Meta分析中二分类数据效应量的选取

摘要

Objective To investigate the selection of the best effect size from binary data in Meta-analysis based on case study.Methods Through systematic retrospective of 551 reviews and 114 Meta-analysis documents, 4 effective indexes were analyzed and compared including odds ratio (OR), risk difference (RD), relative risk of benefit [RR (B)] and relative risk of harmful [RR (H)].Results The evidence from 551 reviews showed thatRR and OR had better efficacy than RD in the aspect of outcome superiority. Meanwhile,OR had an inclination to overestimate the pooled results, even leaded to a qualitative inconsistency in the Meta-analysis. The results of 114 Meta-analysis documents showed that for intervention aimed at preventing reverse events, the highest incidence rate of baseline risk would induce the greatest absolute benefit effect, and RR (H) was optimized at this moment.Conclusion The multiple factors should be considered including cause of baseline risk variation, effect size interpretability and mathematical properties in the selection of effect size. For some special cases, the characteristics of clinical trials should also be taken as one of factors in selecting effect size.%目的 基于实例研究,探讨Meta分析中二分类数据最佳效应量的选取.方法 通过系统回顾551篇系统评价及114篇Meta分析,对比值比(OR)、危险差(RD)、相对获益危险度[RR(B)]及相对有害危险度[RR(H)]4个效应指标进行分析比较.结果 经系统回顾551篇Meta所提供的证据显示:在表示结果优越性方面上,RR与OR较RD具有更好功效;同时,与RR相比,OR可能会夸大Meta分析定量的合并结果,甚至产生定性上的不一致.来自于114篇Meta分析结果分析表明,对于预防不良事件的干预试验,基线风险发生率越大绝对获益功效越大,此时应优选RR(H).结论 效应量的选取需要综合考虑基线风险变化原因、效应量可解释性及其数学特性等多方面因素,对于特殊情形下,临床试验特征也是作为效应量选取因素之一.