|
Activity Number:
|
177
|
|
Type:
|
Topic Contributed
|
|
Date/Time:
|
Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
|
|
Sponsor:
|
Section on Health Policy Statistics
|
| Abstract - #301173 |
|
Title:
|
Instrumental Variable Validation and Covariate Balance
|
|
Author(s):
|
Thomas E. Love*+ and Randall D. Cebul and Charles Thomas and Neal V. Dawson
|
|
Companies:
|
Case Western Reserve University and Case Western Reserve University and Case Western Reserve University and Case Western Reserve University
|
|
Address:
|
Center for Health Care Res. and Policy, Cleveland, OH, 44122,
|
|
Keywords:
|
observational studies ; health services research ; instrumental variables ; hidden bias ; selection bias ; health policy
|
|
Abstract:
|
This paper describes the importance of assumptions related to balancing covariates in instrumental variables analyses to infer causality in observational studies. In a previously published analysis of the effect of right heart catheterization (RHC) among seriously ill patients (SUPPORT study), RHC use significantly increased 30-day mortality rates. Bivariate comparisons suggest a candidate instrument: hospitals had different levels of RHC exposure but similar 30-day mortality rates. If hospital site is a valid instrument, stratification by site permits estimation of a local average treatment effect of RHC on mortality in marginal patients (who would get RHC only at high-RHC hospitals). While bivariate comparisons of the candidate instrument to exposure and outcome are one step in validating site as an instrument, we must also explore whether patients at the two sites differ in terms of observed (and unobserved) health characteristics. Researchers selecting instruments to estimate causal effects need to demonstrate balance in key covariates related to outcome. In particular, bivariate comparisons of instrument to exposure and outcome are insufficient to assess validity.
|
