This is the largest effect on cognition anybody has reported for these drugs, and it points the opposite way from the other study in this section that ran the same comparison. Both cannot be describing the same reality, and working out why is more useful than either number.
What this study found
Elderly outpatients with heart failure with preserved ejection fraction, obesity and type 2 diabetes were given oral semaglutide or a DPP-4 inhibitor as an add-on, matched on age, sex and body mass index, and followed a year. [1] New cognitive impairment appeared in 33 of the semaglutide group and 73 of the comparator group, p < 0.0001, which a logistic regression rendered as a 77% lower risk.
The abstract publishes that 77% without a confidence interval, so how precisely it is estimated cannot be read off the record.
The incidence is the tell
106 new cases among 246 people in twelve months, or 43.1 per 100 patients per year. Nearly half the cohort acquired a cognitive impairment diagnosis inside a year.
That is far above what a general elderly population would produce, which means either the population was already close to the threshold or the threshold was set low. Neither is wrong — a cardiology clinic full of elderly patients with heart failure, obesity and diabetes is a high-risk group — but it does mean the outcome being counted here is not the outcome most readers picture, in the same way an entry criterion set locally changes what a trial is about.
What the matching covers
Age, sex and body mass index. Those are the three variables the abstract names.
The strongest predictors of a new cognitive impairment diagnosis are baseline cognition and education, and neither appears in that list. In a non-randomized comparison where a clinician chose who got which drug, that gap matters — it is the same structural weakness as every real-world comparison assembled after the fact.
How to hold both results
As an open question, not a settled benefit. One study found a very large effect in a small, high-risk, narrowly matched cohort; a much larger one found nothing against the closest available comparator.
Nobody is selling these drugs for cognition and nothing here is a reason to buy one. What the pair demonstrates is that a single striking percentage is worth very little until somebody has asked what else was published on the same question — the habit reading any body of literature requires.