Method and device for constructing prediction model of maternal and infant adverse outcomes, equipment and medium

Abstract

This disclosure relates to a method, apparatus, device, and medium for constructing a predictive model for adverse maternal and infant outcomes in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies. The method includes: acquiring a training set and multiple predictive factors related to adverse maternal and infant outcomes; performing regression analysis on each predictive factor based on the training set using a multivariate logistic regression analysis model to determine the regression coefficients of each predictive factor and construct a predictive model; plotting an ROC curve based on a pre-defined validation set and evaluating the predictive value of the predictive model based on the area under the ROC curve; and plotting a nomogram based on the predictive model if the predictive value reaches a pre-defined value. The nomogram includes: a score axis, a total score axis, and a risk probability axis for each predictive factor. This disclosure enables the construction of a predictive model that comprehensively considers the influence of multiple factors, improving the operability of the predictive model in clinical practice.

Description

Technical Field

[0001] This disclosure relates to the field of medical prediction model technology, and in particular to a method, apparatus, equipment and medium for constructing a prediction model for adverse maternal and infant outcomes. Background Technology

[0002] Currently, clinical risk assessment for adverse pregnancy outcomes or cardiac events in fetuses of pregnant women positive for anti-SSA / Ro and / or SSB / La antibodies relies primarily on physicians' clinical experience or relies on rough assessments based on single antibody tests, lacking systematic and standardized predictive tools. Although existing research has explored the influence of routine clinical factors such as antibody titers and immune status, it is mostly limited to univariate analysis or small sample exploration, failing to achieve a systematic integration of multidimensional risk factors.

[0003] Therefore, constructing a predictive model that can integrate multiple influencing factors and has high predictive accuracy is a key issue that urgently needs to be addressed. Summary of the Invention

[0004] To address the aforementioned technical problems, this disclosure provides a method, apparatus, device, and medium for constructing a predictive model for adverse maternal and infant outcomes.

[0005] According to one aspect of this disclosure, a method for constructing a predictive model for adverse maternal and infant outcomes is provided, the method comprising: Obtain a training set and multiple predictive factors associated with adverse maternal and infant outcomes; Based on the multi-factor logistic regression analysis model, regression analysis is performed on each of the predictive factors according to the training set to determine the regression coefficients of each predictive factor and construct a prediction model. ROC curves are plotted based on a preset validation set, and the predictive value of the prediction model is evaluated based on the area under the ROC curve. When the predicted value reaches a preset value, a nomogram is plotted based on the predicted model; wherein the nomogram includes: the score axis, total score axis, and risk probability axis of each predicted factor.

[0006] According to another aspect of this disclosure, an apparatus for constructing a predictive model for adverse maternal and infant outcomes is also provided, the apparatus comprising: The data acquisition module is used to acquire training sets and multiple predictive factors related to adverse maternal and infant outcomes. The model building module is used to perform regression analysis on each of the predictors based on the training set according to the multi-factor logistic regression analysis model, determine the regression coefficient of each of the predictors, and build a prediction model. The model evaluation module is used to draw ROC curves based on a preset validation set and evaluate the predictive value of the prediction model based on the area under the ROC curve. A visualization module is used to draw a nomogram based on the prediction model when the predicted value reaches a preset value; wherein the nomogram includes: a score axis, a total score axis, and a risk probability axis for each of the predicted factors.

[0007] According to another aspect of this disclosure, an electronic device is also provided, the electronic device comprising: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the above method.

[0008] According to another aspect of this disclosure, a computer-readable storage medium is also provided, the storage medium storing a computer program for performing the above-described method.

[0009] The technical solution provided in this disclosure has the following advantages compared with the prior art: The method for constructing a predictive model for adverse maternal and infant outcomes provided in this disclosure includes: acquiring a training set and multiple predictive factors related to adverse maternal and infant outcomes; performing regression analysis on each predictive factor based on the training set using a multivariate logistic regression analysis model to determine the regression coefficients of each predictive factor and constructing a predictive model; plotting an ROC curve based on a preset validation set and evaluating the predictive value of the predictive model based on the area under the ROC curve; and plotting a nomogram based on the predictive model when the predictive value reaches a preset value; wherein the nomogram includes: the score axis, the total score axis, and the risk probability axis for each predictive factor.

[0010] For scenarios involving adverse maternal and infant outcomes in pregnant women positive for anti-SSA / Ro and / or SSB / La antibodies, this technical solution overcomes the limitations of traditional univariate analysis. By acquiring a training set and integrating multiple predictive factors, it determines the regression coefficients of each predictive factor based on a multivariate logistic regression analysis model, constructing a predictive model that comprehensively considers the influence of multiple factors. This avoids the one-sidedness of single-factor assessment and effectively eliminates confounding interference between predictive factors, achieving a leap from single-dimensional to multi-dimensional systematic integration of predictive factors. Subsequently, ROC curves are plotted using a validation set, and the predictive value of the model is quantitatively evaluated based on the area under the curve, ensuring that the predictive model has validated discriminative ability before clinical application. Only when the predictive value reaches a preset standard is the model applied to the next step, guaranteeing its reliability. Furthermore, this solution transforms the predictive model into a visual nomogram, allowing clinicians to quickly and individually assess the risk of adverse maternal and infant outcomes based on the patient's specific situation without complex calculations, significantly improving the model's operability and ease of use in clinical practice. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0012] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0013] Figure 1 This is a flowchart illustrating the method for constructing a predictive model for adverse maternal and infant outcomes as described in the embodiments of this disclosure. Figure 2 This is a schematic diagram illustrating the process of constructing a predictive model for adverse pregnancy outcomes as described in the embodiments of this disclosure. Figure 3 This is a schematic diagram illustrating the process of constructing a predictive model for the cardiac manifestations of neonatal lupus as described in an embodiment of this disclosure. Figure 4 This is a schematic diagram of ROC curves corresponding to different machine learning models described in the embodiments of this disclosure; Figure 5 This is a schematic diagram of the multivariate logistic regression analysis process described in the embodiments of this disclosure; Figure 6 Calibration curve for the prediction model of adverse pregnancy outcome risk described in the embodiments of this disclosure; Figure 7This is a nomogram showing the risk of adverse pregnancy outcomes as described in the embodiments of this disclosure; Figure 8 The decision curve is the prediction model for the risk of adverse pregnancy outcomes described in the embodiments of this disclosure; Figure 9 This is a nomogram showing the risk of developing neonatal lupus cardiac manifestations as described in the embodiments of this disclosure; Figure 10 ROC curves for predicting cardiac manifestations of neonatal lupus in fetuses using the predictive model described in this embodiment of the present disclosure; Figure 11 This is a model calibration curve for the risk of developing neonatal lupus cardiac manifestations as described in the embodiments of this disclosure; Figure 12 Clinical decision curves for predicting the risk of neonatal lupus cardiac manifestations in fetuses according to embodiments of this disclosure; Figure 13 This is a schematic diagram of the structure of the electronic device described in an embodiment of this disclosure. Detailed Implementation

[0014] To better understand the above-mentioned objectives, features, and advantages of this disclosure, the solutions disclosed herein will be further described below. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0015] Numerous specific details are set forth in the following description in order to provide a full understanding of this disclosure, but this disclosure may also be implemented in other ways different from those described herein; obviously, the embodiments in the specification are only some, and not all, of the embodiments of this disclosure.

[0016] Currently, there is no comprehensive predictive model that can be widely applied in clinical practice to assess the risk of adverse pregnancy outcomes or cardiac events in the fetus of pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies. Existing risk assessment methods have significant limitations, such as: (1) There is a lack of systematic and comprehensive predictive tools that specifically target the special population of anti-SSA / SSB antibody positive and integrate multiple factors (such as immunological indicators, clinical diagnosis, treatment intervention, etc.). The general risk models are not comprehensive enough and cannot accurately reflect the specific risks of this population.

[0017] (2) Existing methods rely heavily on doctors’ subjective experience and lack standardized and quantitative prediction tools, resulting in poor consistency of assessment results among different doctors and limited prediction accuracy.

[0018] (3) For adverse pregnancy outcomes, most existing assessment systems do not include key ultrasound parameters of fetal cardiac function, resulting in limited predictive accuracy. In addition, for fetal neonatal lupus-related cardiac manifestations (Cardiac-NL), the quantitative assessment of the protective effect of specific treatments during pregnancy (such as the use of hydroxychloroquine and aspirin) on the fetal heart is often neglected, which cannot provide accurate data support for the development of individualized monitoring plans and intervention strategies in clinical practice.

[0019] (4) Due to the lack of quantitative and visual assessment methods, the existing methods have poor interpretability in clinical practice and are difficult to help doctors quickly and intuitively develop individualized monitoring and intervention strategies.

[0020] To improve at least one of the above problems, this disclosure provides a method, apparatus, device, and medium for constructing a predictive model for adverse maternal and infant outcomes. This solution can construct a predictive model that integrates multiple influencing factors, has high predictive accuracy, and is easy to operate clinically. This predictive model can be used to predict the risk of adverse maternal and infant outcomes in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies. Specifically, it can include: a predictive model for predicting the risk of adverse pregnancy outcomes at the maternal level, and a predictive model for predicting the risk of neonatal lupus-related cardiac manifestations at the fetal and neonatal level.

[0021] Figure 1 This flowchart illustrates a method for constructing a predictive model for adverse maternal and infant outcomes, as provided in this embodiment. This method can be executed by a device for constructing such a predictive model, which can be implemented using software and / or hardware, specifically, for example, an electronic device or a server. The electronic device can include devices with communication capabilities such as tablets, desktop computers, laptops, and smartphones. The server can be a cloud server or a server cluster, or other devices with storage and computing capabilities.

[0022] like Figure 1 As shown in the figure, the method for constructing a predictive model for adverse maternal and infant outcomes in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies provided in this embodiment may include the following steps S102 to S108.

[0023] S102, Obtain the training set and multiple predictive factors related to adverse maternal and infant outcomes; This embodiment may include: collecting clinical data related to adverse maternal and infant outcomes in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies; preprocessing the clinical data to obtain standardized clinical data; wherein, the preprocessing includes at least: data cleaning, missing value imputation and format conversion; and dividing the standardized clinical data into training set and validation set according to a preset ratio.

[0024] Based on this, obtaining multiple predictors related to adverse maternal and infant outcomes can include: LASSO (Least Absolute Shrinkage and Selection Operator) regression was used to screen multiple predictors associated with adverse maternal and infant outcomes from standardized clinical data in the training set.

[0025] In one application scenario, adverse maternal and infant outcomes include: adverse pregnancy outcomes; correspondingly, multiple predictors associated with adverse pregnancy outcomes include at least: maternal age, number of pregnancies, history of miscarriage or induced labor, assisted reproduction, strong antibody positivity, hydroxychloroquine, global longitudinal strain of the left ventricle (LV_GLS) and global longitudinal strain of the right ventricle (RV_GLS).

[0026] Among them, adverse pregnancy outcomes for pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies include: miscarriage, premature birth, preeclampsia, placental abruption, fetal growth restriction and stillbirth, and other negative events related to the mother or pregnancy process.

[0027] For specific examples of obtaining predictive factors related to adverse pregnancy outcomes in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies, please refer to the following examples.

[0028] Reference Figure 2 By reviewing the hospital's historical medical records, databases, or archives, clinical data of pregnant women diagnosed with positive anti-SSA / Ro and / or SSB / La antibodies within the first predetermined time period were retrospectively collected.

[0029] In the context of adverse pregnancy outcomes at the maternal level, the clinical data for each pregnant woman who is positive for anti-SSA / Ro and / or SSB / La antibodies can include a number of clinical and ultrasound parameters such as maternal age, obstetric history, immune status, medication use, and fetal ultrasound parameters.

[0030] The clinical data undergoes preprocessing including data cleaning, missing value imputation, and format conversion. Data cleaning may include removing abnormal data such as twin pregnancies, ultrasound abnormalities, missing clinical data, or poor ultrasound image quality. After missing value imputation and format conversion, standard clinical data is obtained. The standardized clinical data is then randomly divided into training and validation sets according to a preset ratio (e.g., 7:3).

[0031] In one specific example, the standard clinical data in the training set may include: data from 153 cases of adverse pregnancy outcomes and data from 319 cases of no adverse pregnancy outcomes.

[0032] Because the initially collected clinical data contained multicollinearity and some redundant information, directly incorporating it into a multivariate logistic regression analysis model might lead to overfitting. Therefore, this embodiment uses the LASSO regression algorithm to screen features of the standardized clinical data in the training set to identify multiple predictors most associated with adverse pregnancy outcomes. After LASSO regression screening, the following predictors significantly associated with adverse pregnancy outcomes were finally identified.

[0033] Maternal age: Advanced maternal age is an independent risk factor for pregnancy complications; Number of pregnancies: reflects the cumulative risk of pregnancy exposure; History of miscarriage or induced labor: indicates the impact of previous adverse pregnancy history on the current pregnancy; The application of assisted reproductive technology reflects the difficulty of conception and potential embryo quality or endocrine factors; Strongly positive antibody test: High antibody titers are usually associated with a more active immune response and a higher risk of pathology. Hydroxychloroquine usage: As a protective factor, its use or non-use significantly affects pregnancy outcomes; Global longitudinal strain of the left ventricle (LV_GLS): reflects the contractile function of the left ventricular myocardial fibers. Subclinical cardiac dysfunction may indicate abnormal systemic vascular or placental perfusion. Global longitudinal strain of the right ventricle (RV_GLS): reflects right ventricular function, reflects subclinical cardiac function, and is closely related to pulmonary artery pressure and overall cardiac load.

[0034] Compared to related technologies, this embodiment incorporates eight predictive factors—maternal age, number of pregnancies, history of miscarriage or induced labor, assisted reproduction, strong antibody positivity, hydroxychloroquine, global left ventricular longitudinal strain, and global right ventricular longitudinal strain—into a comprehensive prediction system for adverse pregnancy outcomes in pregnant women with positive anti-SSA / SSB antibodies. This constructs a multi-factor synergistic prediction system, overcoming the limitations of single-dimensional assessment. In particular, this embodiment introduces global left and right ventricular longitudinal strain as predictive factors in the prediction of adverse pregnancy outcomes, which can keenly detect subclinical myocardial damage and significantly enhance the model's physiological basis and early warning capability.

[0035] In another application scenario, adverse maternal and infant outcomes include: fetal neonatal lupus cardiac manifestations (Cardiac-NL); correspondingly, multiple predictors associated with fetal neonatal lupus cardiac manifestations include at least: assisted reproduction, strong antibody positivity, clinical diagnosis of connective tissue disease, hydroxychloroquine, and aspirin.

[0036] Among them, neonatal lupus cardiac manifestations in fetuses of pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies include congenital heart block, cardiomyopathy, and endocardial fibro-elastic tissue hyperplasia.

[0037] For predictive factors related to neonatal lupus cardiac manifestations in fetuses of pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies, specific examples of obtaining these factors can be found below.

[0038] Reference Figure 3 Clinical data of pregnant women diagnosed with positive anti-SSA / Ro and / or SSB / La antibodies were retrospectively collected within a second pre-defined time period.

[0039] In the context of fetal and neonatal lupus cardiac manifestations at the fetal and neonatal level, clinical data for each pregnant woman who is positive for anti-SSA / Ro and / or SSB / La antibodies may include: demographics, diagnosis of autoimmune diseases, reproductive history, treatment interventions and serological markers, clinical diagnosis of connective tissue diseases, use of glucocorticoids, etc.

[0040] Clinical data undergoes preprocessing such as data cleaning, missing value imputation, and format conversion to obtain standardized clinical data. The standardized clinical data is then randomly divided into training and validation sets according to a predetermined ratio.

[0041] In one specific example, the standard clinical data in the training set could include two groups based on whether the offspring of the pregnant woman developed the disease: 26 cases of fetal Cardiac-NL and 643 cases of non-fetal Cardiac-NL; fetal Cardiac-NL was the case group with a positive outcome, and non-fetal Cardiac-NL was the control group with a negative outcome.

[0042] The LASSO regression algorithm was used to screen features of standardized clinical data in the training set to identify multiple predictors most associated with cardiac manifestations of neonatal lupus, including the following.

[0043] Application of assisted reproductive technology: There is a specific association between the risk of cardiac involvement in the fetus of pregnant women who conceive through assisted reproduction, which may be related to advanced maternal age, endocrine environment, or embryo selection process. Strongly positive antibody test: High titers of anti-SSA / Ro or anti-SSB / La antibodies are a direct cause of fetal heart damage, and a strongly positive test significantly increases the risk. Clinical diagnosis of connective tissue diseases: Compared with asymptomatic antibody carriers, pregnant women diagnosed with connective tissue diseases such as systemic lupus erythematosus (SLE) and Sjögren's syndrome (SS) have a higher risk to their fetuses, suggesting the influence of the overall maternal immune load. Use of hydroxychloroquine: As a key protective factor, the standardized use of hydroxychloroquine during pregnancy can significantly reduce the incidence of fetal heart conduction block; Aspirin use: As a synergistic protective factor, aspirin, used alone or in combination with other medications, has shown potential benefits in improving placental microcirculation and reducing fetal cardiac risk.

[0044] Compared to related technologies, this embodiment deeply integrates five predictive factors—assisted reproduction, strong antibody positivity, clinical diagnosis of connective tissue diseases, hydroxychloroquine use, and aspirin use—to construct a comprehensive predictive system for Cardiac-NL based on multiple factors, overcoming the limitations of single-dimensional assessment. In particular, this embodiment quantifies the protective effects of hydroxychloroquine and aspirin, overcoming the limitations of single-indicator assessment by integrating obstetric, immunological, and therapeutic intervention data. This multidimensional integration significantly improves the model's physiological interpretability and predictive accuracy, providing a scientific basis for early and accurate identification of high-risk fetuses and drug optimization.

[0045] S104. Based on the multi-factor logistic regression analysis model, regression analysis is performed on each predictor based on the training set to determine the regression coefficient of each predictor and construct the prediction model.

[0046] The multifactor logistic regression analysis model in this embodiment is a target model selected from various machine learning models after comparing their predictive performance.

[0047] An example of selecting a multivariate logistic regression analysis model from multiple machine learning models may include: evaluating the performance of multiple machine learning models based on predictor factors for predicting adverse maternal and infant outcomes on a pre-defined training set and test set; determining the target model with the best performance evaluation results among the multiple machine learning models; wherein the target model includes: a multivariate logistic regression analysis model.

[0048] It is understandable that the underlying principle for selecting a multivariate logistic regression analysis model from multiple machine learning models is the same for both scenarios involving adverse pregnancy outcomes at the maternal level and scenarios involving neonatal lupus heart disease at the fetal and neonatal level. Therefore, this embodiment will only use the scenario of adverse pregnancy outcomes at the maternal level as an example to introduce the method for selecting a multivariate logistic regression analysis model from multiple machine learning models.

[0049] This embodiment includes: performance evaluation of multiple machine learning models based on predictive factors for predicting adverse pregnancy outcomes on preset training and test sets; determining the target model with the best performance evaluation results among the multiple machine learning models; wherein the target model includes a multivariate logistic regression analysis model. This multivariate logistic regression analysis model is used to predict the risk of adverse pregnancy outcomes in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies.

[0050] Combination Figure 2 This embodiment may specifically include: acquiring various machine learning models, such as: multifactor logistic regression model, random forest model, support vector machine model, decision tree model, XGboost, Naive Bayes model, and multilayer perceptron, etc.

[0051] Each of the aforementioned machine learning models is trained on the training set. For example, for the multivariate logistic regression model, the coefficients are solved using maximum likelihood estimation, and an L2 regularization term is introduced to prevent overfitting caused by multicollinearity. For random forest models, support vector machine models, etc., a grid search combined with 5-fold cross-validation is used to automatically optimize the key hyperparameters of each model, ensuring that each machine learning model participates in the comparison at its best.

[0052] The performance of each trained machine learning model in predicting adverse maternal and infant outcomes based on predictor factors was evaluated using a test set.

[0053] The acquisition method for the test set may include: prospectively collecting clinical data from pregnant women diagnosed with positive anti-SSA / Ro and / or SSB / La antibodies within a third pre-defined time period. Abnormal data, such as data showing intracardiac or extracardiac abnormalities on ultrasound, data with missing clinical data, or data with poor ultrasound image quality, is removed. After preprocessing such as missing value imputation and format conversion, standard clinical data is obtained; this standard clinical data is then added to the test set. As a possible example, the standard clinical data in the test set may include: data from 32 cases of adverse pregnancy outcomes and data from 58 cases without adverse pregnancy outcomes.

[0054] The test set was incorporated into the machine learning model, and the performance of the machine learning model in predicting adverse maternal and infant outcomes based on predictor factors was quantified by plotting ROC curves and calculating the area under the ROC curve (AUC).

[0055] like Figure 4 As shown, the predictive performance of seven machine learning models for adverse pregnancy outcomes is presented in the test set. The horizontal axis represents specificity (false positive rate), and the vertical axis represents sensitivity (true positive rate). The larger the area under the ROC curve, the stronger the predictive performance and discriminative ability of the machine learning model. Figure 4 The results show that the multivariate logistic regression analysis model performs best, exhibiting both high AUC and good interpretability.

[0056] Based on a comprehensive comparison of the performance evaluation results of various machine learning models, and considering the stringent requirements of medical scenarios for model stability, accuracy of predictive performance, and clinical interpretability, this embodiment ultimately determined the multivariate logistic regression analysis model as the target model with the best performance evaluation results.

[0057] This embodiment avoids the bias that may be caused by directly and subjectively selecting a single model through comparison of multiple machine learning models and rigorous validation set evaluation. It ensures that the multifactor logistic regression analysis model finally determined and used as a prediction tool has the accuracy and reliability of prediction, and meets the requirements of transparency and interpretability in clinical medicine.

[0058] The multivariate logistic regression analysis model determined based on the above embodiments is used to handle yes-or-no outcomes, such as adverse pregnancy outcome versus no adverse pregnancy outcome. The multivariate logistic regression analysis model can determine not only the impact of a single predictor on adverse pregnancy outcome, but also the impact of multiple predictors working synergistically on adverse pregnancy outcome.

[0059] Similar to the principles of the aforementioned embodiments, this embodiment, targeting the scenario of fetal and neonatal lupus cardiac manifestations at the fetal and neonatal level, may include: evaluating the performance of multiple machine learning models based on predictive factors for predicting fetal and neonatal lupus cardiac manifestations on preset training and test sets; determining the target model with the best performance evaluation results among the multiple machine learning models; wherein, the target model includes: a multivariate logistic regression analysis model. This multivariate logistic regression analysis model is used to predict the risk of fetal and neonatal lupus cardiac manifestations occurring in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies.

[0060] Based on the above embodiments, referring to Figure 5 This embodiment uses a multi-factor logistic regression analysis model to perform regression analysis on each predictor, determine the regression coefficients of each predictor, and construct a prediction model, which may include the following:

[0061] (1) Based on the univariate logistic regression analysis method, regression is performed on each predictor based on the training set to determine the significant factors of each predictor; (2) Based on the multi-factor logistic regression analysis model, regression analysis is performed on the predictors whose significant factors are less than the preset values ​​according to the training set to obtain the regression coefficients of each predictor; a prediction model is constructed based on the regression coefficients of each predictor.

[0062] Taking the scenario of adverse pregnancy outcomes at the maternal level as an example, a specific implementation method is provided, including: using multiple predictors related to adverse pregnancy outcomes as independent variables, and whether the pregnant woman experiences an adverse pregnancy outcome as the dependent variable. For each predictor, a univariate logistic regression analysis is performed to obtain the significant factors of each predictor, which can be represented by a p-value.

[0063] Because univariate logistic regression analysis does not consider interactions or confounding factors among independent variables, and some independent variables may appear insignificant due to collinearity with other strongly correlated variables, they may become significant after correction in a multivariate logistic regression model. Therefore, this embodiment can identify predictors whose significance is less than a preset value (e.g., P < 0.1).

[0064] If P < 0.1, the predictor is considered to have a potential association with adverse pregnancy outcomes, and the predictor is retained and included in the subsequent multivariate logistic regression analysis model.

[0065] If P ≥ 0.1, the predictor is considered to have a very weak or no association with adverse pregnancy outcomes and is therefore excluded from the multivariate logistic regression analysis model.

[0066] All the selected predictors with P < 0.1 are used as the input variable set and passed to the multivariate logistic regression analysis model. The multivariate logistic regression analysis model is used to fit these selected predictors to the training set and calculate the regression coefficients and model constants for each predictor.

[0067] Subsequently, the regression coefficients and model constants are substituted into the general formula for Logistic regression to construct a predictive model specifically designed to assess adverse pregnancy outcomes.

[0068] This embodiment performs regression analysis from single-factor to multi-factor, which ensures the robustness of the final prediction model.

[0069] For scenarios involving fetal and neonatal lupus cardiac manifestations, the implementation of the predictive model is based on the same principles as the above-described embodiments, and mainly includes: Multiple predictors associated with neonatal lupus cardiac manifestations were used as independent variables, with the occurrence of neonatal lupus cardiac manifestations in the fetus or newborn as the dependent variable. For each predictor, univariate logistic regression analysis was performed to determine the significant factors for each predictor.

[0070] All predictors with P < 0.1 selected were used as the input variable set and passed to a multivariate logistic regression model. The model was then used to fit these predictors to the training set, calculating the regression coefficients and model constants for each predictor. These regression coefficients and model constants were then substituted into the general logistic regression formula to construct a predictive model specifically designed for assessing the cardiac manifestations of lupus in fetuses and newborns.

[0071] S106. Plot the ROC (Receiver Operating Characteristic) curve based on the preset test set, and evaluate the predictive value of the predictive model based on the area under the ROC curve.

[0072] S108. When the predicted value reaches the preset value, draw a nomogram based on the prediction model; wherein, the nomogram includes: the score axis of each predictor, the total score axis, and the risk probability axis.

[0073] In this embodiment, the test set is incorporated into the prediction model, and the discriminative performance and clinical applicability of the prediction model are quantified by plotting the ROC curve and calculating the area under the ROC curve (AUC).

[0074] When the predictive value reaches the preset value, the predictive model is converted into a visual nomogram, thereby translating complex mathematical formulas into charts for a graphical scoring tool. This allows doctors without statistical knowledge to easily use the highly accurate predictive model, facilitating rapid clinical evaluation.

[0075] Based on the above embodiments, the following embodiments will be described in detail for two different scenarios: adverse pregnancy outcomes and neonatal lupus cardiac manifestations.

[0076] For scenarios involving adverse pregnancy outcomes at the maternal level, one can primarily refer to... Figures 6 to 8 .

[0077] like Figure 6 As shown, the calibration curve of the predictive model for the risk of adverse pregnancy outcomes is presented. Figure 6In Figure (A), the calibration curves for the training set are shown, and in Figure (B), the calibration curves for the test set are shown. The vertical axis represents the actual diagnosed risk of adverse pregnancy outcomes; the horizontal axis represents the predicted risk of adverse pregnancy outcomes. The green dashed line (ideal prediction), i.e., the diagonal line, represents the ideal situation where the predicted probability is completely consistent with the observed probability; the blue solid line (actual prediction) represents the relationship between the original predicted probability of the observation model before adjustment and the observation frequency; the orange solid line (predicted after calibration) represents the relationship between the predicted probability adjusted by isotonic regression and the observation frequency. The closer the calibration curve is to the ideal diagonal line, the higher the reliability of the probability output by the observation model.

[0078] Figure 7 This is a nomogram predicting the risk of adverse pregnancy outcomes in pregnant women who are positive for anti-SSA / SSB antibodies, constructed based on a predictive model. The nomogram includes: score axes for eight predictive factors, including maternal age and number of pregnancies, a total score axis, and a risk probability axis.

[0079] In practical applications, we first obtain eight predictive factors for a pregnant woman who is positive for anti-SSA / SSB antibodies before delivery. By drawing a vertical line from the value of each predictive factor to the score line at the top of the nomogram, we find the score corresponding to each predictive factor. We then add up the scores of these eight predictive factors to obtain the total score. We find the total score on the total score axis and draw a vertical line down to the risk probability axis to obtain the probability of adverse pregnancy outcomes.

[0080] Figure 8 Decision curves for predicting the risk of adverse pregnancy outcomes. Figure 8 In the figure, Figure A shows the decision curve for the training set, and Figure B shows the decision curve for the test set. The vertical axis represents net profit, and the horizontal axis represents the risk threshold. The black diagonal line represents the hypothesis that none of the patients had adverse pregnancies, the gray horizontal line represents the hypothesis that all patients had adverse pregnancies, and the red and blue lines represent the performance of the prediction model on the training and test sets, respectively.

[0081] For scenarios involving neonatal lupus-related cardiac manifestations at the fetal and neonatal level, the main reference can be made to... Figures 9 to 12 .

[0082] like Figure 9 The diagram shows a nomogram illustrating the risk of neonatal lupus cardiac manifestations in fetuses of pregnant women who are positive for anti-SSA / SSB antibodies, constructed based on a predictive model. The nomogram includes score axes for five predictive factors: assisted reproduction, strong positive anti-SSA / SSB antibody test, clinical diagnosis of connective tissue disease, hydroxychloroquine, and aspirin; a total score axis; and a risk probability axis. Furthermore, by finding the corresponding point on the total score axis and mapping it vertically downwards to the risk probability, the probability of developing Cardiac-NL can be obtained.

[0083] Figure 10ROC curve for predicting cardiac manifestations of neonatal lupus (NL) in fetuses using a predictive model; this figure shows the model's ability to distinguish between Cardiac-NL: Horizontal axis: 1 - specificity (false positive rate), Vertical axis: sensitivity (true positive rate), Area under the curve (AUC): This model achieves an AUC of 0.872. Significance: The closer the ROC curve is to the upper left corner, the closer the AUC is to 1, indicating a stronger discriminative ability of the model. An AUC of 0.872 indicates that the model has excellent predictive performance and can effectively distinguish between high-risk and low-risk pregnant women.

[0084] Figure 11 Calibration curves for a model of the risk of developing cardiac manifestations of lupus in fetuses and newborns; Figure 11 This graph is used to assess the consistency between model-predicted probabilities and actual observations. The horizontal axis represents the predicted risk probability of fetal neonatal lupus cardiac manifestations; the vertical axis represents the actual observed incidence of fetal neonatal lupus cardiac manifestations. The diagonal (dashed line) represents the ideal prediction line (predicted probability = actual probability); the red solid line (actual prediction) represents the relationship between the unadjusted original predicted probability of the observation model and the observation frequency; the green solid line (calibrated prediction) represents the relationship between the predicted probability adjusted by isotonic regression and the observation frequency. The closer the model calibration curve is to the diagonal, the better the calibration of the prediction model and the more accurate the predicted probability. This graph verifies the reliability of the prediction model output, ensuring that the prediction results have clinical reference value.

[0085] Figure 12 Clinical decision curves for predicting the risk of neonatal lupus cardiac manifestations in fetuses. Figure 12 This is used to evaluate the clinical net benefit of the predictive model at different risk thresholds. The horizontal axis represents the preset risk threshold; the vertical axis represents the net benefit (i.e., the trade-off between the gains and losses from correctly identifying high-risk patients and misdiagnosis). Black diagonal lines indicate the assumption that all patients are free of fetal and neonatal lupus cardiac manifestations; gray horizontal lines indicate the assumption that all patients have fetal and neonatal lupus cardiac manifestations; the colored curves represent the actual performance of this predictive model.

[0086] If the clinical decision curve of the predictive model is above the two reference lines within a wide threshold range, it indicates that using the model to guide clinical decision-making can bring greater net benefits. This figure validates the clinical practical value of the model, proving that it can effectively assist physicians in making intervention decisions.

[0087] In addition, this embodiment may include the following schemes.

[0088] Establish an online learning mechanism to continuously update the parameters of the prediction model as new case data is collected.

[0089] The predictive model can be developed into a mobile app, desktop software, web tool, or integrated into the hospital information system to support automatic data entry and automatic risk calculation.

[0090] External validation was conducted in collaboration with multiple hospitals to improve the universality of the prediction model.

[0091] The predictive model can be extended to other application scenarios, such as predicting immune-related adverse pregnancy outcomes like antiphospholipid antibody syndrome, and predicting fetal heart damage related to autoantibodies.

[0092] Monitoring and treatment recommendations are automatically generated based on risk predictions of adverse maternal and infant outcomes.

[0093] Incorporating more predictive factors, such as fetal echocardiogram parameters and maternal genotype, can improve the accuracy of the predictive model.

[0094] In summary, the method for constructing a predictive model for adverse maternal and infant outcomes provided in this disclosure includes: acquiring a training set and multiple predictive factors related to adverse maternal and infant outcomes; performing regression analysis on each predictive factor based on the training set using a multivariate logistic regression analysis model to determine the regression coefficients of each predictive factor and constructing a predictive model; plotting an ROC curve based on a preset validation set and evaluating the predictive value of the predictive model based on the area under the ROC curve; and plotting a nomogram based on the predictive model when the predictive value reaches a preset value; wherein the nomogram includes: the score axis, the total score axis, and the risk probability axis for each predictive factor.

[0095] This technical solution overcomes the limitations of traditional univariate analysis. By acquiring a training set and integrating multiple predictive factors, it determines the regression coefficients of each predictive factor based on a multivariate logistic regression analysis model, constructing a predictive model that comprehensively considers the influence of multiple factors. This avoids the one-sidedness of single-factor assessment, effectively eliminates confounding interference between predictive factors, and achieves a leap from single-dimensional to multi-dimensional systematic integration of predictive factors. Subsequently, an ROC curve is plotted using a validation set, and the predictive value of the model is quantitatively evaluated based on the area under the curve, ensuring that the predictive model has validated discriminative ability before clinical application. Only when the predictive value reaches a preset standard is it applied to the next step, guaranteeing the reliability of the predictive model. Furthermore, by transforming the predictive model into a visual nomogram, this solution allows clinicians to quickly and individually assess the risk of adverse maternal and infant outcomes based on the patient's specific situation without performing complex calculations, significantly improving the model's operability and ease of use in clinical practice.

[0096] In this embodiment, an apparatus for constructing a predictive model for adverse maternal and infant outcomes is provided. This apparatus is used to implement the method for constructing a predictive model for adverse maternal and infant outcomes provided in the above embodiment. The apparatus for constructing a predictive model for adverse maternal and infant outcomes includes the following modules: The data acquisition module is used to acquire training sets and multiple predictive factors related to adverse maternal and infant outcomes. The model building module is used to perform regression analysis on each of the predictors based on the training set according to the multi-factor logistic regression analysis model, determine the regression coefficient of each of the predictors, and build a prediction model. The model evaluation module is used to draw ROC curves based on a preset validation set and evaluate the predictive value of the prediction model based on the area under the ROC curve. A visualization module is used to draw a nomogram based on the prediction model when the predicted value reaches a preset value; wherein the nomogram includes: a score axis, a total score axis, and a risk probability axis for each of the predicted factors.

[0097] The device provided in this embodiment has the same implementation principle and technical effect as the aforementioned method embodiment. For the sake of brevity, any parts not mentioned in the device embodiment can be referred to the corresponding content in the aforementioned method embodiment.

[0098] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Figure 13 As shown, the electronic device 200 includes one or more processors 201 and memory 202.

[0099] The processor 201 may be a central processing unit (CPU) or other form of processing unit with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 200 to perform desired functions.

[0100] The memory 202 may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor 201 may execute the program instructions to implement the method for constructing a predictive model for adverse maternal and infant outcomes according to the embodiments of this disclosure described above, and / or other desired functions. Various contents such as input signals, signal components, and noise components may also be stored in the computer-readable storage medium.

[0101] In one example, the electronic device 200 may also include an input device 203 and an output device 204, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0102] In addition, the input device 203 may also include, for example, a keyboard, a mouse, etc.

[0103] The output device 204 can output various information to the outside, including determined distance information, direction information, etc. The output device 204 may include, for example, a display, a speaker, a printer, and a communication network and its connected remote output devices, etc.

[0104] Of course, for the sake of simplicity, Figure 13 Only some of the components of the electronic device 200 relevant to this disclosure are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device 200 may include any other suitable components depending on the specific application.

[0105] Furthermore, this embodiment also provides a computer-readable storage medium storing a computer program for executing the above-described method for constructing a predictive model for adverse maternal and infant outcomes.

[0106] The present disclosure provides a computer program product for constructing a predictive model for adverse maternal and infant outcomes, including a method, apparatus, electronic device, and medium. The program product includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0107] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, 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. Without further limitations, 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 said element.

[0108] The above description is merely a specific embodiment of this disclosure, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this disclosure. Therefore, this disclosure is not to be limited to the embodiments described herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for constructing a predictive model for adverse maternal and infant outcomes, characterized in that, The method includes: Obtain a training set and multiple predictive factors associated with adverse maternal and infant outcomes; Based on the multi-factor logistic regression analysis model, regression analysis is performed on each of the predictive factors according to the training set to determine the regression coefficients of each predictive factor and construct a prediction model. ROC curves are plotted based on a preset validation set, and the predictive value of the prediction model is evaluated based on the area under the ROC curve. When the predicted value reaches a preset value, a nomogram is plotted based on the predicted model; wherein the nomogram includes: the score axis, total score axis, and risk probability axis of each predicted factor.

2. The method according to claim 1, characterized in that, The method further includes: Collect clinical data related to adverse maternal and infant outcomes in pregnant women who are positive for anti-SSA / Ro and / or SSB / La antibodies; The clinical data is preprocessed to obtain standardized clinical data; wherein the preprocessing includes at least: data cleaning, missing value imputation, and format conversion; The standardized clinical data is divided into a training set and a validation set according to a preset ratio.

3. The method according to claim 2, characterized in that, Obtain multiple predictors associated with adverse maternal and infant outcomes, including: LASSO regression was used to screen multiple predictors associated with adverse maternal and infant outcomes from standardized clinical data in the training set.

4. The method according to claim 1, characterized in that, The adverse maternal and infant outcomes include: adverse pregnancy outcomes; and the multiple predictors associated with the adverse pregnancy outcomes include at least: maternal age, number of pregnancies, history of miscarriage or induced labor, assisted reproduction, strong antibody positivity, hydroxychloroquine, global longitudinal strain of the left ventricle and global longitudinal strain of the right ventricle.

5. The method according to claim 1, characterized in that, The adverse maternal and infant outcomes include: neonatal lupus cardiac manifestations in the fetus; multiple predictors associated with the neonatal lupus cardiac manifestations in the fetus include at least: assisted reproduction, strong antibody positivity, clinical diagnosis of connective tissue disease, hydroxychloroquine, and aspirin.

6. The method according to claim 1, characterized in that, The multi-factor logistic regression analysis model performs regression analysis on each of the predictor factors based on the training set, determines the regression coefficients of each predictor factor, and constructs a prediction model, including: Based on the univariate logistic regression analysis method, the predictor factors are regressed according to the training set to determine the significant factors of each predictor factor; Based on the multivariate logistic regression analysis model, regression analysis is performed on the predictors whose significant factors are less than the preset values ​​according to the training set to obtain the regression coefficients of each predictor. A prediction model is constructed based on the regression coefficients of each predictor.

7. The method according to claim 1, characterized in that, The method further includes: On the preset training and test sets, the performance of various machine learning models based on the predictive factors for predicting adverse maternal and infant outcomes was evaluated. Among various machine learning models, the target model with the best performance evaluation results is determined; wherein, the target model includes: a multivariate logistic regression analysis model.

8. An apparatus for constructing a predictive model for adverse maternal and infant outcomes, characterized in that, The device includes: The data acquisition module is used to acquire training sets and multiple predictive factors related to adverse maternal and infant outcomes. The model building module is used to perform regression analysis on each of the predictors based on the training set according to the multi-factor logistic regression analysis model, determine the regression coefficient of each of the predictors, and build a prediction model. The model evaluation module is used to draw ROC curves based on a preset validation set and evaluate the predictive value of the prediction model based on the area under the ROC curve. A visualization module is used to draw a nomogram based on the prediction model when the predicted value reaches a preset value; wherein the nomogram includes: a score axis, a total score axis, and a risk probability axis for each of the predicted factors.

9. An electronic device, characterized in that, The electronic device includes: processor; Memory used to store the processor's executable instructions; The processor is configured to read the executable instructions from the memory and execute the instructions to implement the method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a terminal device, cause the terminal device to perform the method as described in any one of claims 1-7.

Images