Is a cutoff of 10% appropriate for the change-in-estimate criterion of confounder identification?

J Epidemiol. 2014;24(2):161-7. doi: 10.2188/jea.je20130062. Epub 2013 Dec 7.

Abstract

Background: When using the change-in-estimate criterion, a cutoff of 10% is commonly used to identify confounders. However, the appropriateness of this cutoff has never been evaluated. This study investigated cutoffs required under different conditions.

Methods: Four simulations were performed to select cutoffs that achieved a significance level of 5% and a power of 80%, using linear regression and logistic regression. A total of 10 000 simulations were run to obtain the percentage differences of the 4 fitted regression coefficients (with and without adjustment).

Results: In linear regression, larger effect size, larger sample size, and lower standard deviation of the error term led to a lower cutoff point at a 5% significance level. In contrast, larger effect size and a lower exposure-confounder correlation led to a lower cutoff point at 80% power. In logistic regression, a lower odds ratio and larger sample size led to a lower cutoff point at a 5% significance level, while a lower odds ratio, larger sample size, and lower exposure-confounder correlation yielded a lower cutoff point at 80% power.

Conclusions: Cutoff points for the change-in-estimate criterion varied according to the effect size of the exposure-outcome relationship, sample size, standard deviation of the regression error, and exposure-confounder correlation.

MeSH terms

  • Computer Simulation*
  • Confounding Factors, Epidemiologic*
  • Humans
  • Linear Models*
  • Logistic Models*
  • Reproducibility of Results