1
$\begingroup$

I am currently conducting a retrospective chart review where I am interested in the efficacy of certain drugs when it comes to treating a certain disease (let's call the disease "A")

Multiple drugs have been used to treat disease A. However, only two drugs have been so frequently used that an inferential method of analysis would generate a meaningful output.

So I have two drugs, drug X and drug Y, which according to previous research are also the most efficacious drugs when treating disease A. The outcomes are coded in two dichotomous variables where 1=Success and 0=Fail. The first dichotomous variable is whether or not the drug achieved complete symptom control, and the second variable whether the symptoms were relieved significantly (defined as greater than 50% reduction of symptom frequency).

So I have drug X and drug Y with outcomes coded in dichotomous variables and I would like to test the hypothesis that these drugs significantly differ in treatment outcomes (both on complete symptom control and >50% symptom reduction).

I would appreciate any and all help with this problem.

Thank you!

$\endgroup$
1
  • 1
    $\begingroup$ Given observational nature/retrospective nature, some form of propensity score is usually used. Annals of Internal Medicine has a couple of articles that could serve as templates. $\endgroup$
    – charles
    Mar 17, 2014 at 14:15

1 Answer 1

1
$\begingroup$

If this is all the data you have, and it isn't organized by patient, then you could do two chi-square tests.

However, you say you are doing a chart review. In that case, for each patient I would have an outcome: Completely recovered, partly recovered, not recovered. Then I would do an ordinal logistic regression. The independent variables would include the drug (X, Y, none) and sensible covariates for the disease you are considering (drawn from the literature). I would also strongly consider combining these into a propensity score, as @Charles suggested (+1 to him)

$\endgroup$
1
  • 1
    $\begingroup$ Thank you for your answer, Peter. It is appreciated. Yes, the data set is organized by patient and I have other variables which are there to record other outcomes (side effects from the drugs) as well as other possible covariates. I will have to read up on how to generate a propensity score as I am not familiar with it. $\endgroup$
    – Abu Zabi
    Mar 17, 2014 at 15:22

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.