Dear Community,
I am currently planning an RCT. We originally planned a research design with an intervention group and a waiting-list control group and two measurement points (t0 at baseline and t1 as a 3-month follow-up).
We planned the control group as a waiting-list primarilly as an incentive to enhance participation in the trial for the control group participants and so we originally did not intend to use (or even measure) post- treatment data from that group in the statistical analysis. For this design we planned to analyze the data using an ANCOVA with time (t0, t1) as a covariate. The calculated sample size is N=128 (f=0.25, two groups, one covariate, alpha=0.05, power=0.8).
Now we we are required to include post- treatment data from all patients that received the intervention. What are your statistical recommendations for using post- treatment data of the waiting-list group?
I considered to use 3 groups in the ANCOVA (A: intervention group with its t0 and t1 measures, B: waiting-list group prior to the intervention with t0 and t1 and C: waiting-list group with their pre- intervention t1 and post- intervention t2). This would not affect the power calculation as only the number of groups would change from 2 to 3. But I do not consider such an approach as suitable because the two groups from the waiting list would not be independent. Additionally, this approach would use an unequal randomization ratio of 1:2.
I think, a mixed linear effects model also would not be suitable as the intervention group has data from two points of measurement and the waiting- list group is measured three times.
Any suggestions?
Best regards,
Chris.
Hi Christian
With respect to the analysis, I would prefer a linear mixed effect model over ANCOVA. I would add the Control T2 data to the Intervention T1 data, recoding them as Intervention. The imbalance in sample size is not of importance. Thus if you started with 50 in each group and there were no drop outs, your sample size would look like
T0 T1
Control 50 50
Intervention 50 100
Hi Christian, what exactely do you mean by "we are required"? An appropriate answer to your question probably depends on if this is was a request from notified body or a journal reviewer.
Hi Christian,
I think that your question is related to study design.
In one hand, if the sample baseline is planned to have no statistical difference between the groups, you will not be able to compare the control group outcomes after the intervention to the treated group.
On the other hand, If the baseline characteristics are not important, you could compare the control group outcomes at T2 to the treated group at T1. However, if you provide the same intervention to the control group, you will have two groups with similar outcomes. Hence, you could think about to include a different intervention in the control group. You could compare this to the treated group. It would be a 3 arms study.
I hope this can be helpful.
Best wishes,
Klaus
Have three timepoints for everyone, t0 at baseline, t1 when the control group switches to treatment, and t2 after about double that time. At t1 you have a between subject treatment comparison of treatment versus control adjusted for the t0 baseline covariate. You also have a within subject comparison within the control-treatment group. The right linear mixed model can combine those estimates under certain assumptions. And stronger assumptions can be made of course if blinding occurs and neither group knows who is on control or treatment to time t1.
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis