Comparison of Three Approaches in the Analysis of Interaction between a Time-fixed Factor and a Time-varying Variable

Abstract

Objective This study aimed at illustrating three approaches for the analysis of interaction between a time-fixed factor and a time-varying variable,including Poisson regression,the case-only approach and case-crossover method.We also compared their statistical performance using a simulation study.Methods Two surrogate time-series of daily counts of mortality for smokers and non-smokers were produced using simulation.We analyzed the interaction between daily concentration of PM10 and smoking using three approaches.Their statistical performances were assessed by power and error.Results The case-only approach had the same performance with Poisson regression. Lower statistical power and bigger error were observed in the results from case-crossover method.Conclusion These three approaches are all useful tool in analysis of interaction between time-fixed factors and time-varying variables.Poisson regression and case-only approach provide better statistical performances. 目的本文將介紹分析固定因素和伴時變量交互作用的三種方法 :時間序列方法、單純病例研究法和病例交叉法,并通過模擬研究比較其優劣。方法計算機模擬吸煙和非吸煙人群的日死亡人數時間序列,用三種不同方法分析每日大氣可吸入顆粒物(PM10)濃度和吸煙對死亡影響的交互作用,比較各方法的檢驗效能和誤差。結果時間序列方法與單純病例研究法的檢驗效能和誤差相一致,而病例交叉法的檢驗效能略低,誤差略高于其他兩種方法。結論三種方法都是分析固定因素和伴時變量交互作用的有效工具。時間序列方法與單純病例研究法的表現略佳。