6
$\begingroup$

This might seem a silly question but I am really confused about it. In theory adjusting for a confounder variable should remove its effect. Is this always true? and does this mean that the effect of this confounder variable we adjusted for has been completely removed?

For example we know that the number of comorbidities increases with age. If we adjust for age does this mean that the age is unlikely to explain increase in comorbidities even partly?

$\endgroup$

1 Answer 1

9
$\begingroup$

I don't have a complete answer but can provide some thoughts:

1) Adjustment does remove the confounding effect, but only if the underlying causal pathways are correctly specified. There are occasions where adjustment can cause bias rather than decreasing biases. For more information on this issue, search for collider bias and directed acyclic graphs.

2) Adjustment does remove the confounding effect, but only if the operationalization is correct. In other words, you have chosen the correct variable to represent the construct. There are multiple reasons why age may not be a good indicator of aging (the actual construct that is related to mortality.) For example, fatal heart disease can be lifestyle-related. It can also be related to immunological response and how body mediates inflammation. All these factors can have substantial difference within the same age. In the reporting side, under-report of one's age tends to go up with age, introducing some error that is correlated with age as well. If you control for age, and thinking you have controlled for age-related factors, chance is this assumption is usually over-ambitious. It's always more important to know what the control variables really means.

3) There are also other dynamics which can cause adjustment alone to be insufficient. For example, interaction between age and other variable(s) in the model can bias the estimate of age. Non-linear relationship between age and mortality can also cause simple adjustment for age alone an imperfect method.

My guess is in epidemiology, it's better to say "no" whenever someone asks if something can completely removed whatever... perhaps except "can randomized controlled trial completely remove biases?" Then "theoretically yes."

$\endgroup$
4
  • $\begingroup$ Thanks Penguin_Knight for these very useful thoughts. I am wondering if you or anyone has a resource/ published example to explain more the second point (i.e. that the operationalization of the relation should be correct)? $\endgroup$
    – user21814
    Mar 10, 2013 at 17:36
  • $\begingroup$ +1, this is a really nice contribution; short, simple, clear. $\endgroup$ Mar 10, 2013 at 19:01
  • $\begingroup$ @James It appears you are using two different (unregistered) accounts. Please, follow the instructions on our help page to merge them and don't forget to register your account once and for all. $\endgroup$
    – chl
    Mar 10, 2013 at 20:04
  • $\begingroup$ @James Unfortunately I don't know of any books in the biomedical domain. But if you search for books on questionnaire design in the field of social science, you can get a good introduction on operationalization. It would really get us think clearly about what a variable is and isn't. That cautiousness is definitely transferable to biomedical field. At the end of the day, the second best my be the best you can get. But still it's good to know its limitations. $\endgroup$ Mar 11, 2013 at 3:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.