A flexible, interpretable framework for assessing sensitivity to unmeasured confounding

Vincent Dorie, Masataka Harada, Nicole Bohme Carnegie, Jennifer Hill

Research output: Contribution to journalArticle

Abstract

When estimating causal effects, unmeasured confounding and model misspecification are both potential sources of bias. We propose a method to simultaneously address both issues in the form of a semi-parametric sensitivity analysis. In particular, our approach incorporates Bayesian Additive Regression Trees into a two-parameter sensitivity analysis strategy that assesses sensitivity of posterior distributions of treatment effects to choices of sensitivity parameters. This results in an easily interpretable framework for testing for the impact of an unmeasured confounder that also limits the number of modeling assumptions. We evaluate our approach in a large-scale simulation setting and with high blood pressure data taken from the Third National Health and Nutrition Examination Survey. The model is implemented as open-source software, integrated into the treatSens package for the R statistical programming language.

Original languageEnglish (US)
Pages (from-to)3453-3470
Number of pages18
JournalStatistics in Medicine
Volume35
Issue number20
DOIs
StatePublished - Sep 10 2016

Fingerprint

Programming Languages
Parameter Sensitivity
Bayes Theorem
Confounding
Nutrition Surveys
Sensitivity Analysis
Software
Hypertension
Regression Tree
Causal Effect
Parametric Analysis
Model Misspecification
Open Source Software
Nutrition
Blood Pressure
Treatment Effects
Posterior distribution
Bayesian Approach
Two Parameters
Health

Keywords

  • Bayesian modeling
  • causal inference
  • nonparametric regression
  • sensitivity analysis
  • unmeasured confounding

ASJC Scopus subject areas

  • Epidemiology
  • Statistics and Probability

Cite this

A flexible, interpretable framework for assessing sensitivity to unmeasured confounding. / Dorie, Vincent; Harada, Masataka; Carnegie, Nicole Bohme; Hill, Jennifer.

In: Statistics in Medicine, Vol. 35, No. 20, 10.09.2016, p. 3453-3470.

Research output: Contribution to journalArticle

Dorie, Vincent ; Harada, Masataka ; Carnegie, Nicole Bohme ; Hill, Jennifer. / A flexible, interpretable framework for assessing sensitivity to unmeasured confounding. In: Statistics in Medicine. 2016 ; Vol. 35, No. 20. pp. 3453-3470.
@article{7e96113ed3814b5ab8c334cd98e5011f,
title = "A flexible, interpretable framework for assessing sensitivity to unmeasured confounding",
abstract = "When estimating causal effects, unmeasured confounding and model misspecification are both potential sources of bias. We propose a method to simultaneously address both issues in the form of a semi-parametric sensitivity analysis. In particular, our approach incorporates Bayesian Additive Regression Trees into a two-parameter sensitivity analysis strategy that assesses sensitivity of posterior distributions of treatment effects to choices of sensitivity parameters. This results in an easily interpretable framework for testing for the impact of an unmeasured confounder that also limits the number of modeling assumptions. We evaluate our approach in a large-scale simulation setting and with high blood pressure data taken from the Third National Health and Nutrition Examination Survey. The model is implemented as open-source software, integrated into the treatSens package for the R statistical programming language.",
keywords = "Bayesian modeling, causal inference, nonparametric regression, sensitivity analysis, unmeasured confounding",
author = "Vincent Dorie and Masataka Harada and Carnegie, {Nicole Bohme} and Jennifer Hill",
year = "2016",
month = "9",
day = "10",
doi = "10.1002/sim.6973",
language = "English (US)",
volume = "35",
pages = "3453--3470",
journal = "Statistics in Medicine",
issn = "0277-6715",
publisher = "John Wiley and Sons Ltd",
number = "20",

}

TY - JOUR

T1 - A flexible, interpretable framework for assessing sensitivity to unmeasured confounding

AU - Dorie, Vincent

AU - Harada, Masataka

AU - Carnegie, Nicole Bohme

AU - Hill, Jennifer

PY - 2016/9/10

Y1 - 2016/9/10

N2 - When estimating causal effects, unmeasured confounding and model misspecification are both potential sources of bias. We propose a method to simultaneously address both issues in the form of a semi-parametric sensitivity analysis. In particular, our approach incorporates Bayesian Additive Regression Trees into a two-parameter sensitivity analysis strategy that assesses sensitivity of posterior distributions of treatment effects to choices of sensitivity parameters. This results in an easily interpretable framework for testing for the impact of an unmeasured confounder that also limits the number of modeling assumptions. We evaluate our approach in a large-scale simulation setting and with high blood pressure data taken from the Third National Health and Nutrition Examination Survey. The model is implemented as open-source software, integrated into the treatSens package for the R statistical programming language.

AB - When estimating causal effects, unmeasured confounding and model misspecification are both potential sources of bias. We propose a method to simultaneously address both issues in the form of a semi-parametric sensitivity analysis. In particular, our approach incorporates Bayesian Additive Regression Trees into a two-parameter sensitivity analysis strategy that assesses sensitivity of posterior distributions of treatment effects to choices of sensitivity parameters. This results in an easily interpretable framework for testing for the impact of an unmeasured confounder that also limits the number of modeling assumptions. We evaluate our approach in a large-scale simulation setting and with high blood pressure data taken from the Third National Health and Nutrition Examination Survey. The model is implemented as open-source software, integrated into the treatSens package for the R statistical programming language.

KW - Bayesian modeling

KW - causal inference

KW - nonparametric regression

KW - sensitivity analysis

KW - unmeasured confounding

UR - http://www.scopus.com/inward/record.url?scp=84964922397&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84964922397&partnerID=8YFLogxK

U2 - 10.1002/sim.6973

DO - 10.1002/sim.6973

M3 - Article

C2 - 27139250

AN - SCOPUS:84964922397

VL - 35

SP - 3453

EP - 3470

JO - Statistics in Medicine

JF - Statistics in Medicine

SN - 0277-6715

IS - 20

ER -