环境流行病学研究中广义可加模型在Python中的实现  

Implementation of generalized additive models in environmental epidemiology research in Python

作　　者：李湘莹 李培政 王静 罗晨曦 张清煜 马露[1] Xiang-Ying LI;Pei-Zheng LI;Jing WANG;Chen-Xi LUO;Qing-Yu ZHANG;Lu MA(School of Public Health,Wuhan University,Wuhan 430071,China)

机构地区：[1]武汉大学公共卫生学院,武汉430071

出　　处：《数理医药学杂志》2023年第4期241-245,共5页Journal of Mathematical Medicine

基　　金：湖北省卫生健康委2021—2022年度科研项目(WJ2021F103)。

摘　　要：目的探索环境流行病学领域中常见的时间序列资料,利用Python以及其他常用统计软件实现广义可加模型(generalized additive models,GAM)的建模,比较各软件建模过程和结果的异同。方法以研究某地PM_(2.5)暴露与呼吸系统疾病入院人次之间的关系作为实例,分别利用Python软件statsmodles库、R软件mgcv库和SAS软件的proc gam语法,构建GAM模型,比较各软件命令代码、参数设置以及参数估计的差异。结果三种软件构建GAM模型的建模逻辑相似,但在内置函数拟合过程、命令代码以及可调用的样条函数等方面有所差别,各软件输出结果基本一致。结论Python软件利用第三方库可实现广义可加模型的构建,为进一步拓展其在流行病学领域的应用提供了参考。Objective To explore the common time series data in the field of environmental epidemiology,using Python and other statistical softwares to realize the modeling of generalized additive models(GAM),and to compare the similarities and differences of the modeling process and results of each software.Method A study of the relationship between PM_(2.5) and the number of hospital admissions of respiratory diseases was taken as an example.Python software used statsmodles library,R software used mgcv library,SAS software used proc gam syntax to build GAM models,and the differences in codes,parameter settings,and parameter estimates were compared.Results The modeling logic of 3 programs is similar,but there are differences in the built-in function fitting process,code using and callable spline function.The outputs are basically consistent.Conclusion Python software can build GAM by using third-party libraries.It provides a reference for further expanding its application in the field of epidemiological scientific research.

关 键 词：PYTHON 广义可加模型 时间序列数据 环境流行病学 

分 类 号：R181.34[医药卫生—流行病学] TP312.1[医药卫生—公共卫生与预防医学]