There is a number in this paper that is statistically significant, measured with an interval so narrow you could not ask for better precision, and completely invisible to the person it was measured in — and the authors say so themselves, in the abstract, which is rarer than it should be.
What was pooled
Fifteen studies covering 1,877 participants with moderate to severe obstructive sleep apnea, comparing SGLT2 inhibitors, GLP-1 drugs and aerobic exercise against each other and against placebo. [1] Thirteen of the fifteen are randomized trials; two are case-control studies, which is a weaker foundation than a network meta-analysis usually rests on, and every indirect comparison inherits that weakness.
On the outcome that matters most, GLP-1 drugs did well: breathing events fell by 15.28 an hour against placebo, 95% CI 8.35 to 22.22 — a large effect, consistent with what the dedicated apnea trial found. Body mass index fell 1.78 kg/m2 and average overnight oxygen saturation rose 0.40%.
The number that is not there
Daytime sleepiness is scored on the Epworth scale, and the pooled improvement was 0.20 points, 95% CI 0.14 to 0.26.
The smallest change on that scale a person actually notices — the minimal clinically important difference — is two points. The measured effect is a tenth of that, and the interval is tight enough to say with confidence that the true effect is also about a tenth of that.
The ranking the authors disown
GLP-1 drugs came out top on breathing events, body mass index and oxygen saturation under a statistical ranking method that scores each treatment across the network. The authors then say the comparisons between active treatments were largely non-significant, rested on indirect evidence, and carried certainty that was moderate at best and low or very low for most pairings, so the hierarchy should be read as hypothesis-generating rather than as a basis for recommendations.
That is an unusual thing for a paper to write about its own headline result, and it is the correct thing, for the same reason a pooled count of studies overstates what has been independently observed. A ranking assembled from thin indirect comparisons will still produce a first place.
What to carry away
That a drug can genuinely fix the thing a machine measures and barely move the thing you feel, and that both facts can appear in one paper without either being wrong.
Also that aerobic exercise sat in this network as a comparator and costs nothing, which does not make it equivalent and is worth noticing before a table of treatments starts looking like a shopping list — the habit that reading any ranked table calls for.