0
$\begingroup$

I am looking at how well test scores can predict disease status (case/control). There are 6 tests total, A, B, C, D, E, F. And for tests A-E, a higher score is worse (i.e, a higher score is associated with increased risk of disease). For test F, a higher score is better (i.e., a lower score is associated with increased risk of disease). I constructed a logistic regression model and looked at the odds ratios. However, a few of them are counter-intuitive.

For tests A and B, where higher scores are associated with higher disease risk, the odds ratios are < 1. Meaning that for a one unit increase in score, the odds of having the disease decreases.

For test F, where higher scores are associated with lower disease risk, the odds ratios are > 1, meaning that for a one unit increase in score, the odds of having the disease increases.

Is there a possible explanation for why this is the case?

$\endgroup$
  • $\begingroup$ How big is your sample? $\endgroup$ – robin.datadrivers Apr 13 '15 at 18:41
  • $\begingroup$ @robin.datadrivers About 300. 100 cases and 200 controls $\endgroup$ – Adrian Apr 13 '15 at 19:08
  • 1
    $\begingroup$ And you have done proper data management and cleaning? Have you checked that the tests are scored in the right direction? That there are not crazy outliers? Are the data representative or is it possible there is some bias in your sample? Have you looked at correlations among your data? If you have done all of that, it could be that your theory is simply not supported by the data. $\endgroup$ – robin.datadrivers Apr 13 '15 at 19:52
  • $\begingroup$ Also, check to make sure you don't have a typo- you say for test E a higher score = higher risk, but then also say that "For test E, where higher scores are associated with lower disease risk..." $\endgroup$ – robin.datadrivers Apr 13 '15 at 19:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.