2

I'm doing the analysis for a study where we need to assess the effect of surgery A (hand surgery) on condition B (trigger digit), adjusting for confounders via logistic regression.

The problem is that we have data from two data sets:

Set1: All patients performed surgery A and we saw whether they developed condition B.

Set2: All patients have condition B and we reported whether they performed surgery A.

By using each data set separately it's not possible to determine the effect of A on B, isn't it? because in one set all patients has the condition and in the other every one has the surgery.

I was wondering, is it correct to join the two sets? That way we would have a typical case-control situation on which to apply the logistic regression B ~ A + confounders.

Any suggestion?

UPDATE: It has been noticed that putting together the two sets wont help much, since we have no cases for the A-B- condition and therefore the OR will be an artificial 0. How can we solve this. I also thought about something more theoretical. Since we miss the A-B- we could just recruit new patients that fill that position. But this would mean to artificially pump the OR, since it's proportional to the A-B- number. How can new patients be selected in order to maintain external validity?

Bakaburg
  • 2,917
  • Probably not useful. I can tell you that the odds ratio is probably 0 because you have an empty cell for no-surgery and no-condition-B in the 2x2 contingency table. – Penguin_Knight Dec 12 '14 at 17:14
  • well, actually is indeterminate in the all A+ case, because it would be: A+B+/A+B- over A-B+/A-B+ with the second being 0/0 and therefore indefinite. In the all B+ case is the same. – Bakaburg Dec 12 '14 at 18:00
  • 1
    No, I was referring to even you have combined them, you'll have cases in A+B+, A+B-, and A-B+, but you will not have any in A-B-. Odds ratio is ((A-B-)(A+B+))/((A+B-)(A-B+)), hence it's likely zero. The proposed logistic regression is not going to help tremendously. – Penguin_Knight Dec 12 '14 at 18:17
  • yep, I realized this just yesterday... ideas on how to solve? – Bakaburg Dec 14 '14 at 01:32

0 Answers0