Machine learning to optimize ovarian stimulation

Abstract

Systems and methods for implementing machine learning models for ovarian stimulation are described herein. In some variations, the computer-implemented method may include optimizing an ovarian stimulation process, which may include receiving patient-specific data associated with a patient, and predicting egg outcomes for the patient for each of a plurality of treatment options for the ovarian stimulation process based on at least one predictive model and the patient-specific data, where the at least one predictive model is trained using previous patient-specific data associated with a plurality of previous patients.

Description

[Technical field]

[0001] (CROSS REFERENCE TO RELATED APPLICATIONS) This application claims priority to and the benefit of U.S. Provisional Application No. 63 / 256,343, filed October 15, 2021, and U.S. Provisional Application No. 63 / 209,327, filed June 10, 2021, the disclosures of each of which are incorporated herein by reference in their entireties.

[0002] The present invention relates generally to the field of optimizing ovarian stimulation. [Background technology]

[0003] In vitro fertilization (IVF) is a widely known assisted reproductive technology that involves several complex steps, such as ovarian stimulation, egg / oocyte retrieval, fertilization, embryo development, and embryo transfer. Each step of IVF treatment can play a key role in the successful development and transfer of embryos, which can lead to successful pregnancy and potential live birth. For example, collecting as many mature eggs as possible during the ovarian stimulation stage can maximize the probability of fertilization and, therefore, embryo development and transfer, which can ultimately lead to live birth. Thus, in order to obtain viable embryos for transfer, which can lead to successful pregnancy, the number of mature eggs may need to be optimized during the ovarian stimulation stage (e.g., eggs may need to be maximized, or eggs may need to otherwise reach an ideal, preferred, desirable, or / and equivalent outcome).

[0004] In order to optimize the number of mature eggs, a reproductive endocrinologist (RE) may generally prescribe a stimulation protocol to the patient, including follicle-stimulating hormone (FSH) and / or luteinizing hormone (LH). Stimulation with FSH and LH may encourage multifollicular growth. This, in turn, may maximize the number of mature eggs that can be collected from the patient. The RE may assess the patient and prescribe a stimulation protocol for the patient based on the RE's experience. Throughout the ovarian stimulation phase, the RE may monitor the patient's response and may modify, extend, shorten, and / or discontinue the stimulation protocol according to the patient's response. Thus, prescribing a stimulation protocol or modifying and / or discontinuing a stimulation protocol may be subjective, based on the RE's assessment and experience. It is possible that two REs may prescribe different stimulation protocols for the same patient based on their individual experience.

[0005] More recently, some existing methods use models generated from certain data of previous patients to assist the RE in making clinical decisions. For example, some existing methods use models that may output the protocol to be selected for a specific patient, the amount of a medication regimen to be prescribed to a specific patient, a suggestion to modify the protocol being prescribed to a specific patient, or the date on which the ovarian stimulation phase should be terminated. More specifically, these methods may predict clinical decisions that the RE may make to maximize the number of mature eggs. Such existing methods may have some shortcomings. For example, the models trained to predict clinical decisions may be optimizations for the most common, but not necessarily correct, decisions that result in the optimal number of mature eggs.

[0006] Thus, there is an unmet need for new and improved methods to optimize the number of mature eggs for patients while standardizing the process of ovarian stimulation. Summary of the Invention [Means for solving the problem]

[0007] In general, a computer-implemented method for optimizing ovarian stimulation may include receiving patient-specific data associated with a patient and predicting an egg outcome for the patient for each of a plurality of treatment options. The prediction may be based on at least one predictive model trained using previous patient-specific data associated with a plurality of previous patients. The method may further include providing a health care professional with the predicted egg outcome for selecting among the plurality of treatment options. The predicted egg outcome may provide useful information regarding characteristics of egg-related results of treatment options related to ovarian stimulation, such as number of eggs retrieved, number of mature eggs, maturity yield, number of mature eggs, number of eggs fertilized, number of embryos, number of blastocysts, number of usable blastocysts, number of euploid blastocysts, fetal heartbeat (e.g., resulting from the egg), and / or birth rate (e.g., resulting from the egg).

[0008] The one or more predictive models used to predict egg outcome may receive various types of patient-specific data as input and may be trained on such patient-specific data associated with previous patients. For example, suitable patient-specific data may include age, body mass index, ethnicity, infertility diagnosis, previous pregnancy history, previous childbirth history, information related to one or more previous IVF treatments (e.g., data retrieved during ovarian stimulation, number of retrieved eggs, number of mature eggs, number of successfully fertilized eggs, number of blastocysts, number of usable blastocysts, pregnancy outcome, and birth outcome, etc.), and / or one or more treatment variables (e.g., type of medication, type of hormone trigger injection to induce follicular maturation, number of cycles, etc. associated with the patient).

[0009] The various predictive models may be configured to predict egg outcomes associated with various types of treatment options. For example, in some variations, at least one predictive model may be configured to provide a predicted egg outcome associated with each of a plurality of candidate doses of one or more ovarian stimulation medication regimens (e.g., FSH, LH, etc.) administered to the patient. In at least these variations, the patient-specific data may include one or more baseline measurements selected from the group consisting of a measurement of estradiol (E2), a measurement of luteinizing hormone (LH), a measurement of progesterone (P4), a measurement of follicle-stimulating hormone (FSH), a measurement of anti-Mullerian hormone (AMH), and a measurement of antral follicle count (AFC). In some variations, at least one predictive model may be configured to provide a predicted egg outcome associated with each of a first dose of ovarian stimulation medication regimen and a second dose of ovarian stimulation medication regimen higher than the first dose (e.g., a low dose and a high dose, respectively). Additionally or alternatively, in some variations, the at least one predictive model may be configured to provide a predicted egg outcome associated with each of a first dose ratio of the plurality of ovarian stimulation dosage regimens and a second dose ratio of the plurality of ovarian stimulation dosage regimens, the first dose ratio being different from the second dose ratio. For example, the at least one predictive model may be used to predict an egg outcome associated with a first FSH / LH dose ratio and an egg outcome associated with a second FSH / LH dose ratio that is different from the first LSH / LH dose ratio.

[0010] In some variations, when predicting egg outcomes associated with different candidate ovarian stimulation medication doses, the method may include identifying a set of previous patients similar to the patient based on patient-specific data via similarity matching techniques (e.g., incorporating a KNN model), and at least one predictive model may include a predicted dose-response curve generated based on previous patient-specific data associated with the previous patients. The dose-response curve may predict egg outcomes that vary with doses of one or more ovarian stimulation medications administered to the patient. In some variations, the method may further include classifying the patient as one of: (i) a dose-responsive patient having a predicted egg outcome that varies substantially with doses of one or more ovarian stimulation medications administered to the patient, and (ii) a dose-non-responsive patient having a predicted egg outcome that does not vary substantially with doses of one or more ovarian stimulation medications administered to the patient, the classification being based at least in part on the shape of the dose-response curve. In some variations, when a patient is classified as a dose-responsive patient, the method recommends an optimal dose of one or more ovarian stimulation medications to be administered to the patient based on the point of the dose-response curve associated with the predicted maximum egg outcome. Additionally or alternatively, the method may include displaying the predicted dose-response curve on a display to assist a medical professional in selecting among the candidate doses of the ovarian stimulation medication. In some variations, the method may further include displaying a confidence interval around the dose-response curve and / or displaying a cost estimate for administering different doses of the one or more ovarian stimulation medications according to the dose-response curve.

[0011] As another example, in some variations, the at least one predictive model may additionally or alternatively include a first predictive model and a second predictive model for use in assessing a candidate hormone induction date (e.g., a date on which a hormone induction injection is administered to the patient to cause follicular maturation in the patient). In these variations, the method may include predicting via the first predictive model a first egg outcome resulting from administering a hormone induction injection on the first candidate hormone induction date, and predicting via the second predictive model a second egg outcome resulting from administering a hormone induction injection on the second candidate hormone induction date. The first and / or second egg outcomes may be displayed on a display to assist the medical practitioner in selecting among the candidate hormone induction dates.

[0012] For example, the first date may be the current date (the date on which the first predictive model is activated) and the second date may be a future date (e.g., the day after the current date or two days after the current date). In these variations, the patient-specific data may include, for example, follicle size and / or other follicle measurements for the current date, estradiol (E2) levels for the patient for the current date, and / or progesterone (P4) for the patient for the current date. The predicted egg outcomes may include, for example, the number of eggs retrieved, the number of mature eggs, the number of eggs fertilized, and / or the number of usable blastocysts. The first predictive model and / or the second predictive model may include, for example, a regression-based neural network or a linear regression model. The regression-based neural network may be configured to, for example, predict E2 levels for the patient and / or predict follicle measurements at a future date. The linear regression model may be configured to, for example, predict the first and / or second outcomes based, at least in part, on the E2 levels and / or follicle measurements for the patient.

[0013] In general, in some variations, a computer-implemented method for optimizing an ovarian stimulation process may include receiving patient-specific data associated with a patient, identifying a set of previous patients similar to the patient based on the patient-specific data via similarity matching techniques, and generating a predicted dose-response curve based on the previous patient-specific data associated with the previous patients, the dose-response curve predicting ovarian outcomes that vary with doses of one or more ovarian stimulation medications (e.g., FSH, LH) administered to the patient. In at least these variations, the patient-specific data may include one or more baseline measurements selected from the group consisting of a measurement of estradiol (E2), a measurement of luteinizing hormone (LH), a measurement of progesterone (P4), a measurement of follicle-stimulating hormone (FSH), a measurement of anti-Mullerian hormone (AMH), and a measurement of antral follicle count (AFC). In some variations, the at least one predictive model may be configured to provide a predicted egg outcome associated with each of a first dose of the ovarian stimulation medication therapy and a second dose of the ovarian stimulation medication therapy that is higher than the first dose (e.g., a low dose and a high dose, respectively). Additionally or alternatively, in some variations, the at least one predictive model may be configured to provide a predicted egg outcome associated with each of a first dose ratio of the multiple ovarian stimulation medication therapies and a second dose ratio of the multiple ovarian stimulation medication therapies, where the first dose ratio is different from the second dose ratio. For example, the at least one predictive model may be used to predict an egg outcome associated with a first FSH / LH dose ratio and an egg outcome associated with a second FSH / LH that is different from the first LSH / LH dose ratio.

[0014] In some variations, when predicting egg outcomes associated with different candidate ovarian stimulation medication doses, the method may include identifying a set of previous patients similar to the patient based on patient-specific data via similarity matching techniques (e.g., incorporating a KNN model), and at least one predictive model may include a predicted dose-response curve generated based on previous patient-specific data associated with the previous patients. The dose-response curve may predict egg outcomes that vary with doses of one or more ovarian stimulation medications administered to the patient. In some variations, the method may further include classifying the patient as one of: (i) a dose-responsive patient having a predicted egg outcome that varies substantially with doses of one or more ovarian stimulation medications administered to the patient, and (ii) a dose-non-responsive patient having a predicted egg outcome that does not vary substantially with doses of one or more ovarian stimulation medications administered to the patient, the classification being based at least in part on the shape of the dose-response curve. In some variations, when a patient is classified as a dose-responsive patient, the method recommends an optimal dose of one or more ovarian stimulation medications to be administered to the patient based on the point of the dose-response curve associated with the predicted maximum egg outcome. Additionally or alternatively, the method may include displaying the predicted dose-response curve on a display to assist a medical professional in selecting among the candidate doses of the ovarian stimulation medication. In some variations, the method may further include displaying a confidence interval around the dose-response curve and / or displaying a cost estimate for administering different doses of the one or more ovarian stimulation medications according to the dose-response curve.

[0015] In general, in some variations, a method for optimizing an ovarian stimulation process may include receiving patient-specific data associated with a patient, predicting via a first predictive model a first egg outcome resulting from administering a hormone induction injection on a first candidate hormone induction date, and predicting via a second predictive model a second egg outcome resulting from administering a hormone induction injection on a second candidate hormone induction date, the hormone induction injection configured to cause follicular maturation in the patient. The first and / or second egg outcomes may be displayed on a display to assist a medical practitioner in selecting among the candidate hormone induction dates.

[0016] For example, the first date may be the current date (the date on which the first predictive model is activated) and the second date may be a future date (e.g., the day after the current date or two days after the current date). In these variations, the patient-specific data may include, for example, follicle size and / or other follicle measurements for the current date, estradiol (E2) levels for the patient for the current date, and / or progesterone (P4) for the patient for the current date. The predicted egg outcome may include, for example, the number of eggs retrieved and / or the number of mature eggs. The first predictive model and / or the second predictive model may include, for example, a regression-based neural network or a linear regression model. The regression-based neural network may be configured, for example, to predict E2 levels for the patient and / or to predict follicle measurements at a future date. The linear regression model may be configured, for example, to predict the first and / or second outcomes based, at least in part, on the E2 levels and / or follicle measurements for the patient.

[0017] In general, the computer-implemented method may include receiving patient-specific data associated with a patient, and predicting at least one egg outcome for the patient based on at least one predictive model and the patient-specific data. The patient-specific data may include follicle sizes categorized into a bin from a plurality of predefined bins, each bin of the plurality of predefined bins may be associated with a distinct range of follicle sizes, and the at least one predictive model may be trained using previous patient-specific data associated with a plurality of previous patients. In some variations, the patient-specific data may include an E2 level for the patient, and the method may further include predicting at least one egg outcome based on the E2 level. In some variations, the method may further include displaying the predicted egg outcome on a display (e.g., to a medical professional).

[0018] The plurality of bins may be associated with any suitable set of follicle size ranges. For example, in some variations, at least one bin of the plurality of bins may be associated with at least a first follicle size range of about 10 mm or less, a second follicle size range of about 11 mm to 13 mm, a third follicle size range of about 14 mm to 15 mm, a fourth follicle size range of about 16 mm to 17 mm, a fifth follicle size range of about 18 mm to 19 mm, a sixth follicle size range above about 20 mm, or any suitable subset thereof.

[0019] Generally, in some variations, the computer-implemented method may include receiving patient-specific data associated with a patient, predicting an egg outcome for the patient based on at least one predictive model and the patient-specific data for each of a plurality of days, where the at least one predictive model is trained using previous patient-specific data associated with a plurality of previous patients, and displaying a trend of the predicted egg outcome over the plurality of days on a display. The predicted egg outcome may include, for example, at least one of the number of eggs retrieved and the number of mature eggs. Additionally or alternatively, the predicted egg outcome may include at least one of the following: maturity yield, number of mature eggs, number of fertilized eggs, number of embryos, number of blastocysts, number of usable blastocysts, number of euploid blastocysts, fetal heartbeat, and birth rate.

[0020] In general, a method of treatment may include providing patient-specific data associated with a patient; receiving predicted egg outcomes associated with each of a plurality of treatment options, where at least one predicted egg outcome is generated using at least one predictive model according to any of the methods described above; selecting a treatment option based on the predicted egg outcomes; and administering an ovarian stimulation medication in accordance with the selected treatment option. [Brief description of the drawings]

[0021] [Figure 1] FIG. 1 illustrates exemplary therapeutic decision points during ovarian stimulation.

[0022] [Diagram 2] FIG. 2 illustrates an overview of an exemplary variation of a system for assisting therapy during ovarian stimulation.

[0023] [Diagram 3] FIG. 3 is an exemplary variation of a linear regression model and a neural network to predict egg outcome for a patient.

[0024] [Figure 4] FIG. 4 is an exemplary variation of a display showing an electronic medical record (EMR) connected to a RE application via a browser plug-in.

[0025] [Diagram 5] FIG. 5 illustrates an exemplary variation of the RE application as displayed on the display.

[0026] [Figure 6] FIG. 6 is a flow diagram illustrating an exemplary variation of a method for predicting egg outcome.

[0027] [Figure 7A] FIG. 7A illustrates predicted egg outcomes for patient 1 for varying levels of FSH dose.

[0028] [Figure 7B] FIG. 7B illustrates predicted egg outcomes for patient 2 for varying levels of FSH dose.

[0029] [Figure 8] FIG. 8 illustrates the implementation of a similarity model to identify the optimal baseline FSH dose for a patient in terms of best egg outcome.

[0030] [Figure 9A] FIG. 9A illustrates predicted egg outcomes for Patient 1 by day of the cycle.

[0031] [Figure 9B] FIG. 9B illustrates predicted egg outcomes for Patient 2 by day of the cycle.

[0032] [Figure 10] FIG. 10 illustrates various predicted egg outcomes for different cycle days by developing a multi-output regression model.

[0033] [Figure 11A] FIG. 11A illustrates predicted egg outcomes for patients by developing an independent regression model.

[0034] [Figure 11B] FIG. 11B illustrates the predicted egg outcomes for patients by developing an independent regression model.

[0035] [Figure 11C] FIG. 11C illustrates the predicted egg outcomes for patients by developing an independent regression model.

[0036] [Figure 12A] FIG. 12A illustrates prediction of follicle number and size at future dates by deploying a neural network.

[0037] [Figure 12B] FIG. 12B illustrates prediction of egg outcome at a future date using a combination of techniques.

[0038] [Figure 13] FIG. 13 is a flow diagram of an exemplary method of treatment using the machine learning models described herein.

[0039] [Figure 14A] 14A-14C illustrate examples of statistically significant variables that may predict egg outcome for a patient. [Figure 14B] 14A-14C illustrate examples of statistically significant variables that may predict egg outcome for a patient. [Figure 14C] 14A-14C illustrate examples of statistically significant variables that may predict egg outcome for a patient.

[0040] [Figure 14D] FIG. 14D illustrates another example of statistically significant variables that may predict egg outcome for a patient.

[0041] [Figure 15] 15A-15C illustrate an example of a similarity model described herein being implemented to determine whether a patient is an FSH dose-responsive or FSH dose-non-responsive patient.

[0042] [Figure 16] 16A-16D illustrate an example of implementing one or more models described herein to predict induction dates for patients to maximize egg outcomes.

[0043] [Figure 17] 17A-17C illustrate another example of implementing one or more models described herein to predict induction dates for patients to maximize egg outcome.

[0044] [Figure 18] FIG. 18 illustrates the prediction of egg outcome on the day and the day after stimulation by implementing two independent regression models.

[0045] [Figure 19] FIG. 19 illustrates the recommendations given to the RE based on predicted egg outcome trends and patient classification after the final challenge injection has been administered.

[0046] [Figure 20] FIG. 20 illustrates the generation of an exemplary curve relating egg outcome to dosage and determining whether a patient is dose-responsive or dose-non-responsive.

[0047] [Figure 21A] FIG. 21A illustrates an exemplary curve relating egg outcome to dosage for dose-responsive patients.

[0048] [Figure 21B]FIG. 21B illustrates an exemplary curve relating egg outcome to dosage for dose-non-responsive patients.

[0049] [Figure 22A] 22A and 22B illustrate an example workflow displayed on a display via the RE application that the RE may view to aid in treatment during ovarian stimulation. [Figure 22B] 22A and 22B illustrate an example workflow displayed on a display via the RE application that the RE may view to aid in treatment during ovarian stimulation.

[0050] [Diagram 23] FIG. 23 illustrates an exemplary patient dashboard that may be displayed on the display of a suitable computing device.

[0051] [Figure 24] FIG. 24 illustrates an exemplary starting dose page, as displayed on the display of a suitable computing device.

[0052] [Diagram 25] FIG. 25 illustrates an exemplary invitation page that may be displayed on the display of a suitable computing device.

[0053] [Figure 26] FIG. 26 is a flow diagram illustrating an exemplary variation of a method for prescribing a stimulation protocol to a patient.

[0054] [Figure 27] FIG. 27 is a flow diagram illustrating an exemplary variation of a method for prescribing a dosage (eg, an FSH dosage) to a patient of interest.

[0055] [Figure 28]FIG. 28 is a flow diagram illustrating an exemplary variation of a method for assisting in the RE determination of the final induction selection date for a patient of interest.

[0056] [Figure 29] FIG. 29 is a flow diagram illustrating an exemplary variation of a method for varying FSH and / or LH dosage for a patient of interest.

[0057] [Diagram 30] FIG. 30 is a flow diagram illustrating an exemplary variation of a method for classifying a patient of interest as at risk for Ovarian Hyperstimulation Syndrome (OHSS). DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0058] Detailed Description Non-limiting examples of various aspects and variations of the present invention are described herein and illustrated in the accompanying drawings.

[0059] In vitro fertilization (IVF) is a complex assisted reproductive technology that involves fertilization of eggs outside the body in a laboratory setting. A typical IVF cycle includes an ovarian stimulation phase. The goal during the ovarian stimulation phase is to collect as many mature eggs as possible. During this phase, the patient may be prescribed medications and / or injections that stimulate the ovaries and encourage multifollicular growth. Each follicle may contain an egg that can potentially mature. A final induction injection given to the patient containing hormones that can mature the growing follicles may mark the end of the ovarian stimulation phase.

[0060] Following ovarian stimulation, a reproductive endocrinologist (RE) and / or physician may retrieve eggs from the patient's ovaries (e.g., egg retrieval phase). The eggs may then be fertilized in a laboratory setting (e.g., fertilization phase). Once fertilized, the embryos may begin to develop (e.g., embryo development phase). The RE may then select the most viable embryos for embryo transfer (e.g., embryo transfer phase).

[0061] During each phase of IVF treatment, there may be a risk of egg or embryo loss due to factors such as failure to mature (e.g., overmaturity, immaturity, etc.), failure to fertilize, developmental arrest, or detection of genetic abnormalities. Viable embryos that remain unaffected by these factors may lead to successful pregnancy and, therefore, live birth. Thus, to maximize the probability of live birth, the ovarian stimulation phase may need to be optimized. For example, by optimizing the number of mature eggs during the ovarian stimulation phase (e.g., maximizing the number of mature eggs, or otherwise obtaining the ideal / most suitable / desired number of mature eggs), the probability of live birth may be maximized.

[0062] Mature eggs are developed from follicles. Follicles are fluid sacs in the ovaries that contain developing eggs. Typically, during a regular menstrual cycle, several follicles (each of which may contain an egg) may develop. However, usually only a single dominant follicle reaches maturity. The dominant follicle may develop to a stage where it is ready to release a mature egg. This usually occurs at about days 12-14 of the monthly menstrual cycle. During the ovarian stimulation phase of the IVF cycle, patients may be prescribed hormones to encourage multifollicular growth so that a large number of mature eggs can be retrieved. The combination of drugs, dosages, and / or injections prescribed to encourage multifollicular growth may constitute a stimulation protocol.

[0063] Some commonly used stimulation protocols include antagonist protocols, long protocols, and flare protocols. Each of these stimulation protocols may share three common features: (1) the use of gonadotropins, such as follicle-stimulating hormone (FSH) and luteinizing hormone (LH), to stimulate multifollicular growth, (2) the use of gonadotropins to release hormone (GnRH) agonists or antagonists to suppress early ovulation, and (3) a final hormone trigger injection to help eggs undergo meiosis and prepare for timely release.

[0064] Traditionally, the RE assesses the patient and prescribes a stimulation protocol for the patient. FIG. 1 illustrates certain decision points faced by the RE during the ovarian stimulation phase. As shown in FIG. 1, at the beginning of an IVF cycle (e.g., planning and preparation phase 102), prior to the ovarian induction phase 104, the RE may make a diagnosis and recommend an IVF cycle. In some variations, the patient undergoes a fertility test, which may indicate basal levels hormones in the patient. Based on the patient's pregnancy history and the results of the fertility test, the RE may make a diagnosis regarding the patient. This may include whether IVF treatment should be recommended for the patient.

[0065] Once IVF treatment and IVF cycles are recommended for the patient, the treatment may proceed to an ovarian stimulation phase. During the ovarian stimulation phase, the RE may be faced with multiple decisions that may affect the outcome of the IVF cycle and the health of the patient. One such decision may include determining the stimulation protocol (e.g., 112a) to be prescribed for the patient. For example, the RE may determine the drugs to be used and the starting dosage of the drugs. After the stimulation protocol (e.g., 112a) is selected, the RE may periodically monitor the patient and assess the patient's response to the stimulation protocol. Based on the patient's response, the RE may modify the stimulation protocol (e.g., 112b) and / or discontinue the IVF cycle. This may be the next clinical decision that the RE may need to make during the ovarian stimulation phase. For example, if the patient's response to the stimulation protocol is lower than expected, the RE may increase the dosage of the gonadotropin. Conversely, if the patient's response to the stimulation protocol is higher than expected, the RE may decrease the dosage of the gonadotropin. Additionally, if the patient's response to the stimulation protocol is too high or too low, the RE may abort the IVF cycle. Final clinical decisions during the ovarian stimulation phase may include determining when to prescribe the patient a final induction injection (e.g., 112c) that helps the eggs undergo meiosis and prepare for release. The final induction injection may mature the growing follicles. The day that the final induction injection (e.g., 112c) is prescribed may be critical to the outcome of the IVF cycle.

[0066] After the final induction injection is administered, eggs may be retrieved during a fertilization phase 106. During the fertilization phase 106, the retrieved eggs may be fertilized. The embryos may be analyzed as they develop to determine one or more viable embryos for transfer. A viable embryo may be transferred during an embryo transfer phase 108, which may subsequently lead to a pregnancy 110.

[0067] Traditionally, clinical decisions made during the ovulation induction phase 102 may be based on the RE's experience and the RE's assessment of the patient. However, these clinical decisions may be subjective and specific to each individual RE's experience. It is possible that two different REs may make different decisions regarding the same patient. For example, two different REs may prescribe different stimulation protocols for the same patient. Similarly, one RE may choose to discontinue the IVF protocol based on the patient's response to the stimulation protocol, while a different RE may choose to modify the stimulation protocol based on the patient's response to the stimulation protocol. These decisions may be highly subjective, making it difficult to standardize the ovarian stimulation phase.

[0068] Some existing methods use data from previous patients to generate one or more models that can predict clinical decisions for RE. Such models can replace decisions that the RE can make. This can be difficult because replacing or superseding the clinical judgment of the RE may not always result in successful IVF outcomes. For example, for complex or rare cases, available data (e.g., data used to generate the model) may be limited. As a result, the predictions made by the model may not be accurate. However, a RE with several years of training and experience may be better equipped to make clinical decisions for such complex and rare cases, resulting in more favorable patient outcomes. Furthermore, other factors that may not be evident in the available data, such as clinic policies, clinic equipment, and variability in the patient's situation, such as unknown genetic diseases that may lead to unexplained conception problems, may also be essential to making the best clinical decision for the patient. In addition, existing methods and models use a black box approach to predict clinical decisions for RE. Such approaches typically do not produce interpretable results.

[0069] Therefore, what is needed is a technique that can enhance or further inform the RE's decisions, as opposed to replacing them. The techniques described herein develop and implement machine learning models (also referred to herein as "predictive models") to enhance the clinical decisions made by the RE. These machine learning models can be trained on diverse and high-quality data. Instead of replacing the RE's decisions, these machine learning models can, for example, provide a useful second opinion and / or enhance the RE's decisions. Furthermore, for rare or complex cases, machine learning models may be used in conjunction with the RE's decisions to provide more accurate predictions that may result in successful IVF outcomes. Predictions are generated so that the results can be easily interpreted by the RE.

[0070] The techniques described herein use machine learning to predict egg outcomes as opposed to clinical decisions for RE. Egg outcomes may be used as additional information to augment the decision for RE. Some non-limiting examples of egg outcomes may include number of eggs, number of mature eggs, maturity yield, number of mature eggs, number of fertilized eggs, number of embryos, number of blastocysts, number of euploid blastocysts, fetal heart rate, chemical pregnancy rate, birth rate, birth rate accumulated across consecutive transfers, combinations thereof, and / or the like.

[0071] In some variations, the computer-implemented method may include receiving patient-specific data, which may be associated with a plurality of previous patients. The patient-specific data may include data related to ovarian stimulation. A predictive model may be trained based on the patient-specific data. The computer-implemented method may include receiving first patient-specific data associated with a first patient undergoing IVF treatment (e.g., an IVF cycle). The computer-implemented method may include predicting an egg outcome for the first patient based on implementation of a predictive model for the first patient-specific data associated with the first patient.

[0072] System Overview FIG. 2 illustrates an overview of an exemplary variation of a system 200 for assisting RE during the ovarian stimulation phase. The system 200 may access and / or retrieve data from a trusted electronic medical record (EMR) 204. The controller 206 may implement a machine learning model using data retrieved from the EMR 204. The machine learning model may predict an egg outcome for the patient. Predictions from the machine learning model may be transmitted to a RE application 208 implemented on a suitable computing device. In some variations, the RE application 208 may interface with the EMR 204. In some variations, predictions from the machine learning model may be stored in a database 216. In some variations, these predictions may be accessed from the database 216 at future times to further improve the accuracy of the machine learning model. The RE may access the predictions on the RE application 208 to augment its clinical decisions.

[0073] EMR204 is an eIVF system from Mellowood Medical et al. TM Patient Portal, ARtisan TM Fertility Portal, Babysentry TM Management system, EPIC TM Patient Portal, IDEAS TM , or any suitable electronic medical record management software, or other trusted database. In some variations, the EMR 204 may be associated with a specific clinic. In such variations, the EMR 204 may be configured to interface with one or more servers associated with the specific clinic. In some variations, the EMR 204 may be hosted on a cloud-based platform (e.g., Microsoft Azure®, Amazon® Web Services, IBM® Cloud Computing, etc.).

[0074] In some variations, the EMR 204 may be associated with a specific clinic. For example, an EMR 204 from a specific clinic may not be shared with other hospitals, pharmacies, practitioners, etc. Additionally or alternatively, the EMR 204 may be configured to access a database associated with each clinic. The EMR 204 may automatically extract relevant information from a patient's chart and match it against a database of de-identified medical records. Thus, relevant data across several entities (e.g., clinics, hospitals, pharmacies, practitioners, etc.) may be extracted from the EMR 204 without compromising patient privacy (e.g., by maintaining Health Insurance Portability and Accountability Act regulations).

[0075] The EMR 204 may be accessed via a computing device. Some non-limiting examples of computing devices include computers (e.g., desktops, personal computers, laptops, etc.), tablets and e-readers (e.g., Apple iPad®, Samsung Galaxy® Tab, Microsoft Surface®, Amazon Kindle®, etc.), mobile devices and smartphones (e.g., Apple iPhone®, Samsung Galaxy®, Google Pixel®, etc.), and the like. For example, the EMR 204 may be stored on a memory associated with the computing device. Alternatively, the EMR 204 may be accessed online through a web browser (e.g., Google®, Mozilla®, Safari®, Internet Explorer®, etc.) rendered on the computing device. In yet another alternative variation, the EMR 204 may be stored on a third party database that may be accessed via the computing device.

[0076] Patient-specific data may be extracted from the EMR 204. Patient-specific data extracted from the EMR 204 may refer to (1) data associated with one or more patients, which may include record descriptions, content, values, combinations thereof, and / or the like, and / or (2) metadata that provides context for the data. For example, patient-specific data extracted from the EMR 204 may include one or both of data and metadata associated with a patient record.

[0077] Some non-limiting examples of patient-specific data that may be extracted from the EMR 204 include: (a) patient information such as age, body mass index, race, ethnicity, infertility diagnosis or cause, previous IVF history, previous uterine surgery information, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth from natural conception, and / or the like; (b) data related to previous IVF cycles and / or treatments such as baseline measurements of drugs and hormones during previous IVF cycles and / or treatments, stimulation protocol, response to stimulation protocol, number of eggs retrieved, number of mature eggs, number of eggs successfully fertilized, pregnancy outcome, and birth outcome; (c) data related to previous IVF cycles and / or treatments such as estradiol (E2), luteinizing hormone (LH), luteinizing hormone (V ... (d) treatment variables such as type and brand of medication (e.g., for gonadotropins (FSH and LH), GnRH agonist and antagonist and final induction injection), amount of medication dosage (e.g., starting dosage, ending dosage, daily dosage, and total drug amount), days in cycle, and / or the like; (e) response to stimulation protocol such as daily measurements of follicular measurements (e.g., follicle number and size), E2 and P4 levels, and / or the like.

[0078] In some variations, the patient-specific data may also include ultrasound images of follicles, uterus, etc. The ultrasound images may provide information such as follicle count, follicle size, presence of fibroids in the uterus, etc.

[0079] The controller 206, communicatively coupled to the EMR 204, may extract patient-specific data (e.g., from the EMR 204). In some variations, the controller 206 may include one or more servers and / or one or more processors running on a cloud platform (e.g., Microsoft Azure®, Amazon® Web Services, IBM® Cloud Computing, etc.). The servers and / or processors may be any suitable processing device configured to run and / or execute a set of instructions or code and may include one or more data processors, image processors, graphic processing units, digital signal processors, and / or central processing units. The servers and / or processors may be, for example, general purpose processors, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), and / or the like.

[0080] In some variations, the controller 206 may include a processor (e.g., a CPU). The processor may be any suitable processing device configured to run and / or execute a set of instructions or code, and may include one or more data processors, image processors, graphic processing units, physical processing units, digital signal processors, and / or central processing units. The processor may be, for example, a general-purpose processor, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), and / or the like. The processor may be configured to run and / or execute application processes and / or other modules, processes and / or functions associated with the system, and / or networks associated therewith. The underlying device technologies may be provided in various component types (e.g., MOSFET technologies such as complementary metal oxide semiconductor (CMOS), bipolar technologies such as emitter coupled logic (ECL), polymer technologies (e.g., silicon conjugated polymer and metal conjugated polymer metal structures), composite analog and digital, and / or the like.

[0081] The controller 206 may train one or more machine learning models using the patient-specific data extracted from the EMR 204. The machine learning models may predict egg outcomes for the patient. For example, the patient-specific data may be used to train a machine learning model to select a stimulation protocol for the patient. The machine learning model may predict egg outcomes for various stimulation protocols (e.g., antagonist protocol, long protocol, flare protocol, etc.) for the patient. This may assist the RE's clinical decision to determine which stimulation protocol to prescribe for the patient. In a similar manner, the patient-specific data may be used to train a machine learning model to determine the starting dose of FSH to be prescribed for the patient. The machine learning model may predict egg outcomes for various starting doses of FSH. This may assist the RE's clinical decision to determine the starting dosage of FSH to be prescribed for the patient. In yet another alternative variation, the patient-specific data may be used to train a machine learning model to determine the day on which the patient may be administered a final induction injection. The machine learning model may predict egg outcomes for different days on which the final induction injection may be administered. The RE may use this information to determine the date when the final challenge injection should be administered for the patient.

[0082] In some variations, the machine learning model may augment the clinical decision of the RE, but does not predict the clinical decision itself. For example, the output of the machine learning model may not be a prediction of the stimulation protocol to be explained, a prediction of the modification to be made to the stimulation protocol, or a prediction of the decision when the final induction injection may be administered. Rather, in some variations, the output of the machine learning model is an egg outcome, and the machine learning model gives the RE the ability to use the egg outcome for various scenarios to make a more informed decision. For example, such a machine learning model, whose output is an egg outcome, may thus assist the RE in verifying its individual assessment of its clinical decision, rather than completely replacing its decision. Alternatively, in some variations, the machine learning model may output one or more clinical decisions itself.

[0083] In some variations, the machine learning model may be a series of regression models. As a non-limiting example, the machine learning model may be a series of regression models for selecting a stimulation protocol for a patient, determining the starting dose of FSH to be prescribed to the patient, determining the day the patient may receive a final induction injection, and / or the like. In some variations, the regression model may be a linear regression model. Additionally or alternatively, the machine learning model may be a feed-forward neural network and / or a recurrent neural network. As described above, the machine learning model may predict egg outcomes for a patient. FIG. 3 is an exemplary variation of a linear regression model 324a and a neural network 324b. The linear regression model 324a and the neural network 324b may predict egg outcomes (e.g., number of eggs, number of mature eggs, etc.) for various patients, as seen in graph 326. Graph 326 shows predicted eggs versus actual eggs retrieved from various patients. As seen in graph 326, the predicted egg outcomes are close to the actual egg outcomes. As an example, the neural network 324b may be a feedforward neural network, for example a three-layer network with ReLU activation and dropout, trained with a gradient descent optimizer to minimize the mean squared error.

[0084] In some variations, the accuracy of the machine learning model may improve as more patient-specific data becomes available to train the machine learning model. For example, referring to Table 1, before the ovarian stimulation phase, after the patient is approved for IVF, the patient-specific data available to train the model may include diagnosis, age, BMI, FSH and estradiol amount associated with various previous patients. The R-squared value for predicting egg outcome at this stage shows a value of 0.28, implying that predicted egg outcome and actual egg outcome may not be very similar. However, during the next stage, before ovarian stimulation, the patient-specific data may additionally include baseline AFC and baseline AMH (e.g., in addition to age, BMI, FSH and estradiol amount, etc.). The R-squared value for predicting egg outcome at this stage shows a value of 0.45. During ovarian stimulation, in addition to age, BMI, amount of FSH and estradiol, baseline AFC, and baseline AMH, patient-specific data may also include type of stimulation protocol, cycle days, and amount of medication dosage associated with various previous patients. The R-squared value for predicting egg outcome at this stage shows a value of 0.50. After ovarian stimulation and before egg retrieval, patient-specific data may additionally include number of follicles. The R-squared value for predicting egg outcome at this stage shows a value of 0.70. Thus, as can be seen in Table 1, prediction improves as more patient-specific data becomes available. [Table 1-1] [Table 1-2]

[0085] In some variations, the output of the machine learning model may be stored in the database 216. More specifically, egg outcomes for the patient and patient-specific data associated with the patient (e.g., patient information, data related to previous IVF cycles and / or treatments, baseline measurements, treatment variables, response to stimulation protocols, etc.) may be stored in the database 216. This data may be incorporated and update the training data of the machine learning model. That is, in addition to already existing patient-specific data, the machine learning model can be trained on data associated with patients currently undergoing IVF treatment. This, in turn, may improve the accuracy of predictions for the machine learning model.

[0086] As discussed above, the output of the machine learning model and the patient-specific data associated with each patient may be stored in a database 216. The database 216 may be accessed at any suitable time to improve the machine learning model implemented by the controller 206. In some variations, the database 216 may be stored in a memory device such as a random access memory (RAM), a memory buffer, a hard drive, an erasable programmable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), a read-only memory (ROM), a flash memory, and the like. In some variations, the database 216 may be stored on a cloud-based platform such as Amazon Web Services®.

[0087] The output of the machine learning model may be accessible to the RE via application software executable on a computing device (referred to herein as the “RE application” 208). In some variations, the computing device may be the computing device through which the EMR 204 is accessed. More specifically, the RE application 208 may be connected to the EMR 204 to extract the patient-specific data and the output of the machine learning model in real time. In some variations, the RE application 208 may be connected to the EMR 204 through a website portal connection. Additionally or alternatively, the RE application 208 may be connected to the EMR 204 as a direct plug-in (e.g., a browser plug-in) to the EMR 204. In some variations, the RE application 208 may be connected to the EMR 204 through a computing device (discussed above) through which the EMR 204 may be accessed. For example, the RE application 208 may be connected to the computing device through an application programming interface (API), which may in turn connect the computing device to the EMR 204. Additionally or alternatively, the RE application 208 may be rendered on a computing device through a web browser (e.g., Google®, Mozilla®, Safari®, Internet Explorer®, etc.). The web browser may also connect the RE application 208 to the EMR 204.

[0088] In some variations, the RE application 208 (e.g., a web app, a desktop app, a mobile app, etc.) may be pre-installed on the computing device. Alternatively, the RE application 208 may be rendered on the computing device in any suitable manner. For example, in some variations, the RE application 208 (e.g., a web app, a desktop app, a mobile app, etc.) may be downloaded onto the computing device from a digital distribution platform, such as an app store or application store (e.g., Chrome® Web Store, Apple® Web Store, etc.). Additionally or alternatively, the computing device may render a web browser (e.g., Google®, Mozilla®, Safari®, Internet Explorer®, etc.) on the computing device. The web browser may include a browser extension, browser plug-in, etc. that may render the RE application 208 on the computing device. In yet another alternative variation, the browser extension, browser plug-in, etc. may include installation instructions for installing the RE application 208 on the computing device.

[0089] The output of the machine learning model may be accessed in real-time by any user (e.g., the patient, the RE, other clinicians, etc.) via the RE application 208. For example, the RE may access the output of the machine learning model in real-time via the RE application 208. In addition, the RE may access, review, and / or edit patient-specific data associated with the patient in real-time through the EMR 204, which is connected to the RE application 208. FIG. 4 is an example variation of a display 432 displaying an EMR 404a. The EMR 404a is connected to the RE application 408 via a browser plug-in 436.

[0090] In FIG. 4, the display 432 may include an EMR 404a associated with a patient, “Jane Smith Doe” 434a, undergoing IVF treatment. In some variations, the EMR 404a may also include information associated with a sperm donor (e.g., “John Smith” 434b). The display 432 may include a browser plug-in 436 (e.g., a widget, a radio button, etc.) that may connect the EMR 404a to the RE application 408. For example, clicking and / or pressing the browser plug-in 436 may open a pop-up window of the RE application 408. The RE application 408 may include one or more outputs of the machine learning model. For example, the RE application 408 may include a prediction of egg outcome for the patient, “Jane Smith Doe” 434a. In this manner, the RE can access the patient-specific data (e.g., EMR 404a) associated with “Jane Smith Doe” 434a and simultaneously access the machine learning model outcomes simply by clicking and / or pressing the browser plug-in 436.

[0091] FIG. 5 illustrates an example variation of the RE application 408 displayed on a display (e.g., display 432 in FIG. 4). Clicking and / or pressing the browser plug-in 436 may open the RE application 408 as a pop-up. The RE application 408 may include patient-specific data associated with “Jane Smith Doe” 434a. In some variations, the RE application 408 may include baseline measurements 552 such as levels of FSH, AMH, AFC, estradiol, etc. In some variations, the RE application may enable the RE to select a preferred amount of inhibitor 554a, a preferred amount of stimulator 554b, a preferred amount of hormone 554c in the final induction injection, and / or the day the final induction injection should be administered. Such selections may be made prior to the start of the ovarian stimulation phase or during the ovarian stimulation phase.

[0092] For example, the RE may click on the "+" button located below the inhibitor 554a to increase the amount of the inhibitor. Alternatively, the RE may click on the "-" button located below the inhibitor 554a to decrease the amount of the inhibitor. By modifying the measurements of the inhibitor 554a, the stimulant 554b, and the hormone 554c, the RE may view, in real time, the egg outcome 556 for the various measurements. More specifically, the RE may be able to view, in real time, the extent to which the egg outcome 556 may be modified for the various measurements of the inhibitor 554a, the stimulant 554b, and the hormone 554c. Additionally or alternatively, the RE may be able to change the day on which the final induction injection is to be administered (e.g., by clicking on the toggle button above the hormone 554c). Changing the day may change the egg outcome 556. In some variations, the RE application 408 may also display a graph 558 illustrating the egg outcome for various days of the menstrual cycle. This may provide the RE with the necessary information to determine the day the final challenge injection should be administered.

[0093] In some variations, the RE may close the RE application 408 at any time. The RE application 408 may be reopened at any time. When reopened, the graph 558 may be updated to reflect the latest egg outcomes. As discussed above, the RE may also modify the patient-specific data and determine the extent to which the modifications may affect the prediction and final outcome.

[0094] Exemplary Methods for Predicting Egg Outcome 6 is a flow diagram illustrating an example variation of a method 600 for predicting egg outcome. At 602, the method includes receiving patient-specific data from an EMR. A controller (e.g., controller 206 in FIG. 2) may receive the patient-specific data from an EMR (e.g., EMR 204 in FIG. 2). The patient-specific data may include a large amount of data associated with previous patients. For example, the patient-specific data may include patient information, ultrasound images, data related to previous IVF cycles and / or treatments, baseline measurements, treatment variables, responses to stimulation protocols, etc. associated with previous patients.

[0095] The patient-specific data may be used to train a machine learning model. In some variations, the machine learning model may be a linear regression model. Alternatively, the machine learning model may include a neural network, such as a feedforward neural network, a recurrent neural network, or the like. Alternatively, the machine learning model may include a K-nearest neighbor (KNN) model.

[0096] When a patient undergoes IVF treatment, patient-specific data associated with the patient may be received at the controller. At 604, the method may include implementing a machine learning model for the patient undergoing IVF treatment. This may include, for example, implementing a linear regression model and / or a neural network that is trained on the patient-specific data obtained from the EMR. The linear regression model and / or neural network may be implemented for the data associated with the patient. For example, the linear regression model and / or neural network may be implemented for the patient's age, race, body mass index, previous IVF history, pregnancy, parity, etc.

[0097] At 606, implementing the machine learning model may cause the method 600 to predict an egg outcome. The egg outcome may include number of eggs, number of mature eggs, maturity yield, number of mature eggs, number of fertilized eggs, number of embryos, number of blastocysts, number of euploid blastocysts, fetal heart rate, chemical pregnancy rate, birth rate, combinations thereof, and / or the like. For example, the machine learning model may predict egg outcomes for various stimulation protocols for the patient. Similarly, the machine learning model may predict egg outcomes for various baseline dosages of medication therapy for the patient. In a similar manner, the machine learning model may predict egg outcomes for various days on which a final induction injection for the patient should be administered. Additionally or alternatively, the machine learning model may predict egg outcomes for various amounts of hormones to be included in the final induction injection.

[0098] As the patient undergoes ovarian stimulation, the patient's response to the stimulation protocol may be observed. Machine learning models may be implemented to take into account the patient's response. The stimulation protocol may be discontinued and / or modified based on the patient's response. This may include varying the dosage of the medication therapy based on the degree to which the patient responds to a baseline amount of the medication therapy.

[0099] Some non-limiting examples of machine learning models and their implementations are further described below.

[0100] Stimulus protocol selection model In some variations, the stimulation protocol selection model may be trained on patient information such as age, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth history from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), body mass index, race, ethnicity, ultrasound images, and / or the like. In some variations, the training data may additionally include contextual information such as the cost of IVF treatment, cost constraints for the patient, other patient-specific needs, etc.

[0101] In some variations, the stimulation protocol selection model may be a linear regression model. In some variations, the stimulation protocol selection model may implement a K-nearest neighbor (KNN) technique. The stimulation protocol selection model may incorporate patient-specific data (e.g., patient information, baseline measurements, context information) discussed above and predict an egg outcome for the patient. The egg outcome may be the number of eggs and / or the number of mature eggs that may be retrieved from the patient.

[0102] Typically, the RE may make a clinical decision regarding the type of stimulation protocol to prescribe to the patient. If the patient is an average responder, the RE will most likely prescribe an antagonist protocol. In some variations, the RE may know that the patient is an average responder if the current IVF cycle is not the first IVF cycle with the RE. However, in variations where the current IVF cycle is the first IVF cycle, the RE may assess the type of responder the patient may be based on the patient information. If the patient has previously had a low response (e.g., in a previous IVF cycle) or is expected to have a low response, the RE may prescribe an alternative protocol.

[0103] To enhance clinical decisions related to the type of stimulation protocol, the stimulation protocol selection model may predict egg outcomes for a specific protocol (e.g., antagonist protocol). If the egg outcome is below a first threshold, the patient may be classified as a low-responder. However, if the egg outcome is between a range that may be higher than the first threshold, the patient may be classified as an intermediate-responder. In contrast, if the egg outcome is higher than a second threshold (e.g., higher than the highest value of the range classified as an intermediate-responder), the patient may be classified as a high-responder. In some variations, the RE may select the first threshold, the range, and / or the second threshold. The RE may thus customize the classification to tailor the stimulation protocol to a specific patient. For example, the RE may select a threshold of less than 5 eggs to indicate a low response in some patients and less than 8 eggs to indicate a low response in other patients.

[0104] If the patient is classified as a low-responder, the RE may select an alternative stimulation protocol. A stimulation protocol selection model may then be implemented for the alternative stimulation protocol. The stimulation protocol selection model may use the alternative stimulation protocol to optimize egg outcomes for the patient. For example, if the patient is a low-responder, the stimulation protocol model may use the alternative stimulation protocol to optimize egg outcomes for the patient.

[0105] In some variations, if the patient has already undergone an IVF cycle, the initial selection of the stimulation protocol RE, the patient's response to the stimulation protocol, and / or modifications to the stimulation protocol may be incorporated into the stimulation protocol selection model. This may provide more accurate predictions during subsequent IVF cycles. In some variations, a separate predictive model may be generated by incorporating detailed previous IVF history of the patient. Such a predictive model may be adjusted for the patient and may become relatively more accurate during the second, third, or subsequent IVF cycles.

[0106] FIG. 26 is a flow diagram illustrating an example variation of a method 2600 for providing a stimulation protocol to a patient. At 2602, the method 2600 may include using patient-specific training data to train a model (e.g., a linear regression model, a KNN model, etc.) to predict egg outcomes for the patient, as described in further detail herein. The patient-specific training data may include patient information for various patients who may have previously undergone one or more IVF cycles. The patient information may include information such as, for example, age, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth history from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), body mass index, race, ethnicity, ultrasound images, previous stimulation protocols selected, egg outcomes associated with the stimulation protocols, costs of IVF treatments, cost restrictions associated with each patient, and / or the like.

[0107] At 2604, method 2600 may include selecting a stimulation protocol for a patient of interest (e.g., a patient initiating and / or undergoing IVF treatment) and predicting egg outcomes for the stimulation protocol. For example, if the patient of interest has already undergone a previous IVF cycle, method 2600 may include selecting an antagonist protocol for the patient. However, if the patient of interest has not undergone any previous IVF cycles, method 2600 may include selecting an alternative protocol (e.g., other than an antagonist protocol) for the patient.

[0108] After selecting a stimulation protocol, at 2606, the method 2600 may include predicting an egg outcome for the selected stimulation protocol by inputting patient-specific data associated with the patient of interest into the trained model and implementing the trained model. At 2608, based on the predicted egg outcome, the method 2600 may include classifying the patient as a low-responder, an average-responder, or a high-responder. For example, the predicted egg outcome may be compared to one or more thresholds and / or threshold ranges associated with a low-responder class, an average-responder class, and a high-responder class. The patient may be classified, at least in part, based on this comparison.

[0109] If, at 2610, method 2600 determines that the patient is classified as a poor responder for the selected stimulation protocol, method 2600 may further include selecting an alternative stimulation protocol (e.g., repeating 2604). Method 2600 may then continue with respect to the alternative stimulation protocol. However, if, at 2610, method 2600 determines that the patient is not classified as a poor responder for the selected stimulation protocol, method 2600 may proceed to prescribing the selected stimulation protocol to the patient (e.g., at 2612).

[0110] FSH Dosage Model In some variations, an FSH dosage model may be generated / constructed to assist the RE in making decisions regarding the amount of FSH dose to be prescribed to the patient (e.g., starting FSH dose, average daily FSH dose, total FSH dose, and / or the like). For example, the FSH dosage model may predict egg outcomes for varying doses of FSH. Once the FSH model is generated and / or trained, the FSH dose may be varied while all other input variables (e.g., anti-Mullerian hormone (AMH) and antral follicle count (AFC) measurements) are held constant to test the model. The predicted egg outcomes may be compared to actual egg outcomes to validate the model.

[0111] Similar to the stimulation protocol model, the FSH dosage model may be trained on patient information such as age, race, ethnicity, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth from natural conception, ultrasound images, and / or baseline measurements of anti-Mullerian hormone (AMH), antral follicle count (AFC), etc. Additionally, the FSH dosage model may be trained on a selected stimulation protocol (e.g., with the aid of the stimulation protocol model).

[0112] Selection of FSH dose may involve a balance. For example, low doses may not encourage follicular growth, while high doses may overstimulate, leading to adverse effects on egg quality and birth rate. In some variations, the FSH dose model may be a linear regression model or a neural network. The linear regression model or neural network may be trained on the patient-specific data described above. The linear regression model or neural network may be implemented by varying the FSH dose for the patient, and a simulation may be run. For example, FIG. 7A illustrates the predicted egg outcome for patient 1 using a neural network for varying levels of FSH dose, and FIG. 7B illustrates the predicted egg outcome for patient 2.

[0113] As can be seen in Figures 7A and 7B, the predicted and actual egg outcomes are identical for Patient 1 and Patient 2 when the FSH dose for Patient 1 and Patient 2 is 300 International Units. Figures 7A and 7B also illustrate the 80% to 95% confidence intervals for each prediction point. Simulations may be run by varying only the FSH dose while holding other patient-specific data constant.

[0114] A challenge with using a linear regression model may be that in some cases, FSH dose may be negatively correlated to egg outcome. In some cases, egg outcome may increase as FSH dose increases. In some scenarios, a specific FSH dose may provide optimal egg outcome. To address this issue, a K-nearest neighbor (KNN) technique may be implemented. KNN may include modeling a smaller subset of patients. For example, multiple linear regression models may be trained. Each model may be trained on a small subset of patients. For a new patient, models trained on patients similar to the new patient may be implemented to predict egg outcome. In other examples, KNN does not use a linear regression model but directly calculates the mean or median number of eggs from a smaller subset of similar patients.

[0115] A set of most similar patients similar to the patient of interest (e.g., a new patient) can be identified by comparing pairwise distances in a feature vector space, where the feature vector includes one or more parameters such as age, BMI, race / ethnicity, diagnosis, AFC, AMH, previous medical history, and / or others. For example, the feature vector may include all of such parameters, or the feature vector may include any suitable subset of such parameters (e.g., age and AFC or AFC and AMH, etc.). Additionally or alternatively, the parameters in the feature vector may be weighted (e.g., with individual coefficients) to reflect the importance of each parameter (e.g., a first weight for a first parameter and a second weight for a second parameter, the first weight being greater than the second weight when the first parameter is more important to establish patient similarity). After identifying a set of similar patients, egg outcomes can be calculated for each FSH dose used for that set of similar patients. Thus, the egg outcomes of a set of most similar patients can help predict the egg outcome for a patient of interest.

[0116] In some variations, optimization may be performed to identify the best performing distance metric (e.g., Euclidean distance, Manhattan distance, and / or equivalent), neighborhood weighting (e.g., uniform weighting, distance weighting, etc.), and number of neighbors. For example, a KNN model may use Manhattan distance, 60-80 neighbors, and distance-based weighting.

[0117] In some variations, similarity matching (e.g., matching a set of similar patients to a patient of interest) may be distance-based. Additionally or alternatively, similarity matching may include implementing a combination of heuristics. For example, patients with similar diagnoses may be grouped together with the patient of interest. Distance calculations in such scenarios may be performed after the heuristics. In some variations, the KNN model may predict egg outcomes (e.g., number of eggs retrieved, number of mature eggs, etc.) by calculating a weighted average from a set of K neighbors that may be most similar to the patient of interest.

[0118] The performance of the KNN model is R-squared (R 2 ) and mean absolute error (MAE). Once a similarity model (e.g., a model implementing the KNN technique) is generated, the similarity model may be implemented for FSH dose selection by (a) querying the K most similar patients and then (b) calculating the most commonly used FSH dose and the most successful FSH dose in terms of best egg outcome (e.g., number of eggs retrieved and mature eggs). The FSH dose can be a starting dose, average daily dose, total dose, and / or equivalent. FIG. 8 illustrates implementation of a similarity model to identify an optimal FSH dose for a patient in terms of best egg outcome (e.g., number of mature eggs retrieved). For the patient shown in FIG. 8, the optimal baseline dose can be 150-200 International Units.

[0119] In some variations, a similarity model may be generated and implemented to determine whether a patient is an FSH dose-responsive patient (responsive to a dose of FSH in that the patient's egg outcome is predicted to change substantially based on a given dose regimen) or an FSH dose-non-responsive patient (non-responsive to a dose of FSH in that the patient's egg outcome is not predicted to change substantially based on a given dose regimen). For example, the similarity model may determine a set of similar patients (e.g., by implementing KNN matching or other suitable matching techniques). The set of similar patients may be a suitable number of K patients, e.g., 10 patients, 25 patients, 50 patients, 75 patients, 90 patients, 100 patients, etc. The K (e.g., suitable number of patients) most similar patients may be queried to determine the number of eggs retrieved for various FSH dosages. A curve may be fitted for a patient for whom an optimal FSH dosage amount has been identified based on data for the K most similar patients (e.g., number of eggs retrieved for various FSH dosages). The curve may then be analyzed to determine whether the patient is an FSH dose-responsive or non-responsive patient. For example, if the number of retrieved eggs increases significantly with increasing FSH dosage based on the curve, the patient may be identified as an FSH dose-responsive patient. However, if the number of retrieved eggs does not vary significantly with increasing FSH dosage based on the curve, the patient may be identified as an FSH dose-non-responsive patient. Thus, the KNN technique may predict egg outcomes for cases with a negative correlation to baseline FSH dose. The KNN technique may also predict optimal FSH doses for maximum egg outcomes. In some variations, a regression line or polynomial may fit the data and visualize the trend between dose and response. In some variations, the regression line may be constrained to have a specific shape. For example, when fitting the data, the regression line may be constrained so that the curve has a concave shape downward. In some variations, multiple curves may be generated for a single patient to relate the patient's response to one or more types of dosage regimens.For example, a curve may be fitted to analyze whether the patient is responsive or non-responsive to FSH. Similarly, another curve may be fitted to analyze whether the patient is responsive or non-responsive to LH. In some variations, a curve may be fitted using the ratio between different dosage regimens and egg outcome. For example, a curve may be fitted using the ratio between FSH and LH and egg outcome. In some variations, a three-dimensional response curve may be fitted to simultaneously relate delayed multiple dosage regimens to egg outcome.

[0120] FIG. 20 illustrates the generation of an exemplary dose-response curve to determine whether a patient is dose-responsive or dose-non-responsive. A similarity model may be implemented for a patient to identify similar patients. For example, similar patients may be identified by comparing feature vectors representing measurements such as age, BMI, race / ethnicity, diagnosis, AFC, AMH, previous medical history, and / or others. As shown in FIG. 20, 100 (or other suitable number) similar patients may be identified for the patient (e.g., the patient of interest) to whom the curve is to be fitted. Trend lines between starting dose and egg outcome may be fitted for these similar patients, and confidence intervals are shown to indicate the uncertainty in the prediction. Although FIG. 20 illustrates a trend line between starting dose and oocytes, it should be readily understood that curves may be fitted for other outcomes such as blastocysts, cumulative births, or for other dose variables such as total dose, dose per day, etc.

[0121] As discussed above, the similarity model may be implemented to determine whether a patient is dose-responsive or whether the patient is dose-non-responsive. For example, a patient may be classified as dose-responsive or dose-non-responsive based on the shape of the fitted curve. In some variations, the model may also recommend an amount of dose (e.g., starting dose, total dose, dose / day, etc.) for a particular patient for RE. The recommendation may be based on the curve. For example, the curve may be used to identify an amount of dose that would lead to maximum egg outcome. Additionally or alternatively, the curve may be used to identify a minimum dose to reach a specific outcome. The patient may be recommended an amount of dose based on these values ​​identified from the curve. The recommendation may be based on whether the patient is classified as dose-responsive or dose-non-responsive. For example, if the patient is dose-responsive, the patient may be recommended a dose that would maximize egg outcome. However, if the patient is dose-non-responsive, the patient may be recommended a minimum dose (e.g., a low dose) to reach a specific egg outcome to reduce patient costs and / or pharmaceutical waste.

[0122] FIG. 21A illustrates an example curve 2102 relating starting dose and egg outcome for a dose-responsive patient. Curve 2102 and curve 2104 may be used to identify recommended dosages for dose-responsive and dose-non-responsive patients. As seen in FIG. 21A, for a dose-responsive patient, the optimal dose from curve 2102 is dosage 2106. This is at the peak of curve 2102, indicating that egg outcome would be greatest for dosage 2106. If the patient were recommended a higher dosage than dosage 2016, such as dosage 2107, for example, the egg outcome would be reduced. Thus, recommending dosage 2107 to the patient would be a non-optimal dose. Thus, a dose-responsive patient may be recommended dosage 2106 to maximize egg outcome.

[0123] FIG. 21B illustrates an example curve 2104 relating starting dose and egg outcome for a dose-non-responsive patient. For a dose-non-responsive patient, the egg outcome does not change significantly between a low dose, such as dosage 2108, and a high dose, such as dosage 2110. In other words, the patient does not respond significantly to changes in the amount of the dosage. Although there is some optimal dose from the curve 2104 at 2109, the predicted egg outcome does not increase significantly with increasing dosage beyond dosages 2018, 2109. If the patient is recommended dosage 2110, the egg outcome is predicted to be substantially the same for dosages 2108 and 2109. Thus, in terms of egg outcome, there is little expected to be any benefit obtained by increasing the dosage beyond dosages 2108 and 2109. Therefore, the recommended treatment for such patients may include administering lower dosages 2108, reducing patient costs or pharmaceutical waste.

[0124] In some variations, multiple curves may be generated iteratively to optimize multiple variables.For example, the model may also identify a specific desired region of the curve.A quadratic curve may be created within the desired region using another variable.For example, within the optimal region of the curve that relates starting FSH dosage to egg outcome, a quadratic curve may be created within that region that relates LH dosage to egg outcome.

[0125] In some variations, the KNN technique described above may be implemented for the stimulation protocol technique (described above) as well.

[0126] As discussed above, in some variations, to address the negative correlation to egg outcome, the FSH dose model may be a neural network. The neural network may be trained on the patient-specific data described above. The neural network may be implemented by varying the FSH dose for the patient, and a simulation may be run. A confidence interval for the neural network may be generated by multiple rounds of performance inference with node dropout. One or more nodes may be randomly dropped from the neural network to generate a confidence interval and measure the performance of the neural network.

[0127] The RE may consider egg outcomes with respect to varying FSH doses to confirm and / or modify the dose for the patient.

[0128] As with stimulation protocols, in some variations, if a patient has already undergone an IVF cycle, the initial selection of RE of FSH dose, the patient's response to FSH dose, and / or modifications to FSH dose may be incorporated into the stimulation protocol selection model. This may provide more accurate predictions during subsequent IVF cycles. In some variations, a separate prediction model may be generated by incorporating detailed previous IVF history of the patient. Such a prediction model may be adjusted for the patient and may be relatively more accurate during the second, third, or subsequent IVF cycles.

[0129] FIG. 27 is a flow diagram illustrating an example variation of a method 2700 for prescribing a dosage of a medication therapy to a patient of interest in association with a stimulation protocol (e.g., FSH dosage). The method 2700 may include generating a subset of patient-specific training data based on similarity matching, as further described herein. For example, the patient-specific training data may include patient information for various patients who may have previously undergone one or more IVF cycles. The patient information may include information such as, for example, age, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth history from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), body mass index, race, ethnicity, ultrasound images, egg outcomes for various doses of FSH, egg outcomes based on a selected stimulation protocol, and / or the like. Similarity matching (e.g., using KNN techniques) may be performed to identify subsets of patients from the training data such that each patient in the subset may be similar to other patients in the subset. In this manner, multiple subsets of patient-specific training data may be generated.

[0130] At 2702, method 2700 may include training one or more models using subsets of the patient-specific training data, as further described herein. For example, a first model may be trained using a first subset of the patient-specific training data and a second model may be trained using a second subset (different from the first subset) of the patient-specific training data. As discussed above, each subset of the patient-specific training data may include patient-specific data of a similar patient.

[0131] At 2704, the method 2700 may include identifying a model from one or more trained models for the patient of interest. For example, to prescribe a medication regimen and / or dosage for the patient of interest, the method 2700 may perform similarity matching to identify a model that may be best suited for the patient of interest. For example, the identified model may be a model that may have been trained with a subset of patient-specific training data from patients that may be similar to the patient of interest. At 2706, a dose-response curve may be generated for the patient of interest using the identified model. The dose-response curve may assist in identifying whether the patient of interest is dose-responsive or dose-non-responsive. For example, the dose-response curve may be fitted for the patient of interest based on egg outcomes for similar patients (e.g., patients from the subset of patient-specific training data used to train the identified model). At 2708, the method 2700 may include prescribing a dosage for the patient of interest using the dose-response curve generated for the patient, as described above.

[0132] Induction date choice model During the course of ovarian stimulation, the patient may be closely monitored. In some variations, the patient's E2, P4 measurements, overall follicle count, overall follicle size, representative measurements of follicle size (e.g., mean, mean, median, etc.), individual follicle counts, bins, and / or ranges for each of a number of sizes, and other follicle measurements may be recorded. Measurements may be taken, for example, every 2 or 3 days, or according to any suitable schedule (e.g., regular or irregular). As discussed above, one of the clinical decisions that the RE may need to make is the determination of the day the final induction injection may be administered. Administering the final induction injection too early may not allow smaller follicles to reach maturity, while administering the final induction injection too late may be detrimental to egg maturation or cause follicular atresia.

[0133] In some variations, the induction day selection model may be trained on patient information such as age, race, ethnicity, ultrasound images, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), and the selected stimulation protocol. Additionally, the induction day selection model may be trained on E2, P4 measurements, and / or follicle measurements, which may be taken on a regular schedule (e.g., daily, every other day, etc.) or an irregular schedule (e.g., days 5, 7, 9, and 10 of the cycle).

[0134] In some variations, the induction date selection model may be a regression model incorporating the patient-specific data discussed above. In some variations, when patient-specific data, including E2 measurements, P4 measurements, and / or follicle measurements, are incorporated for multiple days, the induction selection model may be a recurrent neural network or a long short-term memory (LSTM) neural network to better account for changes over time. The induction date selection model may predict egg outcomes (e.g., number of eggs retrieved and / or number of mature eggs retrieved) for various days. In some variations, follicle sizes may be grouped into different bins. An exemplary set of bins is shown in Table 2 below, but it should be understood that these are merely exemplary in nature and other suitable follicle size ranges may be grouped to define different suitable sets of bins. Additionally, while Table 2 shows a total of six bins, any suitable number of bins may be used (e.g., three, four, five, six, seven, or more than seven bins). For example, in some variations, the follicle size corresponding to bin 1 may be omitted from the set listed in Table 2, such that bins 2-6 are used. As another example, in some variations, the follicle size corresponding to bin 6 may be omitted, such that bins 1-5 are used. As yet another example, in some variations, the follicle sizes corresponding to bins 1 and 6 may be omitted, such that bins 2-5 are used. [Table 2]

[0135] Additionally or alternatively, follicle sizes may be grouped as maximum size, minimum size, average size, median size, etc. Grouping into bins may reduce some of the noise or measurement error associated with the size measurements of each individual follicle. Additionally, grouping may provide the added benefit of model interpretability without compromising model performance.

[0136] The recorded E2, P4, and follicle measurements for various stimulation days before the final challenge injection is administered may be incorporated into the regression model. In some variations, the rate of change in follicle size may be incorporated into the regression model. In scenarios in which ovarian stimulation continues for several more days (e.g., one or two more days), growth trends may be determined.

[0137] Once the induced selection model is deployed, egg outcomes (e.g., number of eggs retrieved, number of mature eggs retrieved, number of eggs successfully fertilized) may be predicted for each day of stimulation (e.g., days on which blood tests and / or ultrasound measurements were recorded). Figures 9A and 9B illustrate predicted egg outcomes for two exemplary patients by day of the cycle. The shape of the growth trend may be used to approximately predict what would have happened if stimulation were to continue for another day or two.

[0138] In some variations, the induced selection model may be a multi-output regression model, or may be multiple independent regression models, each with a different outcome. For example, the multi-output regression model may predict various egg outcomes, such as the number of eggs retrieved, the number of mature eggs, the number of eggs successfully fertilized, etc. Each of the multiple independent regression models may predict one of the number of eggs retrieved, the number of mature eggs, the number of eggs successfully fertilized, etc. Thus, various egg outcomes and their relationships may be captured by developing a multi-output regression model or multiple independent regression models. FIG. 10 illustrates various predicted egg outcomes for different cycle days by developing a multi-output regression model. However, FIG. 11A, FIG. 11B, and FIG. 11C illustrate different egg outcomes by developing independent multiple regression models.

[0139] In some variations, the induction selection model may include a neural network that may predict E2, P4, and follicle measurements one or two days ahead. This helps determine egg outcomes (e.g., number of eggs retrieved and / or number of mature eggs retrieved) at those future dates. In some variations, the neural network may be a recurrent neural network or a long short-term memory (LSTM) neural network. FIG. 12A illustrates prediction of egg outcomes at future dates (e.g., day 11) for various follicle sizes by deploying a neural network. For example, if follicle sizes are recorded for days 5, 7, 9, and 10 of an IVF cycle, the neural network may (1) predict predicted follicle size for day 11 and (2) calculate predicted egg outcomes if the final induction injection is administered on day 10 compared to day 11. This may allow for more accurate predictions. In some variations, if follicle size and hormone levels are measured only every other day or on an irregular schedule, a neural network or linear model may be trained to interpolate the missing data and fill in days where measurements were not taken. Additionally or alternatively, missing data may be filled in by carrying forward the values ​​from the previous day.

[0140] In some variations, the induction selection model may incorporate a combination of techniques. For example, a recurrent neural network may be used to predict follicle measurements and E2 and / or P4 values ​​one day ahead. An interpretable linear regression model may then be used to predict egg outcomes at two time points: (a) using actual current day measurements of follicle measurements and E2 levels, and (b) using predicted next day measurements of follicle measurements and E2 levels. This allows for comparisons regarding egg outcomes if induction occurs on the current day compared to the next day. FIG. 12B illustrates prediction of egg outcomes at a future date using a combination of techniques. At 1252, a patient is predicted to have a higher number of mature eggs if she waits an additional day before induction. At 1254, a different patient is predicted to have fewer mature eggs if she waits an additional day before induction.

[0141] In some variations, the induction selection model may include multiple independent regression models predicting the same outcome on different days. For example, a first regression model may predict egg outcome if the final induction injection is administered on the day of stimulation, while a second regression model may predict egg outcome if the final induction injection is administered on the day after stimulation. The induction selection model may include a generalized linear regression model, such as, for example, a linear regression model, a Poisson regression model, or a negative binomial regression model. In some variations, the inputs 1846 for the first linear regression model 1848a (predicting egg outcome 1850a resulting from induction injection administered on the current day) may include follicle measurements and / or E2 levels for the current day, while the inputs 1846 for the second linear regression model 1848b (predicting egg outcome 1850b resulting from induction injection administered on the next day) may include follicle measurements and / or E2 levels for a previous day. In other words, if induction is performed today, a linear regression model may use follicle measurements and E2 levels measured on the day of induction to predict egg outcome. If induction is performed tomorrow, a separate linear regression model may use follicle measurements and E2 levels measured one day prior to the day of induction to predict egg outcome. Additionally, an E2 predictive model may use follicle measurements and E2 levels measured one day prior to predict E2 levels for the next day. Together, the combination of these models may allow for a comparison of egg outcomes when induction is performed today versus tomorrow.

[0142] For example, FIG. 18 illustrates, in part, prediction of egg outcome on the day and the day after stimulation by implementing two independent regression models. In FIG. 18, the inputs 1846 for the independent models 1848a and 1848b may be follicle measurements and E2. A first linear regression model 1848a may be implemented to predict an egg outcome for the current day (e.g., today) using the inputs 1846 as described above. As seen in FIG. 18, in this example, the linear regression model 1848a may predict an egg outcome 1850a for the current day to be four eggs. A second linear regression model 1848b may be implemented to predict an egg outcome for the next day using the inputs 1846 as described above. As seen in FIG. 18, in this example, the linear regression model 1848b may predict an egg outcome 1850b for the next day to be six eggs. Thus, the two independent regression models 1848a and 1848b may predict that if the final induction injection were to be administered today, the egg outcome would be 4 eggs, and if the final induction injection were to be administered tomorrow, the egg outcome would be 6 eggs. Thus, the induction selection model may provide a tool to support RE clinical decision making regarding determining the appropriate timing for the final induction injection.

[0143] In some variations, the induction selection model may make recommendations to the RE based on egg outcome predictions on consecutive days. For example, the induction selection model may predict egg outcomes on consecutive days of stimulation. Thus, the induction selection model may classify a patient as early, on schedule, or delayed based on whether a final induction injection was administered to the patient on the day the induction selection model recommends administering a final induction injection. This may allow the RE to perform retrospective analysis of the RE decisions and the output from the induction selection model. FIG. 19 illustrates the recommendations given to the RE based on the predicted egg outcome trend after the final induction injection is administered and the classification of the patient as early, on schedule, or delayed. For example, in FIG. 19, at day 8, the induction selection model predicts egg outcomes for day 8 ("MII today") and day 9 ("MII tomorrow"). Based on the egg outcome trend (i.e., fewer eggs on day 8 compared to day 9), the induction selection model recommends continuing the stimulation protocol at day 8. Similarly, at day 11, the induction selection model predicts egg outcomes for day 11 ("MII today") and day 12 ("MII tomorrow") and recommends continuing the stimulation protocol because at day 11, the predicted egg outcome for day 11 is lower than the predicted egg outcome for day 12. At day 12, the induction selection model predicts that the egg outcome at day 12 ("MII today") will exceed the egg outcome at day 13 ("MII tomorrow"). Based on this trend, if continuing the stimulation protocol beyond day 12 is predicted to result in reduced egg outcome, the induction selection model recommends administering the final induction injection at day 12. A retrospective analysis of the recommendations and RE decisions may then be performed. For example, a patient may be classified as early if they decide to administer the actual final induction injection before the day on which RE is recommended (e.g., the day for administering the final induction injection as predicted by the induction selection model). For example, in Figure 19, a patient may be classified as early if the RE determines to administer the final challenge injection before day 12. A patient may be classified as on schedule if the RE determines to administer the actual final challenge injection on the same day as the recommended day.For example, in FIG. 19, if the RE determines to administer the final induction injection on day 12, the patient may be classified as on schedule. The patient may be classified as delayed if the RE determines to administer the actual final induction injection after the recommended day. For example, in FIG. 19, if the RE determines to administer the final induction injection after day 12, the patient may be classified as delayed. Similar to the stimulation protocol, in some variations, if the patient has already undergone an IVF cycle, the RE's initial selection of the final induction date may be incorporated into the induction date selection model. This may provide a more accurate prediction during subsequent IVF cycles. In some variations, a separate prediction model may be generated by incorporating detailed prior IVF history about the patient. Such a prediction model may be adjusted for the patient and may be relatively more accurate during the second, third, or later IVF cycles.

[0144] 28 is a flow diagram illustrating an example variation of a method 2800 for assisting a RE in determining a final induction selection date for a patient of interest. At 2802, the method 2800 may include training one or more models (e.g., one or more linear regression models) to predict egg outcomes for the patient using patient-specific training data, as further described herein. The patient-specific training data may include patient information for various patients, who may have previously undergone one or more IVF cycles. Patient information may include information such as, for example, age, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), body mass index, race, ethnicity, ultrasound images, E2, P4 measurements, and / or follicle measurements, which may be performed on a regular schedule (e.g., daily, every other day, etc.) or on an irregular schedule (e.g., days 5, 7, 9, and 10 of the cycle), and / or the like.

[0145] At 2804, method 2800 may include predicting egg outcomes for the patient of interest on consecutive days using one or more trained models as further described herein. More specifically, method 2800 may include predicting egg outcomes for each consecutive day under the assumption that that particular day will be the day on which a final challenge injection may be administered. For example, a first trained regression model may predict egg outcomes for day 10 using day 10 as the final challenge selection date, a second trained regression model may predict egg outcomes for day 11 using day 11 as the final challenge selection date, a third trained regression model may predict egg outcomes for day 12 using day 12 as the final challenge selection date, and so on.

[0146] At 2806, the method 2800 may include providing the RE with a predicted egg outcome to aid in the RE's determination of a final induction selection date based on the predicted egg outcome. For example, if, based on the prediction, the egg outcome increases from day 10 to day 11 but decreases from day 11 to day 12, the method may provide the RE with information that results in the RE selecting day 11 as the final induction selection date for the patient of interest. At 2808, the method 2800 may also include performing a retrospective analysis of the RE's determination in relation to one or more recommended induction dates provided by the model. For example, the method 2800 may include classifying the patient of interest as early, on time, or delayed based on whether a final induction injection was administered to the patient of interest on a date on which the one or more trained models recommend administering the final induction injection. If the final induction injection is administered before the recommended date, the patient of interest may be classified as early. In a similar manner, a patient of interest may be classified as delayed if the final challenge injection is administered after the recommended date. Thus, a retrospective analysis of RE decisions may be performed.

[0147] Dose adjustment model In some variations, dose adjustments may be performed during ovarian stimulation. For example, increasing FSH and / or LH doses may increase follicular growth. Similarly, decreasing FSH and / or LH doses may slow down dominant follicles and allow smaller follicles to catch up.

[0148] In some variations, the dose adjustment model may be trained on patient information such as age, race, ethnicity, ultrasound images, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth history from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), and the selected stimulation protocol. In addition, the dose adjustment model may be trained on daily or every other day measurements of E2, P4, and / or follicle measurements. In some variations, the dose adjustment model may be a regression model. In some variations, the dose adjustment model may be a neural network (e.g., a recurrent neural network, LSTM, etc.). The dose adjustment model may predict egg outcomes with respect to levels of variation in FSH and / or LH dosage. Egg outcomes may be simulated with respect to levels of variation in FSH and / or LH dosage by deploying the dose adjustment model. Based on the egg outcomes, the RE may determine whether to increase or decrease FSH and / or LH dosage.

[0149] FIG. 29 is a flow diagram illustrating an example variation of a method 2900 for varying FSH and / or LH dosage for a patient of interest. At 2902, the method 2900 may include a model predicting egg outcomes for a patient using patient-specific training data. The patient-specific training data may include patient information for various patients who may have previously undergone one or more IVF cycles. The patient information may include information such as, for example, age, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth history from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), body mass index, race, ethnicity, ultrasound images, E2, P4, follicle metric measurements, and / or the like.

[0150] At 2904, the method 2900 may include predicting an egg outcome for the patient of interest for varying levels of FSH and / or LH dosage using the model, as further described herein. At 2906, the method may include recommending whether levels of FSH and / or LH dosage should be increased or decreased based on the predicted egg outcome.

[0151] Imputation model As discussed above, the models described herein may be trained on patient information (e.g., age, race, ethnicity, ultrasound images, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth history from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), and the selected stimulation protocol), E2, P4, and / or follicle measurement measurements. It is possible that some of the training data may contain erroneous data or missing information. For example, consider a patient whose data will be included to train the models described herein. For example, if a patient receives a final induction selection injection on day 7 and retrospective analysis shows that the patient received a final induction selection injection as scheduled, typically the follicle size for the patient should either increase from day 5 to day 7 or at least remain the same. However, suppose the training data contained an error. That is, suppose the training data is included to erroneously decrease follicle size from day 5 to day 7. If the model were to be trained on this erroneous data, the model could erroneously predict, based on the follicle size trend, to administer the final trigger selection injection on day 5 instead of day 7. Thus, the accuracy of the model could be affected by erroneous or missing data.

[0152] To address this issue, imputation models may be implemented to infer missing or erroneous data. For example, on the day of stimulation, the completeness of follicle measurements and / or follicle measurements may be evaluated using one or more of the follicle measurements of previous days, based on the assumption that the follicle measurements are most likely to either remain the same size or grow over time. For example, if the day is the 7th day, the follicle measurements for day 7 may be evaluated using the follicle measurements for the preceding 6th, 5th, 4th, 3rd, 2nd, and 1st days. If, based on this evaluation, the follicle measurements for the day are deemed not valid (e.g., the follicle measurements for day 7 are less than those for day 6), an imputation model may be implemented to impute the follicle measurements and ensure that the follicle measurements for the day are valid. For example, the imputation model may estimate the value of the follicle measurements for the day based on the measurements of previous days. In this way, the accuracy of the models described herein may be improved. In some variations, the imputation model may include optimization techniques such as linear programming, nonlinear programming, convex optimization, combinations thereof, and / or the like, to impute the required follicle data. For example, suppose the final induction date is predicted using an induction selection day model. As discussed above, the induction selection day model may group follicle sizes into different bins based on their size. As an example, if the current day is day 7 and the total number of follicles for day 7 is less than the total number of follicles for day 6, this may be an indication that the total number of follicles for day 7 may not be valid. Thus, linear programming may be applied to determine the minimum number of follicles to be added to each follicle bin such that the number of follicles for each follicle bin increases and / or remains the same from day 6 to day 7. Thus, an imputed follicle measurement may be determined for day 7. On day 8, the process may be repeated using the imputed follicle measurement from day 7.

[0153] Safety Model Sometimes there may be risks associated with ovarian stimulation. For example, some patients may suffer complications such as ovarian hyperstimulation syndrome (OHSS), an over-response to follicle-stimulating hormone that can cause the ovaries to swell and cause painful conditions. Such complications may lead to the abortion of a stimulation cycle and may adversely affect future IVF cycles. To mitigate complications associated with OHSS, patients at risk for OHSS may be prescribed specific stimulation protocols that may reduce the risk of adverse responses due to OHSS. A safety model may be implemented to prescribe a stimulation protocol for patients at risk for OHSS.

[0154] In some variations, the safety model may be trained on patient information such as age, race, ethnicity, ultrasound images, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), and the selected stimulation protocol. Additionally, the safety model may be trained on daily or every other day measurements of E2, P4, and / or follicle measurements.

[0155] In some variations, the safety model may be a regression model (e.g., a linear regression model). In some variations, the safety model may be a neural network (e.g., a recurrent neural network, LSTM, etc.). The safety model may identify patients at risk for OHSS and may recommend treatment plans to minimize the risk of complications. To identify patients at risk for OHSS, the safety model may use daily measurements of E2, P4, and / or follicle measurements to predict whether a patient is at risk for OHSS. More specifically, measurements of E2, P4, and / or follicle measurements that exceed a threshold and / or fall outside of a normal range may indicate that the patient is at risk for OHSS. The threshold and / or normal range may be patient-specific. That is, different patients may have different thresholds and / or ranges that may be associated with OHSS.

[0156] Therefore, monitoring daily measurements of E2, P4, and / or follicle measurements to ensure that these measurements do not exceed thresholds or fall outside normal ranges may mitigate the risk of OHSS. The safety model may use these daily measurements for the current day to predict measurements of E2, P4, and / or follicle measurements for the next day. The predicted measurements for the next day may be analyzed to determine whether the measurements exceed thresholds or fall outside normal ranges. If the predicted measurements for the next day indicate that the measurements may fall outside normal ranges, the patient may be classified as at risk for OHSS. The stimulation protocol for such patients may then be adjusted to mitigate the risk of OHSS.

[0157] FIG. 30 is a flow diagram illustrating an example variation of a method 3000 for classifying a patient of interest as at risk for ovarian hyperstimulation syndrome (OHSS). At 3002, the method 3000 may include a model predicting egg outcomes for the patient using patient-specific training data. The patient-specific training data may include patient information for various patients, who may have previously undergone one or more IVF cycles, for example. The patient information may include information such as age, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth history from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), body mass index, race, ethnicity, ultrasound images, E2, P4, follicle metric measurements, and / or the like.

[0158] At 3004, the method 3000 may include predicting, for the patient of interest, measurements of E2, P4, and / or follicle measurements for the next day using the model and the measurements of E2, P4, and / or follicle measurements for the current day, as further described herein. At 3006, the method 3000 may include comparing the predicted measurements of E2, P4, and / or follicle measurements to a predetermined threshold and / or threshold range. In response to determining that the predicted measurements of E2, P4, and / or follicle measurements for the patient of interest exceed the threshold, the method may include classifying the patient as at risk for OHSS.

[0159] Clinic workflow model IVF clinics sometimes face the challenge of coordinating many patient visits within a finite schedule and among a finite number of REs. Balancing the clinic's patient throughput and ensuring that the workload is evenly distributed may require determining the appropriate time to start a patient cycle. For example, if multiple patients at a particular RE start their stimulation protocol cycles at the same time, it is possible that the RE may be overloaded with multiple egg retrievals throughout the day, or the embryologist may be overloaded with embryo biopsies throughout the day. Thus, it may be important to predict expected induction dates and expected cycle lengths for a patient population, to help anticipate the retrieval and / or biopsy load, or to identify a start date for a stimulation cycle for each of these patients, in order to distribute (e.g., stagger) the expected induction dates, expected egg retrieval dates, and / or expected embryo biopsy dates for the patient population. The clinic workflow model may predict the expected induction dates and expected cycles for the patient population.

[0160] In some variations, the clinic workflow model may be trained on patient information such as age, race, ethnicity, ultrasound images, previous IVF history, previous intrauterine insemination (IUI) history, previous pregnancy or normal birth from natural conception, and / or baseline measurements such as anti-Mullerian hormone (AMH), antral follicle count (AFC), and selected stimulation protocol. Additionally, the clinic workflow model may be trained on daily or every other day measurements of E2, P4, and / or follicle measurements.

[0161] In some variations, the clinic workflow model may be a regression model (e.g., a linear regression model, or a Poisson regression model). In some variations, the clinic workflow model may be a neural network (e.g., a recurrent neural network, LSTM, etc.). For a patient population, based on each patient's expected induction date and expected induction cycle, the clinic workflow model may predict a start date for each patient. In some variations, the clinic workflow model may additionally or alternatively predict for each patient the necessary days for a clinic visit to reduce the overall number of patient visits. In some variations, the clinic workflow model may prioritize the order of patients who should see the RE to indicate patients who may need the most attention. For example, patients who may need to be administered a final induction injection urgently to avoid reducing egg outcomes in subsequent days may be prioritized over patients who may not need to be induced urgently. In some variations, the clinic workflow model may include optimization techniques (e.g., linear programming, convex optimization, combinations thereof, and / or the like) to predict expected induction dates and expected induction cycles for each patient and then schedule start dates for patient populations.

[0162] The clinic workflow model may additionally or alternatively be used to assist staff within a clinic. For example, in some variations, egg retrieval dates and / or embryo biopsy dates for a patient population associated with a clinic may be predicted, and thus the clinic workflow model for that clinic may be used to predict staff needs for egg retrieval and / or embryo biopsy procedures on particular days, thereby allowing the clinic to appropriately schedule staff to perform the anticipated procedures.

[0163] Viewing predictions based on a model As discussed above, the techniques described herein may be used by the RE to enhance or further inform its decisions. The models disclosed herein may provide interpretable results that the RE may view to inform its decisions. Predictions from the models may be transmitted in interpretable form to an RE application (e.g., RE application 208) implemented on a suitable computing device. As shown in FIG. 22, at 2282, once the RE application is implemented, a login page may be displayed on the display of the suitable computing device. Users (e.g., REs, clinicians, etc.) that may be registered on the RE application may have access to results from the models and data associated with the patient. Users that are not registered may not have access to the RE application, thereby providing limited access to the results and data associated with the RE application. This keeps the data and results associated with the RE application secure. On the login page, a registered RE may enter their registered email address and password so that the RE application may authenticate the user.

[0164] Once the RE application has authenticated the RE, at 2284, the display displays a patient dashboard. FIG. 23 illustrates an example patient dashboard 2384 displayed on a display of a suitable computing device. The patient dashboard 2384 may include a list of the RE's patients. For example, the patient dashboard may include the patient name, patient ID, stimulation status, and updated timestamps (e.g., 2391). As discussed herein, the RE application may interface with an EMR database (e.g., EMR 204). The patient name and associated patient ID (e.g., 2292) may be populated from the EMR database. For example, in FIG. 23, "Lisa Jones" may be associated with patient ID "m3815," which may be populated from the EMR database.

[0165] The EMR database may also capture a stimulation status 2294 associated with the patient. For example, the stimulation status may indicate whether the patient's stimulation protocol has begun. For example, in FIG. 23, "Lisa Jones" is shown to have a stimulation status of "pre-stimulation," indicating that the stimulation protocol has not begun. If the stimulation status indicates a number of days, this may indicate that the stimulation protocol has begun for the patient and that the stimulation process falls on the day shown on the display. For example, in FIG. 23, "Rose Wolfe" is shown to have a stimulation status of "day 12," indicating that Rose Wolfe's stimulation has begun and that the current day of the cycle is the 12th day of the cycle. For patients actively undergoing stimulation, the patient dashboard 2384 may provide a preview 2396 of some predictions. For example, the patient dashboard 2384 may display predicted egg outcomes (e.g., mature oocytes) for today and the next day. For example, in FIG. 23 , the predicted mature oocytes for "Rose Wolfe" actively receiving stimulation for today (e.g., stimulation status shown as "Day 12") are "8.9" and for tomorrow are "10.2". In some variations, the patient dashboard may also display E2 measurements for today and E2 measurement predictions for tomorrow. For example, the E2 measurements for "Rose Wolfe" for today are shown to be "3260" and the predicted E2 measurements for tomorrow are shown to be "4100". The patient dashboard 2384 may also include filters 2398. For example, the patient list may be filterable by clinic, stimulation status, and RE.

[0166] Referring back to FIG. 22, if the patient is not actively receiving stimulation (e.g., stimulation status at 2284 is "pre-stim"), the RE application displays a start dose page at 2286. FIG. 24 illustrates an exemplary start dose page displayed on a display of a suitable computing device. For example, in FIG. 23, "Lisa Jones" was shown not actively receiving stimulation (e.g., stimulation status at 2284 is "pre-stim"). FIG. 24 shows a start dose page for "Lisa Jones". The start dose page 2486 displays the patient's baseline characteristics (e.g., 2451). For example, the start dose page 2486 shows measurements of age, BMI, anti-Mullerian hormone (AMH), and antral follicle count (AFC) associated with "Lisa Jones". These baseline characteristics may be input into the FSH dose model, as discussed above.

[0167] As discussed herein, the FSH dose model may include a similarity model that identifies patients similar to the patient of interest. A dose-response curve (e.g., a curve that is fitted using data of similar patients) may indicate whether the patient of interest is dose-responsive or dose-non-responsive. In FIG. 24, statistics of patients similar to "Lisa Jones" may be displayed (e.g., 2452). As shown in FIG. 24, the FSH dose model may use historical patient cycle data of 100 similar patients. Statistics of these similar 100 patients are shown in 2452. A dose-response curve may be fitted / generated for "Lisa Jones" based on these statistics 2452. More specifically, the FSH dose-response model may use baseline characteristics 2451 for "Lisa Jones" and statistics 2452 of the 100 similar patients to generate a dose-response curve 2453. The dose-response curve 2453 may depict egg outcomes (e.g., number of mature oocytes) for various starting doses of FSH for "Lisa Jones." The shaded areas within the dose response curve 2453 may depict the confidence interval at each dose increment. As seen in FIG. 24, the optimal starting dose for "Lisa Jones" is "300." In addition, the starting dose page 2486 may also display cost estimates 2454 for different starting doses that may be projected through an 11-day stimulation cycle. In FIG. 24, for a starting FSH dose of "300," the cost estimate for "Lisa Jones" is "$3,300."

[0168] In some variations, the starting dose page 2486 may also display whether the patient has received a cycle in the past (e.g., 2455). Clicking on a past cycle may trigger a summary page for that past cycle for the patient.

[0169] Referring back to FIG. 22, if the patient is actively receiving stimulation (e.g., the stimulation status at 2284 is a day such as "Day 12"), the RE application displays an induction page at 2288. FIG. 25 illustrates an exemplary induction page displayed on a display of a suitable computing device. For example, in FIG. 23, "Michelle James" was shown to be actively receiving stimulation (e.g., the stimulation status at 2284 is "Day 12"). FIG. 25 shows an induction page 2588 for "Michelle James". The induction page 2588 includes baseline characteristics associated with "Michelle James", such as age, BMI, anti-Mullerian hormone (AMH), and antral follicle count (AFC) measurements. These baseline characteristics may be input into an induction selection model, as discussed above. The induction page 2588 allows the RE to select the optimal day for administering the final induction injection to maximize egg outcomes (e.g., mature oocyte yield). For example, the induction page 2588 provides egg outcome predictions for different induction days. In FIG. 25, "Michelle James" is shown to be on day 12 of the stimulation protocol. Egg outcomes for days 7, 9, 11, and 12 (e.g., 2561) are shown on the induction page 2588. The induction page 2588 also shows E2 and P4 measurements (e.g., 2562) for each of the different induction days. In addition, the induction page also includes egg outcome predictions and E2 measurements for the next day (e.g., 2563). In FIG. 25, the current day is day 12, so egg outcome predictions and E3 measurements for day 13 are shown. The induction page 2588 also provides a visual display of follicle measurements (e.g., 2564) for the different induction days and the medications administered on each of the different induction days (e.g., 2565).

[0170] Exemplary Methods of Treatment 13 is a flow diagram of an exemplary method of treatment using the machine learning models described herein. The method 1200 includes providing patient-specific data to a controller (e.g., controller 206 in FIG. 2). The patient-specific data may include patient information, data related to previous IVF cycles and / or treatments, baseline measurements, treatment variables, responses to stimulation protocols, combinations thereof, and / or the like. The controller may generate one or more machine learning models, such as those described above, to predict egg outcomes for the patient.

[0171] At 1204, the method includes receiving an egg outcome from a machine learning model using the patient-specific data. For example, the method may include receiving the egg outcome from a first predictive model related to a stimulation protocol selection. Additionally or alternatively, the method may include receiving the egg outcome from a second predictive model related to an FSH dosage model. Additionally or alternatively, the method may include receiving the egg outcome from a third predictive model related to an induction date selection model.

[0172] Egg outcome may be predicted by varying one or more of the patient-specific data. For example, a first predictive model may include a K-nearest neighbor (KNN) technique. The first predictive model may predict a stimulation protocol that may provide optimal egg outcome. Similarly, a second predictive model may include a K-nearest neighbor (KNN) technique. The second predictive model may predict an FSH dose that may provide optimal egg outcome. In some variations, a third predictive model includes a combination of techniques. For example, a recurrent neural network may be used to predict one-day-ahead follicle measurements and E2 and / or P4 values. An interpretable linear regression model may then be used to predict egg outcome.

[0173] At 1206, the method may include administering a drug dosage to the patient based on the egg outcome. For example, a stimulation protocol may be selected for the patient by implementing a first predictive model. The selected stimulation protocol may include an amount of the drug dosage to be administered to the patient on a daily basis. Additionally or alternatively, an amount of a baseline FSH dose may be selected for the patient by implementing a second predictive model. Additionally or alternatively, a day on which a final challenge injection for the patient should be administered may be selected by implementing a third predictive model.

[0174] In some variations, administering the drug dosage may further include monitoring a patient response. For example, the patient's response to the selected stimulation protocol may be monitored. Based on the patient response, the stimulation protocol may be modified and / or discontinued. For example, if the patient exhibits a hypo-response to the selected stimulation protocol, the first predictive model may be updated to take into account the hypo-response. The selected stimulation protocol may be modified based on the updated predictive model.

[0175] Working Example As discussed above, several variables (e.g., amount of FSH dosage, induction date, stimulation protocol, etc.) may be predictive of egg outcome for a patient. Some non-limiting examples of these variables may include patient personal information such as age, BMI, patient past IVF cycle information such as number of cycles, stimulation protocol, diagnosis, baseline dosage measurements such as estradiol (E2) measurements, luteinizing hormone (LH) measurements, progesterone (P4) measurements, follicle stimulating hormone (FSH) measurements, anti-Mullerian hormone (AMH) measurements, and antral follicle count (AFC) measurements, stimulation protocol such as follicle size, induction date, combinations thereof, and / or the like.

[0176] In some variations, it may be possible to identify the most prognostic and / or most statistically significant parameters / variables that may predict egg outcome for a patient. For example, the top variables may be identified by applying recursive feature elimination, where weaker features are identified one by one and removed from the group of candidate features. Figures 14A-14C illustrate examples of the most statistically significant variables that may predict egg outcome for a patient. As seen in Figures 14A and 14B, follicle size less than or equal to about 11 mm, follicle size between about 12 mm and about 14 mm, follicle size between about 15 mm and about 16 mm, follicle size between about 17 mm and about 18 mm, follicle size between about 19 mm and about 20 mm, and estradiol measurements may be the six most statistically significant variables that may predict egg outcome for a patient. These six most statistically significant variables may closely track egg outcome for a patient. For example, an induction date selection model may be implemented to determine an induction date based on these six most statistically significant variables. For example, an induction date may be selected based on follicle size measurements and estradiol measurements for a patient. Figure 14C illustrates validation of predicted versus actual egg outcomes when the six most statistically significant variables are measured on the induction date for a patient.

[0177] However, as explained above, it should be understood that in other preferred variants, there may be different types of preferred regression model parameters. Figure 14D relates to another embodiment in which seven variables are particularly predictive of egg outcome for a patient. Specifically, as shown in Figure 14D, follicle size less than or equal to about 10 mm, follicle size between about 11 mm and about 13 mm, follicle size between about 14 mm and about 15 mm, follicle size between about 16 mm and about 17 mm, follicle size between about 18 mm and about 19 mm, follicle size greater than or equal to about 20 mm, and estradiol measurements may be seven statistically significant variables that may predict egg outcome for a patient.

[0178] 15A-15C illustrate an example of a similarity model implemented to determine whether a patient is an FSH dose-responsive or non-responsive patient. In this example, data to train / generate the regression model was collected from four different IVF clinics. The data was filtered for autologous non-aborted IVF retrievals. This included 7977 cycles between 2014-2020.

[0179] A KNN patient similarity model was generated based on this data, as discussed above. As seen in FIG. 15A, the KNN model was used to identify 100 similar patients for Patient 1 and 100 similar patients for Patient 2 based on parameters such as BMI, age, baseline AMH, and baseline AFC. The 100 similar patients were used to generate dose-response curves predicting the number of eggs retrieved for various starting FSH doses. For example, the dose-response curve for Patient 1 is shown in FIG. 15B, and the dose-response curve for Patient 2 is shown in FIG. 15C. For each of Patient 1 and Patient 2, the dose-response curves were analyzed to determine whether the patient was FSH dose-responsive or FSH dose-non-responsive. For example, in FIG. 15B, the number of retrieved eggs increases with increasing FSH dosage, indicating Patient 1 is dose-responsive. However, in FIG. 15C, the number of retrieved eggs remains unchanged with increasing FSH dosage, indicating Patient 2 is dose-non-responsive.

[0180] In some variations, as discussed above, one or more models described herein may be implemented to predict the induction date for a patient to maximize egg outcome. The benefit of using the models described herein may be calculated by analyzing each patient on a regular basis (e.g., on a daily basis, such as every day during administration of the stimulation protocol, or every day during a subset of days of administration of the stimulation protocol). More specifically, the quality of the models described herein may be estimated based on data obtained from each patient.

[0181] In some variations, the model described herein may be used to predict the induction selection date. For example, the model may recommend continuing the stimulation protocol for the patient if the predicted egg outcome shows an increase after two days. For example, if the egg outcome is predicted to increase from day 5 to day 7, the model may recommend continuing the stimulation protocol for the patient until at least day 7. Similarly, the model may recommend continuing the stimulation protocol for the patient if the predicted egg outcome is less than 15 (e.g., the number of eggs predicted to be retrieved is less than 15) or the predicted amount of estradiol is less than 5,000. However, if the egg outcome is predicted to show a decrease after two days, the model may recommend inducing ovarian stimulation and extracting eggs. Additionally or alternatively, the predicted induction selection date may be compared to the actual induction date to determine whether the actual induction is early or delayed. In some variations, the model may also make a recommendation to continue or stop the stimulation protocol based on an increase or decrease after one day.

[0182] One or more models described herein may be used to predict the induction date for a patient to maximize egg outcome. For example, estradiol and follicle size may be used to predict the number of mature eggs. At each measurement date during the stimulation protocol, the model may predict the egg outcome (e.g., the number of mature eggs) if that day is the induction date, forecast the estradiol and follicle size for the next day, and also predict the egg outcome if the next day is the induction date. Such models may be used throughout the stimulation protocol to guide whether to continue stimulation or to induce ovulation, reducing the likelihood of inducing either too late or too early ovulation, as described below with respect to Figures 16A-16D and 17A-17D.

[0183] 16A-16D illustrate an example of applying such a model to an exemplary patient undergoing ovarian stimulation, where the circular dots represent predicted egg outcomes and the triangular symbols represent amounts of estradiol (for convenience, follicle size is not shown). The dashed lines depict predictions by the model. Estradiol and follicle measurements for this patient begin on day 4 of stimulation. As shown in FIG. 16A, on day 8 of stimulation, the model predicts egg outcomes if day 8 is the day of induction, predicts estradiol and follicle size for day 9, and predicts egg outcomes if day 9 is the day of induction. Specifically, the model predicts that egg outcomes will increase between days 8 and 9, and therefore, based on the model, the recommendation is to continue the stimulation protocol. Thus, the stimulation protocol continues for this patient until day 10, and as shown in FIG. 16B, on day 10, the model predicts that egg outcomes will remain approximately the same between days 10 and 11. As shown in Figure 16C, at day 11, the model predicts a clear decline in egg outcome for day 12 relative to that for day 11. Because this future decrease in egg outcome is predicted, the recommendation at day 11 would be that day 11 should be the induction date. If the induction date is not selected to be day 11, and instead stimulation is continued until day 12 when induction occurs, the actual induction on day 12 will likely be delayed, as shown in Figure 16D, thereby resulting in undesirably reduced egg outcome.

[0184] 17A-17C illustrate another example of implementing one or more models described herein to predict the induction date for a patient to maximize egg outcome, where the circular dots represent predicted egg outcome and the triangular symbols represent the amount of estradiol. The dashed lines depict the predictions by the model. As shown in FIG. 17A, on day 7 of stimulation, the model predicts egg outcome if day 7 is the induction date, predicts estradiol and follicle size for day 8, and predicts egg outcome if day 8 is the induction date. Specifically, the model predicts that egg outcome will increase between day 7 and day 8, and therefore, based on the model, the recommendation is to continue the stimulation protocol. Thus, the stimulation protocol continues until day 9 for this patient, and as shown in FIG. 17B, on day 9, the model predicts that egg outcome will continue to increase between day 9 and day 10. As shown in Figure 17C, at day 10, the model predicts that egg outcome will increase even further between days 10 and 11, and therefore the recommendation is to continue the stimulation protocol and further delay induction. Thus, if the RE selects day 10 as the induction date, this actual induction on day 10 will likely be premature, thereby resulting in undesirably reduced egg outcome.

[0185] Enumerated Embodiments Embodiment A1. A computer-implemented method for optimizing an ovarian stimulation process, comprising: receiving patient-specific data associated with the patient; predicting an egg outcome for the patient for each of a plurality of treatment options for the ovarian stimulation process based on at least one predictive model and the patient-specific data, the at least one predictive model being trained using prior patient-specific data associated with a plurality of prior patients; A method comprising:

[0186] Embodiment A2. The method of embodiment A1, further comprising providing a health care professional with a predicted egg outcome for selecting among multiple treatment options.

[0187] Embodiment A3. The method of embodiment A1, wherein the predicted egg outcome includes at least one of the number of eggs retrieved and the number of mature eggs.

[0188] Embodiment A4. The method of embodiment A1, wherein the predicted egg outcomes include at least one of maturity yield, number of mature eggs, number of fertilized eggs, number of embryos, number of blastocysts, number of usable blastocysts, number of euploid blastocysts, fetal heartbeat, and birth rate.

[0189] Embodiment A5. The method of embodiment A1, wherein the patient-specific data includes at least one of age, body mass index, ethnicity, infertility diagnosis, previous pregnancy history, previous parity, and one or more baseline measurements selected from the group consisting of estradiol (E2) measurements, luteinizing hormone (LH) measurements, progesterone (P4) measurements, follicle stimulating hormone (FSH) measurements, anti-Mullerian hormone (AMH) measurements, and antral follicle count (AFC) measurements.

[0190] Embodiment A6. The method of embodiment A1, wherein the patient-specific data includes information related to one or more previous in vitro fertilization (IVF) treatments associated with the patient.

[0191] The method of embodiment A6, wherein the information relating to the one or more previous IVF treatments includes at least one of data collected during ovarian stimulation, number of eggs collected, number of mature eggs, number of eggs successfully fertilized, pregnancy outcome, and birth outcome.

[0192] Embodiment A8. The method of embodiment A1, wherein the patient-specific data includes one or more treatment variables associated with the patient.

[0193] The method of embodiment A8, wherein the one or more treatment variables include the type of medication, the type of hormone triggering injection to induce follicular maturation, and the number of cycles associated with the patient.

[0194] Embodiment A10. The method of embodiment A1, wherein at least one predictive model is configured to provide a predicted egg outcome associated with each of a plurality of candidate doses of one or more ovarian stimulation medications administered to the patient.

[0195] The method of embodiment A10, wherein the one or more ovarian stimulation medications comprise follicle stimulating hormone (FSH).

[0196] The method of embodiment A10, wherein the one or more ovarian stimulation medications include luteinizing hormone (LH).

[0197] Embodiment A13. The method of embodiment A10, wherein the patient-specific data comprises one or more baseline measurements selected from the group consisting of an estradiol (E2) measurement, a luteinizing hormone (LH) measurement, a progesterone (P4) measurement, a follicle-stimulating hormone (FSH) measurement, an anti-Mullerian hormone (AMH) measurement, and an antral follicle count (AFC) measurement.

[0198] Embodiment A14. The method of embodiment A10, wherein at least one predictive model is configured to provide a predicted egg outcome associated with each of the first dose and the second dose, wherein the second dose is greater than the first dose.

[0199] Embodiment A15. The method of embodiment A10, wherein at least one predictive model is configured to provide a predicted egg outcome associated with each of a first dose ratio of the plurality of ovarian stimulation dosage regimens and a second dose ratio of the plurality of ovarian stimulation dosage regimens, the first dose ratio being different from the second dose ratio.

[0200] Embodiment A16. identifying a set of prior patients similar to the patient based on the patient-specific data via similarity matching techniques; The method of embodiment A10, further comprising: wherein the at least one predictive model comprises a predicted dose-response curve generated based on previous patient-specific data associated with the previous patient, the dose-response curve predicting an egg outcome that varies with doses of the one or more ovarian stimulation medications administered to the patient.

[0201] Embodiment A17. The method of embodiment A16, wherein the similarity matching technique includes a KNN model.

[0202] Embodiment A18. The method of embodiment A16, further comprising classifying the patient as one of: (i) a dose-responsive patient, who has a predicted egg outcome that varies substantially with the dose of the one or more ovarian stimulation medication regimens administered to the patient; and (ii) a dose-non-responsive patient, who has a predicted egg outcome that does not vary substantially with the dose of the one or more ovarian stimulation medication regimens administered to the patient, wherein the classification is based, at least in part, on the shape of the dose-response curve.

[0203] Embodiment A19. The method of embodiment A18, further comprising recommending an optimal dose of one or more ovarian stimulation medications to be administered to the patient based on the point on the dose-response curve that is associated with the predicted maximum egg outcome when the patient is classified as a dose-responsive patient.

[0204] Embodiment A20. The method of embodiment A16, further comprising displaying the predicted dose-response curve on a display.

[0205] Embodiment A21. The at least one predictive model includes a first predictive model and a second predictive model, and the method further comprises: predicting via a first predictive model a first egg outcome resulting from administering a hormone triggering injection on a first candidate hormone triggering date; predicting via a second predictive model a second egg outcome resulting from administering a hormone triggering injection on a second candidate hormone triggering date; wherein the hormone triggering injection is configured to induce follicular maturation in the patient.

[0206] Embodiment A22. The method of embodiment A21, wherein the first date is the current date and the second date is a future date.

[0207] Embodiment A23. The method of embodiment A21, wherein the patient-specific data includes current day follicle measurements.

[0208] Embodiment A24. The method of embodiment A21, wherein the patient-specific data includes a current day estradiol (E2) level for the patient.

[0209] Embodiment A25. The method of embodiment A21, wherein the patient-specific data further comprises a current day progesterone (P4) level for the patient.

[0210] Embodiment A26. The method of embodiment A21, wherein the first and second predicted egg outcomes include at least one of the number of eggs retrieved, the number of mature eggs, the number of eggs fertilized, and the number of usable blastocysts.

[0211] Embodiment A27. The method of embodiment A21, wherein at least one of the first and second predictive models comprises a recurrent neural network or a generalized linear model.

[0212] Embodiment A28. The method of embodiment A27, wherein at least one of the first and second predictive models comprises a generalized linear model selected from the group consisting of a linear regression model, a Poisson regression model, and a negative binomial regression model.

[0213] Embodiment A29. The method of embodiment A28, wherein each of the first and second predictive models comprises a linear regression model.

[0214] Embodiment A30. The method of embodiment A27, wherein at least one of the first and second predictive models is configured to predict E2 measurements at a future date for the patient.

[0215] Embodiment A31. The method of embodiment A27, wherein at least one of the first and second predictive models is configured to predict follicular measurements for the patient at a future date.

[0216] Embodiment A32. The method of embodiment A27, wherein at least one of the first and second predictive models is configured to predict the first or second egg outcome based, at least in part, on at least one of the E2 measurements and the follicle measurements.

[0217] Embodiment A33. The method of embodiment A21, further comprising displaying the first and second egg outcomes on a display.

[0218] Embodiment B1. A computer-implemented method for optimizing an ovarian stimulation process, comprising: receiving patient-specific data associated with the patient; identifying a set of prior patients similar to the patient based on the patient-specific data via a similarity matching technique; generating a predicted dose-response curve based on previous patient-specific data associated with the previous patient, the dose-response curve predicting an oocyte outcome that varies with doses of one or more ovarian stimulation medications administered to the patient; A method comprising:

[0219] Embodiment B2. The method of embodiment B1, wherein the one or more ovarian stimulation medications comprise follicle stimulating hormone (FSH).

[0220] Embodiment B3. The method of embodiment B2, in which the dose-response curve predicts egg outcome, which varies with the starting dose of FSH.

[0221] Embodiment B4. The method of embodiment B1, wherein the one or more ovarian stimulation medications include luteinizing hormone (LH).

[0222] Embodiment B5. The method of embodiment B1, wherein the patient-specific data comprises one or more baseline measurements selected from the group consisting of an estradiol (E2) measurement, a luteinizing hormone (LH) measurement, a progesterone (P4) measurement, a follicle-stimulating hormone (FSH) measurement, an anti-Mullerian hormone (AMH) measurement, and an antral follicle count (AFC) measurement.

[0223] Embodiment B6. The method of embodiment B1, wherein the predicted dose-response curve is configured to provide a predicted egg outcome associated with each of the first dose and the second dose, wherein the second dose is greater than the first dose.

[0224] Embodiment B7. The method of embodiment B1, wherein the predicted dose-response curve is configured to provide a predicted egg outcome associated with each of a first dose ratio of the plurality of ovarian stimulation dosage regimens and a second dose ratio of the plurality of ovarian stimulation dosage regimens, wherein the first dose ratio is different from the second dose ratio.

[0225] Embodiment B8. The method of embodiment B1, wherein the similarity matching technique includes a KNN model.

[0226] Embodiment B9. The method of embodiment B1, further comprising classifying the patient as one of: (i) a dose-responsive patient, who has a predicted egg outcome that varies substantially with the dose of one or more ovarian stimulation medication regimens administered to the patient; and (ii) a dose-non-responsive patient, who has a predicted egg outcome that does not vary substantially with the dose of one or more ovarian stimulation medication regimens administered to the patient, wherein the classification is based, at least in part, on the shape of the dose-response curve.

[0227] Embodiment B10. The method of embodiment B9, further comprising recommending an optimal dose of one or more ovarian stimulation medications to be administered to the patient based on the point on the dose-response curve that is associated with the predicted maximum egg outcome when the patient is classified as a dose-responsive patient.

[0228] Embodiment B11. The method of embodiment B1, further comprising displaying the dose-response curve on a display.

[0229] Embodiment B12. The method of embodiment B11, further comprising displaying confidence intervals around the dose-response curve.

[0230] Embodiment B13. The method of embodiment B11, further comprising displaying cost estimates for administering different doses of the one or more ovarian stimulation medications according to a dose-response curve.

[0231] Embodiment C1. A computer-implemented method for optimizing an ovarian stimulation process, comprising: receiving patient-specific data associated with the patient; predicting via a first predictive model a first egg outcome resulting from administering a hormone triggering injection on a first candidate hormone triggering date; predicting via a second predictive model a second egg outcome resulting from administering a hormone triggering injection on a second candidate hormone triggering date; wherein the hormone triggering injection is configured to induce follicular maturation in the patient.

[0232] Embodiment C2. The method of embodiment C1, wherein the first date is the current date and the second date is a future date.

[0233] Embodiment C3. The method of embodiment C2, wherein the first day and the second day are consecutive days.

[0234] Embodiment C4. The method of embodiment C1, wherein the patient-specific data includes current day follicle measurements.

[0235] Embodiment C5. The method of embodiment C1, wherein the patient-specific data includes a current day estradiol (E2) level for the patient.

[0236] Embodiment C6. The method of embodiment C4, wherein the patient-specific data further comprises current day progesterone (P4) for the patient.

[0237] Embodiment C7. The method of embodiment C1, wherein the first and second predicted egg outcomes include at least one of the number of eggs retrieved, the number of mature eggs, the number of eggs fertilized, and the number of usable blastocysts.

[0238] Embodiment C8. The method of embodiment C1, wherein at least one of the first and second predictive models comprises a recurrent neural network or a generalized linear model.

[0239] Embodiment C9. The method of embodiment C8, wherein at least one of the first and second predictive models comprises a generalized linear model selected from the group consisting of a linear regression model, a Poisson regression model, and a negative binomial regression model.

[0240] Embodiment C10. The method of embodiment C8, wherein each of the first and second predictive models comprises a linear regression model.

[0241] Embodiment C11. The method of embodiment C8, wherein at least one of the first and second predictive models is configured to predict E2 measurements for the patient at a future date.

[0242] Embodiment C12. The method of embodiment C8, wherein at least one of the first and second predictive models is configured to predict follicle metric measurements at a future date for the patient.

[0243] Embodiment C13. The method of embodiment C8, wherein at least one of the first and second predictive models is configured to predict the first or second egg outcome based, at least in part, on at least one of the E2 measurements and the follicle measurements.

[0244] Embodiment C14. The method of embodiment C1, further comprising providing the first and second predicted egg outcomes to a health care professional to select between the first and second candidate hormone induction dates.

[0245] Embodiment C15. The method of embodiment C14, wherein providing the first and second predicted egg outcomes comprises displaying the first and second egg outcomes on a display.

[0246] Embodiment D1. A computer-implemented method comprising: receiving patient-specific data associated with the patient; predicting at least one egg outcome for a patient based on at least one predictive model and patient-specific data, the patient-specific data including follicle sizes categorized into a bin from a plurality of predefined bins, each bin of the plurality of predefined bins being associated with a distinct range of follicle sizes, the at least one predictive model being trained using prior patient-specific data associated with a plurality of prior patients; A method comprising:

[0247] Embodiment D2. The method of embodiment D1, wherein at least one bin of the plurality of bins is associated with a first follicle size range of about 10 mm or less, a second follicle size range of about 11 mm to 13 mm, a third follicle size range of about 14 mm to 15 mm, a fourth follicle size range of about 16 mm to 17 mm, a fifth follicle size range of about 18 mm to 19 mm, and a sixth follicle size range of greater than about 20 mm.

[0248] Embodiment D3. The method of embodiment D2, wherein the plurality of bins includes six bins, each bin associated with a respective one of a first follicle size range, a second follicle size range, a third follicle size range, a fourth follicle size range, a fifth follicle size range, and a sixth follicle size range.

[0249] Embodiment D4. The method of embodiment D1, wherein the patient-specific data includes an E2 level for the patient, and predicting at least one egg outcome includes predicting at least one egg outcome based on the E2 level.

[0250] Embodiment D5. The method of embodiment D1, further comprising displaying the predicted egg outcome on a display.

[0251] Embodiment E1. A computer-implemented method comprising: receiving patient-specific data associated with the patient; For each of the plurality of days, predicting an egg outcome for the patient based on at least one predictive model and the patient-specific data, the at least one predictive model being trained using prior patient-specific data associated with a plurality of prior patients; displaying on the display a trend of predicted egg outcomes over multiple days; A method comprising:

[0252] Embodiment E2. The method of embodiment E1, wherein the predicted egg outcome includes at least one of the number of eggs retrieved and the number of mature eggs.

[0253] Embodiment E3. The method of embodiment E1, wherein the predicted egg outcome includes at least one of maturity yield, number of mature eggs, number of fertilized eggs, number of embryos, number of blastocysts, number of usable blastocysts, number of euploid blastocysts, fetal heartbeat, and birth rate.

[0254] Embodiment F1. A method of treatment comprising: providing patient-specific data associated with the patient; receiving a predicted egg outcome associated with each of a plurality of treatment options, wherein at least one predicted egg outcome is generated using at least one predictive model according to a method according to any one of embodiments A1-A33, B1-B13, C1-C15, D1-D5, and E1-E3; Selecting a treatment option based on a predicted egg outcome; and administering an ovarian stimulation medication in accordance with the selected treatment option; A method comprising:

[0255] The foregoing description, for purposes of explanation, uses specific terminology to provide a thorough understanding of the invention. However, it will be apparent to one skilled in the art that specific details are not required to practice the invention. Thus, the foregoing descriptions of specific embodiments of the invention are presented for purposes of illustration and description. They are not intended to be exhaustive or to limit the invention to the precise forms disclosed. Obviously, many modifications and variations are possible in light of the above teachings. The embodiments have been chosen and described in order to explain the principles of the invention and its practical application, thereby enabling those skilled in the art to utilize the invention and its various embodiments with various modifications as may be suitable for the particular use contemplated. It is intended that the following claims and their equivalents define the scope of the invention.

Claims

1. 1. A computer-implemented method for optimizing an ovarian stimulation process, the method comprising: receiving patient-specific data associated with the patient; predicting an oocyte outcome for the patient for each of a plurality of treatment options for an ovarian stimulation process based on at least one predictive model and the patient-specific data, the at least one predictive model being trained using prior patient-specific data associated with a plurality of prior patients; A method comprising:

2. 10. The method of claim 1, further comprising providing the predicted egg outcome to a medical professional for selecting among the plurality of treatment options.

3. 10. The method of claim 1, wherein the predicted egg outcome comprises at least one of a number of eggs retrieved and a number of mature eggs.

4. 2. The method of claim 1, wherein the predicted egg outcomes include at least one of maturity yield, number of mature eggs, number of fertilized eggs, number of embryos, number of blastocysts, number of usable blastocysts, number of euploid blastocysts, fetal heart rate, and birth rate.

5. 2. The method of claim 1, wherein the patient-specific data includes at least one of age, body mass index, ethnicity, infertility diagnosis, previous pregnancy history, previous childbirth history, and one or more baseline measurements selected from the group consisting of estradiol (E2) measurements, luteinizing hormone (LH) measurements, progesterone (P4) measurements, follicle stimulating hormone (FSH) measurements, anti-Mullerian hormone (AMH) measurements, and antral follicle count (AFC) measurements.

6. The method of claim 1 , wherein the patient-specific data includes information related to one or more previous in vitro fertilization (IVF) treatments associated with the patient.

7. 7. The method of claim 6, wherein the information related to the one or more previous IVF treatments includes at least one of data retrieved during ovarian stimulation, number of eggs retrieved, number of mature eggs, number of eggs successfully fertilized, number of blastocysts, number of usable blastocysts, pregnancy outcome, and birth outcome.

8. The method of claim 1 , wherein the patient-specific data includes one or more treatment variables associated with the patient.

9. 9. The method of claim 8, wherein the one or more treatment variables include a type of medication associated with the patient, a type of hormone triggering injection to induce follicular maturation, and a number of cycles.

10. 2. The method of claim 1, wherein the at least one predictive model is configured to provide the predicted egg outcome associated with each of a plurality of candidate doses of one or more ovarian stimulation medications administered to the patient.

11. 11. The method of claim 10, wherein the one or more ovarian stimulation medications include at least one of follicle stimulating hormone (FSH) or luteinizing hormone (LH).

12. 11. The method of claim 10, wherein the patient-specific data comprises one or more baseline measurements selected from the group consisting of an estradiol (E2) measurement, a luteinizing hormone (LH) measurement, a progesterone (P4) measurement, a follicle-stimulating hormone (FSH) measurement, an anti-Mullerian hormone (AMH) measurement, and an antral follicle count (AFC) measurement.

13. 11. The method of claim 10, wherein the at least one predictive model is configured to provide the predicted egg outcome associated with each of a first dose and a second dose, the second dose being greater than the first dose.

14. 11. The method of claim 10, wherein the at least one predictive model is configured to provide the predicted egg outcome associated with each of a first dose ratio of a plurality of ovarian stimulation medication regimens and a second dose ratio of a plurality of ovarian stimulation medication regimens, the first dose ratio being different from the second dose ratio.

15. further comprising identifying a set of prior patients similar to the patient based on the patient-specific data via similarity matching techniques; 11. The method of claim 10, wherein the at least one predictive model comprises a predicted dose-response curve generated based on the previous patient-specific data associated with the previous patient, the dose-response curve predicting the egg outcome that varies with doses of one or more ovarian stimulation medications administered to the patient.

16. 16. The method of claim 15, further comprising classifying the patient as one of: (i) a dose-responsive patient having a predicted egg outcome that varies substantially with doses of one or more ovarian stimulation medication regimens administered to the patient; and (ii) a dose-non-responsive patient having a predicted egg outcome that does not vary substantially with doses of one or more ovarian stimulation medication regimens administered to the patient, wherein the classification is based at least in part on the shape of the dose-response curve.

17. 17. The method of claim 16, further comprising recommending an optimal dose of the one or more ovarian stimulation medications to be administered to the patient based on the point on the dose-response curve that is associated with a predicted maximum egg outcome when the patient is classified as a dose-responsive patient.

18. A method for predicting a first egg outcome resulting from administering a hormone induction injection on a first candidate hormone induction date; predicting a second egg outcome resulting from administering said hormone induction injection on a second candidate hormone induction date; and Further comprising: The method of claim 1 , wherein the hormone triggering injection is configured to induce follicular maturation in the patient.

19. 20. The method of claim 18, wherein the first date is a current date and the second date is a future date.

20. 20. The method of claim 18, wherein the patient-specific data includes a current day follicle measurement for the patient.

21. 20. The method of claim 18, wherein the patient-specific data further comprises one or more of a current estradiol (E2) level or a current progesterone (P4) level for the patient.

22. 20. The method of claim 18, wherein the first and second predicted egg outcomes include at least one of: number of eggs retrieved, number of mature eggs, number of eggs fertilized, and number of usable blastocysts.

23. 20. The method of claim 18, wherein the at least one predictive model comprises a generalized linear model selected from the group consisting of a linear regression model, a Poisson regression model, and a negative binomial regression model.

24. 24. The method of claim 23, wherein the at least one predictive model is configured to predict E2 levels for the patient at a future date.

25. 24. The method of claim 23, wherein the at least one predictive model is configured to predict follicular measurements for the patient at a future date.

26. 24. The method of claim 23, wherein the at least one predictive model is configured to predict the first or second egg outcome based at least in part on at least one of E2 levels and follicle measurements.

27. 20. The method of claim 18, further comprising displaying the first and second egg outcomes on a display.

28. 1. A computer-implemented method for optimizing an ovarian stimulation process, the method comprising: receiving patient-specific data associated with the patient; predicting via a first predictive model a first egg outcome resulting from administering a hormone induction injection on a first candidate hormone induction date; predicting via a second predictive model a second egg outcome resulting from administering said hormone induction injection on a second candidate hormone induction date; and Including, The method, wherein the hormone triggering injection is configured to induce follicular maturation in the patient.

Images