Child severe mycoplasmal pneumonia risk prediction method and system based on dynamic column diagram
By receiving and calculating the risk of severe mycoplasma pneumonia in children through a dynamic nomogram system, the problems of subjective judgment and cumbersome operation in existing methods are solved, and rapid and accurate risk prediction and decision support are achieved.
Patent Information
- Application Number
- CN202511531152.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-10-24
- Publication Date
- 2026-01-30
AI Technical Summary
Existing methods for predicting the risk of severe mycoplasma pneumonia in children rely on doctors' subjective judgment based on experience, lack quantitative standards, and thus have a high rate of misjudgment. Static rating scales and traditional nomograms are cumbersome to operate, inefficient, and cannot meet the needs of rapid decision-making.
A risk prediction system based on dynamic nomograms is adopted. It receives multiple clinical indicators input by users through a web interface, automatically calculates them using a pre-trained risk prediction model, updates and visualizes the risk probability values in real time, and realizes interactive hypothesis analysis.
It provides precise quantitative risk probability values, reduces misjudgments, improves assessment efficiency and accuracy, lowers the risk of operational errors, enhances clinical applicability, and supports cross-platform access.
Smart Images
Figure CN121439207A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of pediatric critical illness risk prediction technology, specifically a method and system for predicting the risk of severe mycoplasma pneumonia in children based on dynamic nomograms. Background Technology
[0002] When predicting the risk of severe mycoplasma pneumoniae pneumonia (SMPP) in children, the following methods are mainly used: 1. Clinical experience assessment based on a single or a few key indicators: This is currently the most commonly used but also the most traditional method in clinical practice. Doctors mainly rely on a limited number of clinical or laboratory indicators and combine them with personal experience to make a subjective judgment on the risk of the child developing SMPP; 2. Assessment method based on static rating scales: In order to overcome the drawbacks of purely experience-based judgment, some rating scales for assessing the severity of pneumonia have emerged in the research field. Although they are not specifically designed for pediatric MPP, they are sometimes tried to be used or adapted in clinical practice; 3. Predictive model based on traditional static nomogram: Researchers have identified several independent predictors through regression analysis and constructed a static nomogram model.
[0003] However, the above methods still have shortcomings. Clinical experience assessment is highly subjective and lacks quantitative standards: it relies on the doctor's personal experience, the assessment results are inconsistent, and it cannot give a precise risk probability (such as 85%). It can only make a vague judgment of "high, medium, and low", which has low accuracy and is prone to misjudgment. Static scoring scales have poor specificity and inaccurate prediction: they are not designed for children's MPP, and their indicator selection and weights have poor applicability to the prediction of SMPP, with an excessively high error rate (error rate) and limited clinical guidance value. Traditional static nomograms are cumbersome to operate and slow to react: doctors need to manually find, draw lines, and add scores on paper charts. The whole process takes several minutes, which is inefficient and cannot meet the needs of rapid clinical decision-making. Moreover, it is very easy to make calculation errors due to human factors (high error rate). Therefore, it is necessary to improve them. Summary of the Invention
[0004] The purpose of this invention is to provide a method and system for predicting the risk of severe mycoplasma pneumonia in children based on dynamic nomograms, so as to solve the problems mentioned in the background art.
[0005] To achieve the above objectives, the present invention provides the following technical solution: a method for predicting the risk of severe mycoplasma pneumonia in children based on dynamic nomograms, the specific steps of which are as follows:
[0006] Step 1: Receive user input
[0007] The system receives multiple clinical indicators of the child patient from the user through a web interface.
[0008] Step 2: Send data
[0009] The input data is sent to the server automatically without refreshing the page;
[0010] Step 3: Risk Prediction Model Calculation
[0011] The server calls a pre-stored risk prediction model to process and calculate the risk probability value.
[0012] Step 4: Data Return
[0013] The server's application logic controller packages the risk probability value and the intermediate data used to draw the column chart into JSON format and returns it to the client.
[0014] Step 5: Visualization of Risk Probability Values
[0015] The client dynamically draws a line graph and visualizes the risk probability value for the user.
[0016] As a preferred technical solution of the present invention, the specific method of receiving user input in step one is as follows: the system receives multiple clinical indicators of the child patient input by the user through input controls pre-designed on its user interface. The input controls include a numeric input box for inputting continuous values; a slider selector for adjusting values within a predetermined range to provide a more intuitive interactive experience; and a drop-down menu / radio button for selecting categorical variables.
[0017] As a preferred technical solution of the present invention, after receiving multiple clinical indicators of the child input by the user through the Web interface in step one, the changes in the input data are monitored in real time. Once a change is detected, the steps of sending data, calculating the risk prediction model, returning data, and visualizing the risk probability value are automatically re-executed, thereby realizing interactive hypothesis analysis.
[0018] As a preferred technical solution of the present invention, the specific method of sending data in step two is as follows: the client listens for changes in the input control, and when the input is completed or changes, it automatically encapsulates all the input data asynchronously into a JSON format data packet through AJAX technology and sends it to the application logic controller on the server side through the HTTP protocol.
[0019] As a preferred technical solution of the present invention, the specific method of calculating the risk prediction model in step three is as follows: after the application logic controller of the server receives the data, it calls the risk prediction model object preloaded in memory, passes the user input data into the risk prediction model for calculation, and the risk prediction model calculates the corresponding risk probability value.
[0020] As a preferred technical solution of the present invention, the specific method of dynamically drawing the nomogram by the client in step five is as follows: after receiving the response, the client uses a JavaScript visualization library to dynamically draw the nomogram on the webpage based on the returned data, and displays the final risk probability value in a prominent position in the form of numbers and progress bars.
[0021] As a preferred embodiment of the present invention, the specific method for constructing the risk prediction model in step three is as follows:
[0022] Data preparation:
[0023] Anonymized electronic medical record data of children with Mycoplasma pneumoniae pneumonia who met the inclusion criteria were collected. The total sample size was N=1500 cases, including 500 cases in the severe case group (SMPP) and 1000 cases in the non-severe case group (MPP).
[0024] Predictor variables (features):
[0025] Based on clinical experience and literature review, 38 candidate predictive variables were initially selected, including demographic characteristics, clinical symptoms, signs and laboratory test results within 24 hours of admission.
[0026] Ending variable:
[0027] Based on the diagnostic criteria for severe pneumonia, the children were divided into the SMPP group (assigned a value of 1) and the MPP group (assigned a value of 0);
[0028] Data preprocessing:
[0029] Tools and packages: R language (v4.2.0), glmnet package (v4.1-4);
[0030] Data cleaning was performed using R. For continuous variables, the KNN algorithm was used to impute missing values (using the DMwR2::knnImputation function); for categorical variables, the mode was used to impute missing values; and Z-score standardization was performed on all continuous variables to eliminate the influence of units.
[0031] Variable selection and model training:
[0032] The dataset was randomly divided into training and validation sets in a 7:3 ratio. On the training set, Lasso regression was used to select variables and build a model. Lasso regression automatically compressed the coefficients of unimportant variables to 0 through the L1 regularization penalty term, thereby achieving variable selection. Finally, Lasso regression selected 13 variables with the most predictive value. The non-zero coefficients of these variables constituted the final model.
[0033] Model validation and nodal plot drawing:
[0034] Validation: Use the validation set to evaluate the performance of the risk prediction model;
[0035] Nonograph Conversion: Based on the regression coefficients of the final model, use the rms package (v6.3-0) to convert the risk prediction model into a nonograph format;
[0036] At this point, the risk prediction model is complete, and the model object (final_model) and the nodal graph conversion rules are saved for integration into the web system.
[0037] To achieve the aforementioned other objective, the present invention provides the following technical solution: a risk prediction system for severe mycoplasma pneumonia in children based on dynamic nomograms, comprising: a user interaction module (front-end UI), a prediction calculation module (back-end model), and a data communication module (HTTP / JSON interface), wherein:
[0038] The user interaction module (front-end UI) is used to receive clinical indicator data input by the user and dynamically display the prediction results;
[0039] The prediction calculation module (back-end model) is used to store the risk prediction model and calculate the risk probability value based on the received data;
[0040] The data communication module (HTTP / JSON interface) is used for asynchronous data exchange between the user interaction module and the prediction calculation module.
[0041] As a preferred embodiment of the present invention, the user interaction module (front-end UI) includes an input unit for providing slider and numeric input box controls to receive data; and a visualization unit for dynamically drawing column charts and displaying risk probability values.
[0042] The beneficial effects of this invention are as follows:
[0043] 1. This invention uses a pre-trained quantitative prediction model based on multi-dimensional clinical data to perform calculations, completely eliminating the subjective judgment mode that relies on doctors' personal experience. The model is trained on large-scale clinical data, and its weights and structure are fixed and objective. The system outputs a precise quantitative risk probability value for each child, rather than a vague judgment of "high, medium, or low". This greatly reduces the omissions or misjudgments caused by differences in doctors' experience, and provides a reliable objective basis for early and precise intervention in SMPP.
[0044] 2. This invention achieves automated, high-speed evaluation with "instant results upon input," reducing the manual table lookup and calculation process that originally took several minutes to seconds, with no errors in the results. This feature transforms all the tedious manual steps (searching, drawing lines, and adding) into automatic code execution in the background. In busy clinical work, doctors can obtain reliable results in just a few seconds, greatly improving work efficiency. At the same time, it completely eliminates errors that may occur in manual calculation and reading, ensuring the accuracy of each evaluation.
[0045] 3. This invention enables dynamic, real-time interactive hypothesis analysis, allowing users to instantly trigger recalculation and graphical updates for any adjustment to input parameters without refreshing the page. Doctors can intuitively explore "how the risk of a child will change if a certain indicator improves (such as a decrease in CRP)" by dragging a slider, providing a powerful quantitative decision-making tool for the selection of treatment strategies and the prediction of efficacy, greatly enhancing the clinical practical value of the system.
[0046] 4. The system of the present invention is presented in the form of a website, realizing cross-platform, installation-free instant access. Users only need an internet-connected browser device (such as a computer, tablet, or mobile phone) to use the professional tool anytime and anywhere without installing complicated software, which greatly reduces the threshold for use and facilitates its promotion and popularization in hospitals at all levels and even grassroots medical institutions. Attached Figure Description
[0047] Figure 1 This is a flowchart of the method for predicting the risk of severe mycoplasma pneumonia in children according to the present invention;
[0048] Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation
[0049] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0050] like Figure 1 As shown in the figure, this embodiment of the invention provides a method for predicting the risk of severe mycoplasma pneumoniae in children based on dynamic nomograms. The specific steps are as follows:
[0051] Step 1: Receive user input
[0052] The system receives multiple clinical indicators of the child patient from the user through a web interface.
[0053] Step 2: Send data
[0054] Automatically sends input data to the server without refreshing the page;
[0055] Step 3: Risk Prediction Model Calculation
[0056] The server calls a pre-stored risk prediction model to process and calculate the risk probability value.
[0057] Step 4: Data Return
[0058] The server's application logic controller packages the risk probability value and the intermediate data used to draw the column chart into JSON format and returns it to the client.
[0059] Step 5: Visualization of Risk Probability Values
[0060] The client dynamically draws nomograms and visualizes the risk probability values for the user.
[0061] Steps one through five creatively combine traditional medical prediction models with modern web interaction technologies, specifically including: The implementation of dynamic interactive nodal plots: This is not simply digitizing paper nodal plots, but rather creating a dynamic tool that can instantly respond to input changes and visualize output results through event listening, asynchronous communication (AJAX), and real-time graphics rendering technologies, overcoming the fatal flaw of static nodal plots in their inability to perform hypothesis analysis; An automated closed loop of "input-calculation-visualization": Any user operation (input or modification) on the front-end interface (UI) automatically triggers the calculation of the back-end model and automatically updates the results to the front-end interface in real time in graphical and numerical form, creating a seamless connection. A closed-loop system replaces tedious and error-prone manual operations, which is the core of improving efficiency and quality; deep integration of risk prediction models with web systems: the risk prediction model trained with the statistical / machine learning language (R) allows professional mathematical models to be used directly by clinicians in a user-friendly way; model feature engineering based on specific clinical scenarios: although the model algorithm itself is known, for the specific clinical problem of "severe mycoplasma pneumonia in children", the process of selecting the most effective combination of predictive factors (such as fever days, CRP, LDH, etc.) from many clinical indicators and determining their weights is itself a key technical point, which ensures the professionalism and accuracy of the tool.
[0062] The specific steps for receiving user input in step one are as follows: The system receives multiple clinical indicators of the child patient from the user through pre-designed input controls on its user interface. The input controls include a numeric input box for inputting continuous values; a slider selector for adjusting values within a predetermined range to provide a more intuitive interactive experience; and a drop-down menu / radio button for selecting categorical variables.
[0063] The numeric input box accepts continuous numerical values such as age (years), number of days with fever (days), CRP (mg / L), and LDH (U / L); categorical variables include gender and imaging findings. Step one enables structured data collection, providing input for automated calculations.
[0064] In step one, after receiving multiple clinical indicators of the child patient from the user through the web interface, the changes in the input data are monitored in real time. Once a change is detected, the steps of sending data, calculating the risk prediction model, returning data, and visualizing the risk probability value are automatically re-executed, thereby realizing interactive hypothesis analysis.
[0065] The system continuously monitors all input controls. When the user drags the slider or modifies the value in any input box, the system immediately triggers the entire process from step two to step five: sending the new parameters to the server, recalculating, receiving the new results, and finally updating the column chart and risk value on the UI in real time. This process does not require the user to manually submit or refresh the entire webpage, resulting in a smooth experience. Its technical foundation is event-driven programming and asynchronous web communication technology (AJAX), which ensures the real-time operation and the smoothness of the interface.
[0066] In step two, the specific method for sending data is as follows: the client listens for changes in the input control, and when the input is completed or changes, it automatically encapsulates all the input data asynchronously into a JSON format data packet using AJAX technology, and sends it to the application logic controller on the server side via the HTTP protocol.
[0067] AJAX technology refers to asynchronous JavaScript and XML technology, which is a technology that uses JavaScript to exchange data asynchronously with the server in the background. Its core is to partially update web page content: when the user triggers an event (such as clicking a button or entering content), the browser sends an HTTP request to the server through the XMLHttpRequest object or Fetch API. After the server processes the request, it returns data in JSON / XML / text formats, and JavaScript parses the data and dynamically updates the DOM without refreshing the entire page.
[0068] The specific implementation of the risk prediction model calculation in step three is as follows: After receiving the data, the application logic controller of the server calls the risk prediction model object preloaded in memory, passes the user input data into the risk prediction model for calculation, and the risk prediction model calculates the corresponding risk probability value.
[0069] The risk prediction model is encapsulated and deployed on a web server, accepting front-end data input through a defined interface and returning structured calculation results.
[0070] In step five, the client dynamically draws the nodal chart as follows: after receiving the response, the client uses a JavaScript visualization library to dynamically draw the nodal chart on the webpage based on the returned data, and displays the final risk probability value in a prominent position in the form of numbers and progress bars.
[0071] Through data interaction between the web front-end and back-end and the invocation of risk prediction models, the automation of back-end calculations and the visualization of results were achieved.
[0072] The specific method for constructing the risk prediction model in step three is as follows:
[0073] Data preparation:
[0074] Anonymized electronic medical record data of children with Mycoplasma pneumoniae pneumonia who met the inclusion criteria were collected. The total sample size was N=1500 cases, including 500 cases in the severe case group (SMPP) and 1000 cases in the non-severe case group (MPP).
[0075] Predictor variables (features):
[0076] Based on clinical experience and literature review, 38 candidate predictive variables were initially selected, including demographic characteristics, clinical symptoms, signs and laboratory test results within 24 hours of admission.
[0077] Ending variable:
[0078] Based on the diagnostic criteria for severe pneumonia, the children were divided into the SMPP group (assigned a value of 1) and the MPP group (assigned a value of 0);
[0079] Data preprocessing:
[0080] Tools and packages: R language (v4.2.0), glmnet package (v4.1-4);
[0081] Data cleaning was performed using R. For continuous variables, the KNN algorithm was used to impute missing values (using the DMwR2::knnImputation function); for categorical variables, the mode was used to impute missing values; and Z-score standardization was performed on all continuous variables to eliminate the influence of units.
[0082] Variable selection and model training:
[0083] The dataset was randomly divided into training and validation sets in a 7:3 ratio. On the training set, Lasso regression was used to select variables and build a model. Lasso regression automatically compressed the coefficients of unimportant variables to 0 through the L1 regularization penalty term, thereby achieving variable selection. Finally, Lasso regression selected 13 variables with the most predictive value. The non-zero coefficients of these variables constituted the final model.
[0084] Model validation and nodal plot drawing:
[0085] Validation: Use the validation set to evaluate the performance of the risk prediction model;
[0086] Nonograph Conversion: Based on the regression coefficients of the final model, use the rms package (v6.3-0) to convert the risk prediction model into a nonograph format;
[0087] At this point, the risk prediction model is complete, and the model object (final_model) and the nodal graph conversion rules are saved for integration into the web system.
[0088] Predictive variables (features) included age, number of days with fever, highest body temperature, number of days with cough, dyspnea, extrapulmonary complications, WBC, NEUT%, CRP, LDH, ALB, ALT, AST, Ferritin, D-dimer, pleural effusion, and number of lobular consolidations. The diagnostic criteria for severe pneumonia were based on the "Guidelines for the Diagnosis and Treatment of Community-Acquired Pneumonia in Children." The 13 most predictive variables selected by Lasso regression included number of days with fever, highest body temperature, CRP, LDH, and presence of pleural effusion. During validation, the area under the receiver operating characteristic (AUC) curve of the risk prediction model was 0.89, indicating good discrimination, and the p-value of the Hosmer-Lemeshow goodness-of-fit test was 0.45, indicating good calibration.
[0089] like Figure 2 As shown, this embodiment of the invention also provides a risk prediction system for severe mycoplasma pneumonia in children based on dynamic nomograms, including: a user interaction module (front-end UI), a prediction calculation module (back-end model), and a data communication module (HTTP / JSON interface), wherein:
[0090] The user interaction module (front-end UI) is used to receive clinical indicator data input by the user and dynamically display the prediction results;
[0091] The prediction calculation module (back-end model) is used to store the risk prediction model and calculate the risk probability value based on the received data;
[0092] The data communication module (HTTP / JSON interface) is used for asynchronous data exchange between the user interaction module and the prediction calculation module.
[0093] like Figure 2 As shown in the diagram, the client (browser) is accessed by the user through any device with a browser (such as a computer or tablet), and it contains:
[0094] User interface (UI): A web page interface generated by HTML rendering, which is the entry point for users to interact with the system;
[0095] Dynamic input components include, but are not limited to, numeric input boxes, slider selectors, drop-down menus, etc., for users to input or select various clinical indicators of the child (such as age, number of days of fever, CRP value, etc.). These components are tightly integrated with the UI.
[0096] Visualization Results Area: Used to dynamically and graphically display prediction results, including plotting nomograms and displaying digitized risk probability values. This area is tightly integrated with the UI.
[0097] Connection method: User operations on the UI (such as inputting data, sliding sliders) will trigger client logic to assemble the input data into an HTTP request (usually in JSON format), which is then sent to the server via the Internet. After the server finishes processing, it returns an HTTP response (in JSON format). The client receives the response, parses the data, and updates the visualization results area.
[0098] The server-side (Shiny Server) in the diagram is the core computing and processing unit of the system, built using the R Shiny framework, and internally includes:
[0099] Web service layer (R Shiny): Responsible for handling HTTP communication and application flow control;
[0100] Application Logic Controller: This is a server function defined in the R Shiny framework. It receives HTTP requests from clients, parses out the input parameters, calls the backend risk prediction model layer to perform calculations, and finally organizes the JSON data of the calculation results into an HTTP response and returns it to the client. It communicates bidirectionally with the client via the HTTP protocol.
[0101] Risk prediction model layer: This is a pre-trained R language model object that is loaded into the server's memory. It receives data from the application logic controller, performs prediction calculations, and returns the results (predicted probability values) to the application logic controller.
[0102] Data caching layer: Used to store static data required for model operation, such as model coefficients, nodal plot plotting rules, threshold configurations, etc. The risk prediction model layer and application logic controller can read the data in it;
[0103] Connection method: The application logic controller interacts with the prediction model layer and data cache layer through internal function calls and variable references in the R language.
[0104] The user interaction module (front-end UI) includes an input unit for providing slider and numeric input box controls to receive data, and a visualization unit for dynamically drawing column charts and displaying risk probability values.
[0105] The graphical user interface displayed on the terminal device includes: a first area with multiple interactive controls for inputting clinical indicator data of the child; a second area that dynamically displays a nomogram corresponding to the currently input data; and a third area that dynamically displays the risk probability value derived from the nomogram. When the user modifies the input data by operating the controls in the first area, the contents of the second and third areas are updated synchronously in real time.
[0106] It should be noted that, in this document, relational terms such as "first" and "second" are used only 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 process, method, article, or apparatus.
[0107] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A dynamic column chart-based method for predicting the risk of severe mycoplasma pneumonia in children, characterized in that, The specific steps are as follows: Step one: receiving user input Receive user input of multiple clinical indicators of children through the Web interface; Step two: send data The input data is automatically sent to the server without refreshing the page; Step three: risk prediction model calculation The server calls the pre-stored risk prediction model for processing and calculation to obtain the risk probability value; Step four: data return The application logic controller of the server packs the risk probability value and the intermediate data for drawing the nomogram into JSON format and returns it to the client; Step five: visualization of risk probability value The client dynamically draws the nomogram and visualizes the risk probability value to the user.
2. The dynamic column graph-based method for predicting the risk of children with severe mycoplasma pneumonia according to claim 1, characterized in that: The specific method of receiving user input in step one is: the system receives user input of multiple clinical indicators of children through pre-designed input controls on its user interface, including numerical input boxes for inputting continuous numerical values, slider selectors for adjusting numerical values within a predetermined range to provide a more intuitive interactive experience, and drop-down menus / single selection buttons for selecting categorical variables.
3. The dynamic column graph-based method for predicting the risk of children with severe mycoplasma pneumonia according to claim 1, characterized in that: After receiving the multiple clinical indicators of children through the Web interface in step one, the system listens to the changes in input data in real time. Once a change is detected, it automatically re-executes the steps of sending data, risk prediction model calculation, data return, and visualization of risk probability value, thereby realizing interactive hypothesis analysis.
4. The dynamic column graph-based method for predicting the risk of children with severe mycoplasma pneumonia according to claim 1, characterized in that: The specific method of sending data in step two is: the client listens to the changes in input controls. When the input is complete or changes, it automatically encapsulates all input data into a JSON format data package asynchronously through AJAX technology and sends it to the application logic controller on the server side through the HTTP protocol.
5. The dynamic column graph-based method for predicting the risk of children with severe mycoplasma pneumonia according to claim 1, characterized in that: The specific method of risk prediction model calculation in step three is: after receiving the data, the application logic controller of the server calls the risk prediction model object preloaded in the memory and transmits the user input data into the risk prediction model for calculation. The risk prediction model calculates the corresponding risk probability value.
6. The dynamic column graph-based method for predicting the risk of children with severe mycoplasma pneumonia according to claim 1, characterized in that: The specific method of dynamically drawing the nomogram by the client in step five is: after receiving the response, the client uses a JavaScript visualization library to dynamically draw a nomogram on the webpage based on the returned data and display the final risk probability value in the form of a number and a progress bar in a prominent position.
7. The dynamic column graph-based method for predicting the risk of children with severe mycoplasma pneumonia according to claim 1, characterized in that: The specific method of constructing the risk prediction model in step three is: Data preparation: Collect anonymized electronic medical record data of children with mycoplasma pneumoniae pneumonia who meet the inclusion criteria, with a total sample size N=1500, including 500 cases of severe SMPP and 1000 cases of non-severe MPP; Predictive variables (features): Based on clinical experience and literature review, 38 candidate predictive variables are initially selected, including demographic characteristics, clinical symptoms, signs, and laboratory examination indicators within 24 hours after admission; Outcome variable: According to the diagnostic criteria for severe pneumonia, children are divided into SMPP group (assigned value 1) and MPP group (assigned value 0); Data preprocessing: Tools and packages: R language (v4.2.0), glmnet package (v4.1-4); Data cleaning: using R language, for continuous variables, using KNN algorithm to fill in missing values (using DMwR2::knnImputation function); for categorical variables, using mode to fill in missing values; Z-score standardization is performed on all continuous variables to eliminate the influence of dimension; Variable selection and model training: The dataset is randomly divided into training set and validation set in the ratio of 7:
3. In the training set, Lasso regression is used for variable selection and model construction. Lasso regression automatically compresses the coefficients of unimportant variables to 0 through L1 regularization penalty term, thereby realizing variable selection. Finally, Lasso regression selects 13 variables with the most predictive value. The non-zero coefficients of these variables constitute the final model. Model validation and nomogram drawing: Validation: using the validation set to evaluate the performance of the risk prediction model; Nomogram conversion: according to the regression coefficients of the final model, using the rms package (v6.3-0) to convert the risk prediction model into nomogram form; By now, the risk prediction model has been constructed, and the model object (final_model) and nomogram conversion rules are saved for integration into the Web system.
8. A dynamic column chart-based risk prediction system for severe Mycoplasma pneumonia in children, characterized by, It includes: user interaction module (front-end UI), prediction calculation module (back-end model), data communication module (HTTP / JSON interface), wherein: The user interaction module (front-end UI) is used to receive the clinical index data input by the user and dynamically display the prediction results; The prediction calculation module (back-end model) is used to store the risk prediction model and calculate the risk probability value according to the received data; The data communication module (HTTP / JSON interface) is used for asynchronous data exchange between the user interaction module and the prediction calculation module.
9. The dynamic column chart-based severe Mycoplasma pneumoniae pneumonia risk prediction system for children according to claim 8, characterized in that: The user interaction module (front-end UI) includes an input unit for providing a slider, a digital input box control to receive data; a visualization unit for dynamically drawing a nomogram and displaying a risk probability value.