Macrolide non-reactive mycoplasma pneumonia prediction model and construction method thereof

Abstract

The invention discloses a macrolide non-reactive mycoplasma pneumonia prediction model and a construction method, and belongs to the technical field of artificial intelligence. The invention aims to solve the problem of clinical macrolide non-reactive mycoplasma pneumonia (MUMPP) recognition lag. The method comprises the following steps: constructing a standardized data set; preprocessing the data, namely processing missing values by adopting a chain equation multi-interpolation method, and converting continuous variables into classification variables based on clinical significance or an ROC curve; training by adopting a random forest and a neural network model, screening an optimal feature subset in combination with a recursive feature elimination method, and optimizing model hyper-parameters through a grid search algorithm; and finally, evaluating the model from multiple dimensions of distinction degree, calibration degree and clinical effectiveness. Through a systematic modeling process, the constructed prediction model is high in accuracy, early warning of MUMPP can be realized, and time is won for clinically adjusting treatment strategies in time and preventing disease deterioration.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to a macrolide-unresponsive Mycoplasma pneumoniae pneumonia prediction model and a construction method thereof. BACKGROUND

[0002] Mycoplasma pneumoniae (MP) is a common pathogen of community-acquired pneumonia (CAP) in children, and is also not uncommon in infants aged 1 to 3 years. Mycoplasma pneumoniae pneumonia (MPP) as a common type of CAP in children has important public health significance in clinical practice.

[0003] Macrolide drugs are currently recommended as first-line drugs for the treatment of MPP by domestic and foreign guidelines. However, with the widespread use of macrolide drugs, more and more cases of macrolide-unresponsive Mycoplasma pneumoniae pneumonia (MUMPP) have appeared in clinical practice. MUMPP refers to MPP patients who still have fever, and no improvement or further aggravation of clinical signs and lung imaging after 72 hours of standard treatment with macrolide drugs. At present, the incidence of MUMPP in some areas has exceeded 30%, and has shown an increasing trend year by year.

[0004] The emergence of MUMPP has brought serious challenges to clinical treatment. If MUMPP is not treated properly, about 20% of patients may develop into refractory Mycoplasma pneumoniae pneumonia (RMPP). RMPP refers to MPP that still has fever, and the clinical signs and lung imaging are aggravated after 7 days or more of standard treatment with macrolide antibacterial drugs, and extrapulmonary complications occur. Compared with non-RMPP patients, RMPP patients have more severe clinical symptoms, a significantly higher risk of developing severe extrapulmonary complications, a longer hospitalization time, a higher proportion of ICU admission, and even death in severe cases. Recent clinical evidence has shown that antibiotic treatment failure has become one of the important reasons for the increased mortality rate of MP-infected children.

[0005] Currently, the identification of MUMPP in clinic mainly relies on clinical observation after treatment, and most MUMPP cases are diagnosed after 3 days of macrolide treatment, which seriously affects the timely adjustment of treatment strategy. The later the adjustment of treatment strategy for MUMPP, the higher the risk of developing RMPP in patients, and the poorer the prognosis. Therefore, how to accurately identify high-risk patients with MUMPP at the early stage of treatment has become an important problem that needs to be solved in clinic.

[0006] Although some researchers have tried to build a prediction model for RMPP in recent years, the existing technology still has many shortcomings. First, most of the existing prediction models do not fully consider the influence of treatment methods on disease outcome variables, which limits the prediction accuracy of the model. Second, due to the limitations of sample size and algorithm, the generalization ability and prediction accuracy of the existing model need to be improved. Third, the existing model is mostly limited to the risk prediction level and lacks substantive guidance for clinical treatment decisions. Most importantly, there is currently no special model building research for early prediction of MUMPP, which cannot meet the urgent needs of early identification and intervention of MUMPP in clinic.

[0007] Macrolide drug resistance is one of the important reasons for the occurrence of MUMPP. The gradual prevalence of MP drug-resistant strains further expands the MUMPP patient population and increases the risk of RMPP. However, the existing prediction methods fail to fully integrate drug resistance-related information and effectively utilize multi-dimensional clinical data at patient admission for comprehensive analysis. SUMMARY

[0008] The purpose of the present application is to provide a macrolide non-responsive mycoplasma pneumonia prediction model and its construction method, which can accurately identify high-risk MUMPP patients at the early stage of patient admission (such as within 24 hours), provide a scientific basis for clinicians to timely adjust treatment strategies and prevent disease progression to RMPP, thereby improving the prognosis of children and reducing the medical burden.

[0009] To achieve the above purpose, the present application realizes the following technical solutions: A construction method of a macrolide non-responsive mycoplasma pneumonia prediction model, comprising: S1: obtaining clinical data of mycoplasma pneumonia patients, screening prediction variables, and constructing an original data set; S2: preprocessing the original data set, including imputation processing of missing data, classification conversion of continuous variables, and encoding processing of classification variables, to obtain a standardized data set; S3: based on the standardized data set, a machine learning algorithm is used to construct a prediction model, and the optimal prediction model is obtained through feature selection and parameter optimization; S4: performance evaluation is performed on the optimal prediction model to obtain a macrolide non-responsive Mycoplasma pneumoniae pneumonia prediction model.

[0010] Further, in the step S1, the prediction variables include at least two of demographic characteristics variables, clinical symptom variables, laboratory examination variables, and imaging variables; and the acquisition of the clinical data includes: extracting patient data meeting the diagnostic criteria of Mycoplasma pneumoniae pneumonia from a community-acquired pneumonia database, and dividing the patients into a macrolide non-responsive group and a macrolide responsive group according to the clinical response after 72 hours of macrolide treatment.

[0011] Further, the missing data imputation processing in the step S2 adopts a chain equation multiple imputation method. The classification conversion of the continuous variables includes: For continuous variables with clinically recognized thresholds, the classification variables are directly converted based on the clinical thresholds; For continuous variables without clear clinical thresholds, the optimal cut-off value is determined through receiver operating characteristic curve analysis, and the classification variables are converted based on the optimal cut-off value.

[0012] Further, the determination of the optimal cut-off value through the receiver operating characteristic curve analysis includes: taking macrolide non-responsive Mycoplasma pneumoniae pneumonia as an outcome variable, calculating the Youden index corresponding to each possible value of the continuous variable, and selecting the value with the maximum Youden index as the optimal cut-off value; wherein the Youden index = sensitivity + specificity - 1.

[0013] Further, the classification variable encoding processing in the step S2 adopts one-hot encoding, specifically including: When the classification variable has only one option, it is directly converted into a binary variable; When the classification variable contains n options and n≥2, if the n options are mutually exclusive, n independent binary classification variables are converted; if the n options are not mutually exclusive, each option is split into a separate column and converted into a binary variable.

[0014] Further, the machine learning algorithm in the step S3 includes an ensemble learning algorithm and / or a deep learning algorithm; the feature selection adopts a recursive feature elimination method, the performance of different feature subsets is evaluated through cross-validation to determine the optimal feature subset; and the parameter optimization adopts a grid search algorithm to systematically optimize the hyperparameters of the model.

[0015] Further, the ensemble learning algorithm is a random forest algorithm, and the deep learning algorithm is a feedforward neural network algorithm; the hyperparameters of the random forest algorithm include the number of trees, the maximum depth, and the minimum number of samples required for node splitting; the feedforward neural network algorithm includes an input layer, at least one hidden layer, and an output layer, wherein the hidden layer adopts a ReLU activation function, and the output layer adopts a Sigmoid activation function.

[0016] Further, the performance evaluation in the step S4 includes at least two of discrimination evaluation, calibration evaluation, and clinical effectiveness evaluation; the discrimination evaluation is achieved by calculating a C-index or an area under a receiver operating characteristic curve; the calibration evaluation includes at least one of a Brier score, a goodness-of-fit test, or calibration curve analysis; and the clinical effectiveness evaluation adopts a decision curve analysis.

[0017] Further, the demographic characteristic variable includes at least two of age, gender, weight, and body mass index; the clinical symptom variable includes at least two of fever days, maximum body temperature, cough, and wheezing; the laboratory examination variable includes at least three of C-reactive protein, procalcitonin, white blood cell count, neutrophil ratio, lactate dehydrogenase, ferritin, albumin, and interleukin-6; and the imaging variable includes whether there is a large consolidation and / or whether there is pleural effusion.

[0018] The application also provides a macrolide-unresponsive mycoplasma pneumoniae pneumonia prediction model, which is constructed by the above-mentioned construction method.

[0019] Compared with the prior art, the application has the following beneficial effects: Firstly, the application provides a prediction model construction method specifically for macrolide-unresponsive mycoplasma pneumoniae pneumonia, which can accurately identify high-risk patients with MUMPP in the early stage of hospitalization (such as within 24 hours), and the identification time is at least 48 hours earlier than the traditional method which needs to be diagnosed after 72 hours of treatment, thereby winning valuable time for timely adjustment of clinical treatment strategies and effectively reducing the risk of patients progressing to refractory mycoplasma pneumonia.

[0020] Secondly, through a systematic data preprocessing process, including multiple imputation processing of missing values by chain equation, determination of optimal cut point of continuous variable based on ROC curve analysis, processing of classification variable by one-hot encoding, combination of feature selection by recursive feature elimination and hyperparameter optimization by grid search, the prediction model constructed improves the prediction accuracy and greatly improves the accuracy of clinical diagnosis.

[0021] III. Through the multi-dimensional evaluation system of discrimination, calibration and clinical effectiveness, especially the decision curve analysis shows that in the wide risk threshold range of 20%-80%, the net benefit of model-assisted decision is better than that of extreme strategy, which proves that the model not only has statistical significance, but also has practical clinical application value, and can provide doctors with quantitative and reliable treatment decision basis.

[0022] IV. The present application provides a complete technical process from data acquisition, preprocessing, model construction to performance evaluation, each step has clear technical parameters and implementation details, ensuring the reproducibility of the method and the generalizability in different medical institutions, which is conducive to the wide application in clinical practice. BRIEF DESCRIPTION OF DRAWINGS

[0023] Figure 1 The step flowchart schematic diagram of the method for constructing the macrolide non-responsive mycoplasma pneumoniae pneumonia prediction model in an embodiment of the present application is shown. Figure 2 The step flowchart schematic diagram of the original data set constructed in S1 in an embodiment of the present application is shown. Figure 3 The step flowchart schematic diagram of the standardized data set obtained by preprocessing the original data set in S2 in an embodiment of the present application is shown. Figure 4 The step flowchart schematic diagram of the optimal prediction model constructed based on the standardized data set in S3 in an embodiment of the present application is shown. Figure 5 The step flowchart schematic diagram of the macrolide non-responsive mycoplasma pneumoniae pneumonia prediction model obtained by performing performance evaluation on the optimal prediction model in S4 in an embodiment of the present application is shown. DETAILED DESCRIPTION

[0024] The technical solutions of the present application will be described clearly and completely below with reference to the drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0025] In the description of the present invention, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings and are intended solely to facilitate and simplify the description of the present invention. They are not intended to indicate or imply that the devices or components referred to must have, be constructed, or operate in a specific orientation, and therefore should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0026] The present invention provides a method for constructing a prediction model for macrolide-unresponsive Mycoplasma pneumonia. Figure 1 As shown, the method includes the following steps: S1: Obtain clinical data of patients with Mycoplasma pneumoniae pneumonia, screen predictor variables, and construct the original dataset.

[0027] like Figure 2 As shown, in some embodiments, patient medical record data can be extracted from the community-acquired pneumonia database of a medical institution. The patient needs to meet the diagnostic criteria for both community-acquired pneumonia and Mycoplasma pneumoniae infection. Among them, the diagnosis of community-acquired pneumonia requires meeting three conditions: evidence of acute infection, such as fever, hypothermia or abnormal white blood cells; acute respiratory symptoms, such as new cough, dyspnea, shortness of breath, chest pain, etc.; chest X-ray shows new infiltration or consolidation. The diagnosis of Mycoplasma pneumoniae infection requires a positive test for respiratory Mycoplasma pneumoniae DNA or RNA, or a serum Mycoplasma IgM titer ≥1:160.

[0028] In other embodiments, during the data screening process, it is necessary to exclude Mycoplasma pneumoniae carriers, patients with community-acquired pneumonia not infected with Mycoplasma pneumoniae, and patients with underlying diseases such as chronic lung disease, immunodeficiency disease, airway malformations, and cancer to ensure the accuracy and representativeness of the data. The screened predictive variables should cover multiple dimensions, including demographic variables such as age, gender, weight, and body mass index; clinical symptom variables such as the number of days with fever, maximum temperature, cough nature, and the presence of wheezing; physical sign variables such as inspiratory retractions, lung rales, and wheezing; laboratory test variables such as C-reactive protein, procalcitonin, white blood cell count, neutrophil ratio, lactate dehydrogenase, ferritin, albumin, interleukin-6, etc.; and imaging variables such as the presence of large areas of consolidation and the presence of pleural effusion. In addition, if conditions permit, Mycoplasma pneumoniae DNA resistance site data and mixed infection status can also be included.

[0029] S2: Preprocess the original data set, including interpolation of missing data, classification conversion of continuous variables, and coding of categorical variables to obtain a standardized data set Specifically, as shown in Figure 3 After obtaining the original dataset, it needs to be systematically pre-processed to obtain a standardized dataset. In some embodiments, for missing data, a chained equation multiple imputation method is used for processing. This method can generate multiple complete imputed datasets based on the internal relationship of the data, effectively ensuring the integrity and unbiasedness of the data. In actual operation, the mice package of R language can be used to realize this process, usually generating 5 imputed datasets, and the subsequent model training results will be summarized on these datasets.

[0030] The classification conversion of continuous variables is an important step in data preprocessing. In other embodiments, for continuous variables with clinically recognized thresholds, the conversion can be directly based on the clinical thresholds. For example, fever is usually defined as a body temperature higher than 38.5°C, so the continuous variable of body temperature can be directly converted into a binary classification variable of whether to have a fever. For continuous variables without explicit clinical thresholds, the optimal cut-off value needs to be determined through receiver operating characteristic curve analysis. The specific method is to take macrolide-unresponsive mycoplasma pneumonia as the outcome variable, calculate the Youden index corresponding to each possible value of the continuous variable, and the formula for calculating the Youden index is sensitivity plus specificity minus 1. Select the value that maximizes the Youden index as the optimal cut-off value, and then convert the continuous variable into a categorical variable based on the cut-off value. For example, for the C-reactive protein variable, if the optimal cut-off value is determined to be 48.5 mg / L through analysis, it can be converted into a binary classification variable of higher or lower than 48.5 mg / L.

[0031] In other embodiments, the encoding process of categorical variables uses one-hot encoding technology. When a categorical variable has only one option, it is directly converted into a binary variable. When a categorical variable contains multiple options, different processing needs to be performed according to the relationship between the options. If the options are mutually exclusive, they are converted into independent binary variables of the corresponding number; if the options are not mutually exclusive, each option needs to be split into a separate column and converted into a binary variable. For example, for the lung imaging performance variable, if it contains three mutually exclusive options of segmental consolidation, lobar consolidation, and interstitial changes, it needs to be converted into three independent binary variables, each sample having only one of the three variables as 1 and the rest as 0.

[0032] S3: Based on the standardized dataset, a machine learning algorithm is used to construct a prediction model, and the optimal prediction model is obtained through feature selection and parameter optimization.

[0033] Specifically, as shown in Figure 4As shown, after completing data preprocessing, enter the model construction phase. In this phase, the application adopts a rigorous multi-step process, combines advanced feature selection methods and two different types of machine learning models, namely ensemble learning model and deep learning model, and obtains the optimal prediction model through systematic hyperparameter optimization.

[0034] First, feature selection is performed, and recursive feature elimination method is used to screen out the clinical variable combination with the most information quantity for predicting the target. The preprocessed entire feature set is used as the initial feature set, and the random forest algorithm is selected as the base evaluator in the recursive feature elimination process, because the random forest can effectively evaluate the importance of the features, and has good ability to capture nonlinear relationships in the data. The random forest model is trained on the training data using the complete initial feature set. After training, the importance score of each feature is obtained by calculating the average reduction of Gini impurity or permutation importance score, and all features are sorted in descending order of importance. Then one or more features with the lowest importance score are removed from the current feature set, and the remaining feature subset is used to train the model, evaluate and remove the least important features. This process continues to iterate until the number of features remaining in the feature set reaches the preset minimum value, for example, 1 feature. In each iteration process, a feature subset and its corresponding model performance are obtained. 5-fold or 10-fold cross-validation is used to evaluate the performance of the model constructed by each feature subset, and the performance evaluation index is the area under the receiver operating characteristic curve. Finally, the feature subset that makes the cross-validation average AUC value reach the peak is selected as the optimal feature subset.

[0035] Based on the screened optimal feature subset, a random forest prediction model is constructed. Random forest improves the stability and accuracy of the model by constructing multiple decision trees and integrating their prediction results. In specific implementation, first, multiple sample subsets with the same size as the original training set are randomly sampled with replacement from the training data set by bootstrap method. For each sample subset, an independent decision tree is constructed. When splitting at each node of the decision tree, instead of selecting the optimal splitting feature from all features, a subset of features is randomly selected, usually with a size of the square root of the total number of features, and then the optimal feature is selected from the subset for splitting. This double randomness ensures low correlation between the decision trees. For classification tasks, the final prediction result is determined by voting of all decision trees, and the class with the most votes is the prediction output of the model.

[0036] To obtain the best performance of the random forest model, a grid search combined with cross-validation method is used to systematically optimize its key hyperparameters. The hyperparameters to be optimized include the number of decision trees, which can be searched in values such as 100, 300, 500, 800, etc.; the maximum depth of the decision tree, which is searched in values such as 5, 10, 15, 20, or no depth limit; and the minimum number of samples required for node splitting, which is searched in values such as 2, 5, 10, etc. The grid search will traverse all possible combinations of hyperparameters, and use cross-validation to evaluate the performance of each combination, using AUC as the evaluation metric. The final random forest model is constructed using the parameter combination with the best performance.

[0037] To explore deeper nonlinear relationships in the data, the present application also constructs a feedforward neural network model, also known as a multilayer perceptron. The network architecture includes an input layer, a hidden layer, and an output layer. The number of neurons in the input layer is equal to the number of features in the optimal feature subset, and each neuron receives the input value of a feature. For example, if the optimal features are 18, the input layer has 18 neurons. At least one hidden layer is set, usually two fully connected hidden layers to enhance the expression ability of the model. For example, a network structure can be constructed with 32 neurons in the first hidden layer and 16 neurons in the second hidden layer. Each neuron in the hidden layer uses a rectified linear unit as the activation function, which has the form , which can effectively alleviate the gradient vanishing problem and speed up model convergence. The output layer contains 1 neuron, which is used to output the probability of predicting MUMPP, and uses the Sigmoid activation function, which can map any real number output to the interval [0, 1], consistent with the definition of probability.

[0038] The training of the neural network model uses binary cross-entropy as the loss function to measure the difference between the model's predicted probability and the true label. The adaptive moment estimation optimizer is used to update the network's weights and biases. This optimizer combines the advantages of momentum and RMSProp algorithms, and can adaptively adjust the learning rate of each parameter to achieve efficient and stable training. Similarly, a grid search combined with cross-validation method is used to optimize the key hyperparameters of the neural network model, mainly including the learning rate, which is searched in values such as 0.01, 0.001, 0.0001, etc.; and the batch size, which is searched in values such as 16, 32, 64, etc., determining the number of samples used for each weight update.

[0039] ​Through the above steps, the random forest model and the feedforward neural network model after feature selection and parameter optimization are constructed respectively. These two models capture patterns in data from different perspectives. Random forest provides stable prediction by integrating multiple decision trees, while neural network can learn more complex nonlinear relationships in data. In the subsequent evaluation stage, a comprehensive performance comparison will be conducted on these two models, and the model with the best overall performance will be selected as the macrolide non-responsive Mycoplasma pneumoniae pneumonia prediction model.

[0040] S4: performance evaluation of the optimal prediction model, to obtain a macrolide non-responsive Mycoplasma pneumoniae pneumonia prediction model.

[0041] Specifically, as shown in Figure 5 After the model is constructed, a comprehensive performance evaluation is needed. In some embodiments, the discrimination evaluation is achieved by calculating the C-index or the area under the receiver operating characteristic curve, which reflects the ability of the model to distinguish between high-risk and low-risk patients. Calibration evaluation can be performed by Brier score, goodness-of-fit test or calibration curve analysis, which evaluate the consistency of the model's predicted probability with the actual probability. Clinical effectiveness evaluation uses decision curve analysis, which can calculate the net benefit of using the model to assist decision-making compared to the all-treatment or all-no-treatment strategy at different risk thresholds, thereby evaluating the clinical utility of the model.

[0042] In another embodiment, the present application also provides a macrolide non-responsive Mycoplasma pneumoniae pneumonia prediction model, which is constructed by the above construction method The implementation process of the present application is further illustrated by a specific implementation case. From the children's community-acquired pneumonia database of a certain children's hospital, 1200 cases of children's medical record data diagnosed as Mycoplasma pneumoniae pneumonia and hospitalized from January 2020 to December 2024 were retrospectively collected. According to the clinical response of the children after 72 hours of standard treatment with intravenous infusion of macrolide drugs such as azithromycin, they were divided into MUMPP group and non-MUMPP group. Among them, those whose body temperature is still higher than 38.5℃ and clinical symptoms and lung imaging examination have no improvement or aggravation are classified into MUMPP group as positive samples; otherwise, they are classified into non-MUMPP group as negative samples.

[0043] 68 candidate predictors were extracted from the medical records, covering demographic characteristics, clinical symptoms and signs, laboratory test results, imaging findings, and other aspects. The dataset was randomly divided into a training set of 840 cases and a test set of 360 cases in a 7:3 ratio. About 15% of the missing data in the training set was generated into five complete interpolated datasets using the chain equation multiple interpolation method. For continuous variables such as C-reactive protein and lactate dehydrogenase, which have no clear clinical threshold, the optimal cut-off value is determined by ROC curve analysis. For example, the optimal cut-off value of C-reactive protein is determined to be 48.5 mg / L, and the optimal cut-off value of lactate dehydrogenase is determined to be 380 U / L. All categorical variables are processed using one-hot encoding.

[0044] Recursive feature elimination method was used for feature selection, with random forest as the base evaluator, and the performance of different feature subsets was evaluated by 5-fold cross-validation. The results show that when the number of features is 18, the cross-validation AUC of the model reaches a peak of 0.89. The 18 optimal features mainly include the number of days with fever before admission, the highest body temperature, C-reactive protein level, lactate dehydrogenase level, ferritin level, albumin level, whether there is a large consolidation, whether there is pleural effusion, Mycoplasma pneumoniae drug resistance site mutation, etc.

[0045] Based on the selected 18 features, a random forest model and a feedforward neural network model were constructed respectively. The optimal parameters of the random forest model determined by grid search are: the number of trees is 300, the maximum depth is 10, and the minimum number of samples required for node splitting is 5. The architecture of the neural network model is: the input layer has 18 nodes, the first hidden layer has 32 nodes, the second hidden layer has 16 nodes, and the output layer has 1 node; the optimal learning rate is 0.001, and the batch size is 32.

[0046] On the independent test set, the C-index of the neural network model reaches 0.91, with a 95% confidence interval of 0.88-0.94, and the C-index of the random forest model is 0.88. At the optimal probability threshold of 0.42, the accuracy of the neural network model is 89.2%, the sensitivity is 92.5%, and the specificity is 87.9%. The Brier score is 0.11, indicating that the model has good calibration. Decision curve analysis shows that within the risk threshold range of 20% to 80%, the net benefit of using the model to assist decision-making is higher than that of the all-treatment and all-no-treatment strategies, proving that the model has high clinical practical value.

[0047] As a comparison, the traditional Logistic regression model constructed using the same data and features has an AUC of only 0.78, which is significantly lower than the model constructed by the method of the present application, demonstrating the advantages of the present application in handling complex medical data.

[0048] The method of the present application can not only be used for constructing a prediction model, but also the obtained model can be integrated into a hospital information system to provide real-time risk assessment for clinicians. In addition, the method can also be stored in a computer readable storage medium or deployed on a special computer device, facilitating the popularization and application in different medical institutions.

[0049] The above embodiments are only for illustrating the technical concept and characteristics of the present application, and the purpose is to enable those skilled in the art to understand the content of the present application and implement it, and cannot limit the protection scope of the present application. Any equivalent transformation or modification made according to the spirit and essence of the present application shall be covered within the protection scope of the present application.

Claims

1. A method for constructing a prediction model for macrolide-unresponsive Mycoplasma pneumoniae pneumonia, characterized in that: include: S1: Obtain clinical data of patients with Mycoplasma pneumoniae pneumonia, screen predictor variables, and construct the original dataset; S2: Preprocessing the original data set, including interpolation of missing data, categorical conversion of continuous variables, and coding of categorical variables, to obtain a standardized data set; S3: Based on the standardized data set, a prediction model is constructed using a machine learning algorithm, and an optimal prediction model is obtained through feature selection and parameter optimization; S4: Evaluate the performance of the optimal prediction model to obtain a macrolide-unresponsive Mycoplasma pneumonia prediction model.

2. The construction method according to claim 1, characterized in that In step S1, the predictive variables include at least two categories of demographic characteristic variables, clinical symptom variables, laboratory test variables, and imaging variables; the acquisition of clinical data includes: extracting patient data that meets the diagnostic criteria for Mycoplasma pneumonia from a community-acquired pneumonia database, and dividing the patients into a macrolide-unresponsive group and a macrolide-responsive group based on their clinical response after 72 hours of macrolide treatment.

3. The construction method according to claim 2, characterized in that The demographic characteristic variables include at least two of age, gender, weight, and body mass index; the clinical symptom variables include at least two of the number of fever days, maximum body temperature, cough, and wheezing; the laboratory examination variables include at least three of the following: C-reactive protein, procalcitonin, white blood cell count, neutrophil ratio, lactate dehydrogenase, ferritin, albumin, and interleukin-6; and the imaging variables include the presence of large areas of consolidation and / or pleural effusion.

4. The construction method according to claim 1, wherein The missing data interpolation process in step S2 adopts the chain equation multiple interpolation method; The categorical transformation of the continuous variables includes: For continuous variables with clinically recognized thresholds, they were directly converted to categorical variables based on the clinical thresholds; For continuous variables without clear clinical thresholds, the optimal cut-off value was determined by receiver operating characteristic curve analysis, and the variables were converted into categorical variables based on the optimal cut-off value.

5. The construction method according to claim 4, characterized in that The determination of the optimal cut-off value through receiver operating characteristic curve analysis includes: taking macrolide-unresponsive Mycoplasma pneumonia as the outcome variable, calculating the Youden index corresponding to each possible value of the continuous variable, and selecting the value that maximizes the Youden index as the optimal cut-off value; wherein, Youden index = sensitivity + specificity - 1.

6. The construction method according to claim 1, characterized in that The categorical variable encoding process in step S2 adopts one-hot encoding, which specifically includes: When a categorical variable has only one option, it is directly converted into a binary variable; When a categorical variable contains n options and n ≥ 2, if the n options are mutually exclusive, they are converted into n independent binary variables; if the n options are not mutually exclusive, each option is split into a separate column and converted into a binary variable.

7. The construction method according to claim 1, characterized in that The machine learning algorithm in step S3 includes an ensemble learning algorithm and / or a deep learning algorithm; the feature selection adopts a recursive feature elimination method, and the performance of different feature subsets is evaluated through cross-validation to determine the optimal feature subset; the parameter optimization adopts a grid search algorithm to systematically optimize the hyperparameters of the model.

8. The construction method according to claim 7, characterized in that: The ensemble learning algorithm is a random forest algorithm, and the deep learning algorithm is a feedforward neural network algorithm; the hyperparameters of the random forest algorithm include the number of trees, the maximum depth, and the minimum number of samples required for node splitting; the feedforward neural network algorithm includes an input layer, at least one hidden layer, and an output layer, wherein the hidden layer uses a ReLU activation function and the output layer uses a Sigmoid activation function.

9. The construction method according to claim 1, characterized in that: The performance evaluation in step S4 includes at least two of discrimination evaluation, calibration evaluation and clinical effectiveness evaluation; the discrimination evaluation is achieved by calculating the C index or the area under the receiver operating characteristic curve; the calibration evaluation includes at least one of Brier score, goodness of fit test or calibration curve analysis; and the clinical effectiveness evaluation adopts decision curve analysis.

10. A macrolide-unresponsive Mycoplasma pneumonia prediction model, characterized in that: It is constructed by the construction method according to any one of claims 1 to 9.

Images