The main comparison begins after enrollment

A trial can have more than one starting line. The first marks entry into observation; a later one marks assignment to competing study groups. Reading only the final result can hide what happened between those two points and which people the comparison actually represents.

The structure of STEP 4

In this semaglutide trial, 902 adults entered an initial 20-week run-in. The 803 who met the randomization requirements then continued semaglutide or switched to placebo for another 48 weeks. Both groups received lifestyle support. From randomization to the final assessment, average weight changed by −7.9% with continued semaglutide and +6.9% after switching to placebo. Gastrointestinal events were more common in the continued-treatment group.

A useful comparison still has a boundary

Randomization makes the later comparison informative about continuation versus withdrawal among those who reached it. It does not erase the earlier selection. People who could not meet the run-in requirements were not represented in that randomized result in the same way. This distinction matters especially when interpreting tolerability.

Consider an original analogy: a course evaluates two advanced lessons only among students who completed an introductory class. The comparison can fairly test the advanced lessons while leaving a separate question about students who did not reach that stage. The answer is not to discard the comparison; it is to keep its population attached to the conclusion.

Keep the time axis and denominator visible

A percentage change from the day of randomization answers a different question from a change measured at first enrollment. Before comparing two charts, check where each one sets its starting value. Then trace the participant counts forward: enrolled, eligible for comparison and assessed at follow-up. These labels can reveal more than a single headline percentage.

The study did not test repeated on-and-off cycles or establish an optimal duration for every individual. A planned research endpoint is a measurement date, not an instruction to stop or restart. Decisions about prescribed care require an individual clinical assessment rather than copying a timeline from a graph.

Which version was reviewed

This new S7 explanation uses the publisher's current full text of the 2021 paper, which identifies a June 7, 2023 correction to an abstract data typo. The article is about study design and interpretation; it gives no regimen, administration guidance or assurance about research-vial products.