m/m trial metrics reallypeople, you are reading t

m/m trial metrics
reallypeople, you are reading the primary endpoint correctly...a comparison of patients changes in scores from day 0 to day 14. Nothing in between and not taking the difference in the averages at days 0 and 14. In other words, instead of comparing mean at t=0 with mean at t=14 for the two groups, they will likely compare the average difference in scores (between t=0 and t=14) for the two groups (placebo vs leronlimab) . Possibly they do a non-parametric version of that, like Mann-Whitney that looks at ranks rather than raw scores.

This paired approach of comparing means of differences rather than differences of means tends to have greater statistical power. Sometimes it does not, though, so they will probably also calculate statistics for the differences in means.

Your example of patient A going from 7 to a 3 in the placebo group looking better than patient B going from 3 to a 0 in the leronlimab group because of the hard lower bound (you can't get lower than 0!) is a good one. A few things to think about...for every 7 in the placebo group, there's probably a 7 in the leronlimab group too. How does that patient fare? And do that for ALL the patients. Each 3 in the leronlimab group will have approximately two placebo patients at 3. Do those also go to zero? If one goes higher and one goes to 1, that's a huge win for leronlimab. But that doesn't resolve the problem of having a hard lower bound at 0, which skews the variances and argues for running a non-parametric test instead of the normal-based comparison of means of difference or comparison of differences of means. So, instead of 7 - 3 = 4 and 3 - 0 = 3, they may look at changes in the ordering of patient scores. Rank all the initial scores and see how the ranks have changed through the end of the trial.

If you ask a good question about a complicated data set and someone tells you it's not complicated, move along; they don't know what they are talking about.