Background: Swallowing exercises are prescribed to head and neck cancer patients during radiotherapy to prevent swallowing impairment

Aim: Determine impact of swallowing exercise compliance on swallowing impairment

Method: Retrospective review of database between 2015 and 2020. We included everyone which met the criteria within that period


non-compliant compliant
Swallowing impairment 7 25
No swallowing impairment 4 8

OR=0.56 (p=0.50) - using fisher exact test


When examining differences in baseline characteristics (using fisher exact test), we found significant association (p=0.016) between compliance and T-staging. There was a larger proportion of T1 patients in the compliant group vs the non-compliant group. Which is not unexpected as patients with smaller tumors experience less side effects and thus is more likely to comply to the exercises.

So do we still need to account for this if we have already determined there is no significant association between compliance and swallowing impairment?

If so, I understand we can account for it either through regression or stratified analysis. Given the small sample size, regression is not the best option. So I would think stratified analysis using Cochran-Mantel-Haenszel test.

For the Cochran-Mantel-Haenszel test, what do we report? Just the common OR vs the crude OR, the p-value, and whether there is a confounding effect?

  • $\begingroup$ Is the first p-value .05? “So do we still need to account for this if we have already determined there is no [is this correct?] significant association...”. I find confidence interval reporting is usually helpful, and helps the reader to get a quicker understanding. $\endgroup$ Commented Mar 5, 2021 at 7:04
  • $\begingroup$ We used a Fisher exact test to determine the association between compliance and swallowing impairment, and the p-value of that test is 0.50. OR=0.56, 95% CI=0.13-2.42 $\endgroup$
    – user224743
    Commented Mar 5, 2021 at 7:46
  • $\begingroup$ Thanks. Just looking at the table without doing the calculation I wrongly thought was a mistake. $\endgroup$ Commented Mar 5, 2021 at 11:32

1 Answer 1


Just so I get this right:

  • The study is intended to assess the effect of exercise compliance on swallowing impairment.

  • T staging is a confounder (I don't know what T1 is, you're going to have to explain that) because size of the tumor affects the ease of the exercise, and size of the tumor also affects the impairment (larger meaning more likely to be impaired).

First off, I don't think it is a good idea to test for confounding using a statistical test. That T staging is a confounder is something that you should have accounted for even before getting data. So either you suspect this is a confounder (classic confounding case, X<-Z->Y and X->Y) and you should account for it, or you don't suspect it to be a confounder.

If you suspect it to be a confounder then CMH test sounds fine to me (assuming T staging is a truly discrete variable with not too many categories as opposed to some continuous variable which has been dichotomized). If you perform the test, report the group odds ratios with confidence intervals, the test statistic for the CMH test, the degrees of freedom of the test, and the p value.

  • $\begingroup$ Thank you for the help! Yes, you are right on the first bullet point. As for the 2nd bullet point, I agree too that T-staging is a confounder. T-staging is split into 4 levels (T1,T2,T3,T4), depending on the size of the tumor (T1 being the smallest and T4 being the largest). Btw, does "group odds ratio" refer to the common odds ratio or the stratum-specific odd-ratio? $\endgroup$
    – user224743
    Commented Mar 5, 2021 at 7:32
  • $\begingroup$ @user224743 group odds ratio refers to the stratum specific odds ratios $\endgroup$ Commented Mar 6, 2021 at 16:40

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.