While I've said this before, I will say it again. The parametric measures of significance being used depend on the assumption of a Gaussian (normal) underlying distribution. That is not merely falsifiable, but demonstrably false. In checking for possible selection effects, you must consider those who are not in the study. The combined numbers of those who declined to participate and those who dropped out are roughly equivalent to those who completed trials. You also have 31% of the most 'successful' group not completing both 6MWT, the only objective measure presented. One might expect those who did not take a test involving a short walk to be in worse shape than those who did. Plenty of room for non-random selection. The surprise is that they couldn't make this look better. You should also check on evidence of selecting forces, even beyond investigator bias. With 93% of the most 'successful' group reporting one or more adverse events, even after a redefinition of adverse events to make them less likely, these forces are in evidence.