基于随机森林方法的常见人体中农兽药及化学污染物暴露与高尿酸血症的关联性研究

Random forest analysis on the association between hyperuricemia and exposure to common pesticides,veterinary drugs,and chemical contaminants in humans

目的探索高尿酸血症(HUA)的影响因素,尤其是农兽药及化学污染物暴露与高尿酸血症的关联,分析机器学习方法对于农兽药及化学污染物暴露数据的分析效果。方法根据2018—2019年在石家庄和杭州进行的“降低成年超重者营养相关慢性病风险的适宜身体活动量研究”,分别采用传统Logistic回归和随机森林(RF)建立基本人口学变量和农兽药及化学污染物暴露对HUA发病的预测模型。模型区分效果由ROC曲线下面积(AUC)进行评估。结果RF结果显示,对HUA影响重要程度排名前5的因素依次为多西环素、4-氯苯氧乙酸酯、呋喃他酮、咪鲜胺和全氟癸酸浓度。RF模型的区分效果显著高于传统Logistic回归模型(AUC分别为0.934和0.735)。结论多西环素、4-氯苯氧乙酸酯、呋喃他酮、咪鲜胺和全氟癸酸、饮酒史、居住地为杭州、甘油三酯≥2.26 mmol/L等可能是HUA的危险因素。RF模型适用于农兽药及化学污染物暴露数据的分析,且较常规Logistic回归模型对于鉴别HUA患者具有显著提升的区分能力。

Objective To identify the risk factors of developing hyperuricemia(HUA),especially due to exposure to chemical contaminants,common pesticides,and veterinary drugs in humans.Subsequently,the effect of machine learning techniques on exposure data of agricultural and veterinary drugs and chemical pollutants was explored.Methods According to the“Study on Appropriate Physical Activity to Reduce the Risk of Nutrition-related Chronic Diseases in Overweight Adults”program conducted in Shijiazhuang and Hangzhou,China,from 2018 to 2019,traditional logistic regression and random forest(RF)were used to establish prediction models using demographic indicators and exposure to pesticides,veterinary drugs,and chemical contaminantsas covariates on the development of HUA.The discrimination of the models were assessed by the area under the receiver operating characteristic curve(AUC).Results RF analysis revealed that the top five factors affecting the development of HUA were doxycycline,4-chlorophenoxyacetate(4-CPA),furaltadone,prochloraz,and perfluorodecanoic acid(PFDA).The RF model showed better discriminant ability than the logistic regression model(AUC 0.934 vs.0.735).Conclusion Exposure to doxycycline,4-CPA,furaltadone,prochloraz and PFDA,alcohol drinking history,living in Hangzhou,and a level of triglycerides≥2.26 mmol/L may be risk factors for developing HUA.The RF model was suitable to analyze associations of chemical contaminants,pesticides,and veterinary drugs data,and ehibited a significantly improved discriminatory ability for identifying HUA patients compared with the conventional logistic regression model.