A phase 2b trial asked whether a GLP-1 drug slows Alzheimer’s disease. The answer on the question the trial was built to answer is no [1]. Cerebral glucose metabolism, the primary outcome, differed by -0.17 between arms, with a confidence interval from -0.39 to 0.06 and a P value of 0.14. The interval crosses zero — the reading this site applies to every effect estimate, as in sixteen percent lower, six hundredths of a point.
That is worth stating plainly because the trial will be cited for a different sentence. One secondary outcome — the executive domain of the Alzheimer’s Disease Assessment Scale — favored liraglutide by 0.15, with a confidence interval from 0.03 to 0.28. The abstract attaches the word unadjusted to that P value of 0.01, and the word is doing real work.
The safety finding is real and should not be lost. Liraglutide was generally safe and well tolerated over 52 weeks in people with Alzheimer’s disease and without diabetes, a population where nobody had established that. Daily injections in a group with cognitive impairment is not a trivial thing to run, and the tolerability answer is useful in its own right, whatever the efficacy answer turns out to be.
Liraglutide is the older drug in this family, and the trials being run in cognition now use it partly because it has the longest safety record. Whether a newer, stronger agent would produce a different result is untested. This site has covered the adjacent cognitive question in what the oral tablet did for cognitive impairment, which reached a similarly cautious place.
The honest summary is that 204 people spent a year in a well-designed trial and the primary outcome said no. That is a useful result. It is also the kind of result most likely to be reported by its one positive secondary, which is why the interval and the word unadjusted belong in any summary of it — a habit this site applies to every trial, including the ones whose headline is good, as in the trial where everything improved.