Hello,
I am trying to adjust for the confounding effect of a third variable on the association between ethnicity (has multiple categories) and death (binary). I am using fixed effect conditional logistic regression to build multivariable model. I know that for a factor to be considered an important confounder it has to change the crude odds ratio by more than 10% (besides the other criteria of being associated with the exposure and outcome).
However, in case I have many categories for the exposure, how can I know if a third factor is an important confounder? Should it change the odds ratio of "ALL CATEGORIES" by 10% or more, or even a change in "one out of all categories" makes it an important confounder? or is there another more appropriate way to deal with the situation?
Have you tried adding a potential confounder to your regression?. This should give you a idea about that. Without information about your proposed confounder I have no idea what you should do but it is usually a bad idea to adjust regression output blindly. Good luck, David Booth ps if you have data you might try plotting to see if you proposed confounder has any effect.
David Eugene Booth Per Lytsy
Thank you for your replies.
The third variable, e.g. mental illness, is thought to be a confounder, as it is statistically significantly associated with both the exposure ethnicity and outcome death. When the third factor is added to the model, it changes the OR by more than 10% for some, but not all, ethnicity categories.
Danah Alothman wrote: "I know that for a factor to be considered an important confounder it has to change the crude odds ratio by more than 10% (besides the other criteria of being associated with the exposure and outcome)."
That is someone's rule of thumb, not a statistical law. Can you tell us the source?
As to your original question, how much the OR changes in each category will surely depend on which category is used as the reference category. So unless it is very clear to all concerned which category is the reference category, you may need a new rule of thumb for categorical variables with 3 or more categories. HTH.
I think the real question is do you have data on the potential confounding variables? If you do then there are things you can do. If not, then you're pretty much stuck with what you have. Best wishes, David Booth
BTW if you google how do you study confounding variables
You can get some helpful ideas.However, in order to do this the potential confounders must have been included in the experiment.
Best wishes, David Booth
Bruce Weaver
Yes correct, I decided the threshold to be at 10% or more change in OR. It's not a statistical law but a rule of thumb.
I think I will just add a table with which categories were effected and interpret results per category. Thank you for you input. appreciated.
Danah
David Eugene Booth
Yes I do have data about the potential confounders. Thank you for your input. Appreciated.
Best wishes
Danah
As mentioned, you must identify the role of variables. They may be covariate, confounder, moderator, or mediator (both mathematically and biologically). However, if you are sure about their roles, 10% is a rule of thumb. I suggest applying some post-hoc tests and criteria like Bradford-Hill.
- Badges
- Science topic
- Similar topics
- Social Psychology
- Attitude