T1DM patient diabetic nephropathy risk prediction system based on machine learning

Abstract

The invention discloses a machine learning-based diabetic nephropathy risk prediction system for type 1 diabetic patients, and is applied to the technical field of diabetic nephropathy risk prediction. Comprising a data processing module, a feature selection module, a machine learning module, a performance evaluation module and a model interpretation module. The data processing module collects and preprocesses data of a patient, the feature selection module selects optimal features based on a feature selection algorithm, the machine learning module constructs a machine learning model and performs prediction, and the performance evaluation module evaluates the performance of the machine learning model. And the model explanation module is used for explaining contributions of different features to a prediction result. According to the method, the DN risk of the T1DM patient can be accurately evaluated and predicted, the understanding of key features of the model is enhanced, and a clinician is helped to make a better decision when the severity of the disease is evaluated.

Description

Technical Field

[0001] The present invention relates to the technical field of diabetic nephropathy risk prediction, and more particularly to a diabetic nephropathy risk prediction system for T1DM patients based on machine learning. Background Art

[0002] Type 1 diabetes mellitus (T1DM) is a chronic autoimmune disease characterized by the destruction of pancreatic β cells, resulting in lifelong dependence on exogenous insulin. Despite advances in the treatment of diabetes, patients with T1DM are still at risk for serious and common complications, among which diabetic nephropathy (DN) is particularly prominent. DN is the main cause of chronic kidney disease (CKD) and end-stage renal disease (ESKD) in patients with T1DM, seriously affecting the quality of life of patients and resulting in huge medical expenses. Early detection and management of DN is crucial to prevent its progression to ESRD and improve the long-term prognosis of patients with T1DM.

[0003] Proteinuria is an important sign of kidney disease and an early clinical indicator of DN. Higher levels of proteinuria and decreased estimated glomerular filtration rate (eGFR) (<60 mL / min / 1.73 m 2 ) is used to identify and monitor DN. Early proteinuria screening is performed on T1DM patients, especially in adolescence or when they are no older than 10 years old, because the patient may have a history of diabetes for up to five years at this time. Although proteinuria and eGFR are important indicators of renal damage in diabetic patients, there are still limitations. Even under optimal blood sugar and blood pressure control, the incidence of DN in T1DM patients is still very high, reaching 50%, sometimes even exceeding 70%. Therefore, there is an urgent need to develop accurate and complex prediction tools to identify patients at high risk of DN before renal damage occurs. Therefore, providing a diabetic nephropathy risk prediction system for T1DM patients based on machine learning is a problem that technicians in this field urgently need to solve. Summary of the invention

[0004] In view of this, the present invention provides a diabetic nephropathy risk prediction system for T1DM patients based on machine learning, and establishes and verifies an explanatory machine learning (ML) model for predicting DN complications in T1DM patients.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A diabetic nephropathy risk prediction system for T1DM patients based on machine learning, comprising a data processing module, a feature selection module, a machine learning module, a performance evaluation module and a model interpretation module; wherein the data processing module collects patient data and performs preprocessing, the feature selection module selects the best features in the patient data based on a feature selection algorithm, the machine learning module constructs different machine learning models and performs predictions, the performance evaluation module evaluates the performance of the machine learning model, the model interpretation module is used to explain the contribution of different features to the prediction results, and the statistical analysis module performs survival analysis on the prediction results to estimate the cumulative incidence of DN.

[0007] Optionally, the data processing module collects patient data and performs preprocessing, specifically: collect characteristic data of T1DM patients, including gender, age, T1DM duration, age of onset, family history, body mass index, drinking history, smoking history, diastolic blood pressure, systolic blood pressure, fasting blood glucose, postprandial blood glucose, fasting C-peptide, postprandial C-peptide, glycosylated hemoglobin, autoantibodies related to T1DM, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, alanine aminotransferase, aspartate aminotransferase, total protein, serum albumin, total bilirubin, direct bilirubin, thyroid function, serum creatinine, acute and chronic diabetic complications and urine albumin-creatinine ratio; all patients are divided into data set 1 and data set 2 according to the hospital region, data set 1 is divided into training set and test set in a ratio of 7:3 using stratified random sampling, and data set 2 is retained for external validation; features with a missing ratio of more than 30% are eliminated, and for the remaining features, the weighted K nearest neighbor method is used to fill in the missing data, and the minimum-maximum normalization method is applied to standardize the data; the training set, test set and external validation data are processed separately.

[0008] Optionally, the feature selection module uses a recursive feature elimination algorithm of a random forest and a five-fold cross validation algorithm to calculate the AUC values ​​of different features, and selects a feature set with the largest AUC and the least number of features as the best feature.

[0009] Optionally, the machine learning models built by the machine learning module include logistic regression model, Gaussian process classification model, gradient boosting model, extreme gradient boosting model, random forest model, adaptive boosting model, decision tree model, Gaussian naive Bayes model, K nearest neighbor model, support vector machine, multilayer perceptron and neural network.

[0010] Optionally, the performance evaluation module draws an ROC curve based on the prediction results of the machine learning model, selects 5 machine learning models for further evaluation based on the area under the ROC curve, calculates the accuracy, specificity and sensitivity of the 5 machine learning models in the training set and the test set in the further evaluation to select the best performing machine learning model, and performs external validation on the 5 machine learning models through data set 2 to evaluate their generalization ability; the prediction accuracy of the training set, the test set and data set 2 is compared by determining the 95% confidence interval through bootstrap resampling.

[0011] Optionally, the model interpretation module is used to explain the contribution of different features to the prediction results, specifically: the SHAP value of each feature is calculated using the SHapley addition interpretation method to obtain the importance ranking of the features, identify important risk factors associated with diabetic nephropathy in T1DM patients, and determine the main predictive factors for diabetic nephropathy in the T1DM patient population.

[0012] It can be seen from the above technical solution that compared with the prior art, the present invention provides a diabetic nephropathy risk prediction system for T1DM patients based on machine learning, which has the following beneficial effects: the prediction tool based on machine learning of the present invention can accurately evaluate and predict the DN risk of T1DM patients, and by integrating ML and SHAP methods, better risk prediction for different individuals is achieved, and the understanding of the key features of the model is enhanced, which can help clinicians make better decisions when assessing the severity of the disease. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0014] Figure 1 A schematic diagram of a diabetic nephropathy risk prediction system for T1DM patients of the present invention;

[0015] Figure 2 A schematic diagram of the evaluation results of the machine learning model in an embodiment of the present invention;

[0016] Figure 3 This is a schematic diagram of the ranking of feature SHAP values ​​in an embodiment of the present invention;

[0017] Figure 4 Schematic diagram of feature importance analysis results in an embodiment of the present invention;

[0018] Figure 5This is a schematic diagram of the model explanatory power of T1DM patients without DN complications in an embodiment of the present invention;

[0019] Figure 6 Schematic diagram of the explanatory power of the model for T1DM patients with DN complications in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0021] The embodiment of the present invention discloses a diabetic nephropathy risk prediction system for T1DM patients based on machine learning, such as Figure 1 As shown, it includes a data processing module, a feature selection module, a machine learning module, a performance evaluation module and a model interpretation module; wherein the data processing module collects and preprocesses patient data, the feature selection module selects the best features in the patient data based on a feature selection algorithm, the machine learning module constructs different machine learning models and performs predictions, the performance evaluation module evaluates the performance of the machine learning model, the model interpretation module is used to explain the contribution of different features to the prediction results, and the statistical analysis module performs survival analysis on the prediction results to estimate the cumulative incidence of DN.

[0022] Furthermore, the data processing module collects the patient's data and performs preprocessing, specifically: collecting characteristic data of T1DM patients, including gender, age, T1DM duration, age of onset, family history, body mass index (BMI), drinking history, smoking history, diastolic blood pressure (DBP), systolic blood pressure (SBP), fasting blood glucose (FPG), postprandial blood glucose (PPG), fasting C-peptide (FCP), postprandial C-peptide (PCP), glycosylated hemoglobin (HbA1c), autoantibodies related to T1D, total cholesterol (TC), triglycerides (TG), high-density lipoprotein cholesterol (HDL-C), low-density lipoprotein cholesterol (LDL-C), alanine aminotransferase (ALT), aspartate aminotransferase (AST), total protein (TP), serum albumin (ALB), total bilirubin (TBIL), direct bilirubin (DBIL), thyroid function (such as TSH), serum creatinine (Scr), acute and chronic diabetic complications, and urine albumin-creatinine ratio (UACR); all patients were divided into dataset 1 and dataset 2 according to the hospital region. Dataset 1 was divided into training set and test set in a ratio of 7:3 using stratified random sampling, and dataset 2 was retained for external validation; features with a missing ratio of more than 30% were removed, and for the remaining features, the weighted K nearest neighbor method was used to fill the missing data, and the minimum-maximum normalization method was applied for data standardization; the training set, test set, and external validation data were processed separately. The feature data of T1DM patients included 34 clinical variables, namely "features" in machine learning. There was no obvious collinearity between all variables, and the missing data ratio of all features after screening was less than 30%. The test set and validation set data used the same filling and standardization methods, but were processed separately to ensure the fairness of model evaluation.

[0023] In the present embodiment, the collected data included T1DM patients of all ages, regardless of whether they had diabetic complications. The exclusion criteria were as follows: (1) patients with pregnancy, late-onset diabetes in adults (LADA), juvenile diabetes, urinary tract infection, major surgery within the last six months, or non-diabetes-related renal damage; (2) patients with more than 30% missing data.

[0024] Furthermore, the feature selection module uses the recursive feature elimination algorithm of random forest and the five-fold cross validation algorithm to calculate the AUC values ​​of different features, and selects the feature set with the largest AUC and the least number of features as the best feature.

[0025] Furthermore, the machine learning models constructed by the machine learning module include logistic regression model, Gaussian process classification model, gradient boosting model, extreme gradient boosting model, random forest model, adaptive boosting model, decision tree model, Gaussian naive Bayes model, K nearest neighbor model, support vector machine, multilayer perceptron and neural network.

[0026] Furthermore, the performance evaluation module draws the ROC curve based on the prediction results of the machine learning model, selects 5 machine learning models for further evaluation based on the area under the ROC curve, calculates the accuracy, specificity and sensitivity of the 5 machine learning models in the training set and the test set in the further evaluation to select the best performing machine learning model, and performs external validation on the 5 machine learning models through data set 2 to evaluate their generalization ability; the prediction accuracy of the training set, test set and data set 2 is compared by determining the 95% confidence interval through bootstrap resampling.

[0027] Furthermore, the model explanation module is used to explain the contribution of different features to the prediction results. Specifically, the SHapley additive explanation method is used to calculate the SHAP value of each feature, obtain the importance ranking of the features, identify the important risk factors associated with diabetic nephropathy in T1DM patients, and determine the main predictors of diabetic nephropathy in the T1DM patient population. The SHapley additive explanation (SHAP) method provides the model with theory-based local and global interpretability.

[0028] Further, in one embodiment of the present invention, predictions were made using electronic medical record data of T1DM patients hospitalized in 19 hospitals in Gansu Province, China between January 2016 and December 2023. For patients with multiple hospitalizations, only the data of their most recent hospitalization were included. The cohort was divided into two data sets. Data set 1 was divided into a training set (70% of patients) and a test set (30% of patients) using stratified random sampling, while data set 2 was retained for external validation.

[0029] In the present embodiment, a total of 1368 T1DM patients met the analysis criteria, including 324 DN patients (24%). The average age of the whole population was 27.78±15.37 years old. Dataset 1 contained 847 patients with an average age of 27.44±14.67 years old, and Dataset 2 contained 521 patients with an average age of 28.18±16.18 years old.

[0030] In the embodiment of the present invention, the features selected by the feature selection module are T1DM duration, SBP, PPG, HbA1c, LDL-C and Scr levels; 12 machine learning models were developed in the training data set. Among these models, the best performing model AUCs on the test data set were XGBoost (83%, 95% CI: 76-89), random forest (82%, 95% CI: 75-88), gradient boosting (81%, 95% CI: 74-87), Gaussian process classification (81%, 95% CI: 74-87) and logistic regression (80%, 95% CI: 73-87). Figure 2The ROC curves of the above five models are shown. It can be seen that XGBoost has the highest accuracy (75%, 95% CI: 68-81), sensitivity (76%, 95% CI: 64-86) and specificity (75%, 95% CI: 67-82) and was selected as the preferred model for further analysis. In the external validation, 521 patients from other hospitals were included to distinguish between T1DM patients without DN complications and patients with DN complications. The AUC values ​​of the external validation dataset were 76% (95% CI: 70-82), 74% (95% CI: 68-80), 74% (95% CI: 68-80), 74% (95% CI: 68-79) and 74% (95% CI: 68-79), respectively. The XGBoost model also performed best with an accuracy of 71% (95% CI: 66-75), a sensitivity of 70% (95% CI: 60-79), and a specificity of 72% (95% CI: 68-76). These results confirm the generalization ability of the model and its excellent classification performance.

[0031] In this embodiment of the present invention, the SHAP algorithm is used to illustrate how the XGBoost model affects the prediction results. Figure 3 and Figure 4 As shown, the variables were ranked by mean absolute SHAP values, showing that T1DM duration had the highest predictive accuracy in different time periods, followed by SBP, PPG, Scr, HbA1c, and LDL-C. Whether the predictor was positively or negatively associated with the outcome was determined by SHAP values, and factors affecting T1DM were identified.

[0032] Figure 5 and Figure 6 The explanatory power of the model is shown for T1DM patients without DN complications and T1DM patients with DN complications, respectively. The arrows in the figure indicate the impact of each factor on the prediction, where the blue arrow indicates a decrease in the risk of T1DM and the red arrow indicates an increase in the risk. The overall SHAP value represents the combined effect of all factors. For T1DM patients without DN complications, the SHAP value is low (-2.24), while for typical DN patients, the SHAP value is high (0.99).

[0033] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0034] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A diabetic nephropathy risk prediction system for T1DM patients based on machine learning, characterized in that: It includes a data processing module, a feature selection module, a machine learning module, a performance evaluation module and a model interpretation module; among them, the data processing module collects patient data and performs preprocessing, the feature selection module selects the best features in the patient data based on the feature selection algorithm, the machine learning module builds different machine learning models and performs predictions, the performance evaluation module evaluates the performance of the machine learning model, and the model interpretation module is used to explain the contribution of different features to the prediction results.

2. A diabetic nephropathy risk prediction system for T1DM patients based on machine learning according to claim 1, characterized in that: The data processing module collects the patient's data and performs preprocessing. Specifically, the characteristic data of T1DM patients are collected, including gender, age, duration of T1DM, age of onset, family history, body mass index, drinking history, smoking history, diastolic blood pressure, systolic blood pressure, fasting blood glucose, postprandial blood glucose, fasting C-peptide, postprandial C-peptide, glycosylated hemoglobin, autoantibodies related to T1DM, total cholesterol, triglycerides, high-density lipoprotein cholesterol, low-density lipoprotein cholesterol, alanine aminotransferase, aspartate aminotransferase, total protein, serum albumin, total bilirubin, direct bilirubin, thyroid function, serum creatinine, acute and chronic diabetic complications, and urine albumin-creatinine ratio; all patients are divided into data set 1 and data set 2 according to the hospital region. Data set 1 is divided into training set and test set in a ratio of 7:3 using stratified random sampling, and data set 2 is retained for external validation; features with a missing ratio of more than 30% are eliminated, and for the remaining features, the weighted K nearest neighbor method is used to fill in the missing data, and the minimum-maximum normalization method is applied to standardize the data; the training set, test set, and external validation data are processed separately.

3. The diabetic nephropathy risk prediction system for T1DM patients based on machine learning according to claim 1, characterized in that: The feature selection module uses the recursive feature elimination algorithm of random forest and the five-fold cross validation algorithm to calculate the AUC values ​​of different features, and selects the feature set with the largest AUC and the least number of features as the best feature.

4. The diabetic nephropathy risk prediction system for T1DM patients based on machine learning according to claim 1, characterized in that: The machine learning models built by the machine learning module include logistic regression model, Gaussian process classification model, gradient boosting model, extreme gradient boosting model, random forest model, adaptive boosting model, decision tree model, Gaussian naive Bayes model, K nearest neighbor model, support vector machine, multilayer perceptron and neural network.

5. The diabetic nephropathy risk prediction system for T1DM patients based on machine learning according to claim 2, characterized in that: The performance evaluation module draws the ROC curve based on the prediction results of the machine learning model, and selects 5 machine learning models for further evaluation based on the area under the ROC curve. In the further evaluation, the accuracy, specificity and sensitivity of the 5 machine learning models in the training set and the test set are calculated to select the best performing machine learning model. The 5 machine learning models are externally validated through data set 2 to evaluate their generalization ability; the prediction accuracy of the training set, test set and data set 2 is compared by determining the 95% confidence interval through bootstrap resampling.

6. The diabetic nephropathy risk prediction system for T1DM patients based on machine learning according to claim 1, characterized in that: The model interpretation module is used to explain the contribution of different features to the prediction results. Specifically, the SHAP value of each feature is calculated using the SHapley addition interpretation method to obtain the importance ranking of the feature, identify the important risk factors associated with diabetic nephropathy in T1DM patients, and determine the main predictive factors for diabetic nephropathy in the T1DM patient population.

Images