国家神经系统疾病临床医学研究中心数据管理与共享系统
Data Management And Sharing System of China National Clinical Research Center for Neurological Diseases
Estimate risk difference and number needed to treat in survival analysis
第一作者：Zhang Zhongheng
通讯作者：Zhang Zhongheng
基本信息
- 负责人：Zhang Zhongheng
- 数据来源： OTHERS
- 论文编号：OTHERS20190656
- 发表时间：2018-04-01
2018-04-01
已发表文章
发表杂志：Ann Transl Med
发表类型: Big-Data Clinical Trial Column
影响因子(IF): 3.932
PMID: 29955580 Pubmed原文链接 全文下载
主要研究背景及目的
方法
For simplicity, only one measure (RD or NNT) needs to be illustrated, because the other measure is a reverse of the illustrated one (NNT=1/RD). An artificial dataset is composed by using the survsim package. RD and NNT are estimated with Austin method after fitting a Cox-proportional hazard regression model. The confidence intervals can be estimated using bootstrap method. Alternatively, if the standard errors (SEs) of the survival probabilities of the treated and control group are given, confidence intervals can be estimated using algebraic calculations. The pseudo-value model provides another method to estimate RD and NNT. Details of R code and its output are shown and explained in the main text.
结论
The article provides a step-by-step tutorial on the computation of RD and NNT in survival analysis. An artificial dataset containing survival data was employed to illustrate the computation with R. Several methods exist to estimate RD and NNT: the Austin’s method, Altman’s method and pseudo-value procedure. The Altman’s method is used when there is no individual-level data and thus the results may not be comparable to other estimators. The confidence interval can be estimated using bootstrap technique. As compared with RRR, RD or NNT provides more complete information on the risks and benefits of an intervention. In research practice, the rule of thumb is to report both RRR and RD to make the evidence more informative for clinicians and other workers involved in guideline development.