The honest answer is that the effect is large in relative terms and that a large relative reduction from a severe baseline can still leave a treatable condition.
Home sleep apnoea tests systematically underestimate AHI relative to laboratory polysomnography, so a before-and-after comparison should use the same modality or the difference is partly instrumental.
Daytime somnolence scores improve alongside AHI in these trials, but they also improve with placebo, which is why the instrumented endpoint is the one that carries the argument.
Weight loss by any means is long-established as reducing AHI; the surgical literature has shown this for decades, which is the reason the pharmacological result was expected rather than surprising.
Symptom improvement is a poor proxy for index improvement, in both directions.
A two-band improvement is a real result and is not the same as resolution.
edited 23 Nov 2024 by tyndall_haze — tightened the wording; no substantive change