公卫人

 找回密码
 立即注册

QQ登录

只需一步,快速开始

查看: 697|回复: 8

[分享] 观察性研究中的敏感性分析:E值(E-Value)的引入

[复制链接]
alian 发表于 2020-4-7 12:08:23 | 显示全部楼层 |阅读模式

注册后推荐绑定QQ,之后方才可以使用下方的“用QQ帐号登录”。

您需要 登录 才可以下载或查看,没有帐号?立即注册

x
In the 1970s, economists created statistical methods for detecting and correcting for unobserved treatment selection bias. The E-value indicates the “minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need to have with both the treatment and outcome…to fully explain away a specific treatment-outcome association.”9 [1 W! q+ h1 @5 w2 U& t! ]6 l
https://jamanetwork.com/journals ... fullarticle/2763824$ C. H5 c" }2 q; s) U; _

% C9 o0 S9 _$ \引入E值
6 u/ Q, ^% r( aSensitivity Analysis in Observational Research: Introducing the E-Value9 V# w1 b& r3 B
6 ~; M2 U. t$ {! e& N, J! m
Tyler J. VanderWeele, PhD, and Peng Ding, PhD
4 x4 T2 E# e# g% V( A8 V! L* k0 s) q. m; m; S) L
From Harvard T.H. Chan School of Public Health, Boston, Massachusetts, and University of California, Berkeley, Berkeley,California.
( m# B; n+ A  K( {! O5 f/ Y* B
* U5 e+ _! |+ p* v4 d) qSensitivity analysis is useful in assessing how robust an association is to potential unmeasured or uncontrolled confounding. This article introduces a new measure called the “E-value,” which is related to the evidence for causality in observational studies that are potentially subject to confounding. The E-value is defined as the minimum strength of association, on the risk ratio scale, that an unmeasured confounder would need to have with both the treatment and the outcome to fully explain away a specific treatment–outcome association, conditional on the measured covariates. A large E-value implies that considerable unmeasured confounding would be needed to explain away an effect estimate. A small E-value implies little unmeasured confounding would be needed to explain away an effect estimate.
  c9 i, {9 V! I; E2 q0 M4 {% E
) @  U# e2 a+ q4 M9 lThe authors propose that in all observational studies intended to produce evidence for causality, the E-value be reported or some other sensitivity analysis be used. They suggest calculating the E-value for both the observed association estimate (after adjustments for measured confounders) and the limit of the confidence interval closest to the null. If this were to become standard practice, the ability of the scientific community to assess evidence from observational studies would improve considerably, and ultimately, science would be strengthened.
0 [( e+ `5 F6 b) b0 p- ^7 h
' S  v# B9 J, Z0 t- O8 j5 kVanderWeele, T. J., & Ding, P. (2017). Sensitivity Analysis in Observational Research: Introducing the E-Value. Annals of Internal Medicine, 167(4), 268. doi:10.7326/m16-2607 7 v7 Q, D! I3 {5 F
: v% W- k; b5 Q
社论# T0 P5 Y4 h1 c3 k
Localio, A. R., Stack, C. B., & Griswold, M. E. (2017). Sensitivity Analysis for Unmeasured Confounding: E-Values for Observational Studies. Annals of Internal Medicine, 167(4), 285. doi:10.7326/m17-1485
. |, B1 W5 t) ^2 A- f3 f  ]) ?; J8 B9 H( {
不足与误读1 A' M+ \% s4 j! k/ P! N1 |/ S
Limitations and Misinterpretations of E-Values for Sensitivity Analyses of Observational Studies: k  j5 {; H! F- Z. H6 n
Ann Intern Med. doi:10.7326/M18-2159
9 q- d$ f4 `2 [4 h9 k" j+ g) E8 s/ w  y5 q5 s" t
软件实现- V) z2 C* P9 f# q3 S- F7 F) b5 o
3 {/ Y4 m5 C( u
Stata 社区命令:evalue --  Sensitivity analyses for unmeasured confounding in observational studieshttps://ideas.repec.org/c/boc/bocode/s458592.html
1 z$ t2 _  T) R9 f1 K) W1 Zssc install evalue
0 h( O% G! d  X: S( y2 m# |6 e4 [- w1 C7 y. K: S8 y- D0 s! ?% J$ F4 q
另一个社区命令:evalues-- calculate E-values& E/ o2 s8 B7 I+ i* \8 \# [: f# v
https://ideas.repec.org/c/boc/bocode/s458663.html! T/ Y2 r; N* s! t
ssc install evalues3 z2 w7 `0 \* A8 O4 U; c5 i
4 J7 ]' u& z3 a5 V- l
R语言' t5 C# l7 E# X" P# s+ g

& O! j* ~3 f; d5 z. _R代码 https://github.com/mayamathur/evalue 文献:Website and R Package for Computing E-Values
% _* K& |" |% s5 B5 Y; Q! b+ }Epidemiology. 2018 Sep;29(5):e45-e47. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6066405/. J) q. O/ W% H" C

, l( Q* M9 a0 n5 t在线计算 E-value calculator8 K5 X9 E* \' N; y) X
https://mmathur.shinyapps.io/evalue/https://www.evalue-calculator.com4 Z$ D! r! c: K7 }0 w9 \  W
E-value in observational research
2 l$ [4 o. w  b+ d( g, ]http://medical-statistics.dk/MSDS/epi/evalue/evalue.html7 Z' @0 v' ?  C0 s5 p) F) W
2 u' R2 L9 |' K% q0 p
教程9 w/ |! ]' z3 {2 @' Y8 p: Q
哈佛大学 https://www.hsph.harvard.edu/tyl ... ools-and-tutorials/
, R& i8 E. O( q& @' A3 @Sensitivity Analysis in Observational Research:Introducing the E-value
% K# i6 {8 b( K6 S, t" k' Ohttps://biostat.duke.edu/sites/b ... DukeBiostat2018.pdf
& R" w4 L3 f+ k9 _3 {. K3 O9 v/ e5 C$ P7 P6 ]0 N% E2 c

EDITORIAL E-Values for Observational Studies.pdf

84.37 KB, 下载次数: 4, 下载积分: 钢镚 -1 分, 参与 1 贴

Introducing the E-Value.pdf

161.85 KB, 下载次数: 3, 下载积分: 钢镚 -1 分, 参与 1 贴

Limitations and Misinterpretations of E-Values.pdf

103.7 KB, 下载次数: 7, 下载积分: 钢镚 -1 分, 参与 1 贴

EDITORIAL Correcting Misinterpretations of the E-Value.pdf

87.17 KB, 下载次数: 6, 下载积分: 钢镚 -1 分, 参与 1 贴

社论:观察性研究中的E值.pdf

863.81 KB, 下载次数: 13, 下载积分: 钢镚 -1 分, 参与 1 贴

引入E值.pdf

647.33 KB, 下载次数: 9, 下载积分: 钢镚 -1 分, 参与 1 贴

epiman 发表于 2020-4-30 11:39:05 | 显示全部楼层
Using the E-Value to Assess the Potential Effect of Unmeasured Confounding in Observational Studies
  l# ^' v3 O, m* N3 g3 Whttps://jamanetwork.com/journals/jama/article-abstract/2723079( p+ f8 m- T; n% j
JAMA. 2019;321(6):602-603. doi:10.1001/jama.2018.21554! \' @! v+ |2 q7 l& t& o

7 C9 v$ A, M5 }5 R7 A
% o/ U" z: }) ^) s5 ?; g/ t$ h' [! TTo assess how much of a problem unmeasured confounding factors may pose, researchers may conduct a sensitivity or bias analysis.3 Common to most of these sensitivity analysis methods is the use of a formula for which 2 inputs are required:5 A; C+ w7 i% v! T
(1) the strength and direction of the association between the unmeasured confounder and treatment choice and
8 d' T% c; g& u$ E! Y0 a(2) the strength and direction of association between the unmeasured confounder and outcome/ `/ I+ ~6 E4 ?# }. q: b6 ~- |
% u# b- Y5 n4 R
The E-value is an alternative approach to sensitivity analyses for unmeasured confounding in observational studies that avoids making assumptions that, in turn, require subjective assignment of inputs for some formulas. Specifically, an E-value analysis asks the question: how strong would the unmeasured confounding have to be to negate the observed results?
1 ^: v5 _6 N6 c5 N
' h- ]  b/ L4 t+ b+ w( cThe E-value itself answers this question by quantifying the minimum strength of association on the risk ratio scale that an unmeasured confounder must have with both the treatment and outcome, while simultaneously considering the measured covariates, to negate the observed treatment–1 V; D+ o( i8 ^8 k7 t5 R
outcome association. If the strength of unmeasured confounding is weaker than indicated by the E-value, then the main study result could not be overturned to one of “no association” (ie, moving the estimated risk ratio to 1.0) by the unmeasured confounder. E-values can therefore help assess the robustness of the main study result by considering whether unmeasured confounding of this magnitude is plausible. The E-value provides a measure related to the evidence for causality, hence the name “E-value.”$ w0 M/ P! F$ r0 ~- g" u

haneuse2019.pdf

73.39 KB, 下载次数: 0, 下载积分: 钢镚 -1 分, 参与 1 贴

回复

使用道具 举报

anniemao 发表于 2020-4-30 12:09:09 | 显示全部楼层
太赞了!火速收藏!
回复

使用道具 举报

阿达0622 发表于 2020-5-27 09:53:44 触屏版 | 显示全部楼层
来了,领钢蹦儿

评分

参与人数 1钢镚 +2222 收起 理由
epiman + 2222 积极参与论坛活动!

查看全部评分

回复

使用道具 举报

__Abby 发表于 2020-5-27 09:53:46 触屏版 | 显示全部楼层
真的学到了!

评分

参与人数 1钢镚 +4444 收起 理由
epiman + 4444 积极参与论坛活动!

查看全部评分

回复

使用道具 举报

文姬小蚊子 发表于 2020-5-27 11:07:29 触屏版 | 显示全部楼层
这篇真的学到了很多!

评分

参与人数 1钢镚 +4444 收起 理由
epiman + 4444 积极参与论坛活动!

查看全部评分

回复

使用道具 举报

xxs199110 发表于 2020-5-29 15:33:53 | 显示全部楼层
前几天在公卫人公众号看到的文章:观察性研究中的敏感性分析:引入E值  x" }0 S+ P4 j: C4 v4 ^
是一家的
1 `3 X: Z+ g* r
2 G5 Z) j& I* |0 a0 L. n敏感性分析有助于对于评估某种关联对潜在的无法测量或无法控制的混杂的稳健性。本文介绍了一种新的测量方法,称为“E值”,它与观察性研究中因果关系的证据有关,而观察性研究有可能受到混杂的影响。4 F. y: o/ f9 }- t  O
E值是在危险度量表中一种无法测量的混杂因素与治疗和转归所需的最小关联强度,以充分解释一种特定的治疗-结局之间的关联,这取决于可测量的协变量。较大的E值意味着需要相当大的未测量的混杂来解释效应估计值。较小的E值意味着仅需要很小的未测量的混杂来解释效应估计值。
" E! B6 @6 D- A3 w% j7 g作者提出在所有试图为因果关系提供证据的观察性研究中,应该报告E值或采用其他敏感性分析。他们建议计算E值,不仅针对观测到的关联估计值(调整混杂因素后),而且针对最接近于0的置信区间的极限。如果该指标成为标准操作,那么科学界评估观察行研究中证据的能力将大大提高,最终,科学将得以加强。
/ J8 k. `' D, O动机:试图评估治疗和结局之间因果关系的观察性研究可能会受到未测量的混杂因素的影响。
% |* c  B9 `1 f  A+ v+ }0 l基本原理:敏感性分析能够评估未测量的混杂必须有多大的强度才能解释观察到的治疗-结局关系。敏感性分析技术易于使用、呈现和解释,且本身不作出强有力假设的,这是可取的。
: \. v+ X3 o, Y" cE值的定义:E值指在危险比率量表上的最小关联强度,一种未测量的混杂因子需要与治疗和结局同时具有的关联,以测量的协变量为条件,以充分解释一种特定的治疗-结局关联。
1 R) i4 u8 a1 s+ V; _0 t6 k计算:估计E值,以及最接近0的95%置信区间的极限值,可以通过危险比率和其他措施直接计算。
2 \' L* L) l9 ]E值可以让研究者作出如下形式的陈述:9 z1 ?  E" ^; y" N1 j5 m% D
观察到的危险比率3.9,可以由一种未测量的混杂所解释,该混杂分别以7.2倍的危险比率与治疗和结局同时有关联,高于和超过测量到的混杂因素,但较弱的混杂做不到这一点;置信区间可以由于一种未测量的混杂而包含0,该混杂分别以3.0倍的危险比率与治疗和结局同时有关联,高于和超过测量的混杂因素,但是较弱的混杂不能做到这一点。
回复

使用道具 举报

epiman 发表于 2020-5-30 13:25:40 | 显示全部楼层
xxs199110 发表于 2020-5-29 15:33: X1 `5 m0 Q! D- h9 I- ?9 e
前几天在公卫人公众号看到的文章:观察性研究中的敏感性分析:引入E值4 B. r. w1 [# r0 b7 p2 b
是一家的

" P/ ?. |' C5 Z+ i; Y3 v6 d' M阅读原文 就是 这个帖子地址。
' u# v" J1 D- V) W# ]5 R2 S这个帖子的内容更为详尽。
回复

使用道具 举报

wols888 发表于 2020-6-27 13:30:25 | 显示全部楼层
正好审稿人提了这个问题,赶紧学习一下,谢谢楼主!
回复

使用道具 举报

您需要登录后才可以回帖 登录 | 立即注册

本版积分规则

会员|至尊|接种|公卫人 ( 沪ICP备06060850号-3 )

GMT+8, 2020-7-6 19:05 , Processed in 0.150991 second(s), 16 queries , Gzip On, MemCached On.

Powered by Discuz! X3.4

Copyright © 2001-2020, Tencent Cloud.

快速回复 返回顶部 返回列表