RU2809429C1 - Method of predicting effectiveness of assisted reproductive technology programs based on neural networks - Google Patents

Abstract

FIELD: medicine; obstetrics and gynecology.

SUBSTANCE: invention can be used to predict the effectiveness of assisted reproductive technology programs in women with infertility of various origins in medical decision support systems. At the first stage of the implementation of the protocol for in vitro fertilization (IVF) programs, medical history data, an objective examination and the results of clinical, laboratory and instrumental studies are collected, and the option of the protocol for stimulating superovulation, the starting and total doses of follicle-stimulating hormone and the need to include drugs containing luteinizing hormone are also selected. Then, using all the data obtained: the doctor’s work experience, diagnosis according to ICD N10, age, height and weight of the patient, IVF attempt number, presence of human immunodeficiency virus (HIV) in women and men, presence of chronic hepatitis B or C in women and men, the presence of human papillomavirus (HPV) in a woman, a history of arterial hypertension and/or diabetes mellitus types 1, 2, hemoglobin level in the blood, the count of red blood cells, leukocytes, lymphocytes, platelets in the blood formula, fasting glucose concentration, level of alanine aminotransferase (ALT), aspartate aminotransferase (AST), creatinine, cholesterol, bilirubin, concentration of follicle-stimulating hormone (FSH) and anti-Mullerian hormone (AMH) in the patient’s venous blood, the count of antral follicles on the right and left before stimulation, type of stimulation protocol, starting and the total dose of FSH, the presence of luteinizing hormone (LH) in the stimulation protocol as predictors, using a predictive mathematical model of the first stage, the probability of obtaining 5 or less oocytes, from 6 to 19 oocytes, or 20 or more oocytes suitable for fertilization is calculated. Further, when such oocytes are obtained as a result of puncture, at the second stage of the IVF protocol, the qualitative and quantitative characteristics of the sperm intended for fertilization are determined. The proposed options for fertilization and embryo cultivation are selected. Then, using all the data obtained, namely: the type of stimulation protocol performed, the actual starting and total doses of FSH, the presence of LH in the stimulation protocol, the actual number of days of stimulation, the use of a gonadotropin-releasing hormone (GnRH) agonist or human chorionic gonadotropin as a trigger (hCG), the concentration of estradiol (E2) and progesterone in the blood before puncture, the number of follicles and complexes before puncture of the ovaries according to ultrasound (ultrasound), the number of oocytes obtained as a result of puncture, of which the number of mature oocytes at metaphase stage 2 (M-2 ), number of immature oocytes of the metaphase 1 (M-1) stage, number of immature oocytes of the "Germinal Vesular" (GV) stage and number of degenerative oocytes, source of sperm and method of obtaining sperm, use of cryo-freezing of sperm, number of sperm after treatment, percentage of motile sperm category A+B, percentage of morphologically normal sperm, use of intracytoplasmic sperm injection (ICSI) during fertilization, types of media used for fertilization, cleavage and culture, number of 2PN zygotes obtained, in addition to the first stage data, as predictors, and using a predictive mathematical model of the second stage, the probability of obtaining embryos suitable for transfer into the uterine cavity or cryopreservation in an amount of 1 to 2, 3 or more, or the probability of a negative result, i.e. 0 embryos on the fifth day after fertilization is calculated. When such embryos are obtained, at the third stage of the IVF protocol, the thickness of the endometrium is determined using ultrasound. A decision is made to use a GnRH agonist. Then, using all the data obtained, namely: the number of embryos suitable for transfer, the use of embryo cryopreservation, the thickness of the endometrium before transfer, the presence of complications during the embryo transfer procedure, and the use of a GnRH agonist, in addition to the data from the first and second stages, as predictors, using a prognostic mathematical model of the third stage, the probability of pregnancy is calculated as a percentage. Each of the three predictive mathematical models is implemented in the form of an artificial neural network pre-trained on a prepared data set with an architecture in the form of a multilayer fully connected Rumelhart perceptron.

EFFECT: invention provides accurate prediction of the effectiveness of the implementation of each stage of the IVF protocol, through the use of artificial neural networks as a mathematical apparatus, which, in turn, makes it possible to effectively take into account a large number of heterogeneous predictors and the possibility of making changes to the protocol in order to increase overall efficiency.

1 cl, 2 tbl, 3 ex

Description

The present invention relates to medicine, namely to obstetrics and gynecology, and concerns predicting the effectiveness of assisted reproductive technology programs in women with infertility of various origins, and can be used in medical decision support systems and telemedicine.

Every year, 0.5% of children in the world are born using assisted reproductive technologies (ART). In Russia this number reaches 2% [1]. The effectiveness of the in vitro fertilization (IVF) program is ultimately assessed by the occurrence of pregnancy and the birth of a healthy child (“Take Baby Home” parameter). However, the final result of infertility treatment largely depends on the quality of the preliminary stages of IVF (stimulation of ovulation and follicle puncture, in vitro fertilization and embryo cultivation, selection and transfer of high-quality embryos, their implantation into the endometrium). Today, approved standards and clinical protocols for the treatment of various forms of infertility have been developed and applied in clinical practice; however, they cannot take into account all the individual characteristics of the patient in a specific clinical situation. The variability and large volume of clinical and laboratory parameters, as well as the influence on them of both genetically determined factors and environmental factors, do not always allow even experienced gynecologists (reproductologists) and embryologists to accurately predict the outcome of IVF when choosing a particular protocol in a particular clinical case.

In this regard, various electronic application programs and services that help the doctor choose the optimal protocol and parameters for IVF are of particular importance. A promising direction in this area is the use of mathematical models based on artificial neural networks. This approach, in comparison with conventional regression-variance analysis, allows us to effectively consider as predictors that fully influence the formation of the prognosis, a significantly larger number (tens and hundreds) of clinical and laboratory parameters and their combinations obtained at different stages of IVF, as well as increase the reliability of the forecast due to the accumulation of new data during operation and constant “additional training” of the model, which will ultimately help improve the efficiency of infertility treatment using IVF.

Previously described is a method for predicting the outcomes of an IVF program by qualitative and quantitative assessment of the contents of follicles, characterized in that the activity of peptide hydrolases is examined in follicular fluid, aspirated separately from the right and left ovaries, and at a level of at least 75 mg/ml/h in one from the ovaries, a favorable outcome is predicted [2].

There is a method for predicting the results of infertility treatment using IVF, including transvaginal puncture of the follicles under ultrasound control with subsequent examination of the resulting follicular fluid, characterized in that the follicular fluid is subjected to chemiluminescent analysis using a standard method, the free radical activity of the samples is determined and, based on the indicator of total antioxidant activity, a forecast is given for the onset of infertility. pregnancy: with indicators of total antioxidant activity of 0.05 and above, they give an unfavorable prognosis; with indicators of total antioxidant activity below 0.05, they give a favorable prognosis - the probability of pregnancy is more than 50% [3].

The disadvantages of the two methods described above include the following: 1. The methods are implemented only at the stage of follicle puncture, which does not allow changes to be made to the superovulation stimulation scheme; 2. It is necessary to separately obtain and examine two samples of follicular fluid (from the right and left ovaries); 3. There is a high risk of complications associated with performing transvaginal puncture of follicles under general anesthesia; 4. The laboratory tests described are not routine laboratory practices included in existing clinical guidelines and require additional laboratory equipment and consumables.

There is a known method for predicting the outcome of IVF and embryo transfer, which consists in the fact that before the start of ovarian stimulation, the polymorphism of HLAII class genes is determined in leukocytes of peripheral venous blood in women with infertility and when carriers of haplotypes DQA1*0101-DQB1*0501 or DQA1*0102-DQB1* are detected 0502 predict an unfavorable outcome of the IVF program, and in the absence of carriage of the DQA1*0101-DQB1*0501 and DQA1*0102-DQB1*0502 haplotypes, they predict a favorable outcome of the IVF program [4].

This method has the following disadvantages: 1. The mathematical model uses exclusively genetic markers as predictors, therefore, it does not take into account the influence of the patient’s acquired physiological characteristics on the outcome of the IVF protocol, as well as external influences (medicine support); 2. Determination of a large number of histocompatibility complex antigens is not a routine laboratory practice included in existing clinical guidelines and requires additional laboratory equipment and consumables.

The closest to the proposed invention is “Method for predicting the outcomes of assisted reproductive technologies for tubal infertility” [5]. This method includes determining the concentrations of follicle-stimulating and anti-Mullerian hormones and the subsequent calculation of prognostic indicators using a mathematical model, and additionally determining age, duration of infertility, body mass index, and ovarian volume. Then the prognostic index of pregnancy, d _1, is calculated using a logistic regression equation, and if d ₁ <-1.397, pregnancy is predicted, and if d ₁ >1.151, the absence of pregnancy is predicted. Next, the prognostic index of the probability of live birth d ₂ is calculated and if d ₂ <-l.525 a positive result is predicted, and if d ₂ >l.102 a negative result is predicted.

However, this method contains the following significant drawbacks: 1. The described approach allows making a forecast regarding the occurrence of pregnancy and live birth of a child only on a binomial scale, and does not allow quantifying the probability of the occurrence of these events, which significantly reduces the accuracy of the forecast; 2. The method does not allow making a prediction regarding other important events occurring during the implementation of the IVF protocol, for example, obtaining oocytes suitable for fertilization as a result of stimulation of superovulation and puncture, or obtaining embryos suitable for transfer into the uterine cavity as a result of fertilization and cultivation; 3. As predictors, the mathematical model includes data obtained only at the prehospital stage; 4. The scope of application of the model is limited to only one type of pathology - tubal infertility.

Technical result: increasing the universality and accuracy of the forecast, as well as expanding the scope of application by including all stages of the implementation of the IVF protocol in the treatment of infertility of various origins.

This technical result is achieved by combining three related prognostic mathematical models that describe the main stages of the implementation of the IVF protocol: stimulation of superovulation and follicle puncture, fertilization and cultivation of embryos, transfer of embryos into the uterine cavity, and each mathematical model includes, as predictors, all data available at the time of implementation of the selected stage of the protocol, and contains a forecast of the effectiveness of the implemented stage: the probability of obtaining a certain number of oocytes suitable for fertilization, the probability of obtaining a certain number of embryos suitable for transfer or cryopreservation, and the probability of pregnancy.

A method is proposed for predicting the effectiveness of multi-stage in vitro fertilization (IVF) programs in the treatment of infertility of various origins, which consists in the fact that at the first stage of the implementation of the IVF protocol, before the start of stimulation of superovulation and puncture of the follicles, through a survey and examination of the patient, as well as with the help of laboratory and instrumental diagnostic protocols collect data from anamnesis, objective examination and the results of clinical, laboratory and instrumental studies, and also select the option of the superovulation stimulation protocol, the starting and total doses of follicle-stimulating hormone and the need to include drugs containing luteinizing hormone in the protocol, and then, using all the data obtained (doctor’s work experience, diagnosis according to ICD No. 10, age, height and weight of the patient, IVF attempt number, HIV status of the woman and man, presence of chronic hepatitis B or C in the woman and man, presence of HPV infection in the woman, history of arterial hypertension and/or diabetes mellitus types 1, 2, hemoglobin level in the blood, number of red blood cells, leukocytes, lymphocytes, platelets in the blood count, concentration of fasting glucose, ALT, AST, creatinine, cholesterol, bilirubin, FSH and AMH in venous blood of the patient, the number of antral follicles on the right and left before stimulation, the type of stimulation protocol, the starting and total doses of FSH, the presence of LH in the stimulation protocol) as predictors, using a predictive mathematical model of the first stage, the probability of receiving after stimulation and puncture is calculated 5 and less oocytes, from 6 to 19 oocytes, or 20 or more oocytes suitable for fertilization, then, when such oocytes are obtained as a result of puncture, at the second stage of the IVF protocol, before the fertilization procedure, the results of the previous stage are assessed, qualitative and quantitative are determined characteristics of the sperm intended for fertilization, select the proposed options for fertilization and cultivation of embryos, and then, using all the data obtained (the actual number of days of stimulation, the use of a GnRH or hCG agonist as a trigger, the concentration of E2 and progesterone in the blood before puncture, the number of follicles and complexes before puncture of the ovaries according to ultrasound data, the number of oocytes obtained as a result of puncture, of which the number of mature oocytes of stage M-2, the number of immature oocytes of stage M-1, the number of immature oocytes of stage GV and the number of degenerative oocytes, the source of sperm and the method of obtaining sperm, the presence of cryo-freezing of sperm, the number of spermatozoa after treatment, the percentage of motile spermatozoa of category A+B, the percentage of morphologically normal spermatozoa, the use of ICSI procedure during fertilization, the types of media used for fertilization, crushing and culturing, the number of 2PN zygotes obtained), in addition to data of the first stage, as predictors, using a predictive mathematical model of the second stage, calculate the probability of obtaining on the fifth day after fertilization embryos suitable for transfer into the uterine cavity or cryopreservation, in an amount from 1 to 2, from 3 or more, or the probability of a negative result (0 embryos), then, upon receipt of such embryos, at the third stage of the IVF protocol, before the embryo transfer procedure into the uterine cavity, the results of the previous stage are assessed, the thickness of the endometrium is determined using ultrasound, a decision is made on the use of a GnRH agonist, and then, using all the data obtained (the number of embryos suitable for transfer, preliminary cryopreservation of the embryo, endometrial thickness before transfer, the presence of complications during the embryo transfer procedure and the use of a GnRH agonist), in addition to the data of the first and second stages, as predictors, using a predictive mathematical model The third stage calculates the probability of pregnancy as a percentage, while each of the three predictive mathematical models is implemented in the form of an artificial neural network pre-trained on a prepared data set with an architecture in the form of a multilayer fully connected Rumelhart perceptron.

The essence of the invention is as follows. In the process of implementing the IVF protocol, three stages can be conventionally distinguished, each of which ends with a result, on which the possibility of implementing subsequent stages directly depends. The first stage is the stimulation of superovulation and puncture of the follicles, which ends with obtaining a certain number of oocytes suitable for fertilization. The second stage is fertilization and cultivation of embryos, as a result of the successful implementation of which a certain number of embryos suitable for transfer into the uterine cavity or cryopreservation are obtained. The third stage is the transfer of the embryo into the uterine cavity, the successful result of which is the onset of clinical pregnancy. At each of the described stages of IVF protocol implementation, the composition of available predictors and predicted variables changes, which does not allow the entire protocol to be effectively presented in the form of one predictive mathematical model. Taking this into account, we propose to implement a mathematical model of the IVF protocol in the form of three related models corresponding to each of the described stages of the IVF protocol.

Initially, all the signs provided for by the standard of examination of patients [6] and the Clinical Guidelines of the LLC “Russian Society of Obstetricians and Gynecologists” [7] were included as predictors in the developed predictive mathematical models. Further, the list of clinical diagnostic predictors was adjusted in accordance with the level of evidence (1 or 2) of their influence on the outcomes of infertility treatment and the level of strength of recommendations for their inclusion in the analysis before IVF protocols (A or B), as well as previously published data on the influence of some additional parameters on the outcomes of IVF programs [8 - 15].

At the first stage, the main task is to predict the number of oocytes obtained as a result of stimulation of superovulation and follicle puncture. This parameter was selected as a criterion (predicted) variable. To increase the accuracy of the forecast and simplify the interpretation of the results obtained, the scale for measuring the criterion variable (the absolute number of oocytes received) was converted into a ranking scale. There are three ranks (categories): the first - from 0 to 5 oocytes, the second - from 6 to 19 oocytes, the third - from 20 or more oocytes. At this stage, all variables known at the start of the superovulation stimulation protocol (data from anamnesis, objective examination, results of clinical laboratory tests), as well as the proposed scheme of the stimulation protocol itself, act as predictors. There are 42 predictors in total. A detailed description of the predictors is presented in Table 1. As a result of mathematical processing of the values of the predictors used by the predictive mathematical model of the first stage, the user receives a probability for each of the three ranks of the criterion variable.

Table 1
Predictors used in predictive mathematical models No. Description of the predictor Measurement scale type Predictors of the first stage: “Stimulation of superovulation and puncture of follicles” Work experience of the doctor who performed the puncture relations Diagnosis (ICD No. 10) N 97.0 binomial Diagnosis (ICD No. 10) N 97.1 binomial Diagnosis (ICD No. 10) N 97.2 binomial Diagnosis (ICD No. 10) N 97.3 binomial Diagnosis (ICD No. 10) N 97.4 binomial Diagnosis (ICD No. 10) N 97.8 binomial Diagnosis (ICD No. 10) N 97.9 binomial Patient's age at the time of puncture relations IVF attempt number relations Patient's height relations Patient's weight relations HIV status of woman binomial HIV status of a man binomial Chronic hepatitis B in a woman binomial Chronic hepatitis B in men binomial Chronic hepatitis C in a woman binomial Chronic hepatitis C in men binomial HPV infection in a woman binomial History of arterial hypertension binomial History of type 2 diabetes mellitus binomial History of type 1 diabetes mellitus binomial Hemoglobin relations Red blood cells relations Leukocytes relations Lymphocytes relations Platelets relations Fasting glucose relations ALT relations AST relations Creatinine in blood relations Cholesterol relations Bilirubin relations FSH relations AMG relations Number of antral follicles on the right before stimulation relations Number of antral follicles on the left before stimulation relations The stimulation protocol is short binomial The stimulation protocol is long binomial Starting dose of FSH relations Total dose of FSH relations Presence of LH administration in the protocol binomial Predictors of the second stage: “Fertilization and cultivation of embryos” Number of days of stimulation relations Administration of a GnRH agonist trigger binomial HCG trigger injection binomial Level E2 before puncture relations Progesterone level before puncture relations Number of follicles before puncture relations Number of complexes before puncture relations Number of oocytes obtained as a result of puncture relations Number of mature oocytes stage M-2 relations Number of immature oocytes stage M-1 relations Number of immature GV stage oocytes relations Number of degenerative oocytes relations Source of sperm - husband binomial Source of sperm - donor binomial Method of obtaining sperm - masturbation binomial Method of obtaining sperm - surgical method binomial CRYO of sperm (from any source) binomial Sperm count after treatment relations Percentage of motile sperm of category A+B relations Percentage of morphologically normal sperm relations The use of ICSI for fertilization binomial Fertilization medium - sequential binomial Fertilization medium - mono-step binomial Crushing medium - sequential binomial Crushing medium - mono-step binomial Culture medium - sequential binomial Cultivation medium - mono-step binomial Number of 2 PN zygotes relations Predictors of the third stage: “Transfer of the embryo into the uterine cavity” Number of embryos suitable for transfer relations Embryo transfer after cryopreservation binomial Endometrial thickness before embryo transfer relations Presence of complications during the embryo transfer procedure binomial Use of a GnRH agonist during embryo transfer binomial

At the second stage, which begins after puncture of the follicles, the main task is to predict the number of embryos obtained on the fifth day after fertilization, suitable for transfer into the uterine cavity or cryopreservation. This parameter was selected as a criterion variable of the second stage. To increase the accuracy of the forecast and simplify the interpretation of the results obtained, the measurement scale of the criterion variable (the absolute number of embryos obtained) was converted into a ranking scale. There are three ranks: the first - 0 embryos, the second - from 1 to 2 embryos, the third - from 3 or more embryos. At this stage, all variables known at the time of oocyte fertilization, including first-stage predictors, as well as the expected scheme of fertilization and embryo culture, act as predictors. There are 70 predictors in total. A detailed description of the predictors is presented in Table 1. As a result of mathematical processing of the values of the predictors used by the predictive mathematical model of the second stage, the user receives a probability for each of the three ranks of the criterion variable.

At the third stage, which begins from the moment the patient is prepared for embryo transfer into the uterine cavity, the main task is to predict the likelihood of a clinical pregnancy. This parameter was selected as a criterion variable of the third stage. At this stage, all variables known at the time of embryo transfer into the uterine cavity, including predictors of the first and second stages, act as predictors. There are 75 predictors in total. A detailed description of the predictors is presented in Table 1. As a result of mathematical processing of the values of the predictors used by the prognostic mathematical model of the third stage, the user receives the probability of a clinical pregnancy.

Combining in a single system three related models corresponding to each stage of the implementation of the IVF protocol allows for a more detailed assessment of the effectiveness of the implementation of each stage and provides more opportunities for making changes to the implementation of the entire protocol in order to increase overall efficiency.

Each of the three predictive mathematical models is implemented in the form of an artificial neural network of a given architecture and pre-trained on a prepared data set. The architecture of each artificial neural network is presented in the form of a multilayer fully connected perceptron with several hidden layers (Rumelhart multilayer perceptron). The number of input neurons in the input layer of each neural network corresponds to the number of predictors - 42, 70 and 75, respectively, for the networks of the first, second and third stages. The number of output neurons is equal to the number of categories of predicted features, 3, 3 and 1, respectively. The hyperbolic tangent function was used as the activation function in the neurons of the hidden layers of all networks; for the neurons of the output layers of the neural networks of the first and second stages, the softmax function was used; for neurons the output layer of the third stage neural network is a sigmoid logistic function.

To train and validate artificial neural networks, an array of data suitable for processing by machine learning algorithms was previously generated - a dataset (certificate of state registration of the database RU 2021621398). The dataset included data from 5,200 IVF protocols implemented in 2018 - 2021. in three ART clinics in Yekaterinburg and St. Petersburg. All data underwent preliminary mathematical processing in the form of optimization of quantity and quality in accordance with the predicted variables, and normalization by the average value (z-normalization). Previously, the entire dataset was randomly divided into a test set (10% of the total number of cases, 520 cases) and a training set (90% of the total number of cases, 4680 cases). A validation sample (10%, 468 cases) was randomly formed from the training sample and used for intermediate validation during the training process. Neural networks were trained in the classical way, using the backpropagation algorithm. For more efficient learning, stochastic gradient descent was used as an optimization algorithm.

For practical use, trained artificial neural networks are implemented using the high-level programming language Python and standard libraries in the form of a program for a personal computer.

The method we claim is carried out according to a standard method as follows: during a planned hospitalization of a patient in an ART clinic, for the purpose of treating infertility of various origins using the IVF procedure, the doctor at the first stage, before stimulation of superovulation and puncture of the follicles, receives, through a survey and examination of the patient, and, using standard laboratory protocols, history, physical examination, and clinical laboratory results. Next, the doctor selects the proposed protocol for stimulating superovulation (short or long), the starting and total doses of FSH, and the need to include drugs containing LH in the protocol.

Further, as a result of mathematical processing of all received data (42 variables, presented in detail in Table 1), using an artificial neural network, a forecast of the results of stimulation of superovulation and follicle puncture is calculated. The forecast is calculated relative to the number of oocytes obtained that are suitable for fertilization as a percentage probability for each of the three ranks. The most favorable result is considered to be obtaining from 6 to 19 oocytes after puncture. Receiving from 0 to 5 oocytes is considered insufficient and unfavorable, and more than 20 oocytes is excessive and conditionally favorable due to the high risk of developing ovarian hyperstimulation syndrome. Since the prognosis is calculated before superovulation stimulation, the doctor can, based on the results of the prognosis, personal experience and clinical recommendations, change the proposed stimulation scheme in order to increase its effectiveness.

At the second stage, after stimulation of superovulation and puncture of the follicles, upon receipt of oocytes suitable for fertilization, the doctor records the newly obtained data (all parameters of the implemented scheme and additional laboratory data). At this point, the embryologist knows the qualitative and quantitative characteristics of the sperm intended for fertilization of oocytes, as well as the proposed options for fertilization and cultivation of embryos.

Further, as a result of mathematical processing of all data (70 variables, presented in detail in Table 1), using an artificial neural network, a forecast of the results of fertilization and embryo cultivation is calculated. The forecast is calculated regarding the number of embryos obtained suitable for transfer into the uterine cavity or cryopreservation as a percentage probability for each of the three ranks. The most favorable result is the receipt of 3 or more embryos. Obtaining 1 to 2 embryos is a conditionally favorable result. The absence of embryos is an unfavorable result. Since the prognosis is calculated before fertilization, the embryologist can, along with personal experience and clinical recommendations, be guided by the results of the prognosis in order to improve the efficiency of this stage.

At the third stage, upon receipt of embryos suitable for transfer, the doctor decides to transfer them into the uterine cavity. At the time of transfer, the doctor determines the thickness of the endometrium using ultrasound and decides on the use of a GnRH agonist. As a result of mathematical processing of all data (75 variables, presented in detail in Table 1) using an artificial neural network, the probability of pregnancy as a percentage is calculated.

The essence of the claimed method is illustrated by the following examples:

Example 1. Patient S., 34 years old, was examined and treated at the clinic of the Clinical Institute of Reproductive Medicine LLC in Yekaterinburg for infertility with a diagnosis according to ICD10 - No. 97.4 "Female infertility associated with male factors." Before stimulation of superovulation, the following data were collected from anamnesis, objective examination and results of clinical laboratory tests: IVF attempt - first; The patient’s height is 172 cm; the patient's weight is 64 kg; HIV status of the woman and partner is negative; chronic hepatitis B and C were not detected in the woman and partner; HPV infection in a woman - negative result; the woman has no history of arterial hypertension and diabetes mellitus; hemoglobin - 138 g/l; erythrocytes - 4.18 * 10 ¹² /l.; leukocytes - 5.1 * 10 ⁹ /l.; lymphocytes - 39.9%; platelets - 151 * 10 ⁹ /l; fasting glucose - 5.22 mmol/l; ALT - 10.8 U/l; AST - 15.4 U/l; creatinine - 76 µmol/l; cholesterol - 4.67 mmol/l; bilirubin - 27.8 µmol/l; FSH - 8.11 mIU/ml; AMH - 3.5 ng/ml; number of antral follicles on the right - 6; number of antral follicles on the left - 7; The doctor’s work experience at the time of the puncture was 13 years. The doctor decided to stimulate superovulation using a short protocol, the starting dose of FSH was 225 IU/day; the total dose of FSH for the entire protocol is 1575 IU, using drugs containing LH.

As a result of mathematical processing of the available data, using an artificial neural network, the following forecast of the results of stimulation of superovulation and follicle puncture was calculated: the probability of obtaining from 0 to 5 oocytes suitable for fertilization is 13.0%, from 6 to 19 oocytes - 80.4% , 20 or more oocytes - 6.6%. The prognosis is considered favorable.

Stimulation of superovulation lasted 7 days. Human chorionic gonadotropin (hCG) was used as a trigger for the final maturation of oocytes. Before the puncture, the following indicators were recorded: estradiol concentration (E2) - 6977 pmol/ml; progesterone concentration - 0.4 ng/ml; the number of follicles and complexes according to ultrasound data is 11. As a result of successful transvaginal puncture, 10 oocytes were obtained, 8 of which were mature at the M2 stage. For oocyte fertilization, the source of sperm was the husband, the method of obtaining sperm was masturbation, the sperm was not cryopreserved. The following characteristics of sperm quality were recorded: the number of sperm after treatment - at least 2 million; percentage of motile sperm of category A + B - 90%; the percentage of morphologically normal sperm is 5%. The embryologist decided to use the ICSI procedure when fertilizing an oocyte, and sequential media at the stages of fertilization, crushing and cultivation. After fertilization, 7 zygotes with two pronuclei (2PN) were obtained.

As a result of mathematical processing, using an artificial neural network, of all available data points, the following forecast of the results of fertilization and embryo cultivation was calculated: the probability of not obtaining embryos suitable for transfer into the uterine cavity or cryopreservation on the 5th day is 0.1 %, the probability of getting from 1 to 2 embryos is 10.9%, from 3 or more embryos - 89.0%. The prognosis is considered favorable. As a result of cultivation on the 5th day, 5 embryos were obtained, suitable for transfer into the uterine cavity or cryopreservation.

At the time of transfer, the thickness of the endometrium was 10.1 mm. It was decided to additionally use an analogue of the natural gonadotropin-releasing hormone triptorelin acetate, namely diferelin at a dose of 0.1 mg, subcutaneously. The probability of pregnancy, previously calculated as a result of processing all data using an artificial neural network, provided that the transfer procedure goes without complications, was 75.1%. Thus, a favorable prognosis was given. The embryo transfer took place safely and clinical pregnancy was diagnosed on the 14th day after the transfer.

Example 2. Patient N., 38 years old, was examined and treated at the clinic of the Clinical Institute of Reproductive Medicine LLC in Yekaterinburg for infertility with a diagnosis according to ICD10 - No. 97.4 "Female infertility associated with male factors." Before stimulation of superovulation, the following data were collected from anamnesis, objective examination and results of clinical laboratory tests: IVF attempt - first; The patient’s height is 163 cm; weight - 51 kg; HIV status of the woman and partner is negative; chronic hepatitis B and C were not detected in the woman and partner; HPV infection in a woman was not detected; the woman has no history of arterial hypertension and diabetes mellitus; hemoglobin - 130 g/l; erythrocytes - 4.55 * 10 ¹² /l.; leukocytes - 6.36 * 10 ⁹ /l.; lymphocytes - 34.7%; platelets - 226 * 10 ⁹ /l; fasting glucose - 5.5 mmol/l; ALT - 12.0 U/l; AST - 20.0 U/l; creatinine - 72 µmol/l; cholesterol - 5.3 mmol/l; bilirubin - 5.9 µmol/l; FSH - 5.87 mIU/ml; AMH - 3.2 ng/ml; number of antral follicles on the right - 4; number of antral follicles on the left - 4; The doctor’s work experience at the time of the puncture was 14 years. The doctor decided to stimulate superovulation using a short protocol (starting dose of FSH - 225 IU/day; total dose of FSH for the entire protocol - 1875 IU) using drugs containing LH.

As a result of mathematical processing of the available data, using an artificial neural network, the following forecast of the results of stimulation of superovulation and follicle puncture was calculated: the probability of obtaining from 0 to 5 oocytes suitable for fertilization is 11.8%, from 6 to 19 oocytes - 87.5% , 20 or more oocytes - 0.7%. The prognosis is considered favorable.

Stimulation of superovulation lasted 10 days. HCG was used as a trigger for the final maturation of oocytes. Before the puncture, the following indicators were recorded: progesterone concentration - 1.3 ng/ml; the number of follicles and complexes according to ultrasound data is 8. As a result of successful transvaginal puncture, 7 oocytes were obtained, 4 of which were mature at the M2 stage. For oocyte fertilization, the source of sperm was the husband, the method of obtaining sperm was masturbation, the sperm was not cryopreserved. The following characteristics of sperm quality were recorded: the number of sperm after treatment - at least 408 million; percentage of motile sperm of category A + B - 99%; the percentage of morphologically normal sperm is 2%. The embryologist decided to use the ICSI procedure when fertilizing an oocyte, and mono-step media at the stages of fertilization, crushing and cultivation. After fertilization, 3 zygotes with two pronuclei (2PN) were obtained.

As a result of mathematical processing, using an artificial neural network, of all available data points, the following forecast of the results of fertilization and cultivation of embryos was calculated: the probability of not obtaining embryos suitable for transfer into the uterine cavity or cryopreservation on the 5th day is 0.2 %, the probability of getting from 1 to 2 embryos is 83.6%, from 3 or more embryos - 16.2%. The prognosis is considered conditionally favorable. As a result of cultivation on the 5th day, 1 embryo was obtained, suitable for transfer into the uterine cavity or cryopreservation.

At the time of transfer, the thickness of the endometrium was 13.6 mm. It was decided to additionally use an analogue of natural gonadotropin-releasing hormone triptorelin acetate, namely diferelin at a dose of 0.1 mg, subcutaneously. The probability of pregnancy, previously calculated as a result of processing all data using an artificial neural network, provided that the transfer procedure goes without complications, was 36.7%. Thus, an unfavorable prognosis was given. Despite the low probability of pregnancy, but taking into account the late reproductive age of the patient and the receipt of a single embryo, it was decided to transfer it into the uterine cavity. The transfer took place without complications. On the 14th day after embryo transfer, the patient was diagnosed with no clinical pregnancy.

Example 3. Patient K., 36 years old, was examined and treated at the clinic of the Clinical Institute of Reproductive Medicine LLC in Yekaterinburg for infertility with a diagnosis according to ICD10 - No. 97.1 "Female infertility of tubal origin." Before stimulation of superovulation, the following data were collected from anamnesis, objective examination and results of clinical laboratory tests: IVF attempt - first; The patient’s height is 158 cm; patient weight - 55 kg; HIV status of the woman and partner is negative; chronic hepatitis C and B were not detected in the woman and partner; HPV infection in a woman was not detected; the woman has no history of arterial hypertension and diabetes mellitus; hemoglobin - 133 g/l; erythrocytes - 4.68 * 10 ¹² /l.; leukocytes - 7.4 * 10 ⁹ /l.; lymphocytes - 31.0%; platelets - 249 * 10 ⁹ /l; fasting glucose - 4.9 mmol/l; ALT - 15.7 U/l; AST - 20.6 U/l; creatinine - 72.4 µmol/l; cholesterol - 4.2 mmol/l; bilirubin - 5.3 µmol/l; FSH - 14.28 mIU/ml; AMH - 4.56 ng/ml; number of antral follicles on the right - 3; number of antral follicles on the left - 2; The doctor’s work experience at the time of the puncture was 13 years. The doctor decided to stimulate superovulation using a short protocol, the starting dose of FSH was 200 IU/day; the total dose of FSH for the entire protocol is 1400 IU, using drugs containing LH.

As a result of mathematical processing of the available data, using an artificial neural network, the following forecast of the results of stimulation of superovulation and follicle puncture was calculated: the probability of obtaining from 0 to 5 oocytes suitable for fertilization is 96.2%, from 6 to 19 oocytes - 3.7% , 20 or more oocytes - 0.1%. The prognosis was considered unfavorable.

Stimulation of superovulation lasted 7 days. HCG was used as a trigger for the final maturation of oocytes. Before the puncture, the following indicators were recorded: E2 concentration - 6513 pmol/ml; progesterone concentration - 1.07 ng/ml; the number of follicles and complexes according to ultrasound data is 3. As a result of the transvaginal puncture, no oocytes suitable for further fertilization were obtained.

To test the method, test samples were randomly formed from the general data array, including 520 implemented IVF protocols with different outcomes. The results of testing the claimed method are presented in Table 2.

table 2
Method testing results Index Quantity for mathematical models Stage I Stage II Stage III Total cases in the test sample 520 520 520 True Positive 399 371 199 False positive result 110 128 82 True negative result 937 922 171 False negative result 114 139 68 Method sensitivity: 77.8% 72.7% 74.5% Specificity of the method 89.5% 87.8 67.6% Forecast accuracy 85.6% 82.9% 71.2%

Advantages of the proposed method:

1. The method can be used to predict the effectiveness of each stage of the IVF protocol (stimulation of superovulation and puncture of follicles, fertilization and cultivation of embryos, transfer of embryos into the uterine cavity), which provides more opportunities for making changes to the protocol in order to increase overall efficiency;

2. The scope of application of the method is not limited to one type of infertility;

3. Parameters with a high level of evidence reliability (1 or 2) of their influence on the outcomes of infertility treatment and the level of confidence of recommendations (A or B) are used as predictors, many of which are already included in existing national and international clinical guidelines and ART protocols, which does not lead to unnecessary increase in the cost of the procedure when introducing the method;

4. The use of artificial neural networks as a mathematical apparatus makes it possible to effectively take into account a large number of heterogeneous predictors and calculate the exact probability value of each category of the predicted parameter;

5. High accuracy, specificity and sensitivity of the method.

Sources of information taken into account:

1. Korsak V.S., Smirnova A.A., Shurygina O.V. ART register of the Russian Association of Human Reproduction. Report for 2019. Reproduction problems. 2021; 27(6):14-29. https://doi.org/10.17116/repro20212706114

2. A method for predicting the outcomes of an IVF and ET program. RF patent, RU2193777C2, declared 06/15/2000, published 11/27/2002.

3. A method for predicting the results of infertility treatment using in vitro fertilization. RF patent, RU2357673C1 declared 10/01/2007, published 06/10/2009.

4. A method for predicting the outcome of in vitro fertilization and embryo transfer. RF patent, RU2474822C1, declared 09/26/2011, published 02/10/2013.

5. A method for predicting the outcomes of assisted reproductive technologies for tubal infertility. RF patent, RU 2672267 C1, declared 11/13/2018, published 11/13/2018.

6. Order of the Ministry of Health of the Russian Federation dated July 31, 2020 No. 803 n “On the procedure for using assisted reproductive technologies, contraindications and restrictions on their use.” URL: http://zdravalt.ru/upload/iblock/114/803n.pdf

7. Clinical recommendations “Female infertility”, 2021 https://mz.mosreg.ru/dokumenty/informaciya/klinicheskie-rekomendacii/02-08-2021-11-12-30-zhenskoe-besplodie

8. Smirnova A.A. The use of estrogens in ART programs (literature review). Reproduction problems. 2015; 21 (4): 48-59.

9. Irene Kwan https://pubmed.ncbi.nlm.nih.gov/33844275/ - affiliation-1, Siladitya Bhattacharya, Andrea Woolner. Monitoring of stimulated cycles in assisted reproduction (IVF and ICSI). Meta-Analysis. Cochrane Database Syst Rev. 2021; 12;4(4):CD005289. doi: 10.1002/14651858.CD005289.pub4.

10. Marcel Zeelenberg, Seger M Breugelmans. The good, bad and ugly of dispositional greed. Review. CurrOpinPsychol. 2022; 46:101323.doi: 10.1016/j.copsyc.2022.101323.

11. The Vienna consensus: report of an expert meeting on the development of ART laboratory performance indicators. ESHRE Special Interest Group of Embryology and Alpha Scientists in Reproductive Medicine. Reprod Biomed Online. 2017 Nov; 35(5):494-510. doi: 10.1016/j.rbmo.2017.06.015. Epub 2017 Aug 4.

12. The Maribor consensus: report of an expert meeting on the development of performance indicators for clinical practice in ART. ESHRE Clinic PI Working Group; VeljkoVlaisavljevic, Susanna Apter, Antonio Capalbo, Arianna D'Angelo, Luca Gianaroli, Georg Griesinger, Efstratios M Kolibianakis, George Lainas, TonkoMardesic, TatjanaMotrenko, Sari Pelkonen, Daniela Romualdi, Nathalie Vermeulen, Kelly Tilleman. Hum Reprod Open. 2021 Jul 3;2021(3):hoab022.doi: 10.1093/hropen/hoab022. eCollection 2021.

13. Tarín JJ, Pascual E, García-Pérez MA, Gómez R, Hidalgo-Mora JJ, Cano A. A predictive model for women's assisted fecundity before starting the first IVF / ICSI treatment cycle.

J. Assist Reprod Genet. 2020;37(1):171-180. doi:10.1007/s10815-019-01642-3.

14. Revised guidelines for good practice in IVF laboratories (2015).ESHRE Guideline Group on Good Practice in IVF Labs, De los Santos MJ, Apter S, Coticchio G, Debrock S, Lundin K, Plancha CE, Prados F, Rienzi L, Verheyen G, Woodward B, Vermeulen N. Hum Reprod. 2016;31(4):685-6. doi: 10.1093/humrep/dew016.

15. D. Salimov, I. Kazakova, E. Mayasina, Y. Buev, T, Lisovskaya. A retrospective study on combination of vitrification/warming protocols from different manufacturers. ESHRE 2022. Hybrid meeting 3-6 July 2022. Hum Reprod. 2022;37, Supp 1.P. 140.

Claims (1)

A method for predicting the effectiveness of multi-stage in vitro fertilization (IVF) programs in the treatment of infertility of various origins, which consists in the fact that at the first stage of the implementation of the IVF protocol, before the start of superovulation stimulation and follicle puncture, through a survey and examination of the patient, as well as with the help of laboratory and instrumental diagnostic protocols collect data from anamnesis, objective examination and the results of clinical, laboratory and instrumental studies, and also select the option of the protocol for stimulating superovulation, the starting and total dose of follicle-stimulating hormone and the need to include drugs containing luteinizing hormone in the protocol, and then, using all the data obtained: doctor's work experience, diagnosis according to ICD No. 10, age, height and weight of the patient, IVF attempt number, presence of human immunodeficiency virus (HIV) in women and men, presence of chronic hepatitis B or C in women and men, presence of human papillomavirus ( HPV) in a woman, a history of arterial hypertension and/or diabetes mellitus types 1, 2, hemoglobin level in the blood, the number of erythrocytes, leukocytes, lymphocytes, platelets in the blood count, fasting glucose concentration, alanine aminotransferase (ALT) level , aspartate aminotransferase (AST), creatinine, cholesterol, bilirubin, concentrations of follicle-stimulating hormone (FSH) and anti-Mullerian hormone (AMH) in the patient’s venous blood, the number of antral follicles on the right and left before stimulation, type of stimulation protocol, starting and total doses of FSH, presence in protocol for stimulation of luteinizing hormone (LH) as predictors, using a predictive mathematical model of the first stage, the probability of obtaining after stimulation and puncture 5 or less oocytes, from 6 to 19 oocytes or 20 or more oocytes suitable for fertilization is calculated, then, upon receipt as a result of the puncture of such oocytes, at the second stage of the IVF protocol, before the fertilization procedure, the qualitative and quantitative characteristics of the sperm intended for fertilization are determined, the proposed options for fertilization and embryo cultivation are selected, and then, using all the data obtained, namely: the type of stimulation protocol, the actual starting and total doses of FSH, the presence of LH stimulation in the protocol, the actual number of days of stimulation, the use of a gonadotropin-releasing hormone (GnRH) or human chorionic gonadotropin (hCG) agonist as a trigger, the concentration of estradiol (E2) and progesterone before puncture in the blood, the number of follicles and complexes before puncture of the ovaries according to ultrasound examination (ultrasound), the number of oocytes obtained as a result of puncture, of which the number of mature oocytes at metaphase stage 2 (M-2), the number of immature oocytes at metaphase stage 1 (M-1) , the number of immature oocytes of the “Germinal Vesular” (GV) stage and the number of degenerative oocytes, the source of sperm and the method of obtaining sperm, the use of cryo-freezing of sperm, the number of sperm after treatment, the percentage of motile sperm of category A+B, the percentage of morphologically normal sperm, use for fertilization of the intracytoplasmic sperm injection (ICSI) procedure, the types of media used for fertilization, crushing and cultivation, the number of 2PN zygotes obtained, in addition to the data of the first stage, as predictors, using a predictive mathematical model of the second stage, calculate the probability of obtaining on the fifth day after fertilization of embryos suitable for transfer into the uterine cavity or cryopreservation, in quantities from 1 to 2, from 3 or more, or the probability of a negative result - 0 embryos, then, upon receipt of such embryos, at the third stage of the implementation of the IVF protocol, before the transfer procedure embryo into the uterine cavity, determine the thickness of the endometrium using ultrasound, decide on the use of a GnRH agonist, and then, using all the data obtained, namely: the number of embryos suitable for transfer, the use of embryo cryopreservation, the thickness of the endometrium before transfer, the presence of complications during the procedure embryo transfer and the use of a GnRH agonist, in addition to the data of the first and second stages, as predictors, using a predictive mathematical model of the third stage, the probability of pregnancy is calculated as a percentage, with each of the three predictive mathematical models implemented as pre-trained on a prepared set data from an artificial neural network with an architecture in the form of a multilayer fully connected Rumelhart perceptron.