0
$\begingroup$

My project study the association between coinfection by hepatitis virus and survival among HIV-infected patients. Two opposite data exist: some studies show significant worse survival for patients with coinfection by hepatitis virus than those without coinfection; other studies show non-significant of survival between these two groups. I believe it happens not uncommon that inconsistent data regarding the same topic. So, I would like to have some good opinion about how to explain the different results? For instance, different population used in different study could be one factor, the number of event differ, and others?

My question is regarding general problem, not specifically pointing to some studies.

$\endgroup$
2
  • 1
    $\begingroup$ Are you asking about a difference in results between two specific studies, or between the results of survival models in general? In case of the former, please add the references to which studies you refer. In case of the latter, the reasons are/could be legion. My list of suggestions would start with case-mix, outcome incidence and survival time distribution, length of follow-up, confounders included, etc. etc. ...... and end with plain coincidence. $\endgroup$ – IWS May 23 '17 at 13:11
  • $\begingroup$ @IWS I was asking the different results from survival model in general. Although I now have specific articles which showed inconsistent result, I would like to find out on my own based on good suggestions from other people. Thanks. $\endgroup$ – juanli May 23 '17 at 14:01
2
$\begingroup$

There are many possible reasons - pretty much any difference in the samples, the variables, the methods, the sample size and so on will affect results of a survival analysis (or any other method).

But you ask

Two opposite data exist: some studies show significant worse survival for patients with coinfection by hepatitis virus than those without coinfection; other studies show non-significant of survival between these two groups.

These are not opposite results. An opposite result would be if the effect size was positive in one study and negative (and approximately the same magnitude) in the other.

This is a very common error - that is, looking at two studies, one significant and one not - and concluding that they must be opposite or, at least, conflicting. See Andrew Gelman and Hal Stern's article "The Difference between "Significant" and "Not Significant" is not itself, statistically significant". One source for this article is here but Googling it reveals other sources as well.

$\endgroup$

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.