Korean records from 13 hospitals put numbers on GLP-1 weight-loss safety, and the psychiatric signal is the one that stands out

These drugs are now taken by people who are not ill in the usual sense — who want to lose weight and have no diabetes. That changes what counts as an acceptable risk, because the benefit being weighed is different. This is the largest Korean attempt so far to say what the risks are in that population.

The psychiatric signal#

Both drugs showed it and in the same direction. Semaglutide: psychiatric disorders overall 2.02, anxiety 2.39, depression 3.42. Liraglutide: 1.66, 1.68 and 1.51. The semaglutide intervals are wide — 1.51 to 7.74 for depression — because 2,357 initiators is not many for an uncommon outcome. The liraglutide figures are smaller and tighter, resting on 6,953 initiators.

Two agents in the same class producing a consistent signal is harder to dismiss than either alone. What neither can rule out is the most obvious alternative explanation: people who seek pharmacological help for weight are, on average, more distressed about their weight before they start, and distress that was already present can surface as a coded diagnosis once someone is in regular contact with a clinic.

The physical outcomes differ by drug#

Semaglutide's other strong signal was gastrointestinal dysmotility or obstruction at 3.91, which fits the known mechanism — these drugs slow gastric emptying, and that is part of how they work. Liraglutide instead carried hepatic impairment at 1.46, pancreatobiliary disorder at 1.33 and pancreatitis, cholangitis or cholelithiasis at 1.40.

What the authors qualify themselves#

Semaglutide-associated vision impairment and liraglutide-associated hepatic impairment did not hold up across all sensitivity analyses, and the paper says so. Those two should be treated as unresolved rather than established. Ten outcomes were tested across two drugs, which is twenty comparisons, and some positives at conventional thresholds are expected from that alone.

The design's strength is that it is a new-user comparison against matched non-initiators drawn from the same hospital records, which avoids comparing current users against people who already failed on the drug. Its limit is that the comparison group is not taking anything, so every consequence of deciding to treat obesity pharmacologically — the clinic visits, the monitoring, the expectations — sits inside the exposure. The authors' conclusion is the defensible one: careful patient selection and monitoring, not a reason to withhold.