基于机器学习方法的人体农兽药及化学污染物暴露与高血压的相关性研究

Analysis of the association between pesticide and chemical pollutant exposure and hypertension in humans based on machine learning methods

目的基于不同的机器学习方法探究石家庄与杭州成年居民体内农兽药及化学污染物暴露与高血压患病情况之间的关系。方法采用2018—2019年在石家庄与杭州进行的“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”调查数据,选择496名包含人口学资料、体格测量、常规血清检测和血清农兽药及化学污染物暴露信息的成年居民作为研究对象,在Lasso变量筛选后分别使用传统的逻辑回归模型与多种机器学习模型建立高血压的预测模型,利用ROC曲线下面积(AUC)评估模型效果。结果Lasso变量筛选结果显示,农兽药及化学污染物暴露4-氯苯氧乙酸(4-CPA)、全氟辛酸(PFOA)、全氟己烷磺酸(PFHxS)和全氟辛烷磺酸(PFOS)与高血压具有显著的关联。机器学习模型中支持向量机模型预测效果最好(AUC=0.71),优于传统的逻辑回归模型(AUC=0.57)。结论农兽药及化学污染物暴露中4-CPA、PFOA、PFHxS和PFOS是高血压的重要危险因素,机器学习模型在流行病学影响因素研究中具有很好的适应性,在拟合非线性关系的数据时有一定的优势。

Objective The association between chemical pollutant exposure,such as pesticides and chemical pollutants,and hypertension in adult residents of Shijiazhuang and Hangzhou was assessed using various machine learning methods.Methods A cross-sectional study was conducted in Shijiazhuang and Hangzhou,China from 2018 to 2019.A total of 496 participants were selected based on their individual characteristics,including,body measurements and routine blood tests,as well as pesticide and chemical pollutant exposure.Lasso was used to select features,which were fitted with logistic regression models and other machine learning methods to study the factors influencing hypertension.The effects of the different models were compared based on the area under the curve(AUC).Results The Lasso feature selection results showed that pesticides and chemical pollutants,specifically,4-CPA,PFOA,PFHxS and PFOS were significantly associated with hypertension.Among the machine learning models tested,the support vector machine model had the best performance(AUC=0.71),which was better than the traditional logistic regression model(AUC=0.57).Conclusion Exposure to the pesticide chemicals,4-CPA,PFOA,PFHxS and PFOS,are important risk factors for hypertension.Additionally,machine learning models can be used to study epidemiological influencing factors and have an advantage in fitting non-linear relationships.