBT.

6. The method of predicting the cumulative live birth rate based on a plurality of complete cycles of in vitro fertilization according to claim 5, wherein, It also includes a non-live birth probability prediction equation of each cycle, the non-live birth probability is calculated based on the integral points, the integral points are obtained by a polynomial formula based on the female age, the BMI and the AFC, the cumulative live birth probability is 1 minus the non-live birth probability of the corresponding cycle, in particular: The non-live birth probability prediction equation is applicable to the COX regression model of the pre-treatment stage and the COX regression model of the post-treatment stage, the equation takes the integral points points as the core intermediate variable, the non-live birth probability is obtained by substituting the integral points calculated based on the individual characteristics of the patient into the equation of the corresponding cycle, and finally the cumulative live birth probability = 1-non-live birth probability of the corresponding cycle; the early stage of pregnancy model is the Logistic regression, and the live birth probability is directly output, without a separate non-live birth probability equation; The non-live birth probability prediction equation in the COX regression model of the pre-treatment stage is: ； wherein, ； The non-live birth probability prediction equation of the first cycle ； The non-live birth probability prediction equation of the second cycle is ； The non-live birth probability prediction equation of the third cycle ； The non-live birth probability prediction equation of the fourth cycle ； The non-live birth probability prediction equation in the COX regression model of the post-treatment stage is: ； wherein, ； ； ； ； The non-live birth probability prediction equation of the first cycle is ； The non-live birth probability prediction equation of the second cycle is ； The non-live birth probability prediction equation of the third cycle is ； The non-live birth probability prediction equation of the fourth cycle is ； wherein, Female age is abbreviated as F.

7. The method of predicting the cumulative live birth rate based on a plurality of complete cycles of in vitro fertilization according to claim 6, wherein, The constructed IVF multiple complete cycle cumulative live birth rate prediction model is evaluated, in particular: The characteristic variables of the three stages of pre-treatment, post-treatment and early pregnancy of IVF treatment are input into the corresponding COX regression model of the pre-treatment and post-treatment stages and the Logistic regression model of the early pregnancy stage, and the cumulative live birth probability is output; The probability of no live birth in the first N cycles = the probability of non-live birth in the first cycle x the probability of non-live birth in the second cycle x…x the probability of non-live birth in the Nth cycle; The cumulative live birth probability of the first N cycles = 1-the probability of no live birth in the first N cycles, that is, the probability of achieving live birth at least once in the first N cycles; Based on the cumulative live birth probability, the AUC value, C-index, Brier score, calibration curve and decision curve analysis are used as model evaluation indexes, and 10-fold cross-validation and Bootstrap bootstrap sampling method are used for verification. If it does not exceed the set threshold, it is determined that the current IVF multiple complete cycle cumulative live birth rate prediction model meets the practical requirements.

8. The method of predicting the cumulative live birth rate per complete cycle based on in vitro fertilization according to claim 7, wherein, The Nomogram chart is constructed to directly display the contribution weight of each variable to live birth, specifically: The regression coefficients of the COX regression model and the Logistic regression model are standardized to display the score range of each prediction variable on the number axis in the form of risk score. The patient determines the score of each variable according to the actual index and adds up to get the total score. The total score is directly used to read the non-live birth probability or live birth probability of the corresponding cycle, and then the cumulative live birth rate is obtained. The risk score is the different values of each prediction variable in the model, which are converted into a unified scale score according to the influence weight on the non-live birth risk or live birth probability. The greater the contribution of variable value to the non-live birth risk, the higher the corresponding risk score; on the contrary, the greater the contribution to the live birth probability, the lower the corresponding risk score.

9. A system for predicting the cumulative live birth rate based on in vitro fertilization of a plurality of complete cycles, characterized in that, It includes: A preprocessing module that acquires clinical data, embryo information and pregnancy follow-up data of in vitro fertilization (IVF) patients, establishes a screening standard for patient data, and pre-processes the screened data; A determination module that extracts characteristic variables of pre-treatment, post-treatment and early pregnancy of IVF treatment based on the pre-processed data, and determines the optimal prediction variable set of each stage; A processing module that processes the variables with non-linear relationship with live birth probability based on the optimal prediction variable set of each stage to obtain the optimal prediction variable set after RCS processing of each stage; An acquisition module that takes the number of complete cycles as a discrete time variable and takes the optimal prediction variable set after RCS processing of each stage as input to construct a COX regression model for the pre-treatment and post-treatment stages and a Logistic regression model for the early pregnancy stage. The COX regression model and the Logistic regression model together constitute an IVF multiple complete cycle cumulative live birth rate prediction model; A construction module that evaluates the constructed IVF multiple complete cycle cumulative live birth rate prediction model. If the evaluation is qualified, it enters clinical application and constructs a Nomogram chart to directly display the contribution weight of each variable to live birth.

Images