0
$\begingroup$

I have a dataset of patient information and I'm looking to find a way to compare two groups of patients and take into account confounding variables. My dataset has an N of ~1500 and I'm looking for a difference between two populations (treated/untreated) and whether they responded to therapy.

I have been using a Fisher's Exact test to asses the significance of the difference between the two populations. However, there are plenty of potentially confounding factors like age, gender, time since diagnosis, etc. Some of these variables are significantly associated with the response variable.

I'm looking for ways to control for these confounding variables. I know how to do it when my response variable is a continuous variable (do a regression analysis and include confounders as terms in the regression).

Can I do that with binary variables? Should I use a logistic regression? Is there something else I should try? Should I make sub-cohorts of the data in-which the confounding variable is constant (ie. only test the Male patients) but then I'll limit my sample size too much.

Any suggestions would be helpful.

$\endgroup$
1
$\begingroup$

Generalized linear models (of which logistic regression is a subset) will provide you with the best tool I can think of for addressing your problem.

$\endgroup$
0
$\begingroup$

Methods that work for linear regression with continuous variables will, in general, also work for logistic regression for categorical variables. This could include adding terms to the model or using propensity scores in various ways.

Are you sure "respond to treatment" is dichotomous?

$\endgroup$
  • $\begingroup$ I know in reality it is not. However, that's the way i is coded in the data I have. $\endgroup$ – JudoWill Feb 4 '13 at 16:29

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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