Subgroups and sensitivity analysis

If there were some types of participant, intervention or outcome you thought were likely to be quite different to the others, you might plan a subgroup analysis. For example, you may want to know if anti-inflammatory medication reduces pain in shoulder disorders, but not all shoulder disorders will respond the same. You may therefore decide to look at individual disorders (eg rotator cuff tendinitis, frozen shoulder, arthritis) as separate subgroups, then combine them in a pooled analysis to gain overall effect for any shoulder pain. A subgroup analysis will allow you to look at results separately and together for that analysis.

You should specify subgroup analyses in advance. This is because choosing subgroups after you have seen the results of studies may introduce bias.

Sensitivity analyses investigate the effect of our decisions in the way we conduct a review. For example, we choose a threshold for the quality of studies, usually only including randomised controlled trials. A sensitivity analysis would allow the investigation of what would happen to the result of the review if we made this criterion even higher by only accepting randomised controlled trials with excellent allocation concealment.

As with subgroup analyses, you should specify sensitivity analyses in advance, to minimise bias.

Remember, though, that the more subgroup or sensitivity analyses you do, the more likely you are to find a statistically significant result by chance. So you should have good reason for each analysis you do, with, ideally, some independent evidence to support this and to predict the direction of any differences you find in these analyses. Always remember that you may be finding falsely significant results and that all findings of subgroup analyses should usually be seen as hypothesis-generating and not as proof in themselves.

Now return to Module 7 and finish designing your data extraction form, incorporating your planned analysis into the section for extracting the results of studies.