Method and system for constructing heart failure risk prediction model based on medical big data and machine learning algorithm, and medium

Abstract

The invention provides a construction method and system of a heart failure risk prediction model based on medical big data and a machine learning algorithm, and a medium, and the construction method comprises the steps: screening a medical information database, and obtaining a kidney patient group information data set; preprocessing the data set, dividing a training set, a verification set and a plurality of candidate prediction factors related to the heart failure risk, performing importance sorting, and training through a machine learning algorithm according to an importance sorting result and the training set to obtain a plurality of candidate heart failure risk prediction models; and evaluating and comparing the candidate models according to the verification set, and finally obtaining a heart failure risk prediction model. The heart failure risk prediction model specially designed for the chronic kidney patient group is constructed based on the medical big data and the machine learning algorithm, the heart failure risk prediction accuracy for the kidney patient group can be effectively improved, and the method can be widely applied to the technical field of medical information prediction.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of medical information prediction, and particularly relates to a method and system for constructing a heart failure risk prediction model based on medical big data and a machine learning algorithm, and a medium. BACKGROUND

[0002] In recent years, chronic kidney disease (CKD) has become an important challenge in the field of global public health. Research data shows that the risk of heart failure (HF) in patients with chronic kidney disease is significantly higher than that in the general population, with an incidence of about 17%-21%. Once heart failure occurs, the risk of adverse prognosis and death of the patient will significantly increase. Therefore, developing effective tools to identify high-risk individuals of heart failure in patients with chronic kidney disease to provide opportunities for early intervention and treatment is an important way to improve the health outcomes of patients.

[0003] Although various heart failure risk prediction models have been developed, such as the Framingham heart failure prediction model, the ARIC heart failure prediction model, and the Health ABC heart failure prediction model, most of these models are based on data of the general population. However, the formation and development of heart failure in patients with chronic kidney disease is highly complex and multifactorial. Factors related to chronic kidney disease such as reduced glomerular filtration rate, albuminuria, excessive volume overload, metabolic and electrolyte imbalance, anemia, and malnutrition are closely related to the occurrence and prognosis of heart failure. Existing research has shown that heart failure risk prediction models based on the general population have certain limitations in personalized risk prediction in patients with chronic kidney disease. In addition, most existing prediction models are based on traditional algorithms such as Logistic regression or Cox regression model, and have not fully utilized the potential of machine learning algorithms in heart failure risk prediction in patients with chronic kidney disease. SUMMARY

[0004] The purpose of the present application is to at least partially solve one of the problems in the prior art.

[0005] To this end, one purpose of an embodiment of the present application is to provide a method for constructing a heart failure risk prediction model based on medical big data and a machine learning algorithm, which constructs a heart failure risk prediction model designed specifically for patients with chronic kidney disease based on medical big data and a machine learning algorithm, thereby improving the accuracy of heart failure risk prediction for patients with kidney disease.

[0006] The first technical solution adopted by the present application is:

[0007] A method for constructing a heart failure risk prediction model based on medical big data and a machine learning algorithm, comprising:

[0008] screening a medical information database to obtain a kidney disease patient group information dataset;

[0009] preprocessing the kidney disease patient group information dataset to obtain a training dataset, a validation dataset, and a plurality of candidate predictors related to heart failure risk;

[0010] ranking the importance of each of the candidate predictors, and training a plurality of candidate heart failure risk prediction models through a machine learning algorithm according to the importance ranking results and the training dataset;

[0011] model evaluation and comparison of each of the candidate heart failure risk prediction models according to the validation dataset, and obtaining the heart failure risk prediction model according to the model evaluation and comparison results.

[0012] Further, the screening of the medical information database to obtain the kidney disease patient group information dataset comprises:

[0013] constructing a screening condition according to the characteristic indicators of kidney disease, and screening the medical information database according to the screening condition to obtain the medical health information of the kidney disease patient group;

[0014] constructing the kidney disease patient group information dataset according to the medical health information of the kidney disease patient group.

[0015] Further, the preprocessing of the kidney disease patient group information dataset to obtain the training dataset, the validation dataset, and a plurality of candidate predictors related to heart failure risk comprises:

[0016] imputing missing data of the kidney disease patient group information dataset;

[0017] extracting characteristic variables related to heart failure risk in the kidney disease patient group information dataset, and constructing a plurality of candidate predictors according to the characteristic variables;

[0018] dividing the kidney disease patient group information dataset into the training data and the validation dataset according to a preset proportion.

[0019] Further, the characteristic variables include continuous characteristic variables and discrete characteristic variables, and the constructing of a plurality of candidate predictors according to the characteristic variables comprises:

[0020] standardizing the continuous characteristic variables to obtain continuous candidate factors;

[0021] performing one-hot encoding on the discrete characteristic variables to obtain discrete candidate factors;

[0022] The continuous candidate factor and the discrete candidate factor are taken as the several candidate prediction factors.

[0023] Further, the importance ranking of each candidate prediction factor comprises:

[0024] According to a preset importance evaluation algorithm, an importance index value of each candidate prediction factor is calculated.

[0025] According to the importance index value, the importance ranking of each candidate prediction factor is obtained.

[0026] Further, the several candidate heart failure risk prediction models are trained by a machine learning algorithm according to the importance ranking result and the training data set, comprising:

[0027] According to the importance ranking result, several prediction factors are determined from the candidate prediction factors.

[0028] According to the prediction factors, input variables are determined, and the several candidate heart failure risk prediction models are trained by a machine learning algorithm according to the input variables and the training data set.

[0029] Further, the model evaluation and comparison of each candidate heart failure risk prediction model according to the verification data set comprises:

[0030] The discrimination index and the calibration index of each candidate heart failure risk prediction model on the verification data set are calculated.

[0031] The clinical benefit of each candidate heart failure risk prediction model on the verification data set is calculated by a decision curve analysis method.

[0032] The model evaluation and comparison result is determined according to the discrimination index, the calibration index and the clinical benefit.

[0033] Further, the heart failure risk prediction model is obtained according to the model evaluation and comparison result, comprising:

[0034] According to a preset evaluation rule and the model evaluation and comparison result, a heart failure risk prediction model with optimal performance is determined.

[0035] The heart failure risk prediction model with optimal performance is taken as the heart failure risk prediction model.

[0036] The second technical solution adopted by the present application is:

[0037] A heart failure risk prediction model construction system based on medical big data and a machine learning algorithm comprises:

[0038] a data screening module configured to screen a medical information database to obtain a kidney patient information dataset;

[0039] a data preprocessing module configured to preprocess the kidney patient information dataset to obtain a training dataset, a validation dataset, and a plurality of candidate predictors related to heart failure risk;

[0040] a model training module configured to sort the importance of each of the candidate predictors, and train a plurality of candidate heart failure risk prediction models by using a machine learning algorithm according to the importance sorting result and the training dataset;

[0041] a model evaluation and comparison module configured to perform model evaluation and comparison on each of the candidate heart failure risk prediction models according to the validation dataset, and obtain the heart failure risk prediction model according to the model evaluation and comparison result.

[0042] The third technical solution adopted by the present application is:

[0043] The present application also provides a computer readable storage medium, wherein a processor executable program is stored, and the processor executable program is used to execute the above-mentioned method for constructing a heart failure risk prediction model based on medical big data and a machine learning algorithm when executed by a processor.

[0044] The present application has the following beneficial effects:

[0045] The present application can help to construct a prediction model capable of accurately evaluating the heart failure risk of the chronic kidney disease population by screening the existing medical big data and integrating the multi-dimensional clinical data of the chronic kidney disease population; the most referential prediction factors can be screened out for training the model by sorting the importance of the candidate prediction factors, thereby reducing the number of input features and the difficulty of model training; the deep features and long-term dependence of the training data can be captured by training a plurality of machine learning models by using a plurality of different machine learning algorithms and a plurality of different prediction factor combinations, thereby improving the prediction ability of the model; the model with the best comprehensive performance can be screened out from the plurality of models by evaluating and comparing the plurality of candidate models, thereby ensuring that the final heart failure prediction model has the best prediction ability and clinical benefit; the heart failure prediction model constructed can be put into practical use, the possibility of heart failure of a user can be predicted according to the input variables of the user, the risk evaluation result can be obtained by the user at any time, the threshold of real-time monitoring and early warning can be reduced, and the medical burden can be reduced, thereby providing a more accurate and efficient solution for the health management of chronic kidney disease patients. BRIEF DESCRIPTION OF DRAWINGS

[0046] Figure 1 A step flow chart of a construction method of a heart failure risk prediction model based on medical big data and a machine learning algorithm is provided for an embodiment of the present application.

[0047] Figure 2 A work flow schematic diagram of a construction method of a heart failure risk prediction model based on medical big data and a machine learning algorithm is provided for an embodiment of the present application.

[0048] Figure 3 A research population screening flow chart of a kidney disease big data collaboration network database is provided for an embodiment of the present application.

[0049] Figure 4 A research population screening flow chart of an A biological sample bank is provided for an embodiment of the present application.

[0050] Figure 5 An input variable and output result schematic diagram of a heart failure risk prediction model based on medical big data and a machine learning algorithm in a webpage end is provided for an embodiment of the present application.

[0051] Figure 6 A schematic diagram of a construction system of a heart failure risk prediction model based on medical big data and a machine learning algorithm is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0052] Embodiments of the present application are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numbers represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be understood as a limitation of the present application. For the step numbers in the following embodiments, they are only set for the convenience of explanation and description, and the order between the steps is not limited in any way, and the execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0053] In the description of the present application, the meaning of multiple is two or more, and if the first, the second is described, it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or implicitly indicating the sequence of indicated technical features. In addition, unless otherwise defined, all technical and scientific terms used herein have the same meaning as understood by those skilled in the art.

[0054] The concept of the present application is further described below in conjunction with the background art:

[0055] The existing heart failure prediction model has the following deficiencies:

[0056] (1) Existing heart failure prediction models are mainly based on general population data and cannot accurately assess the risk of heart failure in patients with chronic kidney disease when applied to this population.

[0057] (2) Existing models are mostly based on traditional algorithms such as Logistic regression or Cox regression model, and machine learning algorithms have not been applied to the development of heart failure risk prediction models for chronic kidney disease population.

[0058] In addition, the 2024 KDIGO (Kidney Disease: Improving Global Outcomes) Clinical Practice Guideline for the Evaluation and Management of Chronic Kidney Disease recommends the use of externally validated models in chronic kidney disease population or the combination of estimated glomerular filtration rate (eGFR) and urine albumin-to-creatinine ratio (uACR) for cardiovascular risk prediction in chronic kidney disease patients. However, there is currently a lack of heart failure risk prediction models specifically designed for chronic kidney disease population, and a lack of heart failure risk prediction models that incorporate kidney-related characteristics as key predictors.

[0059] To this end, the present application aims to utilize medical big data and machine learning algorithms to develop a heart failure risk prediction model specifically designed for chronic kidney disease patients, filling the research gap in this field and developing a heart failure risk prediction tool for chronic kidney disease patients. This not only helps to improve the quality of life and prognosis of patients, but also optimizes the allocation of medical resources, reduces medical costs, and promotes the development of chronic kidney disease complication management towards a more precise and efficient model.

[0060] Figure 1 The steps of the method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithms according to the present application are shown in the flowchart below. Figure 1 The present application provides a method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithms, which includes:

[0061] S101, screening the medical information database to obtain a kidney disease population information dataset;

[0062] In some alternative embodiments, the medical information database is screened to obtain a kidney disease population information dataset, which includes:

[0063] S1011, constructing a screening condition based on the characteristics of kidney disease, and screening the medical information database based on the screening condition to obtain the medical and health information of the kidney disease population;

[0064] S1012, constructing a kidney patient population information dataset according to the medical health information of the kidney patient population.

[0065] Specifically, a large amount of medical health information is recorded in the existing medical information database. Taking the kidney disease big data collaboration network database and the A biological sample library as examples, the kidney data system of the former covers the medical health information of 24 medical centers, including demographic, laboratory examination, drug use, medical history, surgery and follow-up data longitudinal information, and the latter recruits men and women aged 40-69 years old in several regions, and provides extensive baseline and follow-up data.

[0066] In order to obtain the medical health information of the kidney patient population meeting the research conditions in such a medical information database, it is necessary to construct a screening condition based on the characteristic indicators of kidney disease, such as in the kidney disease big data collaboration network database, chronic kidney disease patients can be identified by international classification of diseases 10th edition (ICD-10) code N18, and can also be defined as kidney disease population by continuously estimating eGFR <60 mL / min and ACR ≥30 mg / g for more than 90 days. In addition to using kidney disease characteristic indicators as screening conditions, in order to ensure the quality of research data, it is also necessary to exclude abnormal conditions such as age <20 years old, pregnancy, previous kidney replacement therapy, history of heart failure (HF) (ICD-10 I50), lack of baseline serum creatinine, presence of tumor, receiving emergency treatment and surgery, or 30 days of follow-up, etc. Based on the above screening conditions, the medical health information of the kidney patient population meeting the requirements can be finally screened out. On the basis of such medical health information, a kidney patient population information dataset containing baseline data of multiple characteristic dimensions and follow-up data indicating whether the patient has heart failure in subsequent follow-up can be constructed, and this dataset is used as the basis for subsequent data preprocessing and machine learning model training.

[0067] S102, preprocessing the kidney patient population information dataset to obtain a training dataset, a validation dataset, and a plurality of candidate predictors related to heart failure risk;

[0068] In some alternative embodiments, preprocessing the kidney patient population information dataset to obtain a training dataset, a validation dataset, and a plurality of candidate predictors related to heart failure risk includes:

[0069] S1021, imputing missing data in the kidney patient population information dataset;

[0070] S1022, extracting characteristic variables related to heart failure risk in the kidney patient population information dataset, and constructing a plurality of candidate predictors according to the characteristic variables;

[0071] S1023、Divide the kidney patient population information dataset into training data and validation data sets according to a preset proportion.

[0072] In some alternative embodiments, the feature variables include continuous feature variables and discrete feature variables, and a plurality of candidate predictors are constructed according to the feature variables, including:

[0073] A1, standardizing the continuous feature variables to obtain continuous candidate factors;

[0074] A2, performing one-hot encoding on the discrete feature variables to obtain discrete candidate factors;

[0075] A3, taking the continuous candidate factors and the discrete candidate factors as a plurality of candidate predictors.

[0076] Specifically, in the present embodiment, the kidney patient population information dataset needs to be preprocessed to obtain better data quality, implement the division of the training set and the validation set, and extract the features in the dataset as candidate predictors.

[0077] The preprocessing of the data can include conventional data cleaning and data augmentation. For the missing feature data in the kidney patient population information dataset, it needs to be imputed to ensure the uniformity of the feature dimension. The extracted candidate predictors can be demographic data, medical history, drug use information, and laboratory examination indicators related to heart failure risk, etc. For the continuous feature variables of the numerical type, the Z-score standardization can be used to standardize them, and each value is converted by subtracting the mean and dividing by the standard deviation to obtain continuous candidate factors. For the discrete feature variables of the category type, one-hot encoding can be used to process them to obtain discrete candidate factors.

[0078] For the kidney patient population information dataset after preprocessing, it can be divided into a training set and a validation set according to a certain proportion, and the proportion can be set to 8:2. The training set is used as the basis for subsequent machine learning model training, and the validation set is used to evaluate and compare the performance of the model to obtain the optimal model for clinical use.

[0079] S103, importance ranking of each candidate predictor, according to the importance ranking result and the training data set, a plurality of candidate heart failure risk prediction models are trained by machine learning algorithm;

[0080] In some alternative embodiments, the importance ranking of each candidate predictor includes:

[0081] B1, calculating the importance index value of each candidate predictor according to a preset importance evaluation algorithm;

[0082] B2, ranking the importance of each candidate predictor according to the importance index value to obtain an importance ranking result.

[0083] In some alternative embodiments, according to the importance ranking result and the training data set, a plurality of candidate heart failure risk prediction models are trained by a machine learning algorithm, including:

[0084] C1, determining a plurality of candidate predictors as predictors according to the importance ranking result;

[0085] C2, determining an input variable according to the predictors, and training a plurality of candidate heart failure risk prediction models by a machine learning algorithm according to the input variable and the training data set.

[0086] Specifically, in the present embodiment, there can be dozens of features as candidate predictors, in order to save training time, reduce training difficulty, and reduce the number of features that need to be input in actual use of the model, thereby reducing the threshold for clinical use, the number of candidate predictors needs to be reduced, and only part of the data is used for machine model training.

[0087] For different machine learning models, the importance ranking of different candidate predictors can be different, and the importance evaluation algorithm needs to be selected according to the selected machine learning algorithm. The following lists different importance evaluation standards when different machine learning algorithms are selected.

[0088] (1) Random Forest (RF), the feature importance of random forest can be evaluated by calculating the average impurity reduction of each feature in all decision trees. Common impurity indicators include Gini impurity and entropy. For each feature j, in each decision tree, the sum of the impurity reduction brought by the feature in all node splits is calculated, and then the average value of all trees is taken. In addition, the importance of the feature can also be evaluated by permutation importance, that is, by randomly disturbing the values of the feature, the difference in prediction performance of the model on the original data and the disturbed data is calculated;

[0089] (1) Extreme Gradient Boosting (XGBoost), the average gain produced by the feature in all trees where it appears is calculated. The gain is calculated by comparing the change of the objective function before and after the split, which reflects the contribution of the feature to the model performance.

[0090] (3) Decision Tree (DT), the feature importance of decision tree is usually calculated based on the impurity reduction (such as Gini impurity or entropy) when the node is split. The importance of each feature is the sum of the impurity reduction in all split nodes, which reflects the contribution of the feature to the construction of the tree structure.

[0091] (4) Neutural network (nnet), the feature importance of neural network is usually evaluated by the absolute value combination of weights. The features with larger weights contribute more to the prediction of the model. This method is based on the weights from the input layer to the hidden layer in the neural network, and measures the importance of the features by weighted combination.

[0092] (5) Support vector machine (SVM), the nonlinear SVM evaluates the importance of the features by the mapping of the kernel function. The kernel function maps the features to a high-dimensional space, and the importance of the features is evaluated by analyzing the weight distribution in the high-dimensional space.

[0093] After determining the importance ranking of different candidate prediction factors under different machine learning algorithms by different importance evaluation algorithms, some candidate prediction factors can be ranked as final prediction factors and used in subsequent model training.

[0094] In the model training process, the hyperparameter adjustment of the training process can be completed by grid search. In order to obtain the best performance of the simplified model for clinical use, the training strategy of the model can be set to add all prediction factors step by step, and the cumulative value of the area under the receiver operating characteristic curve (AUC) is evaluated at each step to measure the performance of the model trained by the selected prediction factors, and to evaluate the performance improvement trend of the model with the increase of the number of prediction factors, so as to identify the simplest model that maximizes the AUC while maintaining clinical usability.

[0095] S104, model evaluation and comparison of each candidate heart failure risk prediction model according to the validation data set, and obtaining a heart failure risk prediction model according to the model evaluation and comparison result.

[0096] In some alternative embodiments, the model evaluation and comparison of each candidate heart failure risk prediction model according to the validation data set comprises:

[0097] D1, calculating the discrimination index and calibration index of each candidate heart failure risk prediction model on the validation data set;

[0098] D2, calculating the clinical benefit of each candidate heart failure risk prediction model on the validation data set by the decision curve analysis method;

[0099] D3, determining the model evaluation and comparison result according to the discrimination index, the calibration index and the clinical benefit.

[0100] In some alternative embodiments, the heart failure risk prediction model is obtained according to the model evaluation and comparison result, comprising:

[0101] E1, determining a heart failure risk prediction model with optimal performance according to a preset evaluation rule and model evaluation and comparison results;

[0102] E2, taking the heart failure risk prediction model with optimal performance as the heart failure risk prediction model.

[0103] Specifically, in the present embodiment, in order to obtain a model with optimal performance from heart failure risk prediction models adopting different machine learning algorithms and using different prediction factors as the final constructed heart failure risk prediction model, the performance of these models needs to be evaluated and compared.

[0104] In order to evaluate the performance of each model, in the field of medical prediction models, the evaluation indexes of model performance are often set as discrimination, calibration and clinical benefit. The AUC and its 95% confidence interval of the model can be calculated by K-fold cross-validation to obtain the discrimination index of the model; the accuracy, sensitivity, specificity, Brier score, calibration curve slope and intercept of the model can be calculated by bootstrap cross-validation, and the closer the Brier score value is to 0, the closer the slope is to 1, and the closer the intercept is to 0, the better the calibration of the model; in order to evaluate the clinical benefit of the model, the threshold needs to be set by the decision curve analysis method to measure the cost-benefit ratio of implementing intervention on users according to the prediction results of the model, so as to evaluate the clinical practical significance of the model in actual application.

[0105] In order to compare the performance of different models, Delong test can be used to evaluate the difference in AUC values between different models, and the prediction performance of different models can be compared by calculating the net reclassification improvement (NRI) and integrated discrimination improvement index (IDI).

[0106] According to the evaluation and comparison results of the comprehensive model, an optimal heart failure risk prediction model can be obtained from several models, and the model is taken as the final constructed heart failure risk prediction model in the present embodiment.

[0107] In the present embodiment, in order to put the heart failure risk prediction model constructed by the foregoing method into actual use, a user interaction interface can be designed in the form of an online webpage, an applet or a mobile phone application to obtain the data input by the user, which is taken as a prediction factor input into the heart failure risk prediction model to obtain the heart failure risk prediction result of the user and present it to the user interaction interface.

[0108] The present application will be further described below in combination with a specific embodiment.

[0109] Figure 2 A working flow diagram of a construction method of a heart failure risk prediction model based on medical big data and machine learning algorithms provided in the present embodiment is shown in Figure 2The workflow of the embodiment of the present application includes data preprocessing, model construction, model verification, and model comparison. In order to obtain the relevant data of the target research population, in the present embodiment, the kidney disease big data collaboration network database and the A biological sample library are selected as the research cohort, and the reference Figure 3 and Figure 4 The research population screening process of the kidney disease big data collaboration network database is as follows: the target population is chronic kidney disease patients, excluding patients aged <20 years, pregnancy, previous kidney replacement therapy, history of heart failure, lack of baseline serum creatinine, history of tumor, receiving emergency treatment and surgery, or missing within 30 days, and finally 74,049 chronic kidney disease patients from 24 hospitals are obtained. Among these populations, patients from 13 hospitals are randomly divided into a training set (used for developing a model) and a verification set (used for internally verifying the performance of the model), and patients from another 11 hospitals are used as a first cohort for externally verifying the performance of the model. The research population screening process of the A biological sample library database is as follows: the target population is chronic kidney disease patients, excluding patients with previous kidney replacement therapy, history of heart failure, and lack of baseline NT-proBNP, and finally 3,167 chronic kidney disease patients are obtained for a second cohort for externally verifying the performance of the model.

[0110] In terms of data preprocessing, the R language version 4.3.1 is used to analyze the data, the mice package is used to impute missing data, and the demographic data, medical history, drug use information, and laboratory examination indicators of the research population are extracted as candidate predictors. For continuous variables in the candidate predictors, standardization is performed by Z-score standardization, that is, each value is converted by subtracting the mean and dividing by the standard deviation; and for categorical variables, one-hot encoding is used for processing to better identify the factor information of each categorical variable.

[0111] The varImp function of the R language is used to calculate the importance of each candidate predictor under different machine learning algorithms, and among the 73 candidate predictors, the top 15 variables are selected for model training. Five machine learning models are constructed using the train function and the corresponding method parameters in the “caret” package, namely DT (method = “rpart”), RF (method = “ranger”), SVM (method = “svmRadial”), XGBoost (method = “xgbTree”), and NNET (method = “nnet”). Hyperparameter adjustment is performed using grid search in the “caret” package based on 10 rounds of 10-fold cross-validation. In order to simplify the best performance model for clinical use, all features are gradually added, and the cumulative value of the area under the receiver operating characteristic curve (AUC) is evaluated at each step to identify the simplified model that maximizes the AUC while maintaining clinical usability.

[0112] After training several heart failure risk prediction models using different machine learning algorithms, the embodiment calculates the AUC and its 95% confidence interval of each model by 10-fold cross-validation; uses 1,000-fold bootstrap method to calculate its accuracy, sensitivity, specificity, Brier score, calibration curve slope and intercept, the closer the Brier score value is to 0, the closer the slope is to 1, and the closer the intercept is to 0, the better the model calibration. The Delong test is used to evaluate the difference in AUC values between different models. Calculate the net reclassification improvement (NRI) and integrated discrimination improvement index (IDI) to compare the prediction performance of different models.

[0113] Based on the above evaluation and comparison results, the model with the best prediction performance is the XGBoost model. The XGBoost model is simplified, and the simplified XGBoost model is deployed as a web application using R Shiny to design the interface and run the logic, and using the rsconnect package to deploy it to the cloud. Figure 5 The input variables and output results of a heart failure risk prediction model based on medical big data and machine learning algorithms provided by the embodiment of the application in the web page are shown in Figure 5 The input variables corresponding to the predictors of the final model include age (Age), glomerular filtration rate (eGFR), urine microalbumin creatinine ratio (Urine ACR), NT-proBNP, coronary heart disease history (Coronary Heart Disease), atrial fibrillation history (Atrial Fibrillation), mitral valve disease history (Mitral Valve Disease), Beta-blockers drug use, RASi drug use, after the user inputs the variables of the final model in the web page, the application can automatically calculate the probability of new heart failure, and return the output results to the web page, referring to Figure 5 It can be seen that under the input conditions, the final heart failure probability prediction result is 22.691%.

[0114] It can be recognized that the embodiment of the present application constructs a prediction model that can accurately assess the risk of heart failure for chronic kidney disease patients based on medical big data of chronic kidney disease patients. By integrating the multi-dimensional clinical data of patients, including medical history, laboratory test results, imaging data, etc., and using the powerful analysis capability of machine learning algorithms, a heart failure risk prediction model specifically designed for chronic kidney disease patients is developed. This model can capture complex variable relationships, improve the accuracy and reliability of the prediction, and provide strong decision support for clinicians. In addition, machine learning algorithms are applied to the development of a heart failure risk prediction model for chronic kidney disease patients to improve the accuracy and predictive ability of the model. Machine learning algorithms can effectively handle multi-dimensional data, capture long-term dependencies and complex features in the data, and thus achieve accurate prediction of heart failure risk. The application of this technology not only improves the performance of the model, but also provides new ideas and methods for chronic disease management.

[0115] Finally, the heart failure prediction model designed for chronic kidney disease patients is deployed as a practical web link, allowing chronic kidney disease patients and clinicians to assess the risk of heart failure in real time. Through this convenient tool, patients and doctors can access risk assessment results anytime, anywhere, thereby achieving early intervention and personalized treatment, improving patient outcomes, and reducing the burden of medical care. This real-time monitoring and early warning function provides a more accurate and efficient solution for the health management of chronic kidney disease patients.

[0116] Reference Figure 6 The embodiment of the present application provides a heart failure risk prediction construction system, comprising:

[0117] A data screening module is configured to screen a medical information database to obtain a kidney patient information dataset;

[0118] A data preprocessing module is configured to preprocess the kidney patient information dataset to obtain a training dataset, a validation dataset, and a plurality of candidate prediction factors related to heart failure risk;

[0119] A model training module is configured to sort the importance of each candidate prediction factor, and train a plurality of candidate heart failure risk prediction models through machine learning algorithms based on the importance sorting result and the training dataset;

[0120] A model evaluation and comparison module is configured to evaluate and compare each candidate heart failure risk prediction model based on the validation dataset, and obtain a heart failure risk prediction model based on the evaluation and comparison result.

[0121] The content in the method embodiments described above is applicable to the system embodiments, the system embodiments specifically implement the same functions as the method embodiments, and achieve the same beneficial effects as the method embodiments.

[0122] The embodiment of the application further provides a computer readable storage medium, wherein a processor executable program is stored, the processor executable program is used for executing the method for constructing the heart failure risk prediction model based on medical big data and a machine learning algorithm when the processor executable program is executed by a processor.

[0123] The computer readable storage medium of the embodiment of the application can execute the method for constructing the heart failure risk prediction model based on medical big data and a machine learning algorithm provided by the method embodiment of the application, can execute the step of any combination of the method embodiment, has the corresponding functions and beneficial effects of the method.

[0124] The embodiment of the application further discloses a computer program product or a computer program, the computer program product or the computer program includes computer instructions, the computer instructions are stored in a computer readable storage medium. The processor of the device can read the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the device executes the method for constructing the heart failure risk prediction model. Figure 1 The method for constructing the heart failure risk prediction model is shown.

[0125] In some alternative embodiments, the functions / operations mentioned in the block diagram can not occur in the order mentioned in the operation diagram. For example, depending on the functions / operations involved, two blocks shown in succession can actually be executed substantially simultaneously or the blocks mentioned above can be executed in reverse order at times. In addition, the embodiments presented and described in the flowcharts of the application are provided by way of example, with the purpose of providing a more comprehensive understanding of the technology. The disclosed method is not limited to the operations and logical flows presented herein. Alternative embodiments are contemplated, in which the order of various operations is changed and in which sub-operations described as part of a larger operation are independently executed.

[0126] Furthermore, although the present application is described in the context of functional modules, it is to be understood that one or more of the functions and / or features described above can be integrated in a single physical device and / or software module, or one or more functions and / or features can be implemented in separate physical devices or software modules. It will also be appreciated that detailed discussion of the actual implementation of each module is unnecessary to an understanding of the present application. Rather, the actual implementation of the modules, in conjunction with their attributes, functions, and internal relationships, are to be understood within the context of the devices disclosed herein. Thus, those skilled in the art with access to the teachings presented herein will be able to devise suitable implementations of the present application without undue experimentation. It is also to be understood that the particular concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is defined by the appended claims and equivalents thereof.

[0127] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or the part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for making a device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0128] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a list of executable instructions for implementing logic functions, which can be specifically embodied in any computer readable medium for use by an instruction execution system, device or apparatus, such as a computer-based system, a system including a processor, or other system that can fetch and execute instructions from the instruction execution system, device or apparatus, or in conjunction with these instructions execution system, device or apparatus. For the purpose of this specification, "computer readable medium" can be any device that can contain, store, communicate, propagate or transport programs for use by an instruction execution system, device or apparatus, or in conjunction with these instruction execution system, device or apparatus.

[0129] More specific examples (a non-exhaustive list) of the computer-readable medium include the following: an electrical connection (electronic) having one or more wires, a portable computer diskette (magnetic), a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium can also be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, for example, via optical scanning of the paper or other medium, then compiled, interpreted, or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

[0130] It should be understood that aspects of the application can be implemented in hardware, software, firmware or combinations thereof. In the above embodiments, various steps or methods can be implemented in software or firmware which is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any of the following techniques, which are known in the art, and combinations thereof, can be used: discrete logic circuitry having logic gates for implementing logic functions upon data signals, application specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field programmable gate arrays (FPGA), and so forth.

[0131] In the above description of the present application, reference has been made to descriptive terms such as "one embodiment / one example", "another embodiment / another example" or "some embodiments / some examples", etc. It is understood that such terms are not intended to limit the scope of the application to the specific embodiments or examples described. Also, the specific features, structures, materials or characteristics described can be combined in any suitable manner in one or more embodiments or examples.

[0132] While embodiments of the present application have been shown and described, it is to be understood that various modifications, substitutions, combinations and variations of the embodiments can be made by those skilled in the art without departing from the spirit and scope of the present application, which is defined by the appended claims and their equivalents.

[0133] The above is a specific description of the preferred embodiments of the present application, but the present application is not limited to the above-described embodiments, and those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of the present application, and these equivalent modifications or substitutions are included in the scope of the claims of the present application.

Claims

1. A method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithms, characterized in that: include: Screen the medical information database to obtain a dataset of kidney disease population information; Preprocessing the kidney disease population information dataset to obtain a training dataset, a validation dataset, and several candidate predictive factors associated with the risk of heart failure; ranking the candidate predictive factors by importance, and obtaining several candidate heart failure risk prediction models through training using a machine learning algorithm based on the importance ranking results and the training data set; Model evaluation and comparison are performed on each of the candidate heart failure risk prediction models based on the validation data set, and the heart failure risk prediction model is obtained based on the model evaluation and comparison results.

2. The method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithm according to claim 1, characterized in that: The medical information database is screened to obtain a kidney disease population information dataset, including: Constructing screening conditions based on characteristic indicators of kidney disease, screening the medical information database based on the screening conditions to obtain medical health information of people with kidney disease; The kidney disease population information data set is constructed based on the medical health information of the kidney disease population.

3. The method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithm according to claim 1, characterized in that: The kidney disease population information dataset is preprocessed to obtain a training dataset, a validation dataset, and several candidate predictive factors associated with the risk of heart failure, including: Interpolate missing data in the kidney disease population information dataset; Extracting characteristic variables related to the risk of heart failure from the kidney disease population information dataset, and constructing a plurality of candidate predictive factors based on the characteristic variables; The kidney disease population information dataset is divided into the training data and the validation dataset according to a preset ratio.

4. The method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithm according to claim 1, characterized in that: The feature variables include continuous feature variables and discrete feature variables, and constructing a plurality of candidate prediction factors according to the feature variables includes: Standardizing the continuous feature variables to obtain continuous candidate factors; Performing one-hot encoding on the discrete feature variables to obtain discrete candidate factors; The continuous candidate factors and the discrete candidate factors are used as the plurality of candidate prediction factors.

5. The method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithm according to claim 1, characterized in that: The ranking of the importance of each candidate prediction factor includes: Calculate the importance index value of each candidate prediction factor according to a preset importance evaluation algorithm; The candidate prediction factors are ranked by importance according to the importance index value to obtain the importance ranking result.

6. The method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithm according to claim 1, characterized in that: According to the importance ranking result and the training data set, several candidate heart failure risk prediction models are obtained through machine learning algorithm training, including: determining a plurality of candidate prediction factors as prediction factors according to the importance ranking result; Input variables are determined based on the predictive factors, and several candidate heart failure risk prediction models are obtained through training using a machine learning algorithm based on the input variables and the training data set.

7. The method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithm according to claim 1, characterized in that: The performing model evaluation and comparison on each of the candidate heart failure risk prediction models based on the validation data set includes: Calculating the discrimination index and calibration index of each candidate heart failure risk prediction model on the validation dataset; Calculating the clinical benefit of each candidate heart failure risk prediction model on the validation dataset by a decision curve analysis method; The model evaluation and comparison results are determined according to the discrimination index, the calibration index and the clinical benefit.

8. The method for constructing a heart failure risk prediction model based on medical big data and machine learning algorithm according to claim 1, characterized in that: The heart failure risk prediction model is obtained according to the model evaluation and comparison results, including: Determining the heart failure risk prediction model with the best performance based on preset evaluation rules and the model evaluation and comparison results; The heart failure risk prediction model with the best performance is used as the heart failure risk prediction model.

9. A system for constructing a heart failure risk prediction model based on medical big data and machine learning algorithms, characterized in that: include: The data screening module is used to screen the medical information database to obtain the kidney disease population information data set; a data preprocessing module, configured to preprocess the kidney disease population information dataset to obtain a training dataset, a validation dataset, and several candidate predictive factors associated with the risk of heart failure; A model training module is used to rank the importance of each of the candidate predictive factors, and obtain several candidate heart failure risk prediction models through machine learning algorithm training based on the importance ranking results and the training data set; The model evaluation and comparison module is used to perform model evaluation and comparison on each of the candidate heart failure risk prediction models according to the verification data set, and obtain the heart failure risk prediction model according to the model evaluation and comparison results.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program, when executed by the processor, is used to execute a method for constructing a heart failure risk prediction model based on medical big data and a machine learning algorithm as described in any one of claims 1 to 8.

Images