Suppose an RCT, with 30 participants in each arm. Group A receives CBT sessions, Group B as a waitlist control does not. Both groups complete a depression questionnaire measure at baseline, and after Group A have had their CBT, both groups complete the questionnaire again. Te change in depression scores is the outcome of interest. However, let's say 5 individuals from Group A drop out, and so do not have the second time-point data, and likewise 2 individuals from group B dropout. If the study is wanting to carry out intention-to-treat analysis, how does this work. Is missing data computed for the dropped out participants (by some method)? Are group means of the depression scores just calculated as normal and the difference examined, despite group size discrepancy at the end time-point? Or are the participants that dropped out excluded entirely from calculations?
I have gotten very confused! Many thanks for any help!
The concept "analysis by intention to treat" refers to computing the results of each individual in the sample according to their initial assignment group, regardless of their degree of completion of the psychotherapy sessions (in the intervention group), or even if there may be received any intervention outside the study (in the control group). You should try to evaluate all patients even if they have not completed the intervention. If this is not possible, estimate the results using statistical methods of imputation of missing values.
In addition, you can provide complementary sensitivity analyzes (1) considering only the available data, without using imputations, and (2) per protocol analysis, that is, analyzing only patients with an established degree of compliance with the sessions.
A nice brief consideration of this very common problem in clinical trials is provided by Shih 2002:
https://trialsjournal.biomedcentral.com/track/pdf/10.1186/1468-6708-3-4.pdf
ITT is by default preferred "gold standard" analysis in most superiority intervention trials that assess efficacy, as it yields conservative estimates. The dictum is "once randomized, always analyzed". But in certain situations, you may want to additionally present the per protocol analysis. Whatever analysis you present, it would be prudent to mention it a priori while registering the protocol to dispel concerns about HARKING.
- Badges
- Science topic
- Similar topics
- Engineering
- Modeling