I found this report from Global News interesting:
A rare form of sudden and potentially permanent vision loss is becoming the latest safety question surrounding blockbuster GLP-1 drugs, with patients filing lawsuits in the U.S. even as regulators overseas have already added warnings.
The condition is called nonarteritic anterior ischemic optic neuropathy, or NAION. It occurs when blood flow to the optic nerve is reduced, potentially causing abrupt vision loss that can be permanent.
The complication appears to be extremely rare. The bigger issue is that researchers still disagree over whether drugs such as Ozempic and Wegovy actually cause it.
That uncertainty is now moving from medical journals into courtrooms and regulatory agencies.
The EU has already taken action:
The European Medicines Agency reviewed clinical trials, post-marketing surveillance, scientific literature and other available evidence before concluding in 2025 that NAION should be classified as a very rare side effect of semaglutide.
European regulators estimated that it could affect up to roughly 1 in 10,000 people taking semaglutide.
Several large epidemiological studies reviewed by European regulators suggested semaglutide exposure among adults with Type 2 diabetes was associated with approximately twice the risk of developing NAION.
Warnings or information about NAION have also appeared on drug labeling in countries including the U.K., Japan and Australia.
I think that as a stop-gap measure the EU action is prudent but I saw a great opportunity for AI. Give AI everything known about both a significant number of patients who developed NAION and patients who didn’t including whether they took GLP-1 or not and whether they went on to develop NAION: retinal/OCT images, visual acuity, refraction, blood pressure and medications, A1c trajectory, renal function, sleep-apnea information, age, BMI, lipids, and so forth.
More importantly, give it the trajectories. What happened after GLP-1 treatment began? How rapidly did the patient lose weight? How rapidly did A1c fall? Did blood pressure fall while antihypertensive treatment remained unchanged? Was there dose escalation, vomiting, dehydration, renal impairment, or some other change shortly before the event?
Then use AI to look for combinations of characteristics and changes that distinguish the tiny number of patients who develop NAION from the thousands who do not. Freeze the associations it discovers and test them prospectively against a separate population it has never seen.
The useful question may not ultimately be whether semaglutide can cause NAION. It may be why semaglutide appears to cause NAION in one particular patient while having no such effect in thousands of apparently similar patients.
That should be doable nearly as quickly as the data can be collected.
