Multi-index comprehensive scoring patient risk assessment method and system

Through integrated learning and Boosting methods, the multi-index comprehensive scoring model is constructed, which solves the problem of poor subjectivity and accuracy of traditional disease risk assessment methods, and achieves higher prediction accuracy and interpretability, which is suitable for diversified medical data. It is excellent in the prediction of a few types of diseases and missing noise data.

CN119049718AInactive Publication Date: 2024-11-29DEYANG HOSPITAL AFFILIATED HOSPITAL OF CHENGDU UNIV OF TRADITIONAL CHINESE MEDICINE (DEYANG HOSPITAL OF INTEGRATED TRADITIONAL CHINESE & WESTERN MEDICINE)
View PDF 0 Cites 9 Cited by

Patent Information

Application Number
CN202411235444.3
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-11-29
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Traditional disease risk assessment methods have problems such as strong subjectivity, poor accuracy and low efficiency, especially when processing and analyzing large amounts of medical data, and machine learning and deep learning models require a large amount of labeled data and lack of capability. Explanation, limiting its application.

Method used

The multi-index comprehensive scoring model is constructed using integrated learning, multiple basic models are constructed through the Boosting method, the sample weight is gradually adjusted, and the prediction results are finally weighted and averaged. Combined with logistic regression, decision tree and random forest model, cross-entropy loss function and Adam optimizer are used. Carry out training to improve the accuracy and interpretability of the model.

Benefits of technology

Improve the accuracy and interpretability of disease risk prediction, enhance the applicability and generalization of the model, can effectively process complex medical data, significantly improve the accuracy and robustness of prediction, and is suitable for diverse patient data, especially Excellent in the prediction of a few types of diseases and missing noise data.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119049718A_ABST
    Figure CN119049718A_ABST
Patent Text Reader

Abstract

The invention provides a multi-index comprehensive scoring patient risk assessment method and system, belongs to the field of disease risk assessment, and aims to accurately assess the disease risk of a patient. The method comprises the steps of collecting a mapping relation between patient index data and disease risks, performing cleaning, standardization and normalization processing on the data, constructing and training a risk assessment model, and performing result analysis and assessment. The system comprises a data collection module, a data processing module, a model construction and training module, a result analysis module and a visualization module, is helpful for doctors and patients to accurately understand disease risks, and provides a scientific basis for diagnosis and treatment management of the patients. The assessment model constructed through an ensemble learning mode can assess the risk more accurately, help to formulate an individualized treatment scheme, optimize medical resource configuration and improve medical service quality.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of disease risk assessment, and more specifically relates to a patient risk assessment method and system with a multi-index comprehensive scoring. Background Art

[0002] In the field of healthcare, predicting a patient's disease risk is an important task that can help doctors make timely and appropriate treatment decisions and improve patients' quality of life and survival rate. Traditional disease risk assessment methods are mainly based on doctor experience and routine medical examination results, but this method has problems such as strong subjectivity, poor accuracy and low efficiency.

[0003] In recent years, with the rapid growth of medical data and the development of data mining technology, disease risk assessment methods based on data analysis have been widely used. These methods can discover hidden patterns and relationships between data through in-depth analysis of large amounts of medical data, thereby improving the accuracy of disease risk prediction. However, how to effectively process and analyze large amounts of medical data and build accurate and reliable disease risk assessment models remains a challenging problem.

[0004] Multi-index comprehensive scoring is a commonly used disease risk assessment method that can comprehensively consider multiple factors that affect disease risk, thereby improving the accuracy of prediction. However, the traditional multi-index comprehensive scoring method mainly scores by manually setting weights. This method has problems such as weight setting being too subjective and unable to process large amounts of data.

[0005] In addition, with the development of machine learning and deep learning technologies, many researchers have begun to explore the application of these technologies to disease risk assessment. However, these technologies require a large amount of labeled data, and the model results often lack interpretability, which limits their promotion in practical applications. Summary of the invention

[0006] The invention proposes a patient risk assessment method and system with a multi-index comprehensive scoring method. The method constructs a multi-index comprehensive scoring model through integrated learning, which not only improves the prediction accuracy of disease risk, but also improves the interpretability of the results, and has great practical value.

[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0008] The method includes:

[0009] Collect a data set D of mapping relationships between patient indicator data and disease risks; the indicator data includes basic information, physiological indicators, biochemical indicators, clinical diagnosis data, and living habits;

[0010] Preprocessing the mapping relationship data set D, including cleaning, standardization, and normalization;

[0011] The patient risk assessment model with multi-index comprehensive scoring is constructed by ensemble learning, including:

[0012] Use Boosting method to build multiple base models;

[0013] Train a base model on the dataset D, then gradually increase the weight of samples classified by the previous base model, and then train a new base model;

[0014] Repeat this process many times, with each new base model trying to correct the errors of the previous base model. Finally, the prediction results of each base model are weighted averaged to construct a patient risk assessment model.

[0015] Training patient risk assessment models;

[0016] The results of the patient risk assessment model after training are analyzed and evaluated, and the patient risk assessment model is used in actual patient risk assessment.

[0017] In one embodiment, the mapping relationship dataset D comes from an electronic medical record system, a Centers for Disease Control and Prevention, other health research organizations, and includes special research data on specific diseases or specific populations.

[0018] In one embodiment, the final prediction result F(x) of the patient risk assessment model is:

[0019]

[0020] in, is the prediction result of the tth base model; α t is the weight of the t-th base model,

[0021] In one embodiment, training a patient risk assessment model includes:

[0022] Define the loss function: Use the cross entropy loss function, which is defined as follows:

[0023]

[0024] Where N is the total number of samples, y i is the true label of the i-th sample, is the predicted value of the i-th sample;

[0025] Select optimizer: Use Adam optimizer for training;

[0026] Initialize model parameters: Initialize each base model and weight coefficient;

[0027] Perform iterative updates: In each round of training, calculate the loss function and its gradient, and then use the optimizer to update the parameters; after each base model is trained, calculate the error rate error on the training set t , according to the error rate err or t Update the weights of the samples;

[0028] The training of each base model in the Boosting method is carried out serially. After completing the training of a base model, the error rate of the model on the training set is used. The error rate ranges from 0 to 1. When the error rate is less than 0.2, it is used to update the data weight, and then the next base model is trained until the training of all base models is completed, and then the base models are combined according to the weights.

[0029] In one embodiment, the analysis and evaluation of the results includes:

[0030] Use ten-fold cross validation to evaluate the performance of the model on the validation set, and use accuracy, precision, recall, F1 score, and AUC indicators to evaluate the model;

[0031] Collect patient indicator data and input them into the patient risk assessment model for risk assessment.

[0032] In one solution, the base models include a logistic regression model, a decision tree model, and a random forest model. The Boosting method is used to construct multiple base models including:

[0033] 1) Initialize data weights. The data are the patient's index data, including basic information, physiological indicators, biochemical indicators, clinical diagnosis data and living habits. Initially, it is considered that all index data of each patient are equally important in model training, so they are given equal initial weights. Afterwards, according to the model's prediction performance of these data in each round of training, the corresponding weights are adjusted. There are N samples, and each sample is given the same weight at the beginning.

[0034] 2) Use the logistic regression model to train on the weighted data and then make predictions on dataset D;

[0035] 3) Calculate the error E of the logistic regression model. The calculation formula of error E is:

[0036] E = Σ(w(i)*e(i));

[0037] Where e(i) is the number of samples whose predicted results are inconsistent with the actual results. If the prediction is correct, e(i) = 0; if the prediction is wrong, e(i) = 1, and w(i) is the weight of each sample;

[0038] 4) Calculate the base model coefficient α, the calculation formula of α is:

[0039]

[0040] 5) Update the weight of each sample. If the prediction is correct, the weight is updated to w(i)*exp(-α). If the prediction is wrong, the weight is updated to w(i)*exp(α). Then, normalize the weights of all samples so that their sum is 1.

[0041] 6) Use the decision tree model to train on the new weighted data and then make predictions on the original dataset D;

[0042] 7) Calculate the error of the decision tree model and calculate the base model coefficient α;

[0043] 8) Update the weight of each sample and then normalize it;

[0044] 9) Use the random forest model to train on the new weighted data and then make predictions on the original dataset D;

[0045] 10) Calculate the error of the third base model and calculate the model coefficient α;

[0046] 11) The final prediction result is the weighted average of the prediction results of the three models, namely:

[0047]

[0048] In one protocol, the basic information includes the patient's age, sex, weight, height, obesity level, race, and family medical history;

[0049] Physiological indicators include the patient’s blood pressure, heart rate, temperature, respiratory rate, blood oxygen saturation, and pulse;

[0050] Biochemical indicators include blood glucose, blood lipids, cholesterol, serum creatine kinase, serum uric acid, liver function indicators, and kidney function indicators;

[0051] Clinical diagnosis data include current diseases, historical diseases, trauma history, surgical history, and allergy history;

[0052] Lifestyle habits include eating habits, exercise habits, smoking conditions, sleeping conditions, and mental stress conditions.

[0053] In another aspect, a patient risk assessment system with a multi-index comprehensive score is provided, the system being applicable to the method described, and the system comprising:

[0054] Data collection module: used to collect the mapping relationship data set D between patient indicator data and disease risk;

[0055] Data processing module: preprocessing the mapping relationship data set D, including cleaning, standardization, and normalization;

[0056] Model building module: Use ensemble learning to build a patient risk assessment model with multi-index comprehensive scoring to assess the patient's risk; the construction of the patient risk assessment model includes: using the Boosting method to build multiple base models;

[0057] Train a base model on the dataset D, then gradually increase the weights of samples misclassified by the previous model, and then train a new base model;

[0058] Repeat this process many times, with each new base model trying to correct the errors of the previous base model. Finally, the prediction results of each base model are weighted averaged to construct a patient risk assessment model.

[0059] Model training module: training the patient risk assessment model;

[0060] Result analysis module: Analyze and evaluate the results of the trained model, and use the patient risk assessment model in actual patient risk assessment.

[0061] In one embodiment, the system further comprises a visualization module for displaying the results of the predicted risk assessment in the form of a chart or report to facilitate understanding and interpretation by doctors and patients.

[0062] Beneficial effects of the present invention:

[0063] 1. Accuracy: Building a multi-index comprehensive scoring model through ensemble learning can reduce the deviation of the model to a certain extent and improve the predictive accuracy of patient risk assessment.

[0064] 2. Interpretability: Compared with black box models such as deep learning, the integrated learning model used in the present invention has better interpretability, is easy to understand and apply, and helps doctors make more accurate clinical decisions.

[0065] 3. Model universality: The method of the present invention is not limited to specific diseases or populations, has strong applicability and generalizability, and can be widely used in the assessment of patient risks.

[0066] 4. Helps preventive work: Predicting patient risks can not only improve treatment outcomes, but can also be used in disease prevention, intervening in high-risk groups to nip problems in the bud.

[0067] 5. This method can integrate the advantages of each individual model by integrating multiple models. It can comprehensively process and analyze the diverse patient data, which greatly improves the accuracy and robustness of the prediction compared to a single model. At the same time, the Boosting method is used to construct multiple base models, and each new model attempts to correct the errors of the previous model, so that the entire evaluation model has better adaptability to complex samples, significantly improving the accuracy of the prediction. Through the present invention, doctors can be assisted to quickly and accurately assess the risk level of patients, improve the quality and efficiency of medical services, and also enhance the self-protection ability of patients.

[0068] 6. For the prediction of a few types of diseases, or a large amount of missing and noisy data, the prediction accuracy and model stability are significantly enhanced by adjusting weights and introducing randomness.

[0069] Therefore, the present invention can not only handle complex medical scenarios, but also effectively improve the effects of disease diagnosis and prevention, and has great implementation value in the medical field. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1 is a flow chart of the method of the present invention;

[0071] Figure 2 A flow chart of a patient risk assessment model for constructing a multi-index comprehensive score for the present invention;

[0072] Figure 3 A flow chart for training the patient risk assessment model of the present invention;

[0073] Figure 4 This is a block diagram of the patient risk assessment system with multi-index comprehensive scoring according to the present invention. DETAILED DESCRIPTION

[0074] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0075] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those understood by those skilled in the art of the present invention. The terms used in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0076] like Figure 1 As shown, the patient risk assessment method of a multi-index comprehensive score includes:

[0077] S1. Collect the mapping relationship dataset D between patient indicator data and disease risk.

[0078] The mapping relationship dataset D comes from the electronic medical record system, the Centers for Disease Control and Prevention, other health research organizations, and special research data for specific diseases or specific populations.

[0079] Data sets usually come from electronic medical record systems of hospitals or other medical institutions, the Centers for Disease Control and Prevention, or other health research organizations. Data sets may include patients’ personal information (such as age, gender, race, etc.), physiological and biochemical indicators (such as blood pressure, blood sugar, cholesterol, etc.), lifestyle habits (such as smoking, drinking, exercise, etc.), and doctors’ diagnosis results.

[0080] In addition, you can also obtain some special research data sets for specific diseases or specific populations, such as cancer gene research data, elderly health status survey data, etc. These data sets usually come from major medical research institutions, public welfare organizations or disease control centers.

[0081] S2. Preprocess the mapping relationship data set D, including cleaning, standardization, and normalization.

[0082] In the preprocessing stage, data cleaning is first required. Data cleaning mainly includes removing duplicate data, filling missing values, processing noise data, etc. Data set D contains n data objects, each of which has m attributes. Use the following method to remove duplicate data: If there are two data objects d1 and d2 in data set D, and their m attributes are the same, then they can be considered duplicates and one of them can be deleted.

[0083] Secondly, perform data standardization and normalization operations to eliminate the impact of data dimension and data size. Data standardization is to scale the data so that it falls into a small specific range. Common standardization methods include Z-score standardization and minimum-maximum standardization.

[0084] The Z-score standardization formula is: z = (x-μ) / σ, where x is the data to be standardized, μ is the mean of all sample data, and σ is the standard deviation of all sample data.

[0085] Minimum-maximum standardization (also known as normalization) is based on the formula: x' = (x-min) / (max-min), where x is the data to be standardized, min and max are the minimum and maximum values ​​of the data, respectively. This method scales the data to the [0,1] interval.

[0086] After obtaining clean and shaped data, it can be used to build models, train and evaluate.

[0087] S3. Construct a patient risk assessment model with a multi-index comprehensive score to assess the patient's risk.

[0088] The patient risk assessment model described adopts an integrated learning approach to construct the model;

[0089] Use Boosting method to build multiple base models;

[0090] First, a base model is trained on the data set D, and then the weights of samples misclassified by the previous base model are gradually increased, and then a new base model is trained;

[0091] This process is repeated many times, with each new base model trying to correct the errors of the previous base model; finally, the prediction results of each base model are weighted averaged;

[0092] Right now in is the prediction result of the tth base model, α t is the weight of the t-th base model,

[0093] The Boosting method described herein constructs multiple base models including:

[0094] Logistic Regression: Logistic regression is a simple model with easy-to-interpret results. It can provide baseline performance in the evaluation model because its simplicity can reduce the risk of overfitting.

[0095] Decision Tree Model: Decision trees have advantages in dealing with nonlinear, non-monotonic, and high-dimensional interaction problems. At the same time, the results of decision trees are very interpretable, and doctors can understand why a patient's risk score is like this through the branch logic of the decision tree.

[0096] Random Forest Model: Random Forest can further improve model performance and reduce the risk of overfitting by integrating the prediction results of a large number of decision trees.

[0097] like Figure 2 The specific steps are as follows:

[0098] S301, initialize data weights, there are N samples, each sample is given the same weight at the beginning The data are the patient's indicator data, including basic information, physiological indicators, biochemical indicators, clinical diagnosis data and living habits. Initially, it is considered that all indicator data of each patient are equally important in model training, so they are given equal initial weights. Afterwards, the corresponding weights are adjusted according to the model's predictive performance of these data in each round of training.

[0099] S302: Use the logistic regression model to train on the weighted data, and then make predictions on the dataset D.

[0100] S303, calculating the error of the logistic regression model, the error calculation formula is:

[0101] E=Σ(w(i)*e(i))

[0102] Where e(i) is the number of samples whose predicted results are inconsistent with the actual results. If the prediction is correct, e(i) = 0; if the prediction is wrong, e(i) = 1, and w(i) is the weight of each sample;

[0103] S304, calculate the model coefficient α, the calculation formula of α is:

[0104]

[0105] S305. Update the weight of each sample. If the prediction is correct, the weight becomes w(i)*exp(-α). If the prediction is wrong, the weight becomes w(i)*exp(α). Then, normalize the weights of all samples so that their sum is 1.

[0106] S306: Use the decision tree model to train on the new weighted data, and then make predictions on the original data set D.

[0107] S307, calculating the error of the decision tree model and calculating the model coefficient α.

[0108] S308: Update the weight of each sample and then perform normalization.

[0109] S309: Use the random forest model to train on the new weighted data, and then make predictions on the original data set D.

[0110] S310, calculating the error of the third model and calculating the model coefficient α.

[0111] S311. The final prediction result is the weighted average of the prediction results of the three models, namely:

[0112]

[0113] The present invention constructs a risk assessment model by adopting an ensemble learning method, which is to complete the learning task by constructing and combining multiple learners. In this process, the Boosting method is used, which is a method of combining multiple weak learners into a strong learner, and each base model is trying to correct the errors of the previous base model. During the training process, the weight of the sample misclassified by the previous base model will gradually increase, so that the model is more inclined to focus on the samples that were misclassified before in subsequent training, in order to improve the final prediction accuracy of the model.

[0114] In addition, by combining multiple models, the bias and variance of the model can be reduced to a certain extent, and the generalization ability of the model can be improved. Bias refers to the error between the predicted value and the true value, that is, the accuracy of the model, while variance refers to the sensitivity of the model to different training data, that is, the stability of the model. Usually, the accuracy and stability of the model are contradictory. Improving accuracy will reduce stability, and improving stability will reduce accuracy. This is the famous bias-variance tradeoff. Ensemble learning is a method designed to solve this problem. By combining multiple models, it not only maintains the stability of the model, but also improves the accuracy of the model, thus solving the bias-variance tradeoff problem to a certain extent.

[0115] The present invention adopts an integrated learning model so that each base model attempts to correct the errors of the previous base model, and the construction process of each base model is explicit, so the model generation process has strong interpretability. For example, assuming that the base model is a decision tree model, it can be understood as a series of judgments on features, and the judgment process is completely transparent, intuitive and easy to understand.

[0116] In addition, while giving the prediction results, the integrated model can also obtain the contribution of each feature to the results, that is, which features are more important and which features have relatively less impact on the results, which increases the interpretability to a certain extent.

[0117] This easy-to-understand feature is very helpful to medical staff. They can understand which factors have a greater impact on a certain disease based on the feedback from the model, so as to provide targeted treatment. In addition, when doctors explain diseases to patients, models with strong interpretability will be more convincing, allowing patients to better understand the potential risks of the disease, improve their compliance with treatment and the prevention and treatment effects of the disease.

[0118] S4. Train the patient risk assessment model.

[0119] like Figure 3 As shown, the patient risk assessment model is trained including:

[0120] S401. Define loss function: Use cross entropy loss function, which is defined as follows:

[0121]

[0122] Where N is the total number of samples, y i is the true label of the i-th sample, is the predicted value of the ith sample.

[0123] S402, select optimizer: use Adam optimizer for training.

[0124] S403, initializing model parameters: each base model and weighting coefficient needs to be initialized.

[0125] S404, perform iterative update: In each round of training, calculate the loss function and its gradient, and then use the optimizer to update the parameters, parameter α t And other related base model parameters need to be obtained by optimizing the loss function; after each base model is trained, its error rate error on the training set needs to be calculated t The weights of the samples are then updated based on this.

[0126] For the Boosting method used, the training of each base model is carried out in series. After training a base model, according to the error rate of the model on the training set, the error rate ranges from 0 to 1. When the error rate is less than 0.2, it is used to update the data weight. Then the next base model is trained until all base models are trained, and then the base models are combined according to the weights.

[0127] S5. Analyze and evaluate the results of the trained model and use it in actual patient risk assessment.

[0128] The analysis and evaluation of the results include:

[0129] Ten-fold cross validation was used to evaluate the performance of the model on the validation set, and the accuracy, precision, recall, F1 score, and AUC indicators were used to evaluate the model.

[0130] Accuracy: number of correctly predicted samples / total number of samples;

[0131] Precision: the number of samples correctly predicted as positive / the number of all samples predicted as positive;

[0132] Recall rate: the number of samples correctly predicted as positive / the number of samples that are actually positive;

[0133] F1 value: 2 precision recall rate / (precision rate + recall rate);

[0134] AUC value: the area under the ROC curve, the larger the value, the better.

[0135] Through these indicators, we can get the performance of the model in predicting the results. So we can analyze where the model does well and where it needs to be improved.

[0136] Collect patient indicator data, including basic information, physiological indicators, biochemical indicators, clinical diagnosis, and living habits, and input them into the patient risk assessment model for risk assessment.

[0137] The basic information described includes the patient's age, gender, weight, height, obesity level, race, and family medical history;

[0138] Physiological indicators include the patient’s blood pressure, heart rate, temperature, respiratory rate, blood oxygen saturation, and pulse;

[0139] Biochemical indicators include blood glucose, blood lipids, cholesterol, serum creatine kinase, serum uric acid, liver function indicators, and kidney function indicators;

[0140] Clinical diagnosis includes current diseases, historical diseases, trauma history, surgical history, and allergy history;

[0141] Lifestyle habits include eating habits, exercise habits, smoking conditions, sleeping conditions, and mental stress conditions.

[0142] The collected patient indicator data is input into the trained patient risk assessment model, and the patient's disease risk assessment results can be output through the model.

[0143] Embodiment 2:

[0144] like Figure 4 As shown, a patient risk assessment system with a multi-index comprehensive score is applicable to the method described above, and the system comprises:

[0145] Data collection module: used to collect the mapping relationship dataset D between patient indicator data and disease risk: The main function of this module is to collect good, clear and complete patient indicator data and disease risk relationship datasets. This is achieved by connecting and extracting medical data from electronic medical record systems, Centers for Disease Control and Prevention, other health research organizations, and obtaining special research data for specific diseases or specific populations. Mainly use data crawling technology and API calls. You can collect public patient datasets on the Internet by writing crawlers, or you can directly pull data from APIs provided by hospitals, Centers for Disease Control and Prevention, etc.

[0146] Data processing module: Preprocess the mapping relationship dataset D, including cleaning, standardization, and normalization. The collected data cannot be directly used for model training and prediction, and needs to be preprocessed. This module is mainly responsible for the following: data cleaning, eliminating invalid, erroneous, and missing data; data standardization, converting data to the same scale to eliminate the impact of data units and dimensions on the model; data normalization, making the data on the same scale to prevent large numerical features from dominating model training. Python's data processing library Pandas is used for data cleaning, and standardization and normalization can be completed using the functions provided by the Scikit-learn library.

[0147] Model building module: responsible for building a patient risk assessment model with a multi-indicator comprehensive score to assess the patient's risk. The main function of this module is to build a model for risk assessment. The model is built using an integrated learning approach, and multiple base models are built using the Boosting method. There is a dependency between the training of the base models, and the training of the latter base model depends on the training results of the previous base model. Improve the final model performance through iterative training and optimization. Use the integrated learning API in Scikit-learn or use more advanced machine learning platforms such as TensorFlow, PyTorch, etc. to build and train the model.

[0148] Model training module: training the patient risk assessment model, which is a key step in machine learning. During the training process, an optimization algorithm is used to adjust the parameters of the model to minimize the loss function, thereby improving the prediction performance of the model. The present invention uses the Adam optimizer for training. In addition, after each training, the weight of the data is adjusted so that the model pays more attention to samples misclassified by the previous model in subsequent training.

[0149] Result analysis module: Analyze and evaluate the results of the trained model and use it in actual patient risk assessment; this module is mainly responsible for evaluating and analyzing the trained model. It uses methods such as 10-fold cross validation to run the model on the validation set and calculates multiple indicators such as accuracy, precision, recall, F1 score, AUC, etc. to evaluate the performance of the model. Use model evaluation methods such as cross validation in Scikit-learn.

[0150] In specific applications, when new patient data is input into the system, the data collection module will collect these data, which may include (but not limited to): patient age, gender, weight, height, blood pressure, heart rate, blood lipids, blood sugar, existing diseases, etc. The data processing module will process these data, such as cleaning, standardization and normalization.

[0151] The processed data will be input into the trained patient risk assessment model, which will calculate and obtain the risk assessment result. This result is obtained based on the input patient data and the model's calculation of the weights of various indicators, and the model's past learning and experience with the disease or population will provide an assessment of the potential risk.

[0152] This risk assessment result not only gives the patient's risk probability of developing a certain disease, but may also provide corresponding explanations and suggestions, such as the need to change lifestyle habits, increase physical examinations, or implement specific drug treatments.

[0153] Finally, the result analysis module will analyze and evaluate the risk assessment results, including the calculation of evaluation indicators such as the accuracy and sensitivity of the model. Then, the visualization module will present these results in the form of charts and reports for doctors and patients to understand and refer to. Therefore, this patient risk assessment system can assist doctors in formulating disease treatment and prevention strategies, and help patients understand their health status and improve their self-protection awareness.

[0154] Visualization module: responsible for displaying the predicted risk assessment results in the form of charts or reports, which are easy for doctors and patients to understand and interpret. The main function of the visualization module is to present the risk assessment results predicted by the model in the form of charts and reports, so that medical staff and patients can better understand the predicted results of the model and make reasonable decisions. Use Matplotlib, Seaborn and other Python visualization libraries to draw graphics. For more complex interactive visualization results, you can use D3.js, Echarts and other front-end technologies for display.

[0155] Embodiment three:

[0156] First, through the data collection module, a series of patient data can be obtained from the electronic medical record system, the Centers for Disease Control and Prevention, etc., including the patient's age, gender, weight, height, obesity level, family medical history, heart rate, blood pressure and other indicators.

[0157] Assume that the following sample data is obtained:

[0158] Patient A: Male, 45 years old, non-obese, with family history of heart disease, heart rate 80bpm, blood pressure 130 / 80mmHg.

[0159] Patient B: female, 35 years old, obese, no family history of heart disease, heart rate 75bpm, blood pressure 120 / 75mmHg.

[0160] Patient C: Male, 52 years old, non-obese, with family history of heart disease, heart rate 85bpm, blood pressure 140 / 90mmHg.

[0161] Then, the data processing module is used to clean, standardize and normalize the data. For example, blood pressure data can be converted into a standard score (z-score). The standard score of blood pressure can be obtained by subtracting the mean blood pressure of the entire sample from the blood pressure value and then dividing it by the standard deviation. The result follows a standard normal distribution (mean is 0 and standard deviation is 1).

[0162] Then, the model building module is constructed. Here, the logistic regression model, decision tree model and random forest model are used as the base model, and the AdaBoost method is used to combine the models. Each time a new model is trained, the weight of the misclassified samples of the previous model is increased.

[0163] The model training module is used to train the model. The cross entropy is defined as the loss function and the Adam optimizer is used for training.

[0164] After training, the result analysis module is used to analyze the performance of the model. The model can be tested on the validation set using the 10-fold cross validation method to calculate evaluation indicators such as accuracy, precision, recall, F1 score, and AUC value.

[0165] Finally, the system’s visualization module will present the final risk assessment results to doctors and patients in the form of charts or reports. For example, for patient A, it will be shown that he has a certain risk of cardiovascular disease and needs regular checkups or changes in his lifestyle.

[0166] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0167] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-index comprehensive scoring method for patient risk assessment. Features: The method includes: Collect a data set D of mapping relationships between patient indicator data and disease risks; the indicator data includes basic information, physiological indicators, biochemical indicators, clinical diagnosis data, and living habits; Preprocessing the mapping relationship data set D, including cleaning, standardization, and normalization; The patient risk assessment model with multi-index comprehensive scoring is constructed by ensemble learning, including: Use Boosting method to build multiple base models; Train a base model on the dataset D, then gradually increase the weight of samples classified by the previous base model, and then train a new base model; Repeat this process many times, with each new base model trying to correct the errors of the previous base model. Finally, the prediction results of each base model are weighted averaged to construct a patient risk assessment model. Training patient risk assessment models; The results of the patient risk assessment model after training are analyzed and evaluated, and the patient risk assessment model is used in actual patient risk assessment.

2. A patient risk assessment method based on a multi-index comprehensive score according to claim 1, Features: The mapping relationship dataset D comes from the electronic medical record system, the Centers for Disease Control and Prevention, other health research organizations, and includes special research data for specific diseases or specific populations.

3. According to the patient risk assessment method of a multi-index comprehensive score according to claim 1, Features: The final prediction result F(x) of the patient risk assessment model is: in, is the prediction result of the tth base model; α t is the weight of the t-th base model, 4. The patient risk assessment method of a multi-index comprehensive score according to claim 1, Features: Training the patient risk assessment model includes: Define the loss function: Use the cross entropy loss function, which is defined as follows: Where N is the total number of samples, y i is the true label of the i-th sample, is the predicted value of the i-th sample; Select optimizer: Use Adam optimizer for training; Initialize model parameters: Initialize each base model and weight coefficient; Perform iterative updates: In each round of training, calculate the loss function and its gradient, and then use the optimizer to update the parameters; after each base model is trained, calculate the error rate error on the training set t , according to the error rate error t Update the weights of the samples; The training of each base model in the Boosting method is carried out serially. After completing the training of a base model, the error rate of the base model on the training set is used. The error rate ranges from 0 to 1. When the error rate is less than 0.2, it is used to update the data weight, and then the next base model is trained until the training of all base models is completed, and then the base models are combined according to the weights.

5. The patient risk assessment method of a multi-index comprehensive score according to claim 1, Features: The analysis and evaluation of the results include: Use ten-fold cross validation to evaluate the performance of the model on the validation set, and use accuracy, precision, recall, F1 score, and AUC indicators to evaluate the model; Collect patient indicator data and input them into the patient risk assessment model for risk assessment.

6. The patient risk assessment method of a multi-index comprehensive score according to claim 1, Features: The base models include logistic regression model, decision tree model, random forest model, and the Boosting method is used to build multiple base models including: 1) Initialize the weight of the indicator data. At the beginning, each indicator data has the same initial weight. Then adjust the corresponding weight according to the model's prediction performance of these data in each round of training. There are N samples, and each sample is given the same weight at the beginning. 2) Use the logistic regression model to train on the weighted data and then make predictions on dataset D; 3) Calculate the error E of the logistic regression model. The calculation formula of error E is: E = Σ(w(i)*e(i)); Where e(i) is the number of samples whose predicted results are inconsistent with the actual results. If the prediction is correct, e(i) = 0; if the prediction is wrong, e(i) = 1, and w(i) is the weight of each sample; 4) Calculate the base model coefficient α, the calculation formula of α is: 5) Update the weight of each sample. If the prediction is correct, the weight is updated to w(i)*exp(-α). If the prediction is wrong, the weight is updated to w(i)*exp(α). Then, normalize the weights of all samples so that their sum is 1. 6) Use the decision tree model to train on the new weighted data and then make predictions on the original dataset D; 7) Calculate the error of the decision tree model and calculate the base model coefficient α; 8) Update the weight of each sample and then normalize it; 9) Use the random forest model to train on the new weighted data and then make predictions on the original dataset D; 10) Calculate the error of the third base model and calculate the model coefficient α; 11) The final prediction result is the weighted average of the prediction results of the three models, namely:

7. The patient risk assessment method of a multi-index comprehensive score according to claim 5, Features: The basic information described includes the patient's age, gender, weight, height, obesity level, race, and family medical history; Physiological indicators include the patient’s blood pressure, heart rate, temperature, respiratory rate, blood oxygen saturation, and pulse; Biochemical indicators include blood glucose, blood lipids, cholesterol, serum creatine kinase, serum uric acid, liver function indicators, and kidney function indicators; Clinical diagnosis data include current diseases, historical diseases, trauma history, surgical history, and allergy history; Lifestyle habits include eating habits, exercise habits, smoking conditions, sleeping conditions, and mental stress conditions.

8. A patient risk assessment system with a multi-index comprehensive score, the system being applicable to the method according to any one of claims 1 to 7, Features: The system comprises: Data collection module: used to collect the mapping relationship data set D between patient indicator data and disease risk; Data processing module: preprocessing the mapping relationship data set D, including cleaning, standardization, and normalization; Model building module: Use ensemble learning to build a patient risk assessment model with multi-index comprehensive scoring to assess the patient's risk; the construction of the patient risk assessment model includes: using the Boosting method to build multiple base models; Train a base model on the dataset D, then gradually increase the weights of samples misclassified by the previous model, and then train a new base model; Repeat this process many times, with each new base model trying to correct the errors of the previous base model. Finally, the prediction results of each base model are weighted averaged to construct a patient risk assessment model. Model training module: training the patient risk assessment model; Result analysis module: Analyze and evaluate the results of the trained model, and use the patient risk assessment model in actual patient risk assessment.

9. A patient risk assessment system with a multi-index comprehensive score according to claim 8, Features: The system also includes a visualization module for displaying the results of the predicted risk assessment in the form of charts or reports to facilitate understanding and interpretation by doctors and patients.

Citation Information

Cited By

  • Tumor patient medication side effect management method and system based on big data analysis

    CN119446400A

  • Method and system for managing side effects of medication for cancer patients based on big data analysis

    CN119446400B

  • Early myocardial injury monitoring system based on medical internet of things

    CN119581038A

  • Data processing method and system for digestive tract tumors

    CN119811661A

  • Cardiovascular case data processing method and system

    CN119943432A