稳健Poisson模型:两水平模型与GEE模型在相对危险度或患病率比估计中的应用比较

A Comparison between Two-level and GEE Based on Robust Poisson Regression Models in the Estimation of Relative Risk or Prevalence Ratio

目的在处理具有层次结构特征的非罕见结局事件资料时,比较基于稳健Poisson模型的两水平模型和GEE模型在估计RR/PR时的应用。方法将两水平稳健Poisson模型及稳健Poisson-GEE应用到2010年欧洲社会调查资料,估计影响居民生活满意度的各因素相关的PR及95%CI,以说明两模型在理论和应用上的区别和联系。结果稳健Poisson-GEE模型的PR估计值与稳健Poisson回归模型相同,但置信区间较宽;两水平稳健Poisson模型的PR值较GEE模型为低,显示了随机效应对解释变量的混杂作用。结论两种方法均可处理具有层次结构特征的非罕见结局事件的RR/PR估计,但两水平模型比GEE可提供更多随机效应的信息,且易于扩展至更高水平或随机系数模型。

Objective To compare two-level and GEE based robust Poisson regression models in estimation of relative risk （ RR ） or prevalence ratio （PR） for common outcome data with intra-class correla- tion. Methods Two-level and GEE based robust Poisson regression models were compared by examing factors associated with life satisfaction using data from the 2010 European Social Survey. Prevalence ratios and 95 % confidence intervals （ 95 % CIs ） were estimated. Results Com- pared to results from regular robust Poisson model, the GEE based robust Poisson model provided the same PR point estimates but wider 95% Cls. The two level robust Poisson model revealed lower point estimates,indica- ting potential confounding effects caused by random effects on the assoca- tion of interest. Conclusion Both two-level and GEE based methods are suitable for estimating relative risk or prevalence ratio for common outcomes with the hierarchical structure. The two-level model is superior when there are random effects, and can be easily extended for higher hierarchical structures.