Bayesian Effect Estimation Accounting for Adjustment Uncertainty

Chi Wang, Giovanni Parmigiani, Francesca Dominici

Research output: Contribution to journalArticle

Abstract

Model-based estimation of the effect of an exposure on an outcome is generally sensitive to the choice of which confounding factors are included in the model. We propose a new approach, which we call Bayesian adjustment for confounding (BAC), to estimate the effect of an exposure of interest on the outcome, while accounting for the uncertainty in the choice of confounders. Our approach is based on specifying two models: (1) the outcome as a function of the exposure and the potential confounders (the outcome model); and (2) the exposure as a function of the potential confounders (the exposure model). We consider Bayesian variable selection on both models and link the two by introducing a dependence parameter, ω, denoting the prior odds of including a predictor in the outcome model, given that the same predictor is in the exposure model. In the absence of dependence (ω= 1), BAC reduces to traditional Bayesian model averaging (BMA). In simulation studies, we show that BAC, with ω > 1, estimates the exposure effect with smaller bias than traditional BMA, and improved coverage. We, then, compare BAC, a recent approach of Crainiceanu, Dominici, and Parmigiani (2008, Biometrika95, 635-651), and traditional BMA in a time series data set of hospital admissions, air pollution levels, and weather variables in Nassau, NY for the period 1999-2005. Using each approach, we estimate the short-term effects of on emergency admissions for cardiovascular diseases, accounting for confounding. This application illustrates the potentially significant pitfalls of misusing variable selection methods in the context of adjustment uncertainty.

Original languageEnglish (US)
Pages (from-to)661-671
Number of pages11
JournalBiometrics
Volume68
Issue number3
DOIs
StatePublished - Sep 2012
Externally publishedYes

Fingerprint

Confounding
Uncertainty
Adjustment
uncertainty
Bayesian Model Averaging
Air Pollution
Weather
Emergencies
Cardiovascular Diseases
exposure models
Predictors
Model
Bayesian Variable Selection
Estimate
Selection of Variables
Odds
Time Series Data
Emergency
air pollution
selection methods

Keywords

  • Adjustment uncertainty
  • Bayesian model averaging
  • Exposure effects
  • Treatment effects

ASJC Scopus subject areas

  • Applied Mathematics
  • Statistics and Probability
  • Agricultural and Biological Sciences(all)
  • Biochemistry, Genetics and Molecular Biology(all)
  • Immunology and Microbiology(all)
  • Medicine(all)

Cite this

Bayesian Effect Estimation Accounting for Adjustment Uncertainty. / Wang, Chi; Parmigiani, Giovanni; Dominici, Francesca.

In: Biometrics, Vol. 68, No. 3, 09.2012, p. 661-671.

Research output: Contribution to journalArticle

Wang, Chi ; Parmigiani, Giovanni ; Dominici, Francesca. / Bayesian Effect Estimation Accounting for Adjustment Uncertainty. In: Biometrics. 2012 ; Vol. 68, No. 3. pp. 661-671.
@article{d9c498c189b947f4ab2fafb9ba02f427,
title = "Bayesian Effect Estimation Accounting for Adjustment Uncertainty",
abstract = "Model-based estimation of the effect of an exposure on an outcome is generally sensitive to the choice of which confounding factors are included in the model. We propose a new approach, which we call Bayesian adjustment for confounding (BAC), to estimate the effect of an exposure of interest on the outcome, while accounting for the uncertainty in the choice of confounders. Our approach is based on specifying two models: (1) the outcome as a function of the exposure and the potential confounders (the outcome model); and (2) the exposure as a function of the potential confounders (the exposure model). We consider Bayesian variable selection on both models and link the two by introducing a dependence parameter, ω, denoting the prior odds of including a predictor in the outcome model, given that the same predictor is in the exposure model. In the absence of dependence (ω= 1), BAC reduces to traditional Bayesian model averaging (BMA). In simulation studies, we show that BAC, with ω > 1, estimates the exposure effect with smaller bias than traditional BMA, and improved coverage. We, then, compare BAC, a recent approach of Crainiceanu, Dominici, and Parmigiani (2008, Biometrika95, 635-651), and traditional BMA in a time series data set of hospital admissions, air pollution levels, and weather variables in Nassau, NY for the period 1999-2005. Using each approach, we estimate the short-term effects of on emergency admissions for cardiovascular diseases, accounting for confounding. This application illustrates the potentially significant pitfalls of misusing variable selection methods in the context of adjustment uncertainty.",
keywords = "Adjustment uncertainty, Bayesian model averaging, Exposure effects, Treatment effects",
author = "Chi Wang and Giovanni Parmigiani and Francesca Dominici",
year = "2012",
month = "9",
doi = "10.1111/j.1541-0420.2011.01731.x",
language = "English (US)",
volume = "68",
pages = "661--671",
journal = "Biometrics",
issn = "0006-341X",
publisher = "Wiley-Blackwell",
number = "3",

}

TY - JOUR

T1 - Bayesian Effect Estimation Accounting for Adjustment Uncertainty

AU - Wang, Chi

AU - Parmigiani, Giovanni

AU - Dominici, Francesca

PY - 2012/9

Y1 - 2012/9

N2 - Model-based estimation of the effect of an exposure on an outcome is generally sensitive to the choice of which confounding factors are included in the model. We propose a new approach, which we call Bayesian adjustment for confounding (BAC), to estimate the effect of an exposure of interest on the outcome, while accounting for the uncertainty in the choice of confounders. Our approach is based on specifying two models: (1) the outcome as a function of the exposure and the potential confounders (the outcome model); and (2) the exposure as a function of the potential confounders (the exposure model). We consider Bayesian variable selection on both models and link the two by introducing a dependence parameter, ω, denoting the prior odds of including a predictor in the outcome model, given that the same predictor is in the exposure model. In the absence of dependence (ω= 1), BAC reduces to traditional Bayesian model averaging (BMA). In simulation studies, we show that BAC, with ω > 1, estimates the exposure effect with smaller bias than traditional BMA, and improved coverage. We, then, compare BAC, a recent approach of Crainiceanu, Dominici, and Parmigiani (2008, Biometrika95, 635-651), and traditional BMA in a time series data set of hospital admissions, air pollution levels, and weather variables in Nassau, NY for the period 1999-2005. Using each approach, we estimate the short-term effects of on emergency admissions for cardiovascular diseases, accounting for confounding. This application illustrates the potentially significant pitfalls of misusing variable selection methods in the context of adjustment uncertainty.

AB - Model-based estimation of the effect of an exposure on an outcome is generally sensitive to the choice of which confounding factors are included in the model. We propose a new approach, which we call Bayesian adjustment for confounding (BAC), to estimate the effect of an exposure of interest on the outcome, while accounting for the uncertainty in the choice of confounders. Our approach is based on specifying two models: (1) the outcome as a function of the exposure and the potential confounders (the outcome model); and (2) the exposure as a function of the potential confounders (the exposure model). We consider Bayesian variable selection on both models and link the two by introducing a dependence parameter, ω, denoting the prior odds of including a predictor in the outcome model, given that the same predictor is in the exposure model. In the absence of dependence (ω= 1), BAC reduces to traditional Bayesian model averaging (BMA). In simulation studies, we show that BAC, with ω > 1, estimates the exposure effect with smaller bias than traditional BMA, and improved coverage. We, then, compare BAC, a recent approach of Crainiceanu, Dominici, and Parmigiani (2008, Biometrika95, 635-651), and traditional BMA in a time series data set of hospital admissions, air pollution levels, and weather variables in Nassau, NY for the period 1999-2005. Using each approach, we estimate the short-term effects of on emergency admissions for cardiovascular diseases, accounting for confounding. This application illustrates the potentially significant pitfalls of misusing variable selection methods in the context of adjustment uncertainty.

KW - Adjustment uncertainty

KW - Bayesian model averaging

KW - Exposure effects

KW - Treatment effects

UR - http://www.scopus.com/inward/record.url?scp=84866768287&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84866768287&partnerID=8YFLogxK

U2 - 10.1111/j.1541-0420.2011.01731.x

DO - 10.1111/j.1541-0420.2011.01731.x

M3 - Article

C2 - 22364439

AN - SCOPUS:84866768287

VL - 68

SP - 661

EP - 671

JO - Biometrics

JF - Biometrics

SN - 0006-341X

IS - 3

ER -