7
$\begingroup$

I am doing data analysis on a clinical trial. I know the rule here is to do analysis according to intention to treat (ITT, ie. analyse according to the group to which the individuals were randomised). But I still need to perform a per-protocol analysis (i.e. analyse as per RECEIPT of intervention).

Suppose the total sample size is 100, with 50 in the intervention group; and 50 in control group.

By the end of trial, for the intervention group, in fact, only 40 patients received intervention. I need to perform a per-protocol analysis here.

It seems like the per-protocol analysis should compare the 40 patients (who really received intervention) and the 50 control group.

The question is can I move the 10 people (who did not receive intervention but in intervention group by randomization) to control group which can improve my statistical power (by increasing the sample size)?

i.e can I compare the 40 patients who really get the intervention with the 60 patients who did not receive the intervention?

Improve this question
$\endgroup$

1 Answer 1

Reset to default
3
$\begingroup$

Your per-protocol analysis should compare the 40 intervention group patients who received the intervention to the 50 control group patients, as randomised. This may diminish statistical power, but it preserves per-protocol validity and reliability. Moving participants between groups after randomisation can bias the results.

Improve this answer
$\endgroup$
1
  • $\begingroup$ Does it really preserve per-protocol validity? I don’t think the protocol would state that some patient should leave a treatment arm after time zero. Post time zero exclusions create major biases most of the time, so either this should not be called per-protocol or we need to do away with per-protocol analyses. Modified intent-to-treat is probably more useful, e.g., include only subjects having had the first opportunity to provide data post-randomization (e.g. made it to Visit 1). $\endgroup$
    – Frank Harrell
    Commented Nov 11, 2023 at 15:38

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.