Regression To The Mean



Hello Group,

Any suggestions to provide evidence against regression to the mean (RTM) effects versus "real" effects when this former phenomenon seems plausible?

For example, suppose a RCT trial is conducted to test whether an outcome significantly differs over time (pre, post) in the intervention versus control arm. However, due to attrition and loss of outcome data at follow up, the initially randomized groups are no longer "truly" randomized, i.e. they're now a subsample of the original, randomized sample.

Further, suppose the two groups from the subsample now significantly differ on the outcome measure at baseline (pre), where the intervention group has a higher initial level than does the control group. Then, it is observed that the outcome improves (declines) over time in the intervention group, but stays the same in the control group. How can one argue against the observed effect being a result of RTM (patients who start high are likely to end up low) versus a "true" effect due to the intervention when, as stated earlier, the groups are no longer truly randomized?

Any help/feedback would be greatly appreciated?