A method and system for screening and malignant grading of a prognostic marker of a child with pneumonia, a marker and application thereof

Abstract

The application discloses a screening and malignant grading method and system of a prognosis risk marker of a child with pneumonia, a marker thereof and application, is based on clinical test index data, including blood, urine and fecal detection, and the optimal prognosis risk marker is screened from the blood, urine and fecal detection, a child pneumonia prognosis model based on machine learning technology is constructed, and the child is classified according to the best cut-off value. The model greatly improves the specificity and sensitivity of the prognosis prediction of the child with pneumonia and realizes efficient and accurate artificial intelligence prognosis malignant classification of the child with pneumonia, and can solve the problems of lack of means for evaluating the prognosis of the child with pneumonia, high price and inability to popularize and popularize.

Description

Technical Field

[0001] This invention belongs to the field of medical artificial intelligence technology, specifically relating to a method, system, biomarkers, and applications for screening prognostic risk markers and classifying malignancy in children with pneumonia. Background Technology

[0002] Pneumonia is a lung infection and inflammation that can cause serious morbidity, affecting approximately 150 million people globally each year. Furthermore, pneumonia has a more severe impact on children. According to the World Health Organization, in 2019, approximately 740,000 children under the age of five died from pneumonia globally, accounting for 14% of all deaths among children under five. The 2019 "China Maternal and Child Health Development Report" indicates that the top five causes of death in children under five are premature birth, pneumonia, birth asphyxia, congenital heart disease, and accidental asphyxia, accounting for 55.7% of all deaths. Currently, pneumonia remains a leading cause of death in children; therefore, the screening of prognostic risk markers and methods for classifying the severity of pneumonia in children have extremely important social significance and value.

[0003] Malnutrition, low birth weight, non-breastfeeding, use of solid fuels, overcrowding, and zinc intake are considered risk factors associated with childhood pneumonia, but predictive factors for mortality in children with pneumonia are currently lacking. Radiological examination is a means of assessing the risk of death in children with pneumonia, but its complexity, susceptibility to physician subjective experience, high cost, and the risk of children being exposed to X-rays limit its widespread adoption. Clinical laboratory tests, including blood, urine, and stool analyses, are convenient and readily available methods. Therefore, using artificial intelligence methods to identify optimal prognostic risk biomarkers for children with pneumonia from clinical laboratory indicators, and further developing a clinically applicable and reliable prognostic model for children with pneumonia, along with malignancy grading, can help support clinical treatment and further improve the survival rate of children with severe pneumonia. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, and its biomarkers and applications for screening prognostic risk biomarkers and classifying the malignancy of children with pneumonia. By using prognostic risk biomarkers based on clinical laboratory indicators, this invention can assist in clinical treatment and improve the survival rate of children with pneumonia.

[0005] Therefore, as one aspect of the present invention, the present invention provides a method for screening prognostic risk markers and classifying the malignancy of pneumonia in children, comprising,

[0006] S1, Data Acquisition: Acquire clinical laboratory data of children diagnosed with pneumonia, including the cured group and the death group;

[0007] S2, Data Preprocessing: Clinical laboratory data of pneumonia patients in the cured group are treated as negative data, and clinical laboratory data of pneumonia patients in the deceased group are treated as positive data.

[0008] S3, Indicator Screening: Screening mortality risk indicators for children with pneumonia based on Cox proportional hazards regression analysis, as prognostic risk markers for children with pneumonia.

[0009] S4, Model Building: Based on the classification labels of positive and negative data, and the combination of indicators selected in step S3, stratified random sampling is used to divide the negative and positive data into training and test sets. A prognostic model for children with pneumonia is built based on the training set.

[0010] S5, Model Performance Evaluation: Evaluate the model's prediction results.

[0011] As a preferred embodiment of the method for screening prognostic risk markers and classifying malignancy in children with pneumonia as described in this invention: step S1 further includes,

[0012] S101, Clinical Laboratory Indicators: Includes blood test, urine test, and stool test results.

[0013] As a preferred embodiment of the method for screening prognostic risk markers and classifying malignancy in children with pneumonia as described in this invention, step S2 further includes:

[0014] S201, Missing value handling: Delete indicators where the number of people tested is less than 80% of all samples.

[0015] As a preferred embodiment of the method for screening prognostic risk markers and classifying the malignancy of pneumonia in children according to the present invention: In step S3, the calculation process for screening the indicators is as follows:

[0016] 1) In the entire dataset, a single-factor Cox proportional hazards regression analysis was used to obtain indicators that showed significant differences in the classification labels of negative and positive data;

[0017] 2) For indicators that show significant differences in univariate Cox proportional hazards regression analysis, multivariate Cox proportional hazards regression analysis is performed to obtain independent factors, which are used as a combination of indicators for prognostic mortality risk.

[0018] 3) The combination of indicators of prognostic mortality risk will be used as the final prognostic risk marker for children with pneumonia.

[0019] As a preferred embodiment of the screening and malignancy classification method for prognostic risk markers of children with pneumonia described in this invention: In step S3, the prognostic risk markers of children with pneumonia consist of four clinical test indicators, namely: oxygen saturation, hemoglobin, urea and uric acid.

[0020] As a preferred embodiment of the method for screening prognostic risk markers and classifying malignancy in children with pneumonia as described in this invention, step S4 further includes:

[0021] S401, Dataset Splitting: The indicator data filtered in step S3 is divided into a training set and a test set by stratified random sampling, with negative data and positive data in a 4:1 ratio.

[0022] S402, Construct the extreme gradient boosting model: Construct an extreme gradient boosting (XGBoost) model based on the metrics selected in step S3, using the training set;

[0023] S403, Training an extreme gradient boosting model: Adjusting parameters in the training set using 10-fold cross-validation to achieve the optimal combination.

[0024] S404, Determination of Prognostic Model for Children with Pneumonia: A prognostic model for children with pneumonia is constructed based on the optimal combination of parameters. The performance of the machine learning algorithm in the training set is evaluated using receiver operating characteristic (ROC) curves, thereby determining the optimal cutoff value for risk scores that distinguish different prognostic outcomes, as well as the area under the curve (AUC), sensitivity, and specificity.

[0025] As a preferred embodiment of the method for screening prognostic risk markers and classifying the malignancy of children with pneumonia as described in this invention, step S5 further includes:

[0026] S501, classifying the malignancy of pneumonia patients in the test set according to the optimal cutoff value of the risk score in the training set;

[0027] S502 uses the same method as S404 to evaluate the performance of the prognostic model for children with pneumonia using a test set.

[0028] As another aspect of the present invention, the present invention provides a system for screening prognostic risk markers and classifying the malignancy of children with pneumonia, comprising,

[0029] The data acquisition module (M1) is used to acquire clinical laboratory indicator data of children clinically diagnosed with pneumonia.

[0030] The data preprocessing module (M2) is used to treat the clinical test data of cured pneumonia children as negative data and the clinical test data of deceased pneumonia children as positive data.

[0031] The indicator screening module (M3) is used to screen indicators that show significant differences between positive and negative data based on univariate Cox proportional hazards regression analysis, and to screen indicators with prognostic mortality risk based on multivariate Cox proportional hazards regression analysis, serving as prognostic risk biomarkers for children with pneumonia.

[0032] The model building module (M4) uses stratified random sampling to divide negative and positive data into training and test sets based on the screened indicators, and uses the training set to build a prognostic prediction model for children with pneumonia.

[0033] The Model Performance Evaluation Module (M5) is based on the prognostic model for children with pneumonia. It classifies the severity of pneumonia according to the optimal cutoff value of the risk score and evaluates the model's prediction results based on the test set.

[0034] As another aspect of the present invention, the present invention provides prognostic risk markers for children with pneumonia obtained by the method, wherein the markers are oxygen saturation, hemoglobin, urea and uric acid.

[0035] As another aspect of the present invention, the present invention provides the application of the method and system for screening prognostic risk markers and classifying malignancy in children with pneumonia, wherein: the extreme gradient enhancement model is constructed with the following parameters: learning rate (eta) of 0.1, sample sampling ratio (subsample) of 80%, maximum subtree depth (max.depth) of 3, feature sampling ratio (colsample_bytree) of 0.5, maximum number of iterations (nround) of 170, and the evaluation index is logloss.

[0036] The beneficial effects of this invention are as follows: This invention provides a method, system, and its biomarkers and applications for screening prognostic risk markers and classifying the malignancy of children with pneumonia. Based on clinical laboratory indicator data, the optimal specific biomarkers for death in children with pneumonia are screened out, thereby constructing a prognostic prediction model for pneumonia based on artificial intelligence machine learning technology. This greatly improves the specificity and sensitivity of prognostic prediction for children with pneumonia, and achieves efficient and accurate artificial intelligence-based prediction of pneumonia mortality risk and malignancy classification. It can solve the problems of lack of prognostic assessment methods for children with pneumonia, high cost, and inability to be widely promoted and popularized. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0038] Figure 1 This is a flowchart of the method for screening prognostic risk biomarkers in children with pneumonia according to an embodiment of the present invention.

[0039] Figure 2 This is a system framework diagram of the prognostic risk biomarker screening system for children with pneumonia according to an embodiment of the present invention.

[0040] Figure 3 This is a multivariate Cox proportional hazards regression forest plot of 24 clinical laboratory indicators that show significant differences between negative and positive classification labels in children with pneumonia according to an embodiment of the present invention.

[0041] Figure 4 This is a receiver operating characteristic curve of the prognosis prediction model for children with pneumonia according to an embodiment of the present invention.

[0042] Figure 5 This is a survival curve of the prognosis prediction model for children with pneumonia in this embodiment of the invention after malignancy grading in external children with pneumonia.

[0043] Figure 6 This is a ranking of the importance of the four prognostic risk markers for children with pneumonia in this invention. Detailed Implementation

[0044] To make the above-mentioned objectives, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to specific examples.

[0045] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0046] Figure 1 This is a flowchart illustrating a method for screening prognostic risk markers and classifying the malignancy of pneumonia in children according to an embodiment of the present invention. Figure 1 As shown, the evaluation method includes the following steps:

[0047] S1, Data Acquisition: Acquire clinical laboratory data of children diagnosed with pneumonia.

[0048] In one embodiment, this method obtains all clinical laboratory data of 631 children clinically diagnosed with pneumonia from the Children's Hospital Affiliated to Zhejiang University School of Medicine between 2010 and 2018.

[0049] S2, Data Preprocessing: Clinical laboratory data of children with pneumonia who recovered were treated as negative data, and clinical laboratory data of children with pneumonia who died were treated as positive data.

[0050] In one embodiment, among the clinical laboratory data of children diagnosed with pneumonia, the data of 536 patients who were cured and discharged were considered negative, while the data of 95 patients who died in hospital were considered positive.

[0051] Preferably, after identifying positive and negative data, data preprocessing is required. This includes the following steps:

[0052] S201, Missing value handling: Delete indicators where the number of people tested is less than 80% of all samples.

[0053] In one embodiment, after handling missing values, 87 indicators remain in the clinical laboratory data of children diagnosed with pneumonia.

[0054] S3, Indicator Screening: Based on Cox proportional hazards regression analysis, indicators of mortality risk in children with pneumonia were screened as prognostic risk markers for children with pneumonia.

[0055] Specifically, the calculation process for indicator selection is as follows:

[0056] 1) In the entire dataset, univariate Cox proportional hazards regression analysis was used to obtain indicators that showed significant differences between negative and positive data classification labels, namely 24 clinical laboratory indicators.

[0057] 2) For indicators that show significant differences after univariate Cox proportional hazards regression analysis, multivariate Cox proportional hazards regression analysis is then performed to obtain four independent factors with prognostic mortality risk, which serve as a combination of prognostic mortality risk indicators.

[0058] 3) Combine indicators with prognostic mortality risk will be used as the final prognostic risk biomarkers for children with pneumonia;

[0059] Figure 3 This is a forest plot of 24 clinical laboratory indicators from an embodiment of the present invention. For example... Figure 3 As shown, after screening by univariate Cox proportional hazards regression analysis, a total of 24 indicators showed significant prognostic risk in both the negative and positive groups. After screening by multivariate Cox proportional hazards regression analysis, four independent factors with prognostic risk were obtained: oxygen saturation (%), hemoglobin (g / dL), urea (mmol / L), and uric acid (μmol / L).

[0060] S4, Model Construction: Based on the classification labels of positive and negative data, and the combination of indicators selected in S3, stratified random sampling is used to divide the negative and positive data into training and test sets. A prognostic model for children with pneumonia is constructed based on the training set.

[0061] In one embodiment, to ensure the model's generalization ability, an extreme gradient boosting (XGBoost) model is used for classification prediction. Specifically, the model construction steps include:

[0062] S401, Dataset Splitting: The indicator data filtered in step S3 is divided into a training set and a test set by stratified random sampling, with negative data and positive data in a 4:1 ratio.

[0063] In one embodiment, the training set contains 505 cases, including 429 negative cases and 76 positive cases; the test set contains 126 cases, including 107 negative cases and 19 positive cases.

[0064] S402, Constructing an extreme gradient boosting model: Construct an XGBoost model based on the metrics selected in step S3, using the training set;

[0065] S403, Training an extreme gradient boosting model: Parameters are continuously adjusted in the training set using 10-fold cross-validation to achieve the optimal combination.

[0066] In one embodiment, the specific parameters of the extreme gradient boosting model are as follows:

[0067]

[0068] S404, Determination of Prognostic Model for Children with Pneumonia: A prognostic model for children with pneumonia is constructed based on the optimal combination of parameters. The performance of the machine learning algorithm in the training set is evaluated using receiver operating characteristic (ROC) curves, thereby determining the optimal cutoff value for risk scores that distinguish different prognostic outcomes, as well as the area under the curve (AUC), sensitivity, and specificity.

[0069] like Figure 4 As shown, in one embodiment, four indicators with independent prognostic mortality risk were used as sample data to train the prognostic risk model for children with pneumonia. The parameters in the evaluation results are as follows: sensitivity = 0.928, specificity = 1.0, AUC = 0.992; the optimal cutoff value is: cutoff = 0.210.

[0070] S5, Model Performance Evaluation: Evaluate the model's prediction results.

[0071] like Figure 4 As shown, in one embodiment, the evaluation result of the test set data validation is as follows: AUC = 0.855.

[0072] like Figure 5 As shown, the test set divided patients into high-risk and low-risk groups according to the optimal cutoff value. The mortality rate within 100 days of the high-risk group was significantly higher than that of the low-risk group (p-value < 0.001), and the hazard ratio (HR) reached 1306.

[0073] Figure 2 This is a system framework diagram of the prognostic risk biomarker screening and malignancy grading system for children with pneumonia, according to an embodiment of the present invention. Figure 2 As shown, the system of the present invention includes:

[0074] The data acquisition module (M1) is used to acquire clinical laboratory indicator data of children clinically diagnosed with pneumonia.

[0075] The data preprocessing module (M2) is used to treat the clinical test data of cured pneumonia children as negative data and the clinical test data of deceased pneumonia children as positive data.

[0076] The indicator screening module (M3) is used to screen indicators that show significant differences between positive and negative data based on univariate Cox proportional hazards regression analysis, and to screen indicators with prognostic mortality risk based on multivariate Cox proportional hazards regression analysis.

[0077] The model building module (M4) uses stratified random sampling to divide negative and positive data into training and test sets based on the screened indicators, and uses the training set to build a prognostic prediction model for children with pneumonia.

[0078] The model performance evaluation module (M5) classifies the malignancy of children with pneumonia according to the optimal cutoff value of the risk score and evaluates the model's prediction results based on the test set.

[0079] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for screening prognostic risk markers and classifying the malignancy of pneumonia in children, characterized in that: include, S1, Data Acquisition: Acquire clinical laboratory data of children diagnosed with pneumonia, including the cured group and the death group; S2, Data Preprocessing: Clinical laboratory data of pneumonia patients in the cured group are treated as negative data, and clinical laboratory data of pneumonia patients in the deceased group are treated as positive data. S3, Indicator Screening: Screening mortality risk indicators for children with pneumonia based on Cox proportional hazards regression analysis, as prognostic risk markers for children with pneumonia. S4, Model Building: Based on the classification labels of positive and negative data, and the combination of indicators selected in step S3, stratified random sampling is used to divide the negative and positive data into training and test sets. A prognostic model for children with pneumonia is built based on the training set. S5, Model Performance Evaluation: Evaluate the model's prediction results; Step S4 also includes: S401, Dataset Splitting: The indicator data filtered in step S3 is divided into a training set and a test set by stratified random sampling, with negative data and positive data in a 4:1 ratio; S402, Construct an extreme gradient boosting model: Construct an extreme gradient boosting model based on the selected indicators in step S3, using the training set; S403, Training an extreme gradient boosting model: Adjusting parameters to achieve the optimal combination using 10-fold cross-validation on the training set; S404, Determination of Prognostic Model for Children with Pneumonia: A prognostic model for children with pneumonia was constructed based on the optimal combination of parameters. The performance of the machine learning algorithm in the training set was evaluated using receiver operating characteristic (ROC) curves, thereby determining the optimal cutoff value for risk scores that distinguish different prognostic outcomes, as well as the area under the curve, sensitivity, and specificity. The extreme gradient boosting model is constructed with the following parameters: learning rate of 0.1, sample sampling ratio of 80%, maximum subtree depth of 3, feature sampling ratio of 0.5, maximum number of iterations of 170, and evaluation metric of logloss. The prognostic risk markers for children with pneumonia consist of four clinical laboratory indicators: oxygen saturation, hemoglobin, urea, and uric acid. In step S3, the calculation process for the index selection is as follows: 1) In the entire dataset, a single-factor Cox proportional hazards regression analysis was used to obtain indicators that showed significant differences in the classification labels of negative and positive data; 2) For indicators that show significant differences in univariate Cox proportional hazards regression analysis, multivariate Cox proportional hazards regression analysis is performed to obtain independent factors, which are used as a combination of indicators for prognostic mortality risk. 3) The combination of indicators with prognostic mortality risk will be used as the final prognostic risk markers for children with pneumonia.

2. The method for screening prognostic risk markers and classifying malignancy in children with pneumonia according to claim 1, characterized in that: Step S1 also includes, S101, Clinical Laboratory Indicators: Includes blood test, urine test, and stool test results.

3. The method for screening prognostic risk markers and classifying malignancy in children with pneumonia according to claim 1, characterized in that: Step S2 also includes: S201, missing value handling: delete indicators where the number of people tested is less than 80% of all samples.

4. The method for screening prognostic risk markers and classifying malignancy in children with pneumonia according to claim 1, characterized in that: Step S5 also includes: S501, classifying the malignancy of pneumonia patients in the test set according to the optimal cutoff value of the risk score in the training set; S502 uses the same method as S404 to evaluate the performance of the prognostic model for children with pneumonia using a test set.

5. A system for screening prognostic risk markers and classifying the malignancy of pneumonia in children, characterized in that: The system for screening prognostic risk markers and classifying the malignancy of children with pneumonia is used to execute the method for screening prognostic risk markers and classifying the malignancy of children with pneumonia as described in claim 1, and includes, The data acquisition module (M1) is used to acquire clinical laboratory indicator data of children clinically diagnosed with pneumonia. The data preprocessing module (M2) is used to treat the clinical test data of children who have recovered from pneumonia as negative data and the clinical test data of children who have died from pneumonia as positive data. The indicator screening module (M3) is used to screen indicators that show significant differences between positive and negative data based on univariate Cox proportional hazards regression analysis, and to screen indicators with prognostic mortality risk based on multivariate Cox proportional hazards regression analysis, serving as prognostic risk biomarkers for children with pneumonia. The model building module (M4) uses stratified random sampling to divide negative and positive data into training and test sets based on the screened indicators, and uses the training set to build a prognostic prediction model for children with pneumonia. The Model Performance Evaluation Module (M5) is based on the prognostic model for children with pneumonia. It classifies the severity of pneumonia according to the optimal cutoff value of the risk score and evaluates the model's prediction results based on the test set.

Images