Application method and system of an intervention model for predicting the health of pregnant women with epilepsy

Abstract

The application belongs to the technical field of medical artificial intelligence, clinical risk prediction and neurology / obstetrics cross management, and particularly relates to an application method and system of an intervention model for predicting the health of pregnant women with epilepsy. The method comprises the following steps: providing an intervention model library; selecting at least one matched candidate model as a first intervention model from the intervention model library according to the personal information of a pregnant woman with epilepsy; using the first intervention model to predict the health of the pregnant woman with epilepsy to obtain corresponding output information; and judging whether the first intervention model needs to be updated according to the output information. The application can realize early identification and hierarchical management of seizures during pregnancy by constructing risk prediction models for different pregnancy periods, thereby providing decision support for clinical doctors to develop individualized diagnosis and treatment and follow-up schemes.

Description

Technical Field

[0001] This invention relates to the fields of medical artificial intelligence, clinical risk prediction, and neurology / obstetrics cross-management technology, specifically to an application method and system for an intervention model that predicts the health of pregnant women with epilepsy. Background Technology

[0002] Epilepsy is a common chronic neurological disorder in women of childbearing age. Seizures during pregnancy are associated with a significantly increased risk of maternal death, premature birth, low birth weight, and fetal growth restriction. While maintenance therapy with antiepileptic medications (ASMs) helps control seizures, pregnant women, out of concern for teratogenicity and fetal safety, may discontinue or reduce their medication on their own, leading to an increased risk of relapse or status epilepticus, further increasing the risk of maternal and fetal hypoxia and serious complications. Therefore, timely prediction and intervention for seizures during pregnancy are crucial.

[0003] However, relying on doctors' experience and past seizure history for qualitative judgment has low predictive efficiency. In response, traditional techniques have attempted to use models to predict the risk of seizures in pregnant women with epilepsy.

[0004] For example, patent application CN120260900A proposes a self-management behavior assistance system for patients with pregnancy complicated by epilepsy, belonging to the field of medical monitoring and diagnostic technology. This assistance system includes: a detection unit for acquiring physiological parameters related to the patient's pregnancy complicated by epilepsy; and a processing unit for analyzing and processing the data acquired by the detection unit to assess the patient's health trends. It also incorporates a predictive model based on historical data and individual differences among pregnant women.

[0005] However, the triggers for seizures in pregnant women with epilepsy are very complex and vary greatly from person to person.

[0006] Therefore, there is an urgent need for a predictive tool that can quantify the risk of epilepsy during pregnancy and can run in real time in hospital information systems to support the development of individualized intervention and follow-up plans. Summary of the Invention

[0007] The purpose of this invention is to provide an application method and system for an intervention model that predicts the health of pregnant women with epilepsy, which partially solves or alleviates the above-mentioned shortcomings in the prior art. It can realize the early identification and hierarchical management of epileptic seizures during pregnancy by constructing a pregnancy-specific risk prediction model, thereby providing decision support for clinicians to formulate individualized diagnosis and treatment and follow-up plans.

[0008] To solve the aforementioned technical problems, the present invention specifically adopts the following technical solution: A first aspect of the present invention is to provide a method for applying an intervention model to predict the health of pregnant women with epilepsy, comprising the steps of: S201, providing an intervention model library, the intervention model library including: multiple candidate models, the input information of the candidate models including: multiple predictive factors, and the predictive factors including: modifiable factors and non-modifiable factors, the output information of the candidate models including: health risks of pregnant women with epilepsy, and the correlation of the predictive factors with health risks; Among them, the interventionable factors refer to factors that can be directly and artificially controlled by the user, the non-interventionable factors refer to objective physiological indicators that reflect the current physical state of pregnant women with epilepsy, and / or, the non-interventionable factors refer to factors that cannot be effective within the set intervention period. S202, Based on the personal information of the pregnant woman with epilepsy, at least one matching candidate model is selected from the intervention model library as the first intervention model. S203, Use the first intervention model to perform health prediction on the pregnant woman with epilepsy to obtain corresponding output information; S204, determine whether the first intervention model needs to be updated based on the output information; wherein, S204 includes: Obtain the head factor, which refers to the predictive factor that ranks first in a predetermined sequence length when the factors are sorted according to the magnitude of their correlation. Calculate the proportion of the modifiable factors among the head factors; When the intervention ratio is less than the set first ratio threshold, a model update suggestion is generated, and S202 is executed again in response to the model update suggestion.

[0009] In some embodiments, the interventionable factors are associated with difficulty in intervention, and S202 includes the following steps: S2021, Select at least one matching candidate model from the intervention model library based on the personal information of the pregnant woman with epilepsy; S2022, Calculate the execution difficulty of the candidate model based on the intervention difficulty; S2023, Assess the compliance of the pregnant woman with epilepsy based on her historical medical records, and set execution limits based on the level of compliance; S2024, Select the candidate model whose execution difficulty is less than or equal to the execution limit as the first intervention model.

[0010] In some embodiments, the interventionable factor is further associated with an intervention waiting period, which is used to assess the necessary time required for the interventionable factor to take effect from implementation; correspondingly, S202 further includes: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; Calculate the second duration between the intervention waiting period and the first duration; When the second duration is less than the effective period, the interventionable factor is updated to an ininterventionable factor.

[0011] In some embodiments, the steps further include: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; When the first duration is less than or equal to the set production cycle, the execution limit will be reduced.

[0012] In some embodiments, it also includes: The difficulty of the intervention is calculated based on the personal data; the difficulty of the intervention is defined by at least one of the following difficulty indicators, which include: expected cost and comorbidity.

[0013] In some embodiments, it also includes: When the intervention ratio of the first intervention model is greater than the second ratio threshold, a combined recommendation is generated. The combined recommendation requires the selection of at least one second intervention model, and requires that the non-intervention ratio of the non-intervention factors in the second intervention model is greater than the set third ratio threshold.

[0014] In some embodiments, it also includes: After the first intervention model has been running for at least one treatment cycle, the first intervention model is used to predict the health risks of the pregnant woman in the next treatment cycle through historical intervention strategies and updated intervention strategies to obtain a first predicted risk; wherein, the historical intervention strategy refers to the intervention methods applied to the patient in the previous treatment cycle, and the updated intervention strategy refers to the new intervention methods proposed by the doctor for the current treatment cycle. Using at least one of the second intervention models, a second predicted risk is obtained by predicting the health risks of pregnant women in the next treatment cycle through local intervention strategies; wherein, the local intervention strategy refers to one or more of the updated intervention strategies; If the risk difference between the first predicted risk and the second predicted risk is greater than a preset difference, then the compliance level is reduced.

[0015] A second aspect of the present invention is to provide an application system for an intervention model that predicts the health of pregnant women with epilepsy, comprising: The model library provides a module for providing an intervention model library, which includes: multiple candidate models, the input information of which includes: multiple predictive factors, and the predictive factors include: modifiable factors and non-modifiable factors, and the output information of which includes: the health risk of pregnant women with epilepsy, and the correlation between the predictive factors and the health risk; Among them, the interventionable factors refer to factors that can be directly and artificially controlled by the user, the non-interventionable factors refer to objective physiological indicators that reflect the current physical state of pregnant women with epilepsy, and / or, the non-interventionable factors refer to factors that cannot be effective within the set intervention period. The model screening module is used to select at least one matching candidate model from the intervention model library as the first intervention model based on the personal information of the pregnant woman with epilepsy. The health prediction module is used to make a health prediction for the pregnant woman with epilepsy using the first intervention model, so as to obtain the corresponding output information; An update judgment module is used to determine whether the first intervention model needs to be updated based on the output information; wherein, the update judgment module is used to perform: Obtain the head factor, which refers to the predictive factor that ranks first in a predetermined sequence length when the factors are sorted according to the magnitude of their correlation. Calculate the proportion of the modifiable factors among the head factors; When the intervention ratio is less than the set first ratio threshold, a model update suggestion is generated, and the system re-enters the model screening module in response to the model update suggestion.

[0016] In some embodiments, the association between the modifiable factors and the difficulty of intervention is addressed by the model screening module, which is used to perform the following: A candidate model screening unit is used to screen at least one matching candidate model from the intervention model library based on the personal information of the pregnant woman with epilepsy. An execution difficulty calculation unit is used to calculate the execution difficulty of the candidate model based on the intervention difficulty. An execution limit setting unit is used to assess the compliance of a pregnant woman with epilepsy based on her historical medical records, and to set an execution limit based on the level of compliance. The model selection unit is used to select the candidate model whose execution difficulty is less than or equal to the execution limit as the first intervention model.

[0017] In some embodiments, the interventionable factor is further associated with an intervention waiting period, which is used to assess the necessary time required for the interventionable factor to take effect from implementation; correspondingly, the model screening module is used to perform: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; Calculate the second duration between the intervention waiting period and the first duration; When the second duration is less than the effective period, the interventionable factor is updated to an ininterventionable factor.

[0018] Beneficial technical effects: (1) This invention selects to evaluate the effectiveness of the first intervention model based on the proportion of intervention in head factors, i.e. whether it is executable. In other words, this invention selects to update the model only when the executableness of the intervention model is low (such as when there are too few human interventionable parts, resulting in the model being unable to guide actual treatment). This makes it easier for doctors to formulate specific intervention measures in the future, rather than just indicating risk information.

[0019] (2) By limiting the proportion of interventions, a highly feasible intervention model can be selected, and the direction of model updates can be restricted as the condition changes during pregnancy. This can minimize the generation of low-feasibility output information by the intervention model (such as "because the patient has a long history of refractory epilepsy, the risk of seizures is high"). From this perspective, the present invention has particularly high clinical value.

[0020] (3) The present invention can also calculate the execution difficulty of the model based on the difficulty of intervention of the interventionable factors, so as to determine whether the execution difficulty of the model is higher than the execution limit of the current patient (i.e., whether the model is suitable for the individual). By doing so, a more feasible intervention model can be selected as soon as possible with fewer conditions, thereby ensuring that the recommended intervention measures can be actually implemented under the current conditions and behavioral tendencies of the patient, reducing the risk of sudden failure or model failure caused by the patient's inability or unwillingness to cooperate.

[0021] (4) When selecting the first intervention model, the present invention can also select the update strategy based on the time dimension. For example, the present invention can also compare the length of time between the intervention waiting period and the remaining time to determine whether to update the interventionable factors. That is, the interventionable factors are updated based on the feasibility in the time cycle, thereby further restricting the direction of model selection.

[0022] (5) When the first intervention model is highly subjective (i.e., the intervention ratio is relatively high), the intervention operation may heavily rely on the doctor's supervision and the patient's active cooperation. In this regard, the present invention chooses to introduce a second intervention model with higher objectivity (i.e., a higher non-intervention ratio), which can balance the output results of the first intervention model with its high subjectivity even if the actual intervention is very difficult. This mechanism of mutual reconciliation between subjectivity and objectivity also helps to improve the overall feasibility of the intervention model.

[0023] (6) This invention uses a combined model application effect verification mechanism to determine whether the first intervention model truly needs to be updated. This also allows for appropriate restrictions on model updates, thereby helping to reduce the cost of model updates. In other words, this invention can identify and verify deceptive behaviors that are not easily identifiable in a low-cost and relatively simple way, thereby distinguishing the source of error in models with high intervention ratios that fail, and thus updating the first intervention model more conservatively. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. The elements or parts in the drawings are not necessarily drawn to scale. Obviously, the drawings described below are some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0025] Figure 1 A flowchart illustrating the application method of an intervention model for predicting the health of pregnant women with epilepsy provided by this invention; Figure 2 A schematic diagram of the structure of an application system for an intervention model for predicting the health of pregnant women with epilepsy, provided by the present invention; Figure 3 This is a schematic diagram of the model evaluation curve provided by the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0027] In this document, suffixes such as "module," "part," or "unit" used to denote elements are used only for the purpose of illustrative purposes and have no specific meaning in themselves. Therefore, "module," "part," or "unit" may be used interchangeably.

[0028] In this document, the terms "upper," "lower," "inner," "outer," "front," "rear," "one end," and "the other end," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the present invention and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0029] In this document, unless otherwise explicitly specified and limited, the terms "installed," "equipped with," "connected," etc., should be interpreted broadly. For example, "connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection, a direct connection, or an indirect connection through an intermediate medium; it can be a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0030] In this document, "and / or" includes any and all combinations of one or more of the listed related items.

[0031] In this article, "multiple" means two or more, that is, it includes two, three, four, five, etc.

[0032] As used in this specification, the term "about" typically means + / -5% of the value, more typically + / -4% of the value, more typically + / -3% of the value, more typically + / -2% of the value, even more typically + / -1% of the value, and even more typically + / -0.5% of the value.

[0033] In this specification, certain embodiments may be disclosed in a range-bound format. It should be understood that this "range-bound" description is merely for convenience and brevity and should not be construed as a rigid limitation on the disclosed range. Therefore, the description of a range should be considered as having specifically disclosed all possible subranges and the individual numerical values ​​within those ranges. For example, a description of the range 1-6 should be considered as having specifically disclosed subranges such as from 1 to 3, from 1 to 4, from 1 to 5, from 2 to 4, from 2 to 6, from 3 to 6, etc., and the individual numbers within those ranges, such as 1, 2, 3, 4, 5, and 6. This rule applies regardless of the breadth of the range.

[0034] Example 1: For pregnant women with epilepsy, hormonal fluctuations (estrogen and progesterone) during early, mid, and late pregnancy and postpartum are significant, with marked genetic differences in hormone sensitivity and drug metabolism. Risk patterns also vary considerably among patients at the same gestational week. Furthermore, changes in hormone levels, sleep, and mood during pregnancy can lead to significant fluctuations in the frequency of epileptic seizures (some pregnant women experience an increase in seizures, while others experience a decrease). Therefore, it is very difficult to dynamically predict and intervene in epileptic seizures based on these changes.

[0035] Furthermore, due to the extremely high time sensitivity and individual variability in fetal nervous system development, even pregnant women with similar gestational ages, medication regimens, and physical conditions may experience differences in the fit of risk prediction models, and these differences can have a decisive impact on prenatal intervention decisions. Therefore, such AI models will face significant adaptability challenges, which also imposes extremely high cost requirements on the deployment and application of these models.

[0036] For example, the physiological state of pregnant women with epilepsy can shift significantly as pregnancy progresses. This shift may stem from multiple factors, including fluctuations in the blood concentration of antiepileptic drugs, changes in the frequency and type of seizures, hormonal remodeling during pregnancy, and alterations in nutritional and metabolic status. When the pregnant woman's state shifts, the distribution of risk factors relied upon by the AI ​​model trained on historical data (such as early pregnancy or previous pregnancies) also changes, leading to decreased predictive accuracy and even risk assessment results inconsistent with the current gestational age and physical signs. Therefore, to ensure the effectiveness of the model, it must be regularly updated and maintained to track and adapt to the pregnant woman's latest condition.

[0037] Therefore, the condition of pregnant women with epilepsy often progresses rapidly, with a very short window from risk detection to action. Furthermore, different patients respond very differently to various treatments (medications, lifestyle modifications, etc.).

[0038] In response, this invention selects different intervention models for different types of pregnant women with epilepsy in order to improve the adaptability of the intervention model to pregnant women with epilepsy.

[0039] However, if the condition of a pregnant woman with epilepsy changes significantly, the modifiable factors output by the intervention model may no longer be applicable, and the model can be updated. However, considering that user compliance can affect the effectiveness of the intervention model's output, we propose a restrictive intervention model update mechanism.

[0040] First, this invention selects to evaluate the effectiveness of the first intervention model based on the proportion of interventions in head factors, i.e., whether it is executable. In other words, this invention chooses to update the model only when the executableness of the intervention model is low (such as when the part that can be intervened by humans is too small, causing the model to be unable to guide actual treatment). This makes it easier for doctors to formulate specific intervention measures (such as adjusting medications or strengthening follow-up) rather than just indicating risk information.

[0041] Specifically, please see Figure 1 This invention provides a method for applying an intervention model to predict the health of pregnant women with epilepsy, comprising the following steps: S201, providing an intervention model library, the intervention model library including: multiple candidate models, the input information of the candidate models including: multiple predictive factors, and the predictive factors including: modifiable factors and non-modifiable factors, the output information of the candidate models including: health risks of pregnant women with epilepsy, and the correlation of the predictive factors with health risks; Among them, the interventionable factors refer to factors that can be directly and artificially controlled by the user, the non-interventionable factors refer to objective physiological indicators that reflect the current physical state of pregnant women with epilepsy, and / or, the non-interventionable factors refer to factors that cannot be effective within the set intervention period. S202, Based on the personal information of the pregnant woman with epilepsy, at least one matching candidate model is selected from the intervention model library as the first intervention model. S203, Use the first intervention model to perform health prediction on the pregnant woman with epilepsy to obtain corresponding output information; S204, determine whether the first intervention model needs to be updated based on the output information; wherein, S204 includes: Obtain the head factor, which refers to the predictive factor that ranks first in a predetermined sequence length when the factors are sorted according to the magnitude of their correlation. Calculate the proportion of the modifiable factors among the head factors; When the intervention ratio is less than the set first ratio threshold, a model update suggestion is generated, and S202 is executed again in response to the model update suggestion.

[0042] It should be understood that, in this embodiment, by limiting the proportion of interventions, a highly feasible intervention model can be selected, and the direction of model updates can be restricted as the condition changes during pregnancy. This approach minimizes the generation of low-feasibility output information by the intervention model (such as "because the patient has a long history of refractory epilepsy, the risk of seizures is high"). From this perspective, the present invention has particularly high clinical value.

[0043] Preferably, the attending physician can select at least one matching candidate model as the first intervention model based on the personal information of the pregnant woman with epilepsy.

[0044] For example, if a pregnant woman with epilepsy is in the first trimester, a candidate model suitable for the first trimester can be selected as the first intervention model.

[0045] In some embodiments, the relevance of a predictor to health risk refers to the degree to which the predictor contributes to the health risk.

[0046] In some embodiments, the intervention percentage = number of interventionable factors / number of head factors.

[0047] For example, if the first intervention model outputs 5 head factors, 2 of which are interventionable factors, then the intervention ratio is 40%.

[0048] In some embodiments, the interventionable factors are associated with difficulty in intervention, and S202 includes the following steps: S2021, Select at least one matching candidate model from the intervention model library based on the personal information of the pregnant woman with epilepsy; S2022, Calculate the execution difficulty of the candidate model based on the intervention difficulty; S2023, Assess the compliance of the pregnant woman with epilepsy based on her historical medical records, and set execution limits based on the level of compliance; S2024, Select the candidate model whose execution difficulty is less than or equal to the execution limit as the first intervention model.

[0049] In some embodiments, the difficulty of execution is proportional to the difficulty of intervention.

[0050] In some embodiments, the execution difficulty of candidate models can be scored. The execution difficulty score can be the sum of the intervention difficulty scores of each modifiable factor.

[0051] For example, if the intervenable factors output by candidate model A include a, b, and c, which are associated with intervention difficulties of 1, 2, and 3 respectively, then the execution difficulty of candidate model A is 6.

[0052] In some embodiments, medication adherence in pregnant women with epilepsy can be assessed by analyzing at least one of the following factors based on their historical medical records: medication refill frequency, follow-up visit rate, monitoring completion rate, and doctor-patient communication records. For example, a higher follow-up visit rate indicates higher adherence.

[0053] In some embodiments, compliance of pregnant women with epilepsy can be scored and rated by healthcare professionals.

[0054] In some embodiments, the execution limit may also be referred to as the execution difficulty threshold.

[0055] In this embodiment, the present invention can also calculate the execution difficulty of the model based on the difficulty of intervention of the interventionable factors, in order to determine whether the execution difficulty of the model is higher than the execution limit of the current patient (i.e., whether the model is suitable for the individual). This approach allows for the rapid selection of a more feasible intervention model with fewer conditions, thereby ensuring that the recommended intervention measures are likely to be implemented given the current patient's actual conditions and behavioral tendencies, reducing the risk of unforeseen events or model failure due to the patient's inability or unwillingness to cooperate.

[0056] Furthermore, when selecting the first intervention model, this invention can also select the update strategy based on the time dimension.

[0057] For example, this invention can also compare the length of time between the intervention waiting period and the remaining time to determine whether to update the interventionable factors. That is, it updates the interventionable factors based on the feasibility in terms of time cycle, thereby further restricting the direction of model selection.

[0058] For example, the execution limits can be updated based on whether the pregnant woman is about to give birth, or the direction of model updates can be restricted from the perspective of the individual pregnant woman's execution ability, so as to narrow the range of model selection.

[0059] Specifically, in some embodiments, the interventionable factor is further associated with an intervention waiting period, which is used to assess the necessary time required for the interventionable factor to take effect from implementation; correspondingly, S202 further includes: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; Calculate the second duration between the intervention waiting period and the first duration; When the second duration is less than the effective period, the interventionable factor is updated to an ininterventionable factor.

[0060] In some embodiments, the steps further include: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; When the first duration is less than or equal to the set production cycle, the execution limit will be reduced.

[0061] In some embodiments, it also includes: The difficulty of the intervention is calculated based on the personal data; the difficulty of the intervention is defined by at least one of the following difficulty indicators, which include: expected cost and comorbidity (such as the number of comorbidities).

[0062] In some embodiments, the lower the expected cost, the more difficult the intervention. The more comorbidities, the more difficult the intervention.

[0063] In some embodiments, different cost expectations and comorbidities can be scored and rated. When the difficulty of intervention is defined by two difficulty indicators, cost expectation and comorbidity, the intervention difficulty score can be the sum of the scores corresponding to cost expectation and comorbidity.

[0064] In some embodiments, it also includes: When the intervention ratio of the first intervention model is greater than the second ratio threshold, a combined recommendation is generated. The combined recommendation requires the selection of at least one second intervention model, and requires that the non-intervention ratio of the non-intervention factors in the second intervention model is greater than the set third ratio threshold.

[0065] For example, in some embodiments, when the prediction results given by the first intervention model and the second intervention model are highly similar, the feasibility of the first intervention model can be verified to some extent.

[0066] The second proportional threshold is greater than the first proportional threshold.

[0067] In this embodiment, if the intervention ratio is high, that is, most or all of the model output consists of modifiable factors, the present invention chooses to introduce a second intervention model to achieve an appropriate balance between modifiable and non-modifiable factors, so as to avoid excessive reliance on manual intervention and thus improve the overall intervention effect.

[0068] From another perspective, this also restricts model replacement to a certain extent, that is, it avoids directly replacing the first intervention model as much as possible.

[0069] Furthermore, if the first intervention model assumes that the risk of seizures is largely determined by modifiable factors, then if the patient does not cooperate (e.g., not taking medication on time or not having regular follow-up visits), the low risk predicted by the model may not hold true, the actual risk will increase, and the model's output may be overly optimistic.

[0070] In clinical practice, adherence to medication by pregnant women with epilepsy may also be unreliable (e.g., discontinuing medication due to concerns about fetal malformation). Therefore, models that rely too heavily on modifiable factors may ultimately lead to unsound predictions.

[0071] Furthermore, the factors that can be intervened may have limitations such as difficulty in intervention and waiting period, meaning that intervention may not be possible for all of them. If the model ignores these limitations, a high intervention rate may mislead doctors into believing that the risks can be easily controlled, when in reality some interventions are either too time-consuming or too costly.

[0072] In other words, when the first intervention model is highly subjective (i.e., has a high intervention ratio), the intervention operation may heavily rely on the doctor's supervision and the patient's active cooperation. To address this, this invention introduces a second intervention model with higher objectivity (i.e., a higher non-intervention ratio). Even if the actual intervention is very difficult, this model can balance the more subjective output of the first intervention model. This mechanism of reconciling subjectivity and objectivity also helps improve the overall feasibility of the intervention model.

[0073] It should be understood that the first intervention model provides patients with an actionable intervention path, while the second intervention model avoids predictive failure due to patient non-compliance, thus providing an objective risk baseline independent of patient behavior. In cases where the first intervention model may fail to predict, a more conservative model combination strategy can achieve a better balance between subjective feasibility and objective real-world risk.

[0074] In some embodiments, when the outputs of the first intervention model and the second intervention model are highly similar (e.g., both are high-risk), it indicates that the first intervention model is relatively reliable. However, because the course of epilepsy in pregnant women is very complex, a model that is highly reliable in the previous course may experience significant fluctuations in the next course. To address this, this invention also proposes a multi-model prediction comparison and verification mechanism.

[0075] Preferably, the input information of the main intervention model (i.e., the first intervention model) includes: the patient's current information (such as current vital signs and the intervention methods given by the doctor for the updated intervention strategy), and also includes historical information (i.e., historical intervention strategies, such as historical medication records). The input information of the auxiliary intervention model (i.e., the second intervention model) mainly includes: the patient's current information (such as the intervention methods given by the doctor for the local intervention strategy).

[0076] For example, in some embodiments, the steps include: After the first intervention model has been running for at least one treatment cycle, the first intervention model continues to be used to predict the health risks of pregnant women in the next treatment cycle through historical intervention strategies (such as historical medication records, exercise recommendations, dietary recommendations, etc.) and updated intervention strategies (such as new plans that doctors will implement in the next treatment cycle, such as increasing the dosage or adjusting the medication time, etc.), to obtain the first predicted risk; wherein, the historical intervention strategy refers to the intervention methods used by the patient in the previous treatment cycle, and the updated intervention strategy refers to the new intervention methods proposed by the doctor for the current treatment cycle; Using at least one of the second intervention models, a second predicted risk is obtained by predicting the health risks of pregnant women in the next treatment cycle through a local intervention strategy (i.e., it only selects one or more pieces of information from the updated intervention strategy, such as only selecting the medication dosage and excluding others such as sleep, stress management, exercise management, etc.); wherein, the local intervention strategy refers to one or more pieces of information from the updated intervention strategy. If the risk difference between the first predicted risk and the second predicted risk is greater than a preset difference, then the compliance level is reduced.

[0077] Furthermore, when the compliance level is lower than a preset level, a data verification suggestion is generated.

[0078] In some embodiments, data verification recommendations are used to advise physicians to reassess the extent of patient compliance.

[0079] In this embodiment, by comparing the difference between the first predicted risk and the second predicted risk, if the output risks of the first intervention model and the second intervention model are similar, it indicates that the data of the historical intervention strategy is highly reliable, that is, the patient may have had good compliance in the past, and the original compliance level can be maintained.

[0080] If the risk output by the first intervention model is significantly lower than that output by the second intervention model, it indicates poor compliance with historical real-world behavior (i.e., there may be underreporting or false reporting in the implementation of historical intervention strategies), and in this case, the compliance level can be lowered.

[0081] Furthermore, multiple second intervention models can be selected to validate the first intervention model multiple times. For example, the updated intervention strategy can include various intervention methods; one or more of different intervention methods can be selected as local intervention strategies, corresponding to different second intervention models.

[0082] It should be understood that different patients have varying levels of acceptance of different interventions. For example, some patients may be resistant to medication, while others may have a stronger belief that medication is more effective. Therefore, local intervention strategies should select interventions with higher adherence probabilities to ensure that the predictive results of the second intervention model are more stable and reliable, thus allowing for comparison and validation with the first intervention model. In this case, the physician can choose the local intervention strategy based on the patient's past medical history.

[0083] The following section will provide an exemplary description of the intervention scenarios to which this embodiment is applicable: First, the doctor provides the patient with the initial intervention strategy based on the first intervention model A (hereinafter referred to as Model A). At this time, the second intervention model B (or simply Model B) will simultaneously perform local validation of the initial intervention strategy. If the results given by the two intervention models are relatively consistent, it indicates that the doctor's intervention strategy has a relatively low risk.

[0084] Subsequently, after one treatment cycle (e.g., 2 weeks), the patient returns for a follow-up visit, at which point the doctor simultaneously runs the first intervention model A and at least one second intervention model B.

[0085] In a subsequent new treatment cycle, the initial intervention strategy from the previous treatment cycle will be considered a historical intervention strategy, and any new intervention strategies proposed by the physician will be considered updated intervention strategies.

[0086] This could lead to several different scenarios: Scenario 1: Both intervention model A and intervention model B predict low risk, and the patient has not experienced a relapse during the current treatment cycle. The results of models A and B are consistent and validated, indicating high reliability of the patient's self-report and good adherence.

[0087] Scenario 2: Intervention model A predicts low risk, while intervention model B predicts high risk. In reality, the patient experiences seizures during the current treatment cycle. This may indicate that objective indicators (such as low blood drug concentration or epileptiform discharges on EEG) already suggest danger. However, the patient may have misrepresented their medication or sleep patterns (e.g., claiming to take medication daily but missing doses for several days, or claiming to sleep 7 hours but only sleeping 4 hours). Doctors can use this information to determine if there is concealment or poor adherence.

[0088] In scenario two, doctors can retrieve blood drug concentration reports and reassess patient compliance.

[0089] Scenario 3: Both the first intervention model A and the second intervention model B predict a high risk, and the patient actually has an epileptic seizure. This may indicate that the patient did not conceal the fact, but the existing treatment plan needs to be adjusted (increase the dosage or change the medication).

[0090] In other words, by comparing a high-intervention model that relies on subjective medication factors with a low-intervention model that relies on objective physiological indicators, doctors can identify potential concealment or underreporting while reducing reliance on patient self-reporting, thereby improving the effectiveness of long-term epilepsy management.

[0091] In this embodiment, the present invention uses a combined model application effect verification mechanism to determine whether the first intervention model really needs to be updated. This can also appropriately restrict model updates, thereby helping to reduce model update costs.

[0092] In summary, by using the restrictive intervention model update strategy provided by this invention, the update direction can be appropriately restricted during the model update process for pregnant women with epilepsy, thereby ensuring model fit while reasonably limiting model updates.

[0093] It is worth noting that if a pregnant woman with epilepsy conceals the fact that she has stopped taking medication, reduced medication, or taken medication irregularly, or exaggerates her adherence to medical advice (such as "I take my medication on time every day"), or if there are errors or omissions when entering patient information, the output results of the first intervention model may not match the actual seizure outcome.

[0094] Especially for the first intervention model with a high intervention rate, its intervention effect is highly dependent on the user's subjective performance. In other words, the prediction results of this type of model mainly come from the patient's self-report or behavioral records (such as "whether medication is taken regularly", "whether there are early follow-up visits", "sleep quality"). This information is very easy to be embellished or concealed by the patient or the recorder, whether consciously or unconsciously.

[0095] For example, when a patient conceals the fact that they have stopped taking medication, the high intervention ratio model will still be based on the input-output of good compliance and low seizure risk, while the actual patient may be about to have a seizure.

[0096] In this regard, the present invention essentially proposes a verification mechanism for a high intervention ratio model based on both subjective and objective factors.

[0097] High intervention ratio models are the most desirable to retain in clinical practice (because they can effectively guide intervention treatment), but they are also the most susceptible to the influence of patients' subjective behavior. The validation mechanism provided in this invention can verify whether the first intervention model is truly suitable for the patient by introducing a second intervention model. This approach attributes model failures in some cases to the aforementioned deceptive behavior of users, rather than model performance issues, thereby avoiding unnecessary model updates. Consequently, the model can possess good interventionability while avoiding excessively frequent updates.

[0098] Specifically, if the output of the second intervention model differs significantly from that of the first intervention model, it may indicate that the user has engaged in deceptive behavior. In this case, it is advisable to retest the blood drug concentration and inquire in detail about the history of drug discontinuation.

[0099] In other words, the present invention can identify and verify deceptive behaviors that are not easy to detect in a low-cost and relatively simple way, thereby distinguishing the source of error of model failure and updating the first intervention model in a more conservative manner.

[0100] Please see Figure 2 The present invention also provides an application system for an intervention model for predicting the health of pregnant women with epilepsy, comprising: The model library provides a module for providing an intervention model library, which includes: multiple candidate models, the input information of which includes: multiple predictive factors, and the predictive factors include: modifiable factors and non-modifiable factors, and the output information of which includes: the health risk of pregnant women with epilepsy, and the correlation between the predictive factors and the health risk; Among them, the interventionable factors refer to factors that can be directly and artificially controlled by the user, the non-interventionable factors refer to objective physiological indicators that reflect the current physical state of pregnant women with epilepsy, and / or, the non-interventionable factors refer to factors that cannot be effective within the set intervention period. The model screening module is used to select at least one matching candidate model from the intervention model library as the first intervention model based on the personal information of the pregnant woman with epilepsy. The health prediction module is used to make a health prediction for the pregnant woman with epilepsy using the first intervention model, so as to obtain the corresponding output information; An update judgment module is used to determine whether the first intervention model needs to be updated based on the output information; wherein, the update judgment module is used to perform: Obtain the head factor, which refers to the predictive factor that ranks first in a predetermined sequence length when the factors are sorted according to the magnitude of their correlation. Calculate the proportion of the modifiable factors among the head factors; When the intervention ratio is less than the set first ratio threshold, a model update suggestion is generated, and the system re-enters the model screening module in response to the model update suggestion.

[0101] In some embodiments, the association between the modifiable factors and the difficulty of intervention is addressed by the model screening module, which is used to perform the following: A candidate model screening unit is used to screen at least one matching candidate model from the intervention model library based on the personal information of the pregnant woman with epilepsy. An execution difficulty calculation unit is used to calculate the execution difficulty of the candidate model based on the intervention difficulty. An execution limit setting unit is used to assess the compliance of a pregnant woman with epilepsy based on her historical medical records, and to set an execution limit based on the level of compliance. The model selection unit is used to select the candidate model whose execution difficulty is less than or equal to the execution limit as the first intervention model.

[0102] In some embodiments, the interventionable factor is further associated with an intervention waiting period, which is used to assess the necessary time required for the interventionable factor to take effect from implementation; correspondingly, the model screening module is used to perform: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; Calculate the second duration between the intervention waiting period and the first duration; When the second duration is less than the effective period, the interventionable factor is updated to an ininterventionable factor.

[0103] It should be understood that the application system of the intervention model for predicting the health of pregnant women with epilepsy provided by the present invention can be used to implement the method steps described in any of the above embodiments.

[0104] In some embodiments, predictors (or candidate predictors) may include the following three aspects, totaling 24 items: 1) Demographic characteristics (5 items); Ethnic groups (Race): such as Han Chinese, ethnic minorities; Residence: such as town or rural area; Education level: such as below junior college, junior college and above; Personal income level: e.g., below 5000, ≥5000; Number of pregnancies (Gravidity): a continuous variable; 2) Epilepsy-related characteristics (11 items) Age of onset of epilepsy: a continuous variable; Age at conception: a continuous variable; Epilepsy duration: a continuous variable; Family history of epilepsy: Yes / No; Etiology of epilepsy: known causes, unknown causes; Epilepsy type: Focal epilepsy, generalized epilepsy, unknown type; Seizure type: with GTCS, without GTCS; Drug-resistant epilepsy: Yes / No; History of preconception surgery: Yes / No; Did you experience seizures in the nine months before pregnancy? Yes / No; Did you experience any seizures in the month before pregnancy? Yes / No; 3) Antiepileptic drug treatment (ASMs treatments, 8 items); Number of ASM exposures: no medication, discontinued during first pregnancy, monotherapy, combination therapy; Type of ASM exposure: Levetiracetam, Oxcarbazepine, Lamotrigine, Carbamazepine, Valproic acid, Topiramate, Phenytoin sodium, Other monotherapy, Combination therapy containing valproic acid, Combination therapy containing topiramate, Combination therapy containing valproic acid and topiramate, Levetiracetam + Oxcarbazepine, Levetiracetam + Lamotrigine, Carbamazepine + Levetiracetam, Other combination therapy, Discontinued during first pregnancy, No medication used; Whether or not levetiracetam was taken; Whether to take oxcarbazepine; Whether to take lamotrigine; Whether to take valproic acid; Should I take carbamazepine? Whether or not topiramate was taken.

[0105] In some embodiments, the interventionable factors may include: (1) Epilepsy-related characteristics For example, refractory epilepsy is one of the most important predictors of seizures during pregnancy. Seizure control can be improved through systematic preconception assessment, appropriate adjustment of antiepileptic drug (ASMs) regimens, and, if necessary, surgical treatment. Interventional approaches for refractory epilepsy include optimizing treatment plans and assessing surgical indications.

[0106] Pre-conception seizure history is the strongest predictor of seizures during pregnancy. If seizures are still occurring within 9 months or 1 month before conception, pregnancy should be delayed, medication dosage adjusted, or treatment strategy optimized until stable seizure control is achieved before attempting pregnancy. A possible intervention is intensive seizure control before pregnancy.

[0107] For patients with a history of preconception surgical treatment: Completing surgical evaluation and treatment before planning pregnancy can significantly improve long-term seizure control, thereby reducing pregnancy risks. A corresponding intervention could be early surgical evaluation.

[0108] (2) ASM treatment For example, the amount and type of ASM exposure. Discontinuing medication during pregnancy significantly increases the risk of flare-ups; multidrug therapy often indicates poor disease control; the recommended intervention is to switch to the lowest effective dose of monotherapy before pregnancy.

[0109] For example, whether or not specific ASMs are being taken. Corresponding interventions could include optimizing the type of medication before pregnancy and dynamically adjusting the dosage.

[0110] In some embodiments, the health risk output by the model can be a risk level (e.g., level 1-3) or a risk score (e.g., 1-10).

[0111] In some embodiments, the relevance of a predictor to health risk refers to the influence of the predictor on the model's prediction results. For example, the greater the influence, the greater the relevance of the predictor to health risk. Exemplarily, the SHAP (Shapley Additive Explanations) method can be used to quantify the influence of predictors on health risk prediction results.

[0112] In some embodiments, this invention preferably uses real-world cohort data from multiple centers in China, with the study subjects being pregnant women with epilepsy aged 20–45 years, obtained from a multicenter hospital cohort registry system. The data includes demographic information, epilepsy history and classification, pre-pregnancy seizure control, ASM medication use and adjustments during pregnancy, and other clinical variables, collected according to pregnancy stages. An external time validation strategy based on the year of enrollment is adopted (early years as the development set, and subsequent years as the prospective validation set) to enhance the model's robustness to real-world clinical time drift.

[0113] In some embodiments, the present invention initially includes 24 candidate variables related to the outcome, all with missing values ​​below 20%. First, collinearity assessment was performed, and the variance inflation factor (VIF) of all variables was less than 10. Given that the duration of epilepsy is the difference between the onset age of epilepsy and the gestational age, indicating a deterministic linear dependence, the onset age of epilepsy (VIF = 7.238), which has the highest VIF, was removed to reduce redundancy and potential collinearity effects. A second round of collinearity assessment was then performed on the remaining 23 candidate variables, and the VIF of all variables remained less than 10.

[0114] In the feature selection stage of the intervention model, this invention employs two complementary methods. The first is recursive feature elimination (RFE), which shows that retaining the top 22 predictors resulted in the highest average AUC (0.875) across nine machine learning models. The second is Boruta-SHAP, used to identify robust and important features: ultimately, four variables were identified as confirmed, three as tentative, and the rest as rejected. Further forward selection was implemented for the tentative variables: the baseline model built with confirmed variables had an AUC of 0.86; after adding each tentative variable one by one, the AUC increased by 0.015, 0.011, and 0.007, respectively. Taking into account the prediction gain, two tentative variables with an AUC improvement of more than 0.01 were retained and together with four confirmed variables to form the final feature set (a total of 6 variables) for modeling. The variables are: whether there was epileptic seizure in 9 months before pregnancy, whether there was epileptic seizure in 1 month before pregnancy, drug-resistant epilepsy, number of exposures to anti-epileptic drugs (ASMs), duration of epilepsy, and type of epilepsy.

[0115] This invention constructs and evaluates 12 machine learning models, covering regression, ensemble algorithms, and artificial neural networks: Logistic Regression, Linear Discriminant Analysis, K-Nearest Neighbors, Decision Tree, Support Vector Machine (SVC), Gaussian Naïve Bayes, Random Forest, AdaBoost, Extra Trees, Multilayer Perceptron (MLP), XGBoost, and Lightweight Gradient Boosting Machine (LightGBM).

[0116] The model performance evaluation system of this invention uses 10-fold stratified cross-validation to evaluate AUC as the main index, while also outputting accuracy, F1 score, sensitivity, specificity, etc.; the threshold is determined by maximizing the Youden index to achieve a controllable balance between false negatives and false positives.

[0117] The ensemble learning optimization process of this invention selects the five best-performing base models as base learners. Based on these, all possible ensemble combinations (2-5 learners) are constructed and evaluated, resulting in 26 combinations. Five metrics (AUC, accuracy, F1 score, sensitivity, and specificity) are used for ranking, and the average ranking is calculated to obtain a comprehensive performance score. Considering overall prediction accuracy, robustness, and model simplicity, the optimal model is selected.

[0118] The reliability of the probability output was assessed using calibration curves, Brier scores, and the Hosmer-Lemeshow test; clinical net benefit was evaluated using DCA to select risk thresholds and intervention strategies that better suit clinical resource allocation.

[0119] In a comparative analysis of 12 machine learning algorithms, XGBoost demonstrated the best overall classification performance. Its area under the curve (AUC) reached 0.874 (95% CI: 0.821 – 0.926), and it outperformed other models in both accuracy (0.831) and F1 score (0.841). Regarding the balance between model diagnostics and performance, XGBoost maintained high specificity (0.921) while also ensuring good sensitivity (0.766), demonstrating more robust classification capabilities.

[0120] Please see Figure 3The x-axis represents Specificity (false positive rate), and the y-axis represents Sensitivity (true positive rate). A systematic evaluation of the model's discriminative performance, calibration capability, and clinical decision value was conducted using multicenter pregnancies included after 2021. ROC (Receiver Operating Characteristic) curve analysis showed that the model's AUC (Area Under Curve) reached 0.890 (95% CI (Confidence Interval): 0.848–0.927), Sensitivity was 0.828, and Specificity was 0.861. The calibration curve showed that the model's predicted probabilities were highly consistent with the actual observed frequencies, closely following the ideal diagonal, with a Brier score of 0.125. DCA (Decision Curve Analysis) assessed the model's clinical applicability. The results show that, over a very wide threshold probability range (approximately 0.0 to 0.95), the net benefit obtained by using the XGBoost model for intervention decisions is significantly higher than the baseline strategies of "full treatment" or "no treatment," demonstrating potential clinical decision-making value in prospective applications.

[0121] With the help of this invention, the model can be embedded in hospital information systems or clinical decision support systems, and automatically invoked to perform risk calculations during patient follow-up during pregnancy or outpatient visits; during the preconception period or first follow-up of epileptic women, the system automatically invokes the pregnancy stage prediction model, calculates epileptic seizures during pregnancy, and generates risk stratification results; during early pregnancy follow-up, the system updates pregnancy seizure and medication information, invokes the delivery stage prediction model, and outputs the risk assessment results of epileptic seizures in mid-to-late pregnancy.

[0122] The model can also output the probability of epileptic seizures, risk level, and main influencing factors; providing doctors with auxiliary information for adjusting follow-up frequency, medication management, and referral of high-risk pregnancies. The model is used for risk warning and evidence aggregation, and does not replace the doctor's final decision on etiology, drug selection, and obstetric management.

[0123] For example, the model output includes: the probability of epileptic seizure risk and the risk stratification level; key factors (SHAP interpretation): such as the frequency of pre-pregnancy epileptic seizures, the amount of medication taken, etc., and indicating at least one positive / negative contribution. Based on this, physicians can: recommend shorter follow-up intervals, strengthen blood drug concentration / adherence assessment, referral to epilepsy / high-risk obstetric clinics when necessary, and perinatal planning for high-risk patients; for intermediate-risk patients: routine follow-up but highlighting key monitoring factors; for low-risk patients: reducing unnecessary interventions and lowering anxiety and resource consumption.

[0124] This invention supports periodic retraining and calibration after model deployment by introducing new clinical follow-up data to adapt to changes in population characteristics and updates in clinical practice in different hospitals, thereby achieving continuous optimization of model performance.

[0125] The data collection of this invention includes: automatic recording of all cases applied by the model: prediction input, model output, doctor's decision, and final outcome, which are then uploaded to the central database after being anonymized.

[0126] Furthermore, in some embodiments, data integrity checks can be performed, including missing value monitoring and outlier identification. Outcome validation can also be performed: obstetricians and pediatricians independently confirm outcome diagnoses; bias assessment: monitoring differences in data distribution across centers to ensure representativeness.

[0127] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0128] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a computer terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0129] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for applying an intervention model to predict the health of pregnant women with epilepsy, characterized in that, Including the following steps: S201, providing an intervention model library, the intervention model library including: multiple candidate models, the input information of the candidate models including: multiple predictive factors, and the predictive factors including: modifiable factors and non-modifiable factors, the output information of the candidate models including: health risks of pregnant women with epilepsy, and the correlation of the predictive factors with health risks; Among them, the interventionable factors refer to factors that can be directly and artificially controlled by the user, the non-interventionable factors refer to objective physiological indicators that reflect the current physical state of pregnant women with epilepsy, and / or, the non-interventionable factors refer to factors that cannot be effective within the set intervention period. S202, Based on the personal information of the pregnant woman with epilepsy, at least one matching candidate model is selected from the intervention model library as the first intervention model. Wherein, the interventionable factors are associated with difficulty in intervention, and S202 includes the following steps: S2021, Select at least one matching candidate model from the intervention model library based on the personal information of the pregnant woman with epilepsy; S2022, Calculate the execution difficulty of the candidate model based on the intervention difficulty; S2023, Assess the compliance of the pregnant woman with epilepsy based on her historical medical records, and set execution limits based on the level of compliance; S2024, Select the candidate model whose execution difficulty is less than or equal to the execution limit as the first intervention model; S203, Use the first intervention model to perform health prediction on the pregnant woman with epilepsy to obtain corresponding output information; S204, determine whether the first intervention model needs to be updated based on the output information; wherein, S204 includes: Obtain head factors, where head factors refer to predictive factors that are at the top of a set sequence length when the factors are sorted according to their relevance. Calculate the intervention percentage of the modifiable factors among the head factors; When the intervention ratio is less than the set first ratio threshold, a model update suggestion is generated, and S202 is executed again in response to the model update suggestion.

2. The method according to claim 1, characterized in that, The modifiable factor is also associated with an intervention waiting period, which is used to assess the necessary time required for the modifiable factor to take effect from implementation; correspondingly, S202 further includes: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; Calculate the second duration between the intervention waiting period and the first duration; When the second duration is less than the effective period, the interventionable factor is updated to an ininterventionable factor.

3. The method according to claim 1, characterized in that, It also includes the following steps: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; When the first duration is less than or equal to the set production cycle, the execution limit will be reduced.

4. The method according to claim 1, characterized in that, Also includes: The difficulty of the intervention is calculated based on the personal data. The difficulty of the intervention is defined by at least one of the following difficulty indicators, which include: expected cost and comorbidity.

5. The method according to claim 1, characterized in that, Also includes: When the intervention ratio of the first intervention model is greater than the second ratio threshold, a combined recommendation is generated. The combined recommendation requires the selection of at least one second intervention model, and requires that the non-intervention ratio of the non-intervention factors in the second intervention model is greater than the set third ratio threshold.

6. The method according to claim 5, characterized in that, Also includes: After the first intervention model has been running for at least one treatment cycle, the first intervention model is used to predict the health risks of the pregnant woman in the next treatment cycle through historical intervention strategies and updated intervention strategies to obtain a first predicted risk; wherein, the historical intervention strategy refers to the intervention methods applied by the patient in the previous treatment cycle, and the updated intervention strategy refers to the new intervention methods proposed by the doctor for the current treatment cycle. Using at least one of the second intervention models, a second predicted risk is obtained by predicting the health risks of pregnant women in the next treatment cycle through local intervention strategies; wherein, the local intervention strategy refers to one or more of the updated intervention strategies; If the risk difference between the first predicted risk and the second predicted risk is greater than a preset difference, the pregnant woman's compliance level will be reduced.

7. An application system for an intervention model to predict the health of pregnant women with epilepsy, characterized in that, include: The model library provides a module for providing an intervention model library, which includes: multiple candidate models, the input information of which includes: multiple predictive factors, and the predictive factors include: modifiable factors and non-modifiable factors, and the output information of which includes: the health risk of pregnant women with epilepsy, and the correlation between the predictive factors and the health risk; Among them, the interventionable factors refer to factors that can be directly and artificially controlled by the user, the non-interventionable factors refer to objective physiological indicators that reflect the current physical state of pregnant women with epilepsy, and / or, the non-interventionable factors refer to factors that cannot be effective within the set intervention period. The model screening module is used to select at least one matching candidate model from the intervention model library as the first intervention model based on the personal information of the pregnant woman with epilepsy. The association between the modifiable factors is difficult to intervene in, and the model screening module is used to perform the following: A candidate model screening unit is used to screen at least one matching candidate model from the intervention model library based on the personal information of the pregnant woman with epilepsy. An execution difficulty calculation unit is used to calculate the execution difficulty of the candidate model based on the intervention difficulty. An execution limit setting unit is used to assess the compliance of a pregnant woman with epilepsy based on her historical medical records, and to set an execution limit based on the level of compliance. A model selection unit is used to select the candidate model whose execution difficulty is less than or equal to the execution limit as the first intervention model. The health prediction module is used to make a health prediction for the pregnant woman with epilepsy using the first intervention model, so as to obtain the corresponding output information; An update judgment module is used to determine whether the first intervention model needs to be updated based on the output information; wherein, the update judgment module is used to perform: Obtain head factors, where head factors refer to predictive factors that are at the top of a set sequence length when the factors are sorted according to their relevance. Calculate the intervention percentage of the modifiable factors among the head factors; When the intervention ratio is less than the set first ratio threshold, a model update suggestion is generated, and the system re-enters the model screening module in response to the model update suggestion.

8. The system according to claim 7, characterized in that, The modifiable factor is also associated with an intervention waiting period, which is used to assess the necessary time required for the modifiable factor to take effect from implementation; correspondingly, the model screening module is used to perform: Calculate the first duration between the current moment and the expected delivery date for the pregnant woman with epilepsy; Calculate the second duration between the intervention waiting period and the first duration; When the second duration is less than the effective period, the interventionable factor is updated to an ininterventionable factor.

Images