A personalized prediction system for clinical efficacy of Tanreqing injection

Abstract

The application provides a personalized prediction system for clinical curative effect of Tanreqing injection, comprising: an input module, a curative effect prediction module and an output module. The input module is used for acquiring basic conditions of a patient to be predicted, combined medication information and injection time of the Tanreqing injection; the curative effect prediction module is used for taking C-reactive protein as an outcome index, using a constructed personalized prediction model for clinical curative effect of the Tanreqing injection to predict a probability value of ideal clinical curative effect of the patient to be predicted under different combined medication treatment schemes; and the output module is used for outputting the probability value of ideal clinical curative effect of the patient to be predicted under the different combined medication treatment schemes. The application uses the XGBoost model to construct the personalized prediction model for clinical curative effect of the Tanreqing injection and the prediction system thereof based on 13 screened indexes, so as to quickly, accurately and efficiently determine an optimal personalized treatment scheme of the Tanreqing injection and combined medication thereof.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of machine learning, in particular to a personalized prediction system for clinical efficacy of Tanreqing injection. BACKGROUND

[0002] Tanreqing injection is composed of five medicinal materials, i.e., bear bile powder, goat horn, Huangqi, Lianqiao and Jinyinhua, and has the functions of clearing heat, reducing phlegm and detoxifying. It is used for treating phlegm-heat obstructing lung syndrome of wind-heat lung-heat disease, with symptoms of fever, cough, phlegm, sore throat, thirst, red tongue and yellow fur. It is also used for treating early-stage pneumonia, acute bronchitis, acute exacerbation of chronic bronchitis and upper respiratory tract infection. At present, there is a lack of personalized prediction for the effectiveness of Tanreqing injection, and the related researches include RCT research and retrospective clinical research. SUMMARY

[0003] In order to overcome the above technical defects, the purpose of the present application is to provide a personalized prediction system for clinical efficacy of Tanreqing injection, which comprises:

[0004] An input module is used to acquire the basic conditions, combined medication information and injection time of Tanreqing injection of a patient to be predicted, wherein the basic conditions include age, gender, baseline value of albumin, baseline value of albumin / globulin, baseline value of mean corpuscular hemoglobin concentration, baseline value of red blood cell distribution width coefficient of variation, hemoglobin amount and baseline value of uric acid; and the combined medication information is a combined medication treatment scheme composed of beta-lactam, quinolone, beta-lactamase inhibitor compound preparation and carbapenem antibacterial drug.

[0005] An efficacy prediction module is used to use the constructed personalized prediction model for clinical efficacy of Tanreqing injection to predict the ideal probability value of clinical efficacy of the patient to be predicted under different combined medication treatment schemes by taking C-reactive protein as an outcome indicator.

[0006] An output module is used to output the ideal probability value of clinical efficacy of the patient to be predicted under different combined medication treatment schemes. The combined medication treatment scheme corresponding to the highest probability value is the optimal combined medication treatment scheme for the patient to be predicted.

[0007] Further, the construction method of the personalized prediction model for clinical efficacy of Tanreqing injection comprises the following steps:

[0008] Step S1: Collecting a plurality of inpatient data of using Tanreqing injection and having C-reactive protein information in the HIS databases of a plurality of hospitals, and dividing them into a training set and a validation set.

[0009] Step S2: Feature engineering: (1) Determine the outcome indicator: select C-reactive protein as the evaluation index of treatment outcome, the last check result before discharge is high and the highest value during hospitalization, which is defined as disease progression, and other patients are defined as disease remission; (2) Analyze the factors affecting the efficacy of Tanreqing injection from three aspects: patients themselves, dosage and administration time of Tanreqing injection and combination drug use, including β-lactamase inhibitors, quinolones, β-lactamase inhibitor compound preparations, carbapenem antibacterial drugs;

[0010] Step S3: Screening prediction model: use GBDT, XGBoost, LIghtGBM, Adaboost and Catboost in machine learning model to perform preliminary calculation on the cleaned data, use exhaustive search GridSearch for parameter tuning, and then use Precision, Recall, F1-score, AUC and Accuracy to evaluate the model effect of GBDT, XGBoost, LIghtGBM, Adaboost and Catboost, and select XGBoost model with the best effect;

[0011] Step S4: Feature selection using feature sequence forward selection algorithm based on random forest algorithm, which starts from an empty set and searches each time by adding a feature to the feature subset to make the evaluation function AUC reach the optimal value, and the model can reach local optimality when the number of introduced variables reaches 8;

[0012] Step S5: Train the XGBoost model using the training set, and select white ball ratio, Tanreqing injection administration time, albumin, mean corpuscular hemoglobin concentration, red blood cell distribution width coefficient of variation, hemoglobin, uric acid and patient age as the 8 indicators with higher SHAP values;

[0013] Step S6: On the basis of the 8 indicators, β-lactam, quinolone, β-lactamase inhibitor compound preparation and carbapenem are added, as well as patient gender, and the XGBoost model is retrained based on the 13 indicators to finally build a personalized prediction model for the clinical efficacy of Tanreqing injection.

[0014] Further, in the input module, the baseline value of albumin, the baseline value of albumin / globulin, the baseline value of mean corpuscular hemoglobin concentration, the baseline value of red blood cell distribution width coefficient of variation, hemoglobin and uric acid are divided into normal, high and / or low based on clinical reference values.

[0015] Further, in the input module, the beta-lactam, quinolone, beta-lactamase inhibitor compound preparation, carbapenem antibacterial drugs are divided into two cases of use and non-use.

[0016] Further, in the input module, the age and the injection time of Tanreqing injection are both continuous variables, and the unit of the injection time of Tanreqing injection is day.

[0017] After the above technical scheme is adopted, compared with the prior art, the following beneficial effects are achieved:

[0018] Based on the screened 13 indexes, the XGBoost model is used to construct the clinical efficacy personalized prediction model and prediction system of Tanreqing injection, which can predict the clinical efficacy of Tanreqing injection under different combined medication treatment schemes for specific patients, so as to quickly, accurately and efficiently determine the optimal personalized treatment scheme of Tanreqing injection and its combined medication. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 Fig. 1 is a module structure diagram of the clinical efficacy personalized prediction system of Tanreqing injection of the present application;

[0020] Figure 2 Fig. 5 is a feature screening result diagram using the feature sequence forward selection (SFS, Sequential Forward Selection) algorithm based on the random forest algorithm;

[0021] Figure 3 Fig. 7 is a SHAP diagram obtained after using the XGBoost model for pre-training;

[0022] Figure 4 Fig. 8 is an example diagram of the user interface of the clinical efficacy personalized prediction system of Tanreqing injection of the present application. DETAILED DESCRIPTION

[0023] The advantages of the present application are further described below in combination with the drawings and specific embodiments. Those skilled in the art should understand that the specific description below is illustrative rather than limiting, and should not limit the protection scope of the present application.

[0024] As shown in Figure 1 The present embodiment provides a clinical efficacy personalized prediction system of Tanreqing injection, which comprises an input module, an efficacy prediction module and an output module.

[0025] The input module is used to obtain the basic situation of the patient to be predicted, the information of combined medication and the injection time of Tanreqing injection. The basic situation includes age, gender, albumin baseline value, albumin / globulin baseline value, mean corpuscular hemoglobin concentration baseline value, red blood cell distribution width coefficient of variation baseline value, hemoglobin amount and uric acid baseline value. The combined medication information is a combined medication treatment plan composed of β-lactam, quinolone, β-lactamase inhibitor compound preparation and carbapenem antibacterial drugs. The albumin baseline value, albumin / globulin baseline value, mean corpuscular hemoglobin concentration baseline value, red blood cell distribution width coefficient of variation baseline value, hemoglobin amount and uric acid baseline value are divided into normal, high and / or low based on clinical reference values. β-lactam, quinolone, β-lactamase inhibitor compound preparation and carbapenem antibacterial drugs are divided into two situations: use and non-use. Age and Tanreqing injection injection time are continuous variables, and the unit of Tanreqing injection injection time is day.

[0026] Using C-reactive protein as the outcome indicator, the constructed clinical efficacy personalized prediction model of Tanreqing injection is used to predict the probability value of the ideal clinical efficacy of the patient to be predicted under different combined medication treatment plans.

[0027] The construction method of the clinical efficacy personalized prediction model of Tanreqing injection includes the following steps:

[0028] Step S1: Collecting a plurality of inpatient data of Tanreqing injection and C-reactive protein information in the HIS databases of a plurality of hospitals, and dividing them into a training set and a validation set.

[0029] Specifically, 83785 cases of inpatient case information in the HIS databases of 10 national third-grade class-A hospitals from 2015 to 2023 are collected, 20489 cases of patient data using Tanreqing injection are screened out, 2378 cases of patient data with missing CRP outcome indicators are excluded, and the remaining 18111 cases of inpatient data using Tanreqing injection and C-reactive protein information are constructed into a data set, which is divided into a training set (14489 cases) and a validation set (3622 cases). Each patient data information includes but is not limited to demographic indicators (including but not limited to age and gender, etc.), patient laboratory examination baseline values (including but not limited to albumin baseline value, albumin / globulin baseline value, mean corpuscular hemoglobin concentration baseline value, red blood cell distribution width coefficient of variation baseline value, hemoglobin amount and uric acid baseline value, etc.), combined use of antibacterial drugs (including but not limited to β-lactam, quinolone, β-lactamase inhibitor compound preparation, carbapenem antibacterial drugs, etc.) and Tanreqing injection medication time (unit: day) and the like.

[0030] Step S2: Feature engineering:

[0031] (1) Define outcome indicators: select C-reactive protein as the evaluation index of treatment outcome, and define the highest value of the last examination result before discharge as disease progression (0), and other patients as disease remission (1).

[0032] (2) Analyze the factors affecting the efficacy of Tanreqing injection from three aspects: patients themselves, dosage and administration time of Tanreqing injection, and combination therapy.

[0033] Baseline status of patients: The demographic characteristics of patients include age and gender. The first laboratory examination result after admission is used as an indicator to evaluate the baseline status of patients, and according to the normal value range specified in the clinical guidelines, it is divided into low, normal and high three groups.

[0034] Since most patients in this study use Tanreqing injection for single medication of 20-40 ml (91.8%), and the medication method is qd in 99.7% of the cases. Therefore, we only include the administration time of Tanreqing injection in the analysis index. In terms of combination therapy, we selected the four types of antibacterial drugs with the highest usage frequency in this study to analyze the combination therapy: β-lactamase inhibitor, quinolone, β-lactamase inhibitor compound preparation, and carbapenem antibacterial drugs. The actual use of combination therapy includes one or more of β-lactamase inhibitors, quinolones, β-lactamase inhibitor compound preparations, and carbapenem antibacterial drugs, or several cases where none of the four drugs are used.

[0035] Step S3: Screening prediction model: Use GBDT, XGBoost, LIghtGBM, Adaboost, and Catboost in machine learning model to perform preliminary calculation on the cleaned data. We choose exhaustive search (GridSearch) in parameter adjustment: through loop traversal, select the best parameters. Then use Precision, Recall, F1-score, AUC, and Accuracy to evaluate the model effect of GBDT, XGBoost, LIghtGBM, Adaboost, and Catboost, as shown in Table 1, to select the XGBoost model with the best effect.

[0036] Table 1 Model performance comparison table

[0037] Model Precision Recall F1-score AUC Accuracy GBDT 0.6961 0.5621 0.6224 0.6896 0.7079 XGBoost 0.6946 0.6243 0.6576 0.7096 0.7219 LightGBM 0 0 0 0.5 0.5722 Adaboost 0.6785 0.5347 0.5981 0.6727 0.6926 Catboost 0.6149 0.5781 0.5959 0.6537 0.6646

[0038] By comparing the five indicators in Table 1, it can be seen that the performance of the XGBoost model is the best. Therefore, the XGBoost model will be used for training and building a personalized prediction model for the clinical efficacy of Tanreqing injection in the future.

[0039] Step S4: Feature selection is performed using a feature sequence forward selection (SFS) algorithm based on a random forest algorithm, which starts from an empty set and adds one feature to the feature subset each time to make the evaluation function (AUC: area under the ROC curve) optimal. If the candidate feature subset is not better than the last round of feature subset, stop iteration and take the last round of feature subset as the optimal feature selection result, and the result is as shown in Figure 2 We found that the model can reach a local optimum when the number of introduced variables reaches 8. Since the XGBoost model can process high-dimensional sparse features in a distributed manner, there is no need to consider the multicollinearity problem between variables.

[0040] Step S5: The XGBoost model is trained using the training set, and the SHAP values are screened to obtain the white ball ratio, the use time of Shire Qing injection, albumin, the average red blood cell hemoglobin concentration, the red blood cell distribution width variation coefficient, the hemoglobin amount, the uric acid, and the patient age as the 8 indicators with higher SHAP values, which have higher contribution to the model. The SHAP graph is shown in Figure 3 : Each point in the graph represents a case. The horizontal coordinate represents the SHAP value, which represents the contribution of the variable to the model (the influence on the outcome indicator), and the SHAP value greater than 0 (the point located on the right side) indicates that the variable has a positive contribution to the model, and vice versa. The vertical coordinate is the true value of the variable, and the color is red, which means that the higher the value of the variable, the bluer the color, which means that the lower the true value of the variable. ALB / GLO (white ball ratio), RDW_CV (red blood cell volume variation coefficient), HB (hemoglobin amount): Red points are concentrated on the left side, and blue points are concentrated on the right side, indicating that patients with lower initial values of the above indicators are more likely to have disease remission. ALB (albumin), MCHC (average red blood cell hemoglobin concentration), UA (uric acid): Blue points are concentrated on the left side, indicating that patients with lower values of the above indicators are more likely to have disease progression. Time (Shire Qing injection use time), Age (patient age): Blue points are concentrated on the left side, indicating that patients with shorter use time and lower age are more likely to have disease progression.

[0041] Step S6: On the basis of the 8 indicators, β-lactam, quinolone, β-lactamase inhibitor compound preparation, and carbapenem are added, as well as the patient gender, the XGBoost model is retrained based on the 13 indicators, and a personalized prediction model of the clinical efficacy of Shire Qing injection is finally constructed, and a personalized prediction system of the clinical efficacy of Shire Qing injection is developed based on the personalized prediction model of the clinical efficacy of Shire Qing injection, and the model is visually presented. For example, the user interface of the personalized prediction system of the clinical efficacy of Shire Qing injection is as shown in Figure 4 .

[0042] Output the probability value of the ideal clinical efficacy of the to-be-predicted patient under different combined medication treatment schemes, and the combined medication treatment scheme corresponding to the highest probability value is the optimal combined medication treatment scheme of the to-be-predicted patient.

[0043] Based on the prediction platform in Figure 4 In the input information, the age and the medication time (unit: "day") are continuous variables; the gender is divided into "male" and "female"; the albumin baseline value, the albumin / globulin baseline value, the mean corpuscular hemoglobin concentration baseline value, the red blood cell distribution width coefficient of variation baseline value, the hemoglobin amount and the uric acid baseline value are divided into "normal", "high" and / or "low" cases based on the clinical reference value; each of the beta-lactam, quinolone, beta-lactamase inhibitor compound preparation and carbapenem antibacterial drug is divided into "use" and "non-use" two cases.

[0044] For a specific patient, Figure 4 The basic information (age and gender) and laboratory examination results (albumin baseline value, albumin / globulin baseline value, mean corpuscular hemoglobin concentration baseline value, red blood cell distribution width coefficient of variation baseline value, hemoglobin amount and uric acid baseline value) in are fixed and unchanged, and by continuously changing the specific conditions of combined medication (beta-lactam, quinolone, beta-lactamase inhibitor compound preparation and carbapenem antibacterial drug) and medication time of Tanreqing injection, the prediction value of ideal efficacy under different treatment schemes is output, wherein the combined medication treatment scheme corresponding to the highest probability value is the optimal combined medication treatment scheme of the to-be-predicted patient, thereby realizing the individualized prediction of the clinical efficacy of Tanreqing injection.

[0045] It should be noted that the embodiments of the present application have better implementation, and do not limit the present application in any form, and any skilled person in the art can change or modify the above disclosed technical content into equivalent effective embodiments, as long as it does not deviate from the content of the technical solution of the present application, and any modification or equivalent change and modification of the above embodiments according to the technical essence of the present application, still belongs to the scope of the technical solution of the present application.

Claims

1. A personalized clinical efficacy prediction system for Tanreqing injection, characterized in that, include: The input module is used to acquire the basic information of the patient to be predicted, the information on combined medication, and the injection time of Tanreqing injection. The basic information includes age, gender, baseline albumin level, baseline albumin / globulin ratio, baseline mean corpuscular hemoglobin concentration, baseline coefficient of variation of red blood cell distribution width, hemoglobin level, and baseline uric acid level. The combined medication information refers to a combined treatment regimen consisting of β-lactams, quinolones, β-lactamase inhibitor combination preparations, and carbapenem antibiotics. The efficacy prediction module is used to predict the probability value of ideal clinical efficacy for patients under different combination drug treatment regimens by using C-reactive protein as the outcome indicator and a pre-constructed personalized clinical efficacy prediction model for Tanreqing injection. The output module is used to output the probability value of the ideal clinical efficacy of the patient under different combination drug treatment regimens. The method for constructing the personalized prediction model for the clinical efficacy of Tanreqing injection includes the following steps: Step S1: Collect data from several hospitalized patients who used Tanreqing injection and had C-reactive protein information from multiple hospital HIS databases, and divide them into training set and validation set; Step S2: Feature engineering: (1) Determine the outcome indicators: Select C-reactive protein as the evaluation indicator of treatment outcome. If the last test result before discharge is high and is the highest value during hospitalization, it is defined as disease progression. Other patients are defined as disease remission; (2) Analyze the factors affecting the efficacy of Tanreqing injection from three perspectives: the patient itself, the dosage and administration time of Tanreqing injection, and combined medication. Combined medication includes β-lactams, quinolones, β-lactamase inhibitor compound preparations, and carbapenem antibacterial drugs; Step S3: Selecting Predictive Models: Use GBDT, XGBoost, LightGBM, Adaboost, and Catboost from the machine learning models to perform preliminary calculations on the cleaned data. Use GridSearch for exhaustive search to tune the parameters. Then, use five metrics—Precision, Recall, F1-score, AUC, and Accuracy—to evaluate the performance of GBDT, XGBoost, LightGBM, Adaboost, and Catboost, respectively, and select XGBoost as the best performing model. Step S4: Use the feature sequence forward selection algorithm based on the random forest algorithm to filter features. This algorithm starts searching from the empty set and adds one feature to the feature subset each time, so that the area under the ROC curve of the evaluation function AUC reaches the optimal value. When the number of variables introduced reaches 8, the model can reach a local optimum. Step S5: Train the XGBoost model using the training set, and select eight indicators with high SHAP values ​​based on the following: albumin / globulin ratio, duration of administration of Tanreqing injection, albumin, mean corpuscular hemoglobin concentration, coefficient of variation of red blood cell distribution width, hemoglobin level, uric acid, and patient age. Step S6: Based on the 8 indicators, add β-lactams, quinolones, β-lactamase inhibitor compound preparations, carbapenems, and patient gender. Retrain the XGBoost model based on these 13 indicators to finally construct a personalized prediction model for the clinical efficacy of Tanreqing injection.

2. The personalized clinical efficacy prediction system for Tanreqing Injection as described in claim 1, characterized in that, In the input module, baseline values ​​for albumin, albumin / globulin, mean corpuscular hemoglobin concentration, red blood cell distribution width coefficient of variation, hemoglobin level, and uric acid are classified as normal, high, and / or low based on clinical reference values.

3. The personalized clinical efficacy prediction system for Tanreqing Injection as described in any one of claims 1 to 2, characterized in that, In the input module, β-lactams, quinolones, β-lactamase inhibitor combination preparations, and carbapenems are categorized into two cases: used and not used.

4. The personalized clinical efficacy prediction system for Tanreqing Injection as described in any one of claims 1 to 2, characterized in that, In the input module, age and the injection time of Tanreqing Injection are both continuous variables, and the unit of the injection time of Tanreqing Injection is days.

Images