3

I work in a medical environment. I have a situation where I do not know which statistical test would fit well with what I want to find out and thought you might be able to give me guidance.

I am investigating the mortality linked to an infection picked up by some patients, and whether the likelihood of site (blood, urine, and sputum) where the bug was found, is significant enough amongst those who died and those who survived.

I am not sure whether a chi-squared test is the right one, not that I remember how to use it either!

Essentially my population size is 74, and there have been 22 deaths, and most of these were ill with other co-morbidities. The infection is essentially an opportunistic infection. Literature suggests however that those who have it in their bloodstream or part of a pneumonia are more likely to die than those who might have it as a wound infection etc. I am keen to see if my small sample size correlates with this finding.

                  Death   (22)                                 Discharged (52)
Sputum            22%      (5)                                 15%        (15)
Blood             31%      (7)                                 13%         (7)
Urine             22%      (5)                                 31%        (16)

Is there any statistical significance linked to site of infection?

kjetil b halvorsen
  • 77,844
Francois Botha
  • 131
  • 1
    Why not start from some simpler and probably easier to understand relative risk or odds ratio? If individual site will be considered as diagnostic information I don't see why pooling them in a chi-square will help. Yes, you will know that site of discovery matters, but the next question will then be which site. – Penguin_Knight Oct 03 '13 at 12:59
  • Interesting, a pity you have so few data, I'm not sure how much you can learn from them. I think you first need to phrase your question in term of a null hypothesis which you would like to test. Maybe the null hypothesis could be the probability of dying via blood P(D|B) is same as probability of dying via sputum or urine P(D|S OR U). Then the job would be to proove that given your data this assertion is rejected (for example at 95% CL). Would you agree with that ? – Mr Renard Oct 03 '13 at 13:40
  • 1
    Yes, you can use a chi-squared test for independence to determine if Death vs. Discharged differs by site. There is a recent discussion of the chi-squared test here: Can you use the chi-squared test when the expected values are not determined? – gung - Reinstate Monica Oct 03 '13 at 13:40
  • A bit overdue, but thanks to everyone for the valuable contributions. – Francois Botha Feb 04 '14 at 11:24
  • Could also be framed as a logistic regression – kjetil b halvorsen Feb 13 '21 at 02:54

0 Answers0