Maybe someone can help me or at least give me some clues.

I have 1057 patients all with different types of prostheses. One of them seems to have a higher revision rate than the others.

total numbers:
prosthesisA 662
prosthesisB 162
prosthesisC 151
other       82

revision rate:
prosthesisA 9   1,36%
prosthesisB 11  6,79%
prosthesisC 3   1,99%
other       4   4,88%

Could anyone tell me how I can show whether the revision rate of prothesisB (6,79%) is significantly higher or not compared to the other prostheses (incl. "other")?

Thank you & kind regards

Edit: Is it possible to compare only the prosthesisB with all others at once? I've created the following contingency table:

        prostB  other   total
Rev     11      16      27
noRev   162     868     1030
        173     884     1057

Odds-Ratio = (11*868)/(16*162) = 3.68
p-Value (Fisher Test) = 0.001

Can I say now that the chances for a revision with prosthesisB are ~3.5 times higher compared to all others? Is this plausible since the p-Value of the Fisher's exact test is highly significant?

  • $\begingroup$ When you say "compared to the others", do you mean "compared to prosthesisA, prosthesisC and other", or "compared to other"? $\endgroup$
    – Henry
    Commented Sep 17, 2011 at 23:03
  • $\begingroup$ Yeah, sorry. I should have stated it clearer. I mean compared to all other prosthesis including "other". $\endgroup$
    – TheLostOne
    Commented Sep 17, 2011 at 23:19

1 Answer 1


You should be careful about formulating your hypotheses after seeing the data. Instead of this, what I would do is a logistic regression with the IV being body part.

  • $\begingroup$ I can probably only perform the logistic regression if I have the raw data, correct? $\endgroup$
    – TheLostOne
    Commented Sep 20, 2011 at 17:34
  • $\begingroup$ I've added an own approach to the original answer. Do you think it's correct? Thank you! $\endgroup$
    – TheLostOne
    Commented Sep 20, 2011 at 19:03
  • $\begingroup$ Yes, you need the raw data to do the logistic. Your approach does what I said to be careful about. You can DO it, but it's a bit of fishing, and the resultant p-values are not correct. $\endgroup$
    – Peter Flom
    Commented Sep 23, 2011 at 11:29
  • $\begingroup$ Thank you Peter! I think my solution should do it for now. $\endgroup$
    – TheLostOne
    Commented Sep 25, 2011 at 22:00

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.