Method for predicting treatment cost of chronic diseases

Abstract

The invention discloses a chronic disease treatment cost prediction method, which comprises the steps of setting a data acquisition rule, and acquiring multi-source chronic disease medical data based on the data acquisition rule; obtaining a standardized medical data set based on the multi-source chronic disease medical data; obtaining an optimized feature set based on expert opinions and the standardized medical data set; optimizing the feature set based on recursive feature elimination and regularization processing to obtain an optimal feature subset; obtaining a final data set based on the optimal feature subset; respectively training the plurality of candidate models by adopting the final data set to obtain an optimal model set; and obtaining a chronic disease treatment cost prediction value based on the optimal model set. According to the method, the chronic disease treatment cost can be accurately predicted, powerful support is provided for medical insurance cost control and medical decision, and the method has important practical value and social benefits.

Description

Technical Field

[0001] The invention belongs to the field of statistical technology, and in particular relates to a method for predicting chronic disease treatment costs. Background Art

[0002] Predicting the cost of chronic disease treatment is a complex technical issue involving multiple links and factors. First, it is necessary to obtain a large amount of medical data of chronic disease patients, including medical history, diagnosis and treatment plans, medication records, examination results, etc. These data usually come from the hospital's electronic medical record system, but the systems of different hospitals may be heterogeneous and incompatible, and data standardization and integration are a major problem. Secondly, after obtaining the data, it is necessary to preprocess and extract features to find out the key factors that are highly correlated with the cost of treatment, such as disease severity, complication type, treatment method selection, etc., which requires professional medical knowledge and data analysis technology. Thirdly, to establish a cost prediction model, it is necessary to comprehensively consider the individual differences of patients and select appropriate algorithms and models. How to balance the complexity and generalization ability of the model to avoid overfitting and underfitting is also a technical difficulty. Finally, after the model is established, it is necessary to verify and tune it, evaluate its prediction performance, and continuously iterate and update it to adapt to changes in medical technology and cost standards. In short, the prediction of chronic disease treatment costs involves multiple disciplines such as medicine, public health, statistics, and computer science. It is a complex and challenging subject that requires overcoming a series of technical difficulties such as data collection, feature selection, and model construction. Summary of the invention

[0003] In order to solve the above technical problems, the present invention proposes a method for predicting chronic disease treatment costs to solve the problems existing in the above prior art.

[0004] To achieve the above object, the present invention provides a method for predicting chronic disease treatment costs, comprising:

[0005] Setting data collection rules, acquiring multi-source chronic disease medical data based on the data collection rules; obtaining a standardized medical data set based on the multi-source chronic disease medical data;

[0006] Obtaining an optimized feature set based on expert opinions and the normalized medical data set; processing the optimized feature set based on recursive feature elimination and regularization to obtain an optimal feature subset;

[0007] Obtain the final data set based on the optimal feature subset;

[0008] Using the final data set to train several candidate models respectively to obtain an optimal model set;

[0009] Obtain the predicted value of chronic disease treatment costs based on the optimal model set.

[0010] Optionally, the process of setting data collection rules includes: respectively designing the data interface of each data collection object and predefining data mapping rules; and obtaining data collection specifications and processes based on data collection attributes.

[0011] Optionally, the process of obtaining a normalized medical dataset includes:

[0012] An association rule mining algorithm is used to analyze multi-source chronic disease medical data to obtain the correlation between chronic diseases and treatment costs; a patient medical portrait is constructed based on the medical history records, diagnosis and treatment plans, medication conditions and examination result attributes in the multi-source chronic disease medical data; and structured supplementary data is obtained using natural language processing technology based on the medical data and electronic medical record attributes in the multi-source chronic disease medical data, wherein the multi-source chronic disease medical data is cleansed before analyzing the correlation.

[0013] Optionally, the process of obtaining the optimal feature subset includes:

[0014] A recursive feature elimination algorithm is adopted to recursively build a model, and a preset number of features that do not meet the requirements are eliminated each time to obtain a first feature subset; a regularization method is adopted to introduce a regularization term in the loss function, and the feature coefficients that do not meet the preset conditions are eliminated by controlling the regularization parameter to obtain a second feature subset; if the first feature subset is consistent with the second feature subset, it is determined to be the optimal feature subset; if not, the first feature subset and the second feature subset are merged to obtain the optimal feature subset through cross-validation.

[0015] Optionally, the process of obtaining the final data set includes:

[0016] Based on the optimal feature subset, historical treatment costs and feature data sets of chronic disease patients are obtained to obtain an initial data set; data cleaning is performed on the initial data set; and the initial data set after data cleaning is screened according to business knowledge and data analysis results to obtain a final data set.

[0017] Optionally, the process of obtaining the optimal model set includes:

[0018] The final data set is divided into training set, validation set and test set in the ratio of 6:2:2; the training set is used to train several candidate models; the hyperparameters of several trained candidate models are optimized based on the validation set and grid search method; the test set is input into the candidate model after hyperparameter optimization to obtain the prediction results, and the linear regression model is used as the benchmark model to evaluate the performance of the candidate model to determine whether it exceeds the prediction error threshold. If so, the candidate model will continue to be optimized.

[0019] Optionally, several candidate models are constructed based on decision trees, support vector machines, and neural networks.

[0020] Optionally, it also includes: acquiring updated multi-source chronic disease medical data as an incremental data set based on preset time intervals and optimal feature subsets; after processing the incremental data set for outliers and missing values, converting the format and inputting it into the optimal model, and calculating the error between the predicted output of the model and the actual cost through forward propagation; based on the calculated error, using the gradient descent algorithm to calculate the gradient direction of the model parameters, and updating the model parameters according to the set learning rate until the prediction error of the model on the incremental data set converges to below a preset threshold or reaches the maximum number of iterations.

[0021] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-mentioned method for predicting chronic disease treatment costs.

[0022] Compared with the prior art, the present invention has the following advantages and technical effects:

[0023] The present invention collects patient medical data from the hospital's electronic medical record system, extracts key features through data standardization and feature engineering, and constructs a cost prediction model. In view of the heterogeneity and dynamic change characteristics of medical data, the present invention designs a cross-hospital data interface and incremental learning strategy to achieve generalized application and continuous optimization of the model. Through multidisciplinary cross-integration, the present invention continuously improves the prediction technology and forms a set of standardized and systematic solutions. This method can accurately predict the cost of chronic disease treatment, provide strong support for medical insurance cost control and medical decision-making, and has important practical value and social benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The drawings constituting a part of the present application are used to provide a further understanding of the present application. The illustrative embodiments and descriptions of the present application are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:

[0025] Figure 1 The figure is a flow chart of a method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0027] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0028] Embodiment 1

[0029] like Figure 1 As shown, this embodiment provides a method for predicting chronic disease treatment costs, including:

[0030] Setting data collection rules, acquiring multi-source chronic disease medical data based on the data collection rules; obtaining a standardized medical data set based on the multi-source chronic disease medical data;

[0031] Furthermore, the process of setting data collection rules includes: respectively designing the data interface of each data collection object and predefining data mapping rules; and obtaining data collection specifications and processes based on data collection attributes.

[0032] Specifically, we obtain the database structure and field information of the electronic medical record systems of different hospitals, design corresponding data extraction interfaces for each hospital database, and obtain the original medical data from the hospital database through the interfaces. We convert the obtained original medical data into a standardized data format using predefined data mapping rules, eliminate the heterogeneity of data formats of different hospitals, and obtain unified standardized medical data.

[0033] For example, the electronic medical record system of Hospital A may use a MySQL database, including a patient basic information table, a diagnosis information table, a medication record table, etc.; while Hospital B may use an Oracle database, with different field naming and data storage methods. For each hospital's database, design a corresponding data extraction interface, such as developing an SQL query script for Hospital A and writing an API interface for Hospital B to ensure that the original medical data can be efficiently obtained from the hospital database. The obtained original medical data is standardized using predefined data mapping rules. Assuming that the diagnosis code of Hospital A uses ICD-9 coding, while Hospital B uses ICD-10 coding, the mapping rules are used to uniformly convert the two into ICD-10 coding to eliminate the heterogeneity of data formats. For example, the "heart disease" code of Hospital A is 410, which is converted to "I21" of Hospital B through mapping rules to ensure data consistency.

[0034] Specifically, according to the data collection attributes, formulate data collection standards and processes to ensure that the acquired medical data is complete, accurate and timely. Standardize the data collection process and clarify the scope, frequency, source and collection method of data collection.

[0035] For example, it is required that outpatient and inpatient medical record data be entered into the system within 24 hours after the patient's visit to ensure the timeliness of the data; set up data entry templates, standardize field filling, and reduce data entry errors. At the same time, establish a data quality monitoring mechanism, regularly check the integrity and accuracy of the data, and correct any problems found in a timely manner. Good data collection specifications can ensure data quality and are the basis for building an effective prediction model.

[0036] Furthermore, the process of obtaining a standardized medical dataset includes:

[0037] An association rule mining algorithm is used to analyze multi-source chronic disease medical data to obtain the correlation between chronic diseases and treatment costs; a patient medical portrait is constructed based on the medical history records, diagnosis and treatment plans, medication conditions and examination result attributes in the multi-source chronic disease medical data; and structured supplementary data is obtained using natural language processing technology based on the medical data and electronic medical record attributes in the multi-source chronic disease medical data, wherein the multi-source chronic disease medical data is cleansed before analyzing the correlation.

[0038] Specifically, based on the attributes of chronic diseases and treatment costs, data mining algorithms are used to extract cost-related features from standard data. Based on the attributes of medical history records, treatment plans, medication records, and examination results, patient medical portraits are constructed to depict changes in patient status at different stages of treatment. Based on the attributes of medical data and electronic medical records, natural language processing technology is used to extract structured information and enrich data dimensions. Based on the attributes of heterogeneous data and data integration, data cleaning and conversion processes are designed to achieve unified representation of data from different sources. Based on the attributes of data collection, data collection specifications and processes are formulated to ensure that the acquired medical data is complete, accurate, and timely.

[0039] Exemplarily, from the attributes of chronic diseases and treatment costs, data mining algorithms are used to extract features related to costs. Association rule mining algorithms, such as the Apriori algorithm, can be used to analyze the correlation between different chronic diseases and treatment costs. For example, the data shows that the cardiovascular drug expenditures of hypertensive patients are significantly higher than the average level, and the insulin and blood glucose monitoring costs of diabetic patients account for a large proportion. At the same time, the treatment costs of diabetic patients with renal complications will increase significantly, which shows that complications are an important factor leading to increased costs. This method can identify key chronic diseases and their complications that have a great impact on costs, and provide important indicators for cost prediction. Through this mining, it is possible to identify which chronic diseases are the main factors leading to high costs and how they affect each other. According to the attributes of medical history records, treatment plans, medication records and examination results, a patient medical portrait is constructed to characterize the changes in patient status at different treatment stages. Taking diabetes as an example, patients may only need oral hypoglycemic drugs to control blood sugar in the early stage. At this time, the patient portrait is mainly manifested as a slight increase in blood sugar, a slightly higher glycated hemoglobin index than the normal range, and a relatively low treatment cost. As the disease progresses, the patient may need insulin injections for treatment, and the patient portrait will be updated to insulin dependence, with large blood sugar fluctuations, and may be accompanied by abnormal test results of complications such as retinopathy, diabetes, and nephropathy. The cost of treatment will also increase significantly due to the increase in drug types and doses. At this stage, the portrait will record the insulin dosage, blood sugar fluctuation frequency, and examination indicators of complications. If serious complications such as diabetic foot occur in the later stage, the patient portrait will add records such as wound treatment and anti-infection treatment, and the cost will increase further. In this way, the development trend of the patient's condition and cost changes can be grasped as a whole. According to the attributes of medical data and electronic medical records, natural language processing technology is used to extract structured information and enrich data dimensions. For example, named entity recognition technology can be used to extract key information such as disease names, drug names, and surgical operations from medical record texts. For example, from the text "The patient has a history of coronary heart disease and has been taking aspirin for a long time. He was recently hospitalized due to worsening chest tightness. The electrocardiogram showed ST segment depression, and coronary angiography was performed", entities such as "coronary heart disease", "aspirin", "chest tightness", "electrocardiogram", "ST segment depression", and "coronary angiography" can be extracted. Further combined with context analysis, it can be determined that "coronary heart disease" is a past medical history, "aspirin" is medication information, and "chest tightness" and "ST segment depression" are symptoms and examination results, thereby converting these unstructured text information into structured data, enriching the patient information dimension, and helping to improve the prediction accuracy of the model. Through natural language processing, key information can be mined from massive medical record texts to provide more features for the model.

[0040] Obtaining an optimized feature set based on expert opinions and the normalized medical data set; processing the optimized feature set based on recursive feature elimination and regularization to obtain an optimal feature subset;

[0041] Furthermore, the process of obtaining the optimal feature subset includes:

[0042] A recursive feature elimination algorithm is adopted to recursively build a model, and a preset number of features that do not meet the requirements are eliminated each time to obtain a first feature subset; a regularization method is adopted to introduce a regularization term in the loss function, and the feature coefficients that do not meet the preset conditions are eliminated by controlling the regularization parameter to obtain a second feature subset; if the first feature subset is consistent with the second feature subset, it is determined to be the optimal feature subset; if not, the first feature subset and the second feature subset are merged to obtain the optimal feature subset through cross-validation.

[0043] For example, the data of a diabetic patient is extracted from the electronic medical record system, including blood sugar level, medication type, frequency of visits, complications, etc. These data constitute the original feature set, which provides a basis for subsequent feature selection and model construction. According to professional medical knowledge, highly relevant key features such as disease severity, complication type, and treatment method selection are extracted from the original features to form an optimized feature set. Taking diabetic patients as an example, the severity of the disease can be judged by the level of glycosylated hemoglobin (HbA1c), and the types of complications include retinopathy, diabetes, nephropathy, etc. The treatment methods include oral medication, insulin injection, etc. These key features can more directly reflect the patient's condition and treatment costs. The recursive feature elimination algorithm is used to recursively build the model, and several least important features are eliminated in each recursive process, and finally the optimal feature subset affecting cost prediction is obtained. Assuming that the initial feature set contains 20 features, the recursive feature elimination algorithm is used to eliminate 5 least important features each time. After several iterations, the 5 most important features such as blood sugar level, HbA1c, diabetes, and nephropathy complications may be retained. The regularization method is used to introduce a regularization term into the loss function. By controlling the regularization parameter, some feature coefficients are eliminated to achieve feature selection and obtain the optimal feature subset. For example, using Lasso regression (L1 regularization), by adjusting the regularization parameter λ, the coefficients of some features become 0, and finally retain features such as blood sugar level, medication type, diabetes, and renal complications. The optimal feature subsets obtained by recursive feature elimination and regularization methods are merged. If the feature subsets selected by the two methods are consistent, they are determined as the final feature subsets; if they are inconsistent, the two subsets are merged, and the optimal feature subset is determined by cross-validation. Assuming that the feature subset selected by recursive feature elimination is {blood sugar level, HbA1c, renal complications}, and the feature subset selected by the regularization method is {blood sugar level, medication type, renal complications}, the two subsets are merged into {blood sugar level, HbA1c, medication type, renal complications}, and the final optimal feature subset is determined by cross-validation.

[0044] Obtain the final data set based on the optimal feature subset;

[0045] Furthermore, the process of obtaining the final data set includes:

[0046] Based on the optimal feature subset, historical treatment costs and feature data sets of chronic disease patients are obtained to obtain an initial data set; data cleaning is performed on the initial data set; and the initial data set after data cleaning is screened according to business knowledge and data analysis results to obtain a final data set.

[0047] Specifically, historical treatment cost data and related feature data of patients with chronic diseases are obtained, and the data are preprocessed, including missing value processing, outlier processing, and data standardization, to obtain a cleaned data set; based on business knowledge and data analysis results, features that meet the requirements for correlation with treatment costs are screened out from the data set, and feature selection algorithms such as recursive feature elimination are used to further optimize the feature subset to obtain the final data set.

[0048] For example, for the prediction of treatment costs for diabetic patients, it is first necessary to collect the patient's age, gender, course of disease, blood sugar level, complications (such as retinopathy, nephropathy, etc.), medication (such as insulin dosage), frequency of visits, and other characteristic data, as well as the corresponding historical treatment cost data. Preprocessing the data is a key step to ensure the accuracy of the model. Missing value processing can be achieved by interpolation (such as mean interpolation, median interpolation) or deleting records with more missing values. Outlier processing can be achieved by box plot analysis to identify and remove outliers. Data standardization normalizes features of different dimensions, such as using the Z-score standardization method to convert feature values ​​into a standard normal distribution with a mean of 0 and a standard deviation of 1. Based on business knowledge and data analysis results, screen out features with a high correlation with treatment costs. For example, through correlation analysis, it is found that the blood sugar level, number of complications, and treatment methods (such as whether to use an insulin pump) of diabetic patients are highly correlated with treatment costs, and these features are constructed into feature subsets. The recursive feature elimination (RFE) method was further used to recursively construct a decision tree model, eliminating several of the least important features each time to determine the data contained in the final data set, such as blood sugar level, number of complications, insulin dosage, etc.

[0049] Using the final data set to train several candidate models respectively to obtain an optimal model set;

[0050] Furthermore, several candidate models are constructed based on decision trees, support vector machines, and neural networks.

[0051] Furthermore, the process of obtaining the optimal model set includes:

[0052] The final data set is divided into training set, validation set and test set in the ratio of 6:2:2; the training set is used to train several candidate models; the hyperparameters of several trained candidate models are optimized based on the validation set and grid search method; the test set is input into the candidate model after hyperparameter optimization to obtain the prediction results, and the linear regression model is used as the benchmark model to evaluate the performance of the candidate model to determine whether it exceeds the prediction error threshold. If so, the candidate model will continue to be optimized.

[0053] Specifically, machine learning algorithms such as decision tree, support vector machine and neural network are used to build prediction models. Multiple candidate models are trained by setting different model hyperparameters, such as the maximum depth of the decision tree, the kernel function type of the support vector machine, the number of hidden layers and neurons of the neural network, etc.; the candidate models are tuned on the validation set using hyperparameter optimization methods such as grid search and random search, and the performance of the model under different hyperparameter combinations is evaluated through cross-validation. The hyperparameter combination with the best generalization ability is selected to obtain the optimal model; the prediction performance of the optimal model is evaluated on the test set, and the prediction accuracy and robustness of the model are evaluated using evaluation indicators such as mean square error and mean absolute error, and the model is compared with the benchmark model to verify its effectiveness.

[0054] Exemplarily, the data set is divided into a training set, a validation set, and a test set, with a common ratio of 6:2:2. The training set is used for model training, the validation set is used for model tuning, and the test set is used to evaluate model performance. For example, out of 1000 data, 600 are used for training, 200 are used for validation, and 200 are used for testing. For the optimal feature set, prediction models are constructed using algorithms such as decision tree, support vector machine (SVM), and neural network. The decision tree model explores the optimal structure by setting different maximum depths (such as 3, 5, and 7); the SVM model optimizes the classification effect by selecting different kernel functions (such as linear kernel and radial basis function); and the neural network model finds the optimal network structure by adjusting the number of hidden layers and the number of neurons (such as 50 neurons in 1 layer and 30 neurons in 2 layers). The candidate model is tuned on the validation set using hyperparameter optimization methods such as grid search and random search. For example, through grid search, the maximum depth and minimum number of sample splits of the decision tree are combined and tried, and the model performance under different combinations is evaluated through cross-validation, and the hyperparameter combination with the smallest mean square error is selected. Random search randomly samples in the hyperparameter space and finds the optimal solution through multiple iterations. The prediction performance of the optimal model is evaluated on the test set, using evaluation indicators such as mean square error (MSE) and mean absolute error (MAE). For example, the MSE of a decision tree model on the test set is 5000 and the MAE is 600, indicating that the model has a small prediction error and good robustness. Compared with the benchmark model (such as simple linear regression), if the MSE and MAE of the new model are significantly lower than the benchmark model, the effectiveness of the model is verified. The trained optimal model is deployed to the production environment, and the characteristic data of new patients is input. The model automatically predicts their treatment costs in a certain period of time in the future.

[0055] Obtain the predicted value of chronic disease treatment costs based on the optimal model set.

[0056] Furthermore, the cost prediction method also includes: obtaining updated multi-source chronic disease medical data as an incremental data set based on a preset time interval and an optimal feature subset; after processing the incremental data set for outliers and missing values, performing format conversion and inputting it into the optimal model, calculating the error between the predicted output of the model and the actual cost through forward propagation; based on the calculated error, using a gradient descent algorithm to calculate the gradient direction of the model parameters, and updating the model parameters according to a set learning rate until the prediction error of the model on the incremental data set converges to below a preset threshold or reaches a maximum number of iterations.

[0057] Specifically, in order to adapt to the dynamic changes in medical technology and cost standards, an incremental learning strategy is adopted to regularly obtain new patient medical data, update and iteratively optimize the cost prediction model online, and fine-tune the parameters of the model through algorithms such as gradient descent, so that it can continuously learn new cost distribution and change patterns, and maintain the timeliness and accuracy of the prediction.

[0058] According to the pre-set time interval, the latest patient medical data and cost data are regularly obtained from the medical system database as incremental data sets. The incremental data sets are preprocessed to clean outliers and missing values, extract feature variables related to cost prediction, and construct data that conforms to the model input format. The preprocessed incremental data are input into the trained cost prediction model, and the error between the model's predicted output and the actual cost is calculated through forward propagation. According to the calculated error, the gradient descent algorithm is used to calculate the gradient direction of the model parameters, and the model parameters such as weights and bias are updated according to the set learning rate. Multiple rounds of iterative optimization are performed until the prediction error of the model on the incremental data set converges to below the preset threshold or reaches the maximum number of iterations. The optimized model parameters are saved and replaced with the original model so that it can adapt to the latest medical technology and cost standards and improve the timeliness of cost prediction. The prediction accuracy of the updated model on new patient data is regularly evaluated. If it is lower than expected, the incremental data is re-obtained to trigger the model's online learning and optimization process again.

[0059] For example, in the process of continuous optimization of the medical cost prediction model, it is crucial to regularly obtain the latest patient medical data and cost data from the medical system database. Assuming that the set time interval is once a month, the patient data and cost records added in the previous month are extracted from the database at the beginning of each month to form an incremental data set. For example, the medical data of 300 patients are added in a certain month, including characteristics such as age, gender, medical history, treatment method, medication, and the corresponding actual treatment costs. Preprocessing the incremental data set is a key step to ensure data quality. First, clean up abnormal values, such as abnormally high treatment costs for a patient, which may be a data entry error and need to be removed or corrected. Secondly, deal with missing values, such as missing medication records for a patient, which can be filled by interpolation or using the mean of similar patients. Finally, extract feature variables related to cost prediction, such as patient age, course length, treatment method, etc., to construct data that conforms to the model input format. The preprocessed incremental data is input into the trained cost prediction model, and the error between the predicted output of the model and the actual cost is calculated by forward propagation. Assume that the actual treatment cost of a patient is 6,000 yuan, the model predicts 5,800 yuan, and the error is 200 yuan. According to the calculated error, the gradient descent algorithm is used to calculate the gradient direction of the model parameters, and the model parameters such as weights and biases are updated according to the set learning rate (such as 0.01). In this way, the model can be gradually adjusted to better fit the new data distribution. Repeat the above steps and perform multiple rounds of iterative optimization until the prediction error of the model on the incremental data set converges to below the preset threshold or reaches the maximum number of iterations (such as 100 times). For example, after 50 iterations, the mean square error of the model on the incremental data set decreases from the initial 500 to below 100, indicating that the model has adapted well to the new data. Save the optimized model parameters and replace the original model so that it can adapt to the latest medical technology and cost standards and improve the timeliness of cost prediction. For example, with the use of new drugs and the promotion of new treatment technologies, the original model may not be able to accurately predict new cost trends. By regularly updating the model parameters, the accuracy of the prediction results can be ensured. Regularly evaluating the prediction accuracy of the updated model on new patient data is an important part of ensuring model performance. Assume that an evaluation is conducted every quarter and it is found that the prediction accuracy of the model on the latest data has dropped from the original 90% to 85%, which is lower than expected. At this point, it is necessary to re-acquire the incremental data and trigger the model's online learning and optimization process again. In this way, the model can continue to learn and adapt to changes in data and maintain a high level of prediction accuracy. This regular update and optimization mechanism can not only improve the model's prediction performance, but also promptly discover and solve problems in the actual application of the model. For example, during a certain period of time, the model's cost prediction error for a certain type of disease was large. By analyzing the incremental data, it was found that the diagnosis and treatment methods of this type of disease had changed, resulting in a change in the cost structure.By updating the model in time, the prediction strategy can be adjusted quickly to improve the accuracy of the prediction. In addition, regular model updates can also help cope with changes in data distribution. Over time, the age structure and disease spectrum of patients may change, causing the original model to no longer be applicable. By regularly introducing new data, the model can continuously learn and adapt to these changes and maintain a high generalization ability. In actual operation, a variety of technical means can be combined to further improve the performance of the model. For example, an ensemble learning method is used to integrate the prediction results of multiple models to improve the robustness of the prediction; a deep learning model is used to mine the deep features in the data and enhance the nonlinear fitting ability of the model. Through the comprehensive application of these methods, an efficient and accurate medical cost prediction model can be constructed to provide strong support for resource planning and cost control of medical institutions. In short, through a series of steps such as regular acquisition of incremental data, preprocessing, model updating and evaluation, the continuous optimization and efficient operation of the medical cost prediction model can be ensured, providing a scientific basis for medical management decisions and improving the quality and efficiency of medical services.

[0060] On the basis of interdisciplinary cross-integration, the prediction model is further optimized by comprehensively considering medical knowledge and statistical methods, improving the model's interpretability and generalization ability, and forming a standardized chronic disease treatment cost prediction scheme. For example, combined with the opinions of medical chronic disease management experts, the weights of certain features in the model are adjusted to make the model's prediction results more in line with clinical reality. At the same time, the bootstrap sampling method is used to enhance the robustness of the model under different data distributions. The standardized scheme is integrated into the medical information system to achieve real-time prediction and monitoring of chronic disease treatment costs, and dynamically adjust medical insurance policies and medical resource allocation according to the prediction results to achieve the goal of medical insurance cost control. For example, a prefecture-level city medical insurance bureau integrates the diabetes treatment cost prediction model into the medical insurance information system to monitor the patient's treatment costs in real time. When abnormal cost growth is found, timely intervention is made to adjust the medication plan or hospitalization days to effectively control cost expenditure. Interdisciplinary cross-integration can further improve the practicality and generalization ability of the model. After integration into the medical information system, the real-time prediction and monitoring functions can detect cost anomalies in a timely manner and help hospitals and medical insurance departments to accurately control costs. Continuous iterative optimization is an important guarantee to ensure the long-term effectiveness of the model. Through the comprehensive application of these steps, the chronic disease treatment cost prediction model can not only provide accurate prediction results, but also provide strong support for the rational allocation of medical resources and the scientific formulation of medical insurance policies, ultimately promoting the improvement of the chronic disease prevention and control system and the improvement of medical service levels.

[0061] This embodiment also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned chronic disease treatment cost prediction method are implemented.

[0062] The above are only preferred specific implementations of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for predicting chronic disease treatment costs, characterized in that: The following steps are involved: Setting data collection rules, and acquiring multi-source chronic disease medical data based on the data collection rules; Obtaining a standardized medical data set based on the multi-source chronic disease medical data; Obtaining an optimized feature set based on expert opinions and the normalized medical data set; Processing the optimized feature set based on recursive feature elimination and regularization to obtain an optimal feature subset; Obtain the final data set based on the optimal feature subset; Using the final data set to train several candidate models respectively to obtain an optimal model set; Obtain the predicted value of chronic disease treatment costs based on the optimal model set.

2. The method for predicting chronic disease treatment costs according to claim 1, characterized in that: The process of setting data collection rules includes: designing the data interface of each data collection object and predefining data mapping rules; obtaining data collection specifications and processes based on data collection attributes.

3. The method for predicting chronic disease treatment costs according to claim 1, characterized in that: The process of obtaining a normalized medical dataset includes: An association rule mining algorithm is used to analyze multi-source chronic disease medical data to obtain the correlation between chronic diseases and treatment costs; a patient medical portrait is constructed based on the medical history records, diagnosis and treatment plans, medication conditions and examination result attributes in the multi-source chronic disease medical data; and structured supplementary data is obtained using natural language processing technology based on the medical data and electronic medical record attributes in the multi-source chronic disease medical data, wherein the multi-source chronic disease medical data is cleansed before analyzing the correlation.

4. The method for predicting chronic disease treatment costs according to claim 1, characterized in that: The process of obtaining the optimal feature subset includes: A recursive feature elimination algorithm is adopted to recursively build a model, and a preset number of features that do not meet the requirements are eliminated each time to obtain a first feature subset; a regularization method is adopted to introduce a regularization term in the loss function, and the feature coefficients that do not meet the preset conditions are eliminated by controlling the regularization parameter to obtain a second feature subset; if the first feature subset is consistent with the second feature subset, it is determined to be the optimal feature subset; if not, the first feature subset and the second feature subset are merged to obtain the optimal feature subset through cross-validation.

5. The method for predicting chronic disease treatment costs according to claim 1, characterized in that: The process of obtaining the final dataset includes: Based on the optimal feature subset, historical treatment costs and feature data sets of chronic disease patients are obtained to obtain an initial data set; data cleaning is performed on the initial data set; and the initial data set after data cleaning is screened according to business knowledge and data analysis results to obtain a final data set.

6. The method for predicting chronic disease treatment costs according to claim 1, characterized in that: The process of obtaining the optimal model set includes: The final data set is divided into training set, validation set and test set in the ratio of 6:2:2; the training set is used to train several candidate models; the hyperparameters of several trained candidate models are optimized based on the validation set and grid search method; the test set is input into the candidate model after hyperparameter optimization to obtain the prediction results, and the linear regression model is used as the benchmark model to evaluate the performance of the candidate model to determine whether it exceeds the prediction error threshold. If so, the candidate model will continue to be optimized.

7. The method for predicting chronic disease treatment costs according to claim 1, characterized in that: Several candidate models are constructed based on decision trees, support vector machines, and neural networks.

8. The method for predicting chronic disease treatment costs according to claim 1, characterized in that: It also includes: obtaining updated multi-source chronic disease medical data as an incremental data set based on a preset time interval and an optimal feature subset; after processing the incremental data set for outliers and missing values, converting the format and inputting it into the optimal model, and calculating the error between the predicted output of the model and the actual cost through forward propagation; according to the calculated error, using a gradient descent algorithm to calculate the gradient direction of the model parameters, and updating the model parameters according to a set learning rate until the prediction error of the model on the incremental data set converges to below a preset threshold or reaches a maximum number of iterations.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of a method for predicting chronic disease treatment costs described in any one of claims 1-8 are implemented.

Images