Original article
Bias due to non-differential misclassification of polytomous confounders

https://doi.org/10.1016/0895-4356(93)90009-PGet rights and content

Abstract

This paper addresses potential effects of non-differential misclassification of polytomous confounders on adjusted exposure-disease associations. Although the degree of confounder-misclassification bias heavily depends on the relative distribution of the confounding variable among the compared exposure groups and the misclassification pattern, in most cases the bias is in the same direction (though to a lesser degree) than the confounding, i.e. the observed adjusted measures lie between the crude and the fully adjusted measures. In some instances, however, the confounder misclassification bias may be in the opposite direction. This is in contrast to previous understanding that non-differential confounder misclassification always tends to bias adjusted effect estimates towards the crude estimates and that the extent of this bias has a stable relationship to the degree of misclassification. Consequently, conclusions on the potential effects of non-differential misclassification of a polytomous confounder in any given study should only be made after careful sensitivity analyses which consider plausible ranges of misclassification rates.

References (13)

  • R. Cederlöf et al.

    Air pollution and cancer: risk assessment methodology and epidemiological evidence

    Environ Health Perspect

    (1978)
  • J.L. Kelsey et al.

    Methods in Observational Epidemiology

    (1986)
  • K.H. Rothman

    Modern Epidemiology

    (1986)
  • D.A. Savitz et al.

    Estimating and correcting for confounder misclassification

    Am J Epidemiol

    (1989)
  • S. Greenland

    The effect of misclassification in the presence of covariates

    Am J Epidemiol

    (1980)
  • A. Tzonou et al.

    Misclassification in case-control studies with two dichotomous risk factors

    Rev Epidemiol Sante Publique

    (1986)
There are more references available in the full text version of this article.

Cited by (37)

  • Noncollapsibility in studies based on nonrepresentative samples

    2015, Annals of Epidemiology
    Citation Excerpt :

    Often, in a specific population, the risk factor that is introducing noncollapsibility problems is also a confounder (still assuming no effect modification). Again, in this scenario, the best approach is to control for the risk factor, but if this is not possible, a study selected on the risk factor is likely to be less affected by confounding because of partial control of the confounder and therefore, at least when the risk factor is binary [14,15], is expected to produce a marginal estimate closer to the true conditional effect than the corresponding unselected study. As we have illustrated, the overall gain in validity depends on the combination of the effects that the selection has on noncollapsibility and control of confounding.

  • Outcome Measures, Interim Analyses, and Bayesian Approaches to Randomized Trials. Answers to the September 2009 Journal Club Questions

    2010, Annals of Emergency Medicine
    Citation Excerpt :

    Misclassification bias occurs when a patient passes a stone but fails to recognize spontaneous stone passage and is erroneously classified in the “No stone passed” group or vice versa. If the misclassification occurs equally between the tamsulosin and control groups, the intervention's effect is diluted toward the null hypothesis, or no difference.13 If it occurs more commonly in one group, then the study might not detect a true association or report a spurious one.14

  • Accounting for Expected Adjusted Effect

    2020, Frontiers in Psychology
View all citing articles on Scopus
View full text