Here is a number that should end the argument. In patients with reduced-ejection-fraction heart failure and type 2 diabetes, adding a GLP-1 receptor agonist to guideline-directed medical therapy dropped 18-month all-cause death from 14.4% to 10.5% — a hazard ratio of 0.71, confidence interval 0.62 to 0.83. Nearly a third fewer deaths. If you were writing the headline, you'd write "GLP-1 drugs extend survival in the sickest heart patients," and move on.
Don't move on.
Because two paragraphs into the same study, the survival signal collides with something inconvenient: hospitalization didn't budge. In the same cohort, HF hospitalization ran 63.7% with the drug versus 60.1% without — hazard ratio 1.05, confidence interval crossing 1.0. The drug supposedly saving lives did nothing measurable to keep those lives out of the hospital.
A mortality benefit with a flat hospitalization curve, in HFrEF specifically, is exactly the fingerprint of confounding — not a drug reshaping disease course.
That contradiction is the whole story. And it points in the opposite direction from the headline.
Last week this newsletter dismantled the GLP-1 muscle-loss panic. This week the pivot is bigger: the claim that these drugs are graduating out of metabolism entirely — into cardiology's hardest cases and neurology's most feared disease. The claim is running ahead of the evidence in both organs. Badly.
Start with the heart. The survival figure above comes from a retrospective TriNetX analysis — an electronic-health-record dataset, propensity-matched, reportedly down to roughly 3,158 patients per group from an initial pool of over half a million. Propensity matching is a serious tool. It is not randomization. It cannot balance what wasn't recorded, and in a database study, plenty isn't recorded.
Here is the tell that should make any cardiologist pause: a drug that genuinely rescues a failing heart should reduce the thing that failing hearts do — decompensate and land in the hospital. This one didn't. A mortality benefit with a flat hospitalization curve, in HFrEF specifically, is exactly the fingerprint of confounding — sicker or better-managed patients sorted by something the algorithm couldn't see, not a drug reshaping disease course.
Now put that observational signal next to the randomized evidence, and the collision becomes a pileup.
When investigators combined the HFrEF patients from the FIGHT and EXSCEL trials — 769 people randomized, not matched — liraglutide and exenatide reportedly increased the risk of heart-failure hospitalization. Per the JACC editorial, odds ratio 1.49, confidence interval 1.05 to 2.10, p equals 0.02. Not neutral. Worse.
The SELECT trial, the largest apparent HFrEF cohort ever given a GLP-1 RA in a randomized setting, told the same story where it counts. In the investigator-identified HFrEF subgroup, semaglutide showed reduced cardiovascular mortality (HR 0.77) and all-cause mortality (HR 0.75) — but in the investigator-identified HFrEF subgroup, it did not reduce HF hospitalizations, with a point estimate above 1.0. The same directional signal as exenatide and liraglutide before it.
The JACC editorial commentators cast doubt on whether those "HFrEF" patients even had HFrEF: reportedly the placebo event rate was roughly 4% per year, four to five times lower than the 15.6% to 21.0% seen in bona fide symptomatic HFrEF cohorts like DAPA-HF and EMPEROR-Reduced. In other words, obesity may have been driving the breathlessness, not a failing ventricle.
And here is the part the "expansion" narrative buries. The umbrella review — twelve systematic reviews, 29 unique randomized trials — did find GLP-1 RAs reportedly reduced major adverse cardiovascular events (HR 0.86) and worsening heart failure (HR 0.56), plus a 14-meter gain on the six-minute walk. But it found no benefit for all-cause mortality, cardiovascular mortality, or HF hospitalization. And the positive signals, its authors state plainly, were "primarily driven by studies in HF with mildly reduced or preserved ejection fraction," while HFrEF evidence "showed no benefit."
So the strongest evidence base points at HFpEF — the preserved-ejection-fraction phenotype, often obesity-driven — and away from the reduced-ejection-fraction patients the TriNetX headline is about. The observational study is bragging about the exact population the randomized data flags for caution.
The umbrella review's own conclusion is the one worth quoting: these findings "do not support a definitive change in clinical practice but rather highlight the urgent need for dedicated, phenotype-specific HF trials." That is not a drug arriving. That is a drug being asked, politely, to prove itself in the right patients.
Now cross to the brain, because Parkinson's is the same movie in a different theater.
The mechanistic case is genuinely attractive. GLP-1 receptor signaling engages cAMP/PKA, PI3K/Akt and ERK pathways that regulate mitochondrial function, oxidative stress, and autophagy-lysosomal dynamics — precisely the machinery that fails in Parkinson's. Reportedly, lixisenatide showed neuroprotective effects against the propagation of α-synuclein pathology, the protein cascade at the disease's core. On paper, this is one of the more biologically plausible repurposing stories in neurology.
Plausibility is not proof. And the human number is where the story deflates.
The updated systematic review and meta-analysis of GLP-1 RAs in Parkinson's found a reportedly statistically significant motor improvement: MDS-UPDRS Part III in the ON state, mean difference −2.88 points, p equals 0.01, with modest heterogeneity. Then the sentence that has to travel with that figure: the drugs "were associated with a higher incidence of adverse events across all safety outcomes."
Weigh those against each other. UPDRS-III runs across dozens of points; a 2.88-point shift sits near the edge of what patients or clinicians would notice, and it's the kind of effect size that can be flattered by open-label enthusiasm or the wrong patient mix. Buying it with more gastrointestinal and systemic adverse events is not an obvious trade.
That is not a disease-modifying therapy. It is a small symptomatic blip with a cost, sitting on top of a mechanism that deserves a real trial. The reviewers themselves frame GLP-1 signaling as "biologically plausible" and call for "rigorous evaluation" — the exact language cardiology's umbrella review used. Two organs, one verdict: promising lead, unproven label.
The common failure is a category error the headlines keep making — treating a signal as an indication.
A propensity-matched survival curve is a hypothesis. A 2.88-point motor delta is a hypothesis. A conserved intracellular pathway is a hypothesis. What separates a hot repurposing narrative from an approved use is a specific, boring, expensive thing: a dedicated randomized trial in the right phenotype, powered for hard endpoints, that reads out the way the mechanism predicts.
For HFrEF, that trial doesn't merely not exist — the randomized evidence that does exist leans the wrong way, toward more hospitalizations, not fewer. For Parkinson's, the definitive disease-modifying trials are precisely what everyone concedes is still missing.
So here is the correction to the headline. The story is not "GLP-1 agonists expand into heart failure and Parkinson's." The story is that clinicians are being invited to prescribe on the strength of observational artifacts and mechanistic hope, against randomized data that in one case actively counsels caution.
For the cardiologist facing a HFrEF patient with diabetes, the honest read is that the survival signal is retrospective, the hospitalization signal is neutral-to-adverse, and the trial-grade evidence points at a different phenotype. For the neurologist, it's that the biology is real, the human benefit so far is small and bought with side effects, and the confirmatory trials are still running.
Leads, not labels. Until the phenotype-specific randomized trials read out, that distinction is the entire margin between medicine and marketing.