The GLP-1 Mood Signal Is Real — But So Is the Psychiatric Worsening Risk

A pattern I’ve been tracking across my dual-diagnosis patient panel this year: the patients who come in asking about semaglutide are not the ones struggling with weight alone. They’re the ones carrying both 40 extra pounds and a decade of undertreated depression — often with a substance use history layered underneath. They’ve read something online, or their primary care doctor mentioned it, and now they’re in my office asking whether Ozempic might help their mood. The honest answer, until recently, was that I didn’t know. The data is starting to catch up.

A national cohort study published in Lancet Psychiatry this month followed close to 100,000 Swedish patients with depression or anxiety who also carried a diagnosis of type 2 diabetes or obesity. Those treated with semaglutide showed lower rates of psychiatric deterioration compared to those on liraglutide, and both outperformed non-GLP-1 comparators on several mood-related outcomes. The headline finding is encouraging — but buried in the subgroup data is a signal that deserves more attention than it’s been getting.

Not Everyone Responds the Same Way

A meaningful subset of patients on GLP-1 therapy experienced psychiatric worsening, not improvement. The Swedish investigators were appropriately measured in their interpretation, noting residual confounding and the observational design’s inherent limits, but the signal is consistent enough to prompt serious mechanistic questions. GLP-1 receptors are expressed throughout the limbic system — in the hypothalamus, hippocampus, and ventral tegmental area. Semaglutide’s central penetrance is substantially greater than liraglutide’s, which likely explains both its stronger mood signal and its greater potential for adverse psychiatric effects in vulnerable populations.

In addiction psychiatry, that neuroanatomy is not incidental. The VTA is the origin of the mesolimbic dopamine pathway — the same circuit that underlies reward, craving, and anhedonia. We already know from preclinical data that GLP-1 receptor activation suppresses dopamine release in the nucleus accumbens, which is part of why these drugs reduce food-seeking behavior. In a patient with co-occurring major depressive disorder and alcohol use disorder, suppressing that pathway might attenuate craving. It might also deepen anhedonia. Both outcomes are pharmacologically plausible from the same mechanism.

The counterintuitive read on this data is that the patients most likely to benefit metabolically — those with severe obesity and long-standing insulin resistance — may carry the highest psychiatric risk from central GLP-1 agonism. Metabolic syndrome is independently associated with neuroinflammation and HPA axis dysregulation, two substrates that modulate GLP-1 receptor sensitivity in ways that observational studies cannot yet stratify.

What the Protocol Architects Are Missing

Novo Nordisk’s Phase 3 program for semaglutide in MDD is underway, and Eli Lilly has initiated exploratory work with tirzepatide in mood disorders following the cardiovascular and metabolic outcomes data from the SURMOUNT and SURPASS trial families. Neither program, as publicly designed, appears to be powered to detect psychiatric worsening as a primary or co-primary endpoint. That is a design choice with consequences.

Having run SUD trials where dual-diagnosis patients were quietly screened out by eligibility criteria that no one defended at the investigator meeting, I recognize this pattern. The patients most likely to show a divergent psychiatric response to GLP-1 therapy — those with comorbid SUD, prior mood episode history, or active anhedonia at baseline — are exactly the patients who get excluded for “psychiatric instability.” The result is a Phase 3 dataset that proves efficacy in a population cleaner than anyone sitting in an actual psychiatric clinic.

The Swedish data suggests that baseline psychiatric severity and comorbidity profile are likely moderators of GLP-1 response. Randomized trials need to be designed to test that hypothesis directly, not inadvertently suppress it through exclusion criteria.

A Site Signal Worth Reading

Across my telehealth panel — which spans 5 states, with a significant proportion of Spanish-speaking patients who carry high rates of both metabolic disease and undertreated depression — GLP-1 uptake is accelerating faster than the psychiatric monitoring infrastructure around it. Primary care is prescribing semaglutide. Psychiatry is not always in the loop. That gap between prescriber and mental health monitor is where adverse psychiatric signals get lost before they reach a DSMB.

A prospective RCT with dual metabolic and psychiatric endpoints, enrolling patients with active depression or anxiety plus diabetes or obesity, stratified by SUD comorbidity and baseline anhedonia severity — that trial is buildable, and the recruitment population is not hypothetical. The Swedish cohort study identified 107,000 eligible patients in one country’s registry. The U.S. dual-diagnosis population is an order of magnitude larger, and a substantial fraction of them are already on these drugs without psychiatric oversight.

The readout I’m watching next is the Novo Nordisk semaglutide MDD program’s interim data, expected in late 2026 — specifically whether the protocol captures anhedonia and reward-processing endpoints with enough granularity to detect the worsening signal the Swedish investigators flagged. If it doesn’t, the safety question will outlast the efficacy answer by years.