Two findings sit awkwardly together in this study, and the awkwardness is the interesting part: a large relative benefit, and no sign of it growing with more drug — the opposite of what a dose-dependent effect looks like.
What was compared
Adults with a documented COVID-19 or influenza infection, classified by whether they had been taking semaglutide beforehand or metformin without prior GLP-1 exposure. [1] Matching covered demographics, body mass index, glycated hemoglobin, vaccination status, prior infection, antiviral use, healthcare utilization and baseline comorbidities, leaving 7,092 matched pairs for COVID and 1,920 for influenza.
That is a more thorough matching list than most studies of this kind manage, and it is why the results are worth taking seriously rather than dismissing — it is the opposite end of the spectrum from a matched comparison that let follow-up differ by two hundred days.
The numbers, with their bases
COVID 30-day mortality ran 0.3% against 0.8%, relative risk 0.39. Hospitalization 5.3% against 7.4%. The composite severity outcome 8.2% against 10.6%. In influenza, mortality 0.2% against 0.7%, hospitalization 6.5% against 9.6%.
Five fewer deaths per thousand infections is what a relative risk of 0.39 means on a base of 0.8%. Intensive care admission and mechanical ventilation were not significantly different in either cohort, which are the outcomes least sensitive to who ends up coded as hospitalized.
Where it held up best
Acute kidney injury was significantly lower in both cohorts after adjustment for multiple comparisons, at relative risks of 0.73 and 0.58. Acute coronary syndrome and acute respiratory failure were not.
The benefit also held among unvaccinated patients and among adults over 65, and remained among people who received standard COVID therapeutics — which is the kind of consistency across subgroups that a confounding explanation has to work harder against, though it does not eliminate it.
What a buyer takes from it
Nothing to act on. Nobody should buy a GLP-1 to reduce their risk from influenza, and the authors ask for prospective evaluation rather than claiming the case is made.
What is worth carrying is the check. When a benefit does not increase with more of the drug, that is either a clue about mechanism or a clue about confounding, and the study design decides which — the same question behind any comparison run at doses somebody chose. The absolute figures here are also a reminder that a large relative risk on a small base is a small number of people.