Effects of Adjusting for Instrumental Variables on Bias and Precision of Effect Estimates

Division of Pharmacoepidemiology and Pharmacoeconomics, Department of Medicine, Brigham and Women’s Hospital and Harvard Medical School, Boston, MA 02120, USA.
American journal of epidemiology (Impact Factor: 5.23). 12/2011; 174(11):1213-22. DOI: 10.1093/aje/kwr364
Source: PubMed


Recent theoretical studies have shown that conditioning on an instrumental variable (IV), a variable that is associated with exposure but not associated with outcome except through exposure, can increase both bias and variance of exposure effect estimates. Although these findings have obvious implications in cases of known IVs, their meaning remains unclear in the more common scenario where investigators are uncertain whether a measured covariate meets the criteria for an IV or rather a confounder. The authors present results from two simulation studies designed to provide insight into the problem of conditioning on potential IVs in routine epidemiologic practice. The simulations explored the effects of conditioning on IVs, near-IVs (predictors of exposure that are weakly associated with outcome), and confounders on the bias and variance of a binary exposure effect estimate. The results indicate that effect estimates which are conditional on a perfect IV or near-IV may have larger bias and variance than the unconditional estimate. However, in most scenarios considered, the increases in error due to conditioning were small compared with the total estimation error. In these cases, minimizing unmeasured confounding should be the priority when selecting variables for adjustment, even at the risk of conditioning on IVs.

Download full-text


Available from: Marshall M Joffe,
1 Follower
37 Reads
    • "Methodological development in the application of instrumental variables and propensity score approaches is an area of active ongoing research (Austin, 2013; McCaffrey et al., 2013; Myers et al., 2011). As with any statistical technique, these methods can introduce bias if implemented incorrectly or if the instrumental variables are not well selected (Austin, 2008; Myers et al., 2011). Thus, investigators are encouraged to consult the most recent statistical literature when applying these methods. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Over the past 60 years, the view that "alcoholism" is a disease for which the only acceptable goal of treatment is abstinence has given way to the recognition that alcohol use disorders (AUDs) occur on a continuum of severity, for which a variety of treatment options are appropriate. However, because the available treatments for AUDs are not effective for everyone, more research is needed to develop novel and more efficacious treatments to address the range of AUD severity in diverse populations. Here we offer recommendations for the design and analysis of alcohol treatment trials, with a specific focus on the careful conduct of randomized clinical trials of medications and nonpharmacological interventions for AUDs. This paper provides a narrative review of the quality of published clinical trials and recommendations for the optimal design and analysis of treatment trials for AUDs. Despite considerable improvements in the design of alcohol clinical trials over the past 2 decades, many studies of AUD treatments have used faulty design features and statistical methods that are known to produce biased estimates of treatment efficacy. The published statistical and methodological literatures provide clear guidance on methods to improve clinical trial design and analysis. Consistent use of state-of-the-art design features and analytic approaches will enhance the internal and external validity of treatment trials for AUDs across the spectrum of severity. The ultimate result of this attention to methodological rigor is that better treatment options will be identified for patients with an AUD. Copyright © 2015 by the Research Society on Alcoholism.
    Alcoholism Clinical and Experimental Research 08/2015; 39(9). DOI:10.1111/acer.12800 · 3.21 Impact Factor
  • Source
    • "Moreover, when the outcome of interest is not rare and the number of potential confounders is moderate, they tend to provide results often indistinguishable from those provided by traditional multivariate techniques [20]. The results may be even more biased than those provided by other methods when, in presence of relatively strong unmeasured confounders, variables associated with the exposure but not associated with the outcome are erroneously included in the model used to estimate the PS [21,22]. "
    [Show abstract] [Hide abstract]
    ABSTRACT: Epidemiological and other evidence strongly supports the hypothesis that problem drinking is causally related to the incidence of active tuberculosis and the worsening of the disease course. The presence of a large number of potential confounders, however, complicates the assessment of the actual size of this causal effect, leaving room for a substantial amount of bias. This study aims to contribute to the understanding of the role of confounding in the observed association between problem drinking and tuberculosis, assessing the effect of the adjustment for a relatively large number of potential confounders on the estimated prevalence odds ratio of tuberculosis among problem drinkers vs. moderate drinkers/abstainers in a cross-sectional, nationally representative sample of the South African adult population. A propensity score approach was used to match each problem drinker in the sample with a subset of moderate drinkers/abstainers with similar characteristics in respect to a set of potential confounders. The prevalence odds ratio of tuberculosis between the matched groups was then calculated using conditional logistic regression. Sensitivity analyses were conducted to assess the robustness of the results in respect to misspecification of the model. The prevalence odds ratio of tuberculosis between problem drinkers and moderate drinkers/abstainers was 1.97 (95% CI: 1.40 to 2.77), and the result was robust with respect to the matching procedure as well as to incorrect adjustment for potential mediators and to the possible presence of unmeasured confounders. Sub-population analysis did not provide noteworthy evidence for the presence of interaction between problem drinking and the observed confounders. In a cross-sectional national survey of the adult population of a middle income country with high tuberculosis burden, problem drinking was associated with a two fold increase in the odds of past TB diagnosis after controlling for a large number of socio-economic and biological confounders. Within the limitations of a cross-sectional study design with self-reported tuberculosis status, these results adds to previous evidence of a causal link between problem drinking and tuberculosis, and suggest that the observed higher prevalence of tuberculosis among problem drinkers commonly found in population studies cannot be attributed to the confounding effect of the uneven distribution of other risk factors.
    BMC Public Health 09/2013; 13(1):871. DOI:10.1186/1471-2458-13-871 · 2.26 Impact Factor
  • Source
    • "The researcher who eliminates some unmeasured confounders through the use of an instrumental variable such as physician identity may also remove major sources of pseudorandom variation through the same maneuver [5]. If unmeasured confounders remain after matching on an instrument , they may be the only remaining determinants of variation in treatment within the matched group. "
    [Show abstract] [Hide abstract]
    ABSTRACT: To illustrate that matching on provider may exacerbate, not remove, bias. The degree of confounding bias depends in part on the proportions of treatment variation that can be ascribed to confounders and to instruments, respectively. This commentary raises the specific example of bias by matching on hospital induced in a study of coronary artery bypass graft surgery patients and illustrates the effect of matching on provider in a constructed example. Matching on provider removes a "benign" source of treatment variability, leaving unmeasured confounders as potentially the most important determinants of treatment. Researchers need to articulate the presumed source of pseudorandom variation in observational studies and need to take care not to reduce their effect through unnecessary control.
    Journal of clinical epidemiology 08/2013; 66(8 Suppl):S65-8. DOI:10.1016/j.jclinepi.2013.02.012 · 3.42 Impact Factor
Show more