Three Domains Improved, Then One

A randomized trial in people living with HIV found semaglutide improving three cognitive domains at 32 weeks. Then the authors adjusted for sex and immune status, and one was left — at P = .05, exactly on the line.^[1] Both halves of that sentence are the finding, and only the first half travels.

Why this population

People living with HIV, even on effective antiretroviral therapy, carry higher rates of visceral fat accumulation, chronic inflammation and cognitive decline than people without HIV. Those three things are plausibly connected, and a drug that reduces the first two is a reasonable thing to test against the third.

The same group had already shown that semaglutide reduced weight, visceral fat and several inflammatory markers in this population. This trial asked whether cognition followed.

What was found, in order

Participants on antiretroviral therapy were randomized evenly to 32 weeks of semaglutide or placebo, double-blind, with cognitive function as the primary outcome, measured using a computerized screening instrument.

Adjustment is not a trick played on a result. Sex and CD4 count both plausibly relate to cognition and could differ between arms by chance in a 108-person trial; accounting for them is the more careful analysis. Two of the three findings did not survive it.

The adjusted result is the one the authors believed, and it is a single domain sitting on the threshold.

The paper’s own conclusion says semaglutide may have a beneficial impact on visuospatial cognitive function — hedged, and singular. That is an accurate description of what the table shows. The risk here is entirely in the retelling.

The question the abstract does not answer

It matters because of how thin the surviving margin is. A finding at exactly .05 in one of several domains, in 108 people, is the kind of result that exists to justify a larger trial rather than to change anything. That is also what a phase 2b study is for.

The inflammation route

The authors ran a mediation analysis and reported that semaglutide’s effect on the visuospatial score persisted while accounting for changes in two inflammatory markers — C-reactive protein and soluble CD163 — at P = .04, which they read as the effect running through inflammation.

Mediation analysis tests whether a hypothesis is consistent with the data. It does not establish a causal pathway, because it rests on assumptions that trial data cannot verify. We set that out at more length in is it just the weight loss, and the same caution applies here with less data behind it.

What to take from it

• It is a phase 2b signal, not a treatment finding. One domain, at the threshold, in 108 people over 32 weeks.
• The screening instrument is not a full neuropsychological assessment. It is designed to detect change efficiently, not to characterize it.
• The question is worth pursuing — the biology is coherent and the population has a real unmet need.
• Anyone quoting “three cognitive domains improved” is quoting the unadjusted analysis and stopping before the sentence that follows it.