| Objectives:The purpose of this study is to investigate the epidemic status and influencing factors of hyperuricemia(HUA)in adult residents of Guangdong Province,build a prediction model of HUA based on non-invasive indicators for a large sample population,provide decision-making assistance for diagnosis and screening of HUA in Guangdong Province,promote early screening,early diagnosis and early treatment of patients with HUA,reduce the disease burden,and also provide ideas for the subsequent development of relevant prediction models.Methods:The data were collected from the Chronic Disease and Nutrition Surveillance of Residents in Guangdong province in 2015 and 2018,and the survey subjects were selected from 14 monitoring sites in Guangdong province by multi-stage stratified cluster sampling and the respondents were investigated by questionnaire,medical examination and laboratory examination.1.Chi-square test was used to compare the differences between groups of different demographic characteristics,living habits and prevalence of the subjects,and binary logistic regression model was used to analyze the relevant influencing factors of HUA.The degree of association was described by the odds ratio and its 95%confidence interval,and P<0.05 believed that the difference was statistically significant.It is suggested that the situation of prevention and control of HUA is severe and needs attention of governments at all levels and health administrative departments.2.In terms of the HUA prediction model based on non-invasive indicators,the dietary survey households in the chronic disease and nutrition monitoring data of Guangdong residents in 2015 were selected as the research object.Factor analysis was used for feature extraction and dimensionality reduction of dietary intake characteristics;Use random forest algorithm for feature selection;Build a HUA prediction model based on non-invasive indicators and five machine learning algorithms:random forest,decision tree,gradient lifting decision tree,quadratic discriminant analysis and logistic regression;The effectiveness of the model is evaluated by cross-validation with ten fold;The prediction performance of the five models was evaluated by using accuracy,accuracy,specificity,recall,receiver operating characteristic curve and area under the curve,and the best HUA prediction model was selected.Results:1.This study found that the detection rate of HUA among adult residents in Guangdong Province is 30.00%.Among them,male,80-year-old,urban,Pearl River Delta region,high school/technical school education,unmarried,and commercial/service industry professionals have a higher detection rate of HUA.2.Binary logistic regression analysis found that gender(male),age(80～),region(Pearl River Delta region),occupation of business/service industry,occupation of household/retirees,annual family income of 30000～,annual family income of 100000～,unmarried,overweight,obesity,central obesity,alcohol consumption,high intensity physical activity,excessive red meat intake per day,hypertension,diabetes,dyslipidemia are risk factors for HUA;Low weight is a protective factor for HUA.3.The cross-validation results of the HUA prediction model show that the model with the highest accuracy and accuracy is the gradient lifting decision tree model,which is up to 87.24%and 75.97%respectively;The model with the highest specificity and recall rate is the random forest model,up to 92.30%and 78.84%respectively.The AUC value of random forest is the largest,indicating that the performance of the HUA prediction model based on this algorithm is relatively better than the other four models.Conclusion:1.The detection rate of adult HUA in Guangdong Province is 30.00%.Gender(male),age(80～),region(Pearl River Delta region),occupation of business/service industry,occupation of household/retirees,annual family income of 30000～,annual family income of 100000～,unmarried,overweight,obesity,central obesity,alcohol consumption,high intensity physical activity,excessive red meat intake per day,hypertension,diabetes,dyslipidemia are risk factors for HUA;Low weight is a protective factor for HUA.It is suggested that the situation of prevention and control of HUA is severe and needs attention of governments at all levels and health administrative departments.2.The HUA prediction model based on non-invasive indicators and random forest algorithm has good prediction efficiency,which can provide decision-making assistance for the diagnosis of HUA,and also provide ideas for scholars who develop relevant prediction models. |
PDF Full Text Request