AI Analysis of 400,000 Posts Surfaces Side Effects Patients Report and Trials Did Not

An analysis of roughly 400,000 public forum posts written by people taking GLP-1 drugs has surfaced effects that feature little in the trial literature, including menstrual changes and hot flushes. The work is a demonstration of what patient-reported data can add to a drug's safety picture -- and of how carefully it has to be read.

What was done

Researchers applied language models to a large volume of posts from people describing their experience on semaglutide and related medications, clustering what was reported and how often. The nausea, vomiting and constipation that dominate the prescribing information came through as expected. Less expected were recurring accounts of changes to menstrual cycles, hot flushes, hair thinning, and shifts in mood and in taste.

Why trials can miss this

Clinical trials measure what they set out to measure. Adverse events are collected, but the structured questionnaires are built around anticipated effects, and something nobody thought to ask about is recorded only if a participant volunteers it and an investigator writes it down. Trials also run for a defined period in a selected population, whereas these drugs are now taken for years by a far broader group than was ever studied.

Some of the signals are biologically plausible. Substantial weight loss alters oestrogen production in adipose tissue, which is a credible route to menstrual change and vasomotor symptoms. Rapid weight loss is an established trigger for temporary hair shedding.

Why the method has real limits

People post when something goes wrong. There is no denominator -- no way to know how many took the drug without the effect -- and no way to confirm that posters were taking what they say, at what dose, or that the symptom was caused by the drug rather than by coincidence, by the weight loss, or by another medication. Menstrual changes and hot flushes are common in the age groups taking these drugs regardless. This kind of analysis generates hypotheses; it does not test them.

How it should be used

As a pointer at what to study and what to ask about. A clinician who knows that patients report menstrual change is more likely to raise it, and a patient who hears that others have experienced it is more likely to report it rather than assume it is unrelated.

Bottom line

If you are taking a GLP-1 drug and something has changed that you had not expected, it is worth mentioning -- including things that seem unconnected to appetite or digestion.