Reader's guide to critical appraisal of cohort studies: 2. Assessing potential for confounding
BMJ 2005; 330 doi: https://doi.org/10.1136/bmj.330.7497.960 (Published 21 April 2005) Cite this as: BMJ 2005;330:960
All rapid responses
Rapid responses are electronic comments to the editor. They enable our users to debate issues raised in articles published on bmj.com. A rapid response is first posted online. If you need the URL (web address) of an individual response, simply click on the response headline and copy the URL from the browser window. A proportion of responses will, after editing, be published online and in the print journal as letters, which are indexed in PubMed. Rapid responses are not indexed in PubMed and they are not journal articles. The BMJ reserves the right to remove responses which are being wilfully misrepresented as published articles or when it is brought to our attention that a response spreads misinformation.
From March 2022, the word limit for rapid responses will be 600 words not including references and author details. We will no longer post responses that exceed this limit.
The word limit for letters selected from posted responses remains 300 words.
Mamdani et al. did an excellent job providing an introduction to confounding.
However, I feel they should have been clear that they were explaining
confounders as defined by the classical criteria. By the operational definition,
a confounding variable is one that appreciably changes the association between
the explanatory and outcome variables. This is particularly important for
dichotomous outcomes. As an example, in a hypothetical study of lung cancer,
smoking and gender, investigators report the following data1:
MALE:
| Smoking | |||||||||||||||||
| Lung Cancer |
|
Odds Ratio = 3.0
FEMALE:
| Smoking | |||||||||||||||||
| Lung Cancer |
|
Odds Ratio = 3.0
By the “classic” definition (and that provided by Mamdani et al.), gender is
not a confounder in the relationship between smoking and lung cancer. This is
because there are the same proportion of smokers in both males and females.
However, if one pools the data across gender, the following results are
seen:
BOTH GENDERS:
| Smoking | |||||||||||||||||
| Lung Cancer |
|
Odds Ratio = 2.3
The crude odds ratio (2.3) is 22% lower than the OR adjusted for
gender, even though gender does not meet the classic definition of confounding.
Using the classic definition, gender would not be included in a risk-adjusting
model to measures the independent association between smoking and lung cancer.
Propensity scores are a method to try to account for differences between
individuals in the comparative groups using multiple covariates. However, an
explanation of the operational definition of confounding provides for greater
risk-adjustment and completes the explanation of confounding.
References:
1. Hauck WW, Neuhaus JM, Kalbfleisch JD,
Anderson S. 1991. A consequence of omitted covariates when estimating odds
ratios. J Clin Epidemiol, 44(1): 77-81.
Competing interests:
None declared
Competing interests: No competing interests
assessing differences in distribution
On p. 960 of their article, the authors correctly note that a
potential confounder that only differs in variation can still be a
confounder. However, when they present their suggestion on how to assess
differences, they only present a statistic (the standardized difference)
that will not show a problem if the difference is primarily in the
variability of the distribution. Since the authors, correctly, do not
like standard tests of statistical significance, presumably they have
something in mind other than a test of whether the variances are equal. I
would be interested in knowing what they have in mind for this situation.
Competing interests:
None declared
Competing interests: No competing interests