A method for establishing a hyperuricemia prediction model using machine learning algorithms

Abstract

The present invention provides a method for establishing a hyperuricemia prediction model using a machine learning algorithm, which belongs to the field of health assessment technology. The embodiment of the present invention uses a UKBB cohort based on a large population as training, and uses an internal test set and a dataset of healthy physical examinees from Southern Hospital as an external test set to establish and verify a reliable and practical stacked multimodal machine learning model trained based on genetic and clinical data, which is used to timely identify HUA phenotypes and early predict gout and metabolic-related outcome risks. The model has excellent performance, exceeding each single machine learning model, and shows satisfactory efficacy. The ISHUA marker can quantify gout risk and metabolic-related results at an early stage, can provide active guidance for daily health examinations and the management of risk factors, and provide valuable information support for clinicians to appropriately manage HUA risk factors after routine health screening.

Description

Technical Field

[0001] The present invention relates to the technical field of health assessment, and in particular to a method for establishing a hyperuricemia prediction model using a machine learning algorithm. Background Art

[0002] Hyperuricemia (HUA) represents elevated serum uric acid (SUA) levels, which may be due to increased uric acid production or decreased excretion in the body. Sustained hyperuricemia promotes the long-term deposition of uric acid crystals, with potentially deleterious effects on joint integrity and a significant impact on an individual's quality of life. HUA is the second most prevalent metabolic disorder after diabetes and has gained increasing attention as a major global public health issue.

[0003] Currently, risk assessment of HUA or gout mainly relies on clinical parameters or polygenic risk scores, lacking a comprehensive prediction model that combines genetic and clinical characteristics. In addition, existing models mainly act as classifiers and can only predict the status of HUA within a binary framework, rather than quantitatively assessing the disease on a continuous scale. Quantitative assessment of HUA may optimize personalized care strategies.

[0004] In view of this, there is an urgent need in this field for a prediction model for HUA and its adverse outcomes based on multimodal machine learning, including clinical and genetic variables, which can achieve early detection of HUA, early intervention, reduce the occurrence of adverse outcomes, and promote population health.

[0005] The information disclosed in this background technology section is only intended to enhance the understanding of the overall background of the invention and should not be regarded as an acknowledgment or any form of suggestion that the information constitutes the prior art already known to a person skilled in the art. Summary of the invention

[0006] The purpose of the present invention is to provide a multimodal prediction model for hyperuricemia and its adverse outcomes, so as to achieve early screening and early management of patients with hyperuricemia in the population, reduce the occurrence of adverse outcomes, improve prognosis, and thus improve the health level of the population.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for establishing a hyperuricemia prediction model using a machine learning algorithm, comprising:

[0009] S1: Collect and organize data to obtain training sets and test sets;

[0010] S2: Screening clinical characteristics and SNPs with the most significant impact on hyperuricemia;

[0011] S3: Based on the features screened out in S2, a multimodal machine learning stacking model is constructed using a variety of machine learning algorithms, and a cross-validation method is used to ensure the stability of the stacking model and prevent overfitting, thereby obtaining a hyperuricemia phenotype prediction model;

[0012] S4: Obtain the probability score of each participant suffering from HUA from the prediction model as their ISHUA;

[0013] S5: The ROC curve was used in the training set to evaluate the ability of ISHUA to discriminate whether gout occurred. The critical value was selected based on the principle of the maximum Youden index, and participants with different ISHUA scores were divided into low-risk and high-risk groups;

[0014] S6: using an internal test set and an external test set to evaluate the prediction performance of the hyperuric acid phenotype prediction model constructed in S3 to verify the effectiveness of the hyperuric acid phenotype prediction model;

[0015] S7: Use the internal test set to verify the ability of the high-risk group and low-risk group divided in S5 to predict the adverse prognosis of hyperuricemia, so as to verify the effectiveness of the hyperuricemia adverse prognosis prediction model.

[0016] Preferably, the S1 specifically includes:

[0017] The UKBB and Southern Hospital databases were sorted, the physical examination data and genetic data of each participant were integrated and data cleaning was performed, and the participants with some missing data were eliminated to obtain the original data set. The UKBB data was used as the training set and the Southern Hospital data was used as the test set.

[0018] Preferably, the S2 specifically includes:

[0019] A combination of literature method and expert knowledge-based method was used to screen clinical characteristics, and LASSO regression was used to screen the SNPs with the most significant impact on hyperuricemia.

[0020] Preferably, the multiple machine algorithms in S3 include: LightGBM, XGBoost, CAB, random forest, ADB, logistic regression and KNN.

[0021] Preferably, the multimodal machine learning stacking model includes: a basic module and a meta-module; the basic module and the meta-module are connected in a cascade manner.

[0022] Preferably, in S6 and S7, the performance evaluation indicators include AUC, accuracy, sensitivity and specificity.

[0023] Preferably, the basic module includes seven basic separators: LightGBM, XGBoost, CAB, Random Forest, ADB, Logistic Regression and KNN; these classifiers run in parallel, independently predict input features and summarize the prediction results, and then input the summarized prediction results into the meta-module.

[0024] Preferably, the meta-module is composed of a single XGBoost classifier, which is used to further process the aggregated results from the capital module classifier to obtain the final prediction result.

[0025] Preferably, the ISHUA ranges from a value from 0 to 1, where 0 indicates that the participant has the lowest probability of having HUA and 1 indicates that the participant has the highest probability of having HUA.

[0026] Preferably, the clinical characteristics in S2 include: age, gender, BMI, TG, CHO, LDL-C, HDL-C, glucose, blood uric acid nitrogen and sCr.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The present invention first cleans and integrates data from UKBB and hospital databases to obtain training sets and test sets; then uses literature methods and expert knowledge guidance to screen out important clinical features, and uses LASSO regression to screen the SNPs that have the most significant impact on hyperuricemia; then, based on the important features screened out, multiple machine learning algorithms are used to build models, and a 5-fold cross-validation method is used to ensure the stability of the model and prevent overfitting, thereby obtaining a hyperuricemia phenotype prediction model; then, the probability score of each participant suffering from HUA is obtained from the prediction model as their ISHUA, and the ROC curve is used in the training set to evaluate the ability of ISHUA to distinguish whether gout occurs, and the critical value is selected based on the principle of the maximum Youden index, and participants with different ISHUA scores are divided into low-risk groups and high-risk groups; finally, the constructed optimal model is verified to ensure the prediction performance of the model. In this way, the method can more easily and quickly predict the risk of an individual suffering from HUA and the risk of a poor prognosis, thereby helping healthcare professionals to conduct early intervention and treatment and improve the quality of life of patients. In addition, this method provides a powerful tool for medical research to conduct risk analysis and prediction of HUA on large-scale datasets to promote progress in the field of disease prevention and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] 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 will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0030] Figure 1 A simplified flowchart of a method provided by an embodiment of the present invention.

[0031] Figure 2 A schematic diagram of a flow chart provided for an embodiment of the present invention.

[0032] Figure 3 A in FIG. 1 is a schematic diagram of mean square error corresponding to different SNP numbers revealed by the LASSO model in the training set provided by the embodiment of the present invention. A dotted line is drawn at the optimal λ value according to the principle of minimum mean square error, and the λ value at this time is 1378.

[0033] Figure 3 B is the LASSO coefficient distribution diagram of the SNP provided in the embodiment of the present invention.

[0034] Figure 3 The CDE in the figure are ROC curves of the prediction model constructed by the present invention in the training set, internal test set and external test set respectively.

[0035] Figure 4 The figure is a schematic diagram of the cumulative risk of various adverse prognoses in the high-risk group and the low-risk group in the training set for the prognosis prediction model of the present invention. Figure A represents gout, Figure B represents coronary heart disease, Figure C represents atrial fibrillation, Figure D represents heart failure, Figure E represents hypertension, Figure F represents end-stage renal disease, Figure G represents type 2 diabetes, and Figure H represents all-cause mortality. DETAILED DESCRIPTION

[0036] 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.

[0037] The purpose of the present invention is to provide a method for establishing a hyperuricemia prediction model using a machine learning algorithm, which can more conveniently and accurately predict the risk of an individual suffering from hyperuricemia, thereby helping healthcare professionals to conduct early intervention and treatment and improve the quality of life of patients.

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] Embodiment 1:

[0040] See also Figure 1 and Figure 2 The embodiment of the present invention provides a method for establishing a diabetes prediction model using a machine learning algorithm, the method comprising:

[0041] Step 1: Organize UKBB (UK Biobank) and Southern Hospital databases, integrate the physical examination data and genetic data of each participant and perform data cleaning, remove some participants with missing data to obtain the original data set, use UKBB data as the training set, and Southern Hospital data as the test set;

[0042] Step 2: Clinical characteristics were screened using a combination of literature and expert knowledge-based methods. The final clinical characteristics screened were age, gender, BMI (body mass index), TG (triglycerides), CHO (cholesterol), LDL-C (low-density lipoprotein), HDL-C (high-density lipoprotein), glucose, blood uric acid nitrogen, and sCr (serum creatinine). LASSO (least absolute shrinkage and selection operator) regression was used to screen the SNPs (single nucleotide polymorphisms) that had the most significant effect on hyperuricemia. The number and specific content of the screened SNPs are shown in Figure 3 A. Figure 3 B;

[0043] Step 3: Based on the important features screened in step 2, a multimodal machine learning stacking model consisting of two parts (basic module and meta-module) was constructed using a variety of machine learning algorithms (LightGBM, XGBoost, CAB, random forest, ADB, logistic regression, and KNN). The two modules are connected to each other in a cascade manner. The basic module includes seven basic separators (LightGBM, XGBoost, CAB, random forest, ADB, logistic regression, and KNN). These classifiers run in parallel, independently predict the input features and summarize the prediction results, and then input the summarized prediction results into the meta-module. The meta-module is composed of a single XGBoost classifier, which can further process the aggregated results from the capital module classifier to obtain the final prediction result. The 5-fold cross-validation method was then used to ensure the stability of the model and prevent overfitting, and finally a high uric acid phenotype prediction model was obtained;

[0044] Step 4: Obtain each participant’s probability score of having HUA from the prediction model as their ISHUA (ISHUA ranges from 0 to 1, with 0 indicating that the participant has the lowest probability of having HUA and 1 indicating that the participant has the highest probability of having HUA).

[0045] Step 5: Use the ROC curve in the training set to evaluate the ability of ISHUA (probability of hyperuricemia) to discriminate whether gout occurs. Select the critical value based on the principle of the maximum Youden index, and divide participants with different ISHUA scores into low-risk and high-risk groups;

[0046] Step 6: Use the internal and external test sets to evaluate the prediction performance of the hyperuric acid phenotype prediction model constructed in step 3 to verify the effectiveness of the hyperuric acid phenotype prediction model. The evaluation results are shown in Figure 3 D. Figure 3 E.

[0047] Step 7: Use the internal test set to verify the high-risk group and low-risk group divided in step 5 to evaluate the ability of high uric acid to predict adverse prognosis outcomes. The results are as follows Figure 4 As shown, the high-risk group had a higher incidence of gout, coronary heart disease, atrial fibrillation, heart failure, hypertension, end-stage renal disease, type 2 diabetes, and all-cause mortality than the low-risk group.

[0048] In steps 6 and 7, commonly used performance evaluation indicators include accuracy, precision, ROC curve and AUC value.

[0049] The embodiment of the present invention uses the UKBB cohort based on a large population as training, and uses the internal test set and the Southern Hospital health checkup data set as the external test set to establish and verify a reliable and practical stacked multimodal machine learning model trained based on genetic and clinical data, which is used to timely identify HUA phenotypes and early predict gout and metabolic-related outcome risks. The performance of this model is excellent, exceeding that of each single machine learning model, showing satisfactory efficacy. The ISHUA marker can quantify gout risk and metabolic-related results at an early stage, can provide active guidance for daily health examinations and risk factor management, and provide valuable information support for clinicians to appropriately manage HUA risk factors after routine health screening.

[0050] In summary, the machine learning model provided by the present invention can more easily and quickly predict the risk of an individual suffering from HUA and the risk of poor prognosis, thereby helping healthcare professionals to conduct early intervention and treatment and improve the quality of life of patients. In addition, this method also provides a powerful tool for medical research, which can be used to perform risk analysis and prediction of HUA on large-scale data sets to promote progress in the field of disease prevention and management.

[0051] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0052] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for establishing a hyperuricemia prediction model using a machine learning algorithm, characterized in that: include: S1: Collect and organize data to obtain training sets and test sets; S2: Screening clinical characteristics and SNPs with the most significant impact on hyperuricemia; S3: Based on the features screened out in S2, a multimodal machine learning stacking model is constructed using a variety of machine learning algorithms, and a cross-validation method is used to ensure the stability of the stacking model and prevent overfitting, thereby obtaining a hyperuricemia phenotype prediction model; S4: Obtain the probability score of each participant suffering from HUA from the prediction model as their ISHUA; S5: The ROC curve was used in the training set to evaluate the ability of ISHUA to discriminate whether gout occurred. The critical value was selected based on the principle of the maximum Youden index, and participants with different ISHUA scores were divided into low-risk and high-risk groups; S6: using an internal test set and an external test set to evaluate the prediction performance of the hyperuric acid phenotype prediction model constructed in S3 to verify the effectiveness of the hyperuric acid phenotype prediction model; S7: Use the internal test set to verify the high-risk group and low-risk group divided in S5 to evaluate the prediction ability of high uric acid adverse prognosis outcomes, so as to verify the effectiveness of the high uric acid adverse prognosis prediction model; The S1 specifically includes: Organize the UKBB and Nanfang Hospital databases, integrate the physical examination data and genetic data of each participant, perform data cleaning, remove some participants with missing data to obtain the original data set, use the UKBB data as the training set, and the Nanfang Hospital data as the test set; The S2 specifically includes: A combination of literature and expert knowledge-based methods was used to screen clinical characteristics, and LASSO regression was used to screen the SNPs with the most significant impact on hyperuricemia; The clinical characteristics in S2 include: age, gender, BMI, TG, CHO, LDL-C, HDL-C, glucose, serum uric acid nitrogen and sCr; The various machine algorithms in S3 include: LightGBM, XGBoost, CAB, Random Forest, ADB, Logistic Regression, and KNN; The multimodal machine learning stacking model includes: a basic module and a meta-module; the basic module and the meta-module are connected in a cascade manner; The basic module includes seven basic classifiers: LightGBM, XGBoost, CAB, Random Forest, ADB, Logistic Regression and KNN; these classifiers run in parallel, independently predict input features and aggregate the prediction results, and then input the aggregated prediction results into the meta-module; The meta-module consists of a single XGBoost classifier, which is used to further process the aggregated results from the capital module classifier to obtain the final prediction result; The ISHUA ranges from 0 to 1, with 0 indicating that the participant has the lowest probability of having HUA and 1 indicating that the participant has the highest probability of having HUA; In S6 and S7, the performance evaluation indicators include AUC, accuracy, sensitivity and specificity.

Images