Prediction method for onset risk of preeclampsia

Abstract

The invention relates to the technical field of early pregnancy disease prediction, in particular to a preeclampsia onset risk prediction method, which comprises the following steps: S1, determining contribution degrees of different clinical information to preeclampsia; s2, collecting clinical information of multiple groups of pregnant women; s3, based on the acquired clinical information and the contribution degree of different clinical information to preeclampsia, the preeclampsia incidence probability corresponding to each group of data is determined, and each group of clinical information and the corresponding preeclampsia incidence probability form a data set; s4, constructing and training an XGBoost prediction model by taking the clinical information in the data set as an input feature and taking the preeclampsia incidence probability as a target variable; and S5, inputting clinical information of a new case into the XGBoost prediction model, and outputting the preeclampsia morbidity probability and morbidity risk level by the XGBoost prediction model. The prediction result is more detailed and accurate, the prediction result can be rapidly and dynamically output according to the clinical information of the new case, and compared with a traditional method, the prediction efficiency is higher.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of early pregnancy disease prediction, and more particularly to a method for predicting the risk of preeclampsia. BACKGROUND

[0002] Preeclampsia is a pregnancy-specific hypertension disease, which belongs to a kind of hypertensive disorder in pregnancy, usually occurs after 20 weeks of gestation, and is characterized by new-onset hypertension (≥140 / 90mmHg) and proteinuria (≥0.3g / 24h), and may be accompanied by multiple organ dysfunction. If not timely intervention, it may progress to eclampsia (seizures or coma), which seriously threatens the safety of mother and infant. The "Guidelines for Hypertensive Disorders in Pregnancy (2020)" emphasizes that high-risk factors should be screened, evaluated and prevented in early pregnancy and any prenatal examination.

[0003] In the traditional prediction method, one of them is to make risk judgments based on the medical history and basic clinical indicators (such as age, BMI, blood pressure, medical history, etc.) of pregnant women. This method relies on subjective experience and judgment, and the judgment criteria of each doctor is not completely consistent, resulting in insufficient prediction accuracy. Another is the biomarker detection method, which requires the use of equipment to complete and has high difficulty and cost, making it difficult to promote in resource-limited areas. In addition, the efficiency of the above two methods is also low, and it is difficult to meet the requirements of high-efficiency and accurate prediction in clinical practice. SUMMARY

[0004] In view of the problems of low prediction accuracy and low efficiency in the prior art, the present application provides a method for predicting the risk of preeclampsia, which can improve the prediction accuracy and efficiency and meet the requirements of high-efficiency and accurate prediction in clinical practice.

[0005] To solve the above technical problems, the technical solution provided by the present application is: A method for predicting the risk of preeclampsia, comprising the following steps: S1: determining the contribution of different clinical information to preeclampsia.

[0006] S2: Collecting the clinical information of a plurality of pregnant women.

[0007] S3: Based on the collected clinical information and the contribution of different clinical information to preeclampsia, determining the preeclampsia incidence probability corresponding to each group of data, and each group of clinical information and the corresponding preeclampsia incidence probability form a data set.

[0008] S4: Taking the clinical information in the data set as the input feature and the preeclampsia incidence probability as the target variable, constructing and training an XGBoost prediction model.

[0009] S5: inputting the clinical information of the new case into the XGBoost prediction model, and outputting the preeclampsia incidence probability and incidence risk level from the XGBoost prediction model; wherein the risk level of the incidence risk is determined according to a set threshold rule.

[0010] The technical solution quantifies the prediction contribution of different clinical information, and can output the corresponding preeclampsia incidence probability and incidence risk based on the clinical information in the early stage of pregnancy through machine automatic learning of the influence of different clinical information. Compared with the prediction results of the traditional method, the prediction results are more detailed and accurate, which can provide strong decision support for clinicians, help to identify high-risk pregnant women early, and reduce the probability of occurrence of pregnancy complications. Moreover, the prediction results can be quickly and dynamically output according to the clinical information of the new case, and the prediction efficiency is higher than that of the traditional method.

[0011] Preferably, in the step S1, the clinical information includes age, mean arterial pressure, BMI, parity, smoking history, mode of conception, history of chronic hypertension, history of diabetes, history of autoimmune disease, history of preeclampsia, and family history of preeclampsia.

[0012] Preferably, in the step S3, the collected clinical information is used as a classification variable, and the classification variable is one-hot encoded to obtain a binary feature; the input features of the XGBoost prediction model are all encoded classification variables, wherein the encoding method of different classification variables is: encoding 0 when the age is less than a set value, and encoding 1 when the age is greater than or equal to the set value; encoding 0 when the mean arterial pressure is less than a set value, and encoding 1 when the mean arterial pressure is greater than or equal to the set value; encoding 0 when the BMI is less than a set value, and encoding 1 when the BMI is greater than or equal to the set value; encoding 0 when the parity is primipara, and encoding 1 when the parity is multipara; encoding 1 when there is a smoking history, and encoding 0 when there is no smoking history; encoding 1 when the mode of conception is artificial conception, and encoding 0 otherwise; encoding 1 when there is a history of chronic hypertension, diabetes, autoimmune disease, preeclampsia, and family history of preeclampsia, and encoding 0 when there is no history.

[0013] Preferably, in the step S3, before one-hot encoding the classification variable, the collected data is cleaned to process the abnormal values and missing values in the data.

[0014] Preferably, in the step S3, 90.04 mmHg is used as a critical value, and the mean arterial pressure less than 90.04 mmHg is encoded as 0, and the mean arterial pressure greater than or equal to 90.04 mmHg is encoded as 1.

[0015] Preferably, in the step S3, all abnormal values in the data are marked as missing values, and then the k-nearest neighbor imputation method is used to perform multiple imputation processing on all missing values in the data.

[0016] Preferably, in the step S4, the specific steps of model training are as follows: S41: 5-fold stratified cross-validation is adopted, that is, stratified sampling is first performed on the data set; then the stratified data is randomly divided into 5 mutually exclusive subsets, in each iteration, one subset is selected as the validation set, and the remaining four subsets are combined as the training set, which is repeated 5 times to ensure that all samples participate in the validation.

[0017] S42: The optimal hyperparameter combination is searched in the hyperparameter space by grid search, including learning rate, maximum depth of tree and minimum sample weight of child node.

[0018] S43: The age, mean arterial pressure, BMI, parity, smoking history, mode of conception, history of chronic hypertension, history of diabetes, history of autoimmune disease, history of preeclampsia, family history of preeclampsia in the training set are used as input features; the incidence probability of preeclampsia in the training set is used as the target variable, and the optimal hyperparameter combination is configured to fit the XGBoost prediction model.

[0019] S44: The AUC value, sensitivity and specificity index at a specific false positive rate corresponding to each validation set are calculated, if these indexes all reach the set threshold, the model is determined as the final XGBoost prediction model; otherwise, return to step S41 to adjust the data or return to step S42 to re-adjust the model parameters.

[0020] Preferably, in the step S44, after the final XGBoost prediction model is determined, the SHAP values of each input feature are calculated using the entire data of cross-validation, and the SHAP contribution of each input feature is output from the XGBoost prediction model; in the step S5, the XGBoost prediction model also inputs the SHARP values of each input feature, thereby enhancing the interpretability of the model and facilitating doctors to further understand and explain the prediction results.

[0021] Preferably, in the step S2, the mean arterial pressure data is measured at the 11-13th week of pregnancy.

[0022] Preferably, in the step S5, the set threshold rule is that when the incidence probability of preeclampsia is greater than 0.54, the corresponding incidence risk level is high risk, and when the incidence probability of preeclampsia is less than or equal to 0.54, the corresponding incidence risk level is low risk.

[0023] The present application quantifies the prediction contribution of different clinical information, learns the influence of different clinical information automatically through machine, and outputs the corresponding preeclampsia incidence probability and incidence risk based on the clinical information in the early pregnancy period. Compared with the prediction results of the traditional method, the prediction results are more detailed and accurate, which can provide strong decision support for clinicians, help to identify high-risk pregnant women early, and reduce the probability of pregnancy complications. Moreover, the present application can quickly and dynamically output prediction results according to the clinical information of new cases, and has higher prediction efficiency than the traditional method. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1 is a flowchart of a preeclampsia incidence risk prediction method; Figure 2 is a nonlinear relationship diagram between mean arterial pressure and preeclampsia incidence rate. Figure 3 is a receiver operating characteristic (ROC) curve diagram. DETAILED DESCRIPTION

[0025] The drawings are only used for illustrative description, and should not be understood as limiting the present patent; in order to better illustrate the present embodiment, some components in the drawings may be omitted, enlarged or reduced, and do not represent the actual product size; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted. The positional relationship described in the drawings is only used for illustrative description, and should not be understood as limiting the present patent.

[0026] The same or similar reference numerals in the drawings of the embodiments of the present application correspond to the same or similar components; in the description of the present application, it should be understood that if the terms "upper", "lower", "left", "right", "long", "short" and the like indicate the orientation or positional relationship shown in the drawings, only for the convenience of describing the present application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, therefore the terms describing the positional relationship in the drawings are only used for illustrative description, and should not be understood as limiting the present patent, for those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0027] The technical solutions of the present application will be further described in detail below through specific embodiments and in conjunction with the drawings: Embodiment 1 The present embodiment is a first embodiment of a preeclampsia incidence risk prediction method, as shown in Figure 1 , which comprises the following steps: S1: determining the contribution of different clinical information to preeclampsia.

[0028] S2: collecting clinical information of multiple pregnant women.

[0029] S3: determining the probability of preeclampsia corresponding to each group of data based on the collected clinical information and the contribution of different clinical information to preeclampsia, and each group of clinical information and the corresponding probability of preeclampsia form a data set.

[0030] S4: taking the clinical information in the data set as the input feature and the probability of preeclampsia as the target variable, constructing and training an XGBoost prediction model.

[0031] S5: inputting the clinical information of a new case into the XGBoost prediction model, and the XGBoost prediction model outputs the probability of preeclampsia and the risk level of preeclampsia; wherein the risk level of preeclampsia is determined according to the set threshold rule.

[0032] Further, in step S1, the clinical information includes age, mean arterial pressure, BMI, parity, smoking history, mode of conception, history of chronic hypertension, history of diabetes, history of autoimmune disease, history of preeclampsia, and family history of preeclampsia.

[0033] Further, in step S2, the mean arterial pressure data is measured at the 11th-13th week of pregnancy.

[0034] Further, in step S3, the collected clinical information is taken as a classification variable, and the classification variable is one-hot encoded to obtain a binary feature; the input features of the XGBoost prediction model are all the encoding of the classification variable, wherein the encoding method of different classification variables is: encoding as 0 when the age is less than a set value, and encoding as 1 when the age is greater than or equal to the set value; encoding as 0 when the mean arterial pressure is less than a set value, and encoding as 1 when the mean arterial pressure is greater than or equal to the set value; encoding as 0 when the BMI is less than a set value, and encoding as 1 when the BMI is greater than or equal to the set value; encoding as 0 when the parity is primipara, and encoding as 1 when the parity is multipara; encoding as 1 when there is a smoking history, and encoding as 0 when there is no smoking history; encoding as 1 when the mode of conception is artificial conception, and otherwise encoding as 0; encoding as 1 when there is a history of chronic hypertension, diabetes, autoimmune disease, preeclampsia, and family history of preeclampsia, and encoding as 0 when there is no history of chronic hypertension, diabetes, autoimmune disease, preeclampsia, and family history of preeclampsia.

[0035] The beneficial effects of the present embodiment are: the prediction contribution of different clinical information is quantified, the influence of different clinical information is learned by machine automatically, the corresponding probability of preeclampsia and the risk of preeclampsia can be output based on the clinical information in the early stage of pregnancy, the prediction result of the traditional method is more detailed and accurate, which can provide strong decision support for clinicians, help to identify high-risk pregnant women early, and reduce the probability of occurrence of pregnancy complications. Moreover, the prediction result can be quickly and dynamically output according to the clinical information of a new case, and the prediction efficiency is higher than that of the traditional method.

[0036] Embodiment 2 This embodiment is a second embodiment of a method for predicting the risk of preeclampsia, which further supplements step S3 on the basis of embodiment 1.

[0037] In step S3, 90.04 mmHg is taken as the critical value, and the average arterial pressure less than 90.04 mmHg is coded as 0, and the average arterial pressure greater than or equal to 90.04 mmHg is coded as 1. This embodiment initially screened 15888 cases of singleton pregnancy, excluded 8389 cases, and finally included 7499 participants. Among them, 301 cases were diagnosed as preeclampsia, with an overall incidence of 4.0%; 61 cases of preterm preeclampsia (0.8%), and 240 cases of term preeclampsia (3.2%). The remaining 7198 women (96.0%) did not develop preeclampsia. The association between mean arterial pressure and preeclampsia risk in pregnant women at 11-13 weeks of gestation was analyzed by logistic regression, as shown in Table 1.

[0038] Table 1

[0039] As can be seen from Table 1, after controlling for confounding factors, the corrected odds ratio (OR) of mean arterial pressure is 1.09 (95% confidence interval: 1.08-1.11). The mean arterial pressure is divided into four subgroups at 70, 90 and 110 mmHg, and the same covariate-adjusted stratified logistic regression analysis is performed, and the results are shown in Table 2.

[0040] Table 2

[0041] The results show that the OR value of the mean arterial pressure > 90 mmHg group is significantly increased, reaching 5.62 (95% CI: 2.51-16.03) and 4.78 (95% CI: 1.02-21.18), respectively. At the same time, the non-linear relationship between mean arterial pressure and preeclampsia incidence was analyzed by a restricted cubic spline, and the results are shown in Figure 2 Therefore, the optimal critical value of mean arterial pressure was further evaluated by the receiver operating characteristic curve. As shown in Figure 3 The area under the curve of this model is 0.742 (95% CI: 0.711-0.772), and the optimal critical value is 90.04 mmHg. Therefore, mean arterial pressure ≥ 90.04 mmHg has significant predictive value for preeclampsia in early pregnancy, especially in identifying preterm preeclampsia.

[0042] Further, in step S3, before the categorical variables are one-hot encoded, the collected data is cleaned to handle outliers and missing values in the data.

[0043] Further, in step S3, the abnormal values in the data are all marked as missing values, and then the k-neighbor imputation method is used to perform multiple imputation processing on all missing values in the data.

[0044] The other features, working principles and beneficial effects of this embodiment are consistent with those of embodiment 1.

[0045] Embodiment 3 This embodiment is a third embodiment of a method for predicting the risk of preeclampsia, which further supplements steps S4 and S5 on the basis of embodiment 1.

[0046] In step S4, the specific steps of model training are as follows: S41: 5-fold stratified cross-validation is used, that is, stratified sampling is first performed on the data set; then the stratified data is randomly divided into 5 mutually exclusive subsets, in each iteration, one subset is selected as the validation set, and the remaining four subsets are combined as the training set, which is repeated 5 times to ensure that all samples participate in the validation.

[0047] S42: The optimal hyperparameter combination is found in the hyperparameter space through grid search, and the hyperparameters include learning rate, maximum depth of tree and minimum sample weight of child node.

[0048] S43: The age, mean arterial pressure, BMI, parity, smoking history, mode of conception, history of chronic hypertension, history of diabetes, history of autoimmune disease, history of preeclampsia, and family history of preeclampsia in the training set are used as input features; the preeclampsia incidence probability in the training set is used as the target variable, and the optimal hyperparameter combination is configured to fit the XGBoost prediction model.

[0049] S44: The AUC value, sensitivity and specificity index at a specific false positive rate corresponding to each validation set are calculated, and if these indexes all reach the set threshold, the model is determined as the final XGBoost prediction model; otherwise, return to step S41 to adjust the data or return to step S42 to re-adjust the model parameters.

[0050] Further, in step S44, after the final XGBoost prediction model is determined, the SHAP values of each input feature are calculated using the entire data of cross-validation, and the SHAP contribution of each input feature is output from the XGBoost prediction model; in step S5, the XGBoost prediction model also inputs the SHARP values of each input feature. Thus, the interpretability of the model is enhanced, which facilitates doctors to further understand and explain the prediction results. Further, in step S5, the set threshold rule is that when the preeclampsia incidence probability is greater than 0.54, the corresponding incidence risk level is high risk, and when the preeclampsia incidence probability is less than or equal to 0.54, the corresponding incidence risk level is low risk.

[0051] Other features, principles, and advantages of this embodiment are consistent with those of Embodiment 2.

[0052] In the specific contents of the above specific embodiments, each technical feature can be combined arbitrarily without contradiction. In order to make the description simple, all possible combinations of the above technical features are not described, but as long as the combination of the technical features does not exist contradiction, it should be considered as the scope of the present disclosure.

[0053] Obviously, the above embodiments of the present application are only examples for clearly illustrating the present application, and are not intended to limit the implementation modes of the present application. Based on the above description, other different forms of changes or variations can be made by those skilled in the art, and here it is not necessary and also impossible to exhaust all the implementation modes. Any modification, equivalent replacement, and improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the claims of the present application.

Claims

1. A method of predicting the risk of pre-eclampsia onset, characterized by, The method comprises the following steps: S1: determining the contribution of different clinical information to preeclampsia; S2: collecting clinical information of multiple groups of pregnant women; S3: determining the preeclampsia incidence probability corresponding to each group of data based on the collected clinical information and the contribution of different clinical information to preeclampsia, and each group of clinical information and the corresponding preeclampsia incidence probability forming a data set; S4: taking the clinical information in the data set as an input feature and the preeclampsia incidence probability as a target variable to construct and train an XGBoost prediction model; S5: inputting the clinical information of a new case into the XGBoost prediction model, and the XGBoost prediction model outputs the preeclampsia incidence probability and the incidence risk level; wherein the risk level of the incidence risk is determined according to a set threshold rule.

2. The method of predicting the risk of preeclampsia according to claim 1, wherein In the step S1, the clinical information includes age, mean arterial pressure, BMI, parity, smoking history, mode of conception, history of chronic hypertension, history of diabetes, history of autoimmune disease, history of preeclampsia, and family history of preeclampsia.

3. The method of predicting the risk of preeclampsia according to claim 2, wherein In the step S3, the collected clinical information is one-hot encoded as a classification variable to obtain a binary feature; and the input features of the XGBoost prediction model are all encoded classification variables, wherein the encoding method of different classification variables is: encoding 0 when the age is less than a set value, and encoding 1 when the age is greater than or equal to the set value; encoding 0 when the mean arterial pressure is less than a set value, and encoding 1 when the mean arterial pressure is greater than or equal to the set value; encoding 0 when the BMI is less than a set value, and encoding 1 when the BMI is greater than or equal to the set value; encoding 0 when the parity is primipara, and encoding 1 when the parity is multipara; encoding 1 when there is a smoking history, and encoding 0 when there is no smoking history; encoding 1 when the mode of conception is artificial conception, and encoding 0 otherwise; and encoding 1 when there is a history of chronic hypertension, diabetes, autoimmune disease, preeclampsia, and family history of preeclampsia, and encoding 0 when there is no history. In the step S3, 90.04 mmHg is used as a critical value, and the encoding corresponding to the mean arterial pressure less than 90.04 mmHg is 0, and the encoding corresponding to the mean arterial pressure greater than or equal to 90.04 mmHg is 1.

4. The method of predicting the risk of preeclampsia according to claim 3, wherein In the step S3, before one-hot encoding the classification variables, the collected data is first cleaned to handle abnormal values and missing values in the data.

5. The method of predicting the risk of preeclampsia according to claim 3, wherein In the step S3, all abnormal values in the data are marked as missing values, and then a k-nearest neighbor imputation method is used to perform multiple imputation processing on all missing values in the data.

6. The method of predicting the risk of preeclampsia according to claim 5, wherein In the step S4, the specific steps of model training are as follows:

7. The method of predicting the risk of preeclampsia according to claim 1, wherein S41: 5-fold stratified cross-validation is used, that is, stratified sampling is first performed on the data set; then the stratified data is randomly divided into 5 mutually exclusive subsets, and in each iteration, one subset is selected as the validation set and the remaining four subsets are combined as the training set, and the process is repeated 5 times; S42: the optimal hyperparameter combination is found in the hyperparameter space through grid search, and the hyperparameters include learning rate, maximum depth of tree, and minimum sample weight of child node; ​ S43: Use age, mean arterial pressure, BMI, parity, smoking history, method of conception, history of chronic hypertension, history of diabetes, history of autoimmune diseases, history of preeclampsia, and family history of preeclampsia as input features from the training set. Using the incidence probability of preeclampsia in the training set as the target variable, the optimal hyperparameter combination is configured to fit the XGBoost prediction model. S44: Calculate the AUC value, sensitivity and specificity indicators at a specific false positive rate for each validation set. If these indicators all reach the set threshold, the model is determined as the final XGBoost prediction model; otherwise, return to step S41 to adjust the data or return to step S42 to readjust the model parameters.

8. The method for predicting the risk of preeclampsia according to claim 7, characterized in that, In step S44, after determining the final XGBoost prediction model, the SHAP value of each input feature is calculated using all the cross-validated data, and then the SHAP contribution of each input feature is output from the XGBoost prediction model; in step S5, the XGBoost prediction model also inputs the SHARP value of each input feature.

9. The method of predicting the risk of preeclampsia according to claim 1, wherein In step S2, the mean arterial pressure data are measured during the 11th-13th week of pregnancy.

10. A method for predicting the risk of preeclampsia according to any one of claims 1 to 9, characterized in that, In step S5, the set threshold rule is that when the probability of preeclampsia is greater than 0.54, the risk level is high, and when the probability of preeclampsia is less than or equal to 0.54, the risk level is low.

Images