Article — "GLP-1 Reality Check: Heart Failure Signal Strengthens While Oral Agents Match Them for Diabetes": Article — "GLP-1 Reality Check: Heart Failure Signal Strengthens While Oral Agents Match Them for Diabetes": Two studies landed on my desk this cycle, and they cannot both be the moral of the same story.
The first is a fresh semaglutide heart-failure meta-analysis: four RCTs, 6,109 patients, and a headline that reads like validation—cardiovascular death cut 25% (RR 0.75, 95% CI 0.61–0.92), all-cause mortality down 19% (RR 0.81, 0.67–0.98), serious adverse events reportedly nearly halved. Pair that with a blooming Alzheimer's literature where GLP-1 agonists preserve cerebral glucose metabolism in early trials, and you get the industry's favorite narrative: these drugs are quietly good for everything.
The second study is the one nobody wanted. Yale and VA Connecticut researchers asked a blunt, testable question—do GLP-1 receptor agonists actually let type 2 patients on basal insulin quit insulin better than a generic pill? The answer was a flat no.
Last week this column tracked the expanding evidence base—new indications, new signals. This week the delta is a comparator challenge: put GLP-1s head-to-head against cheaper agents on the exact claims people use to justify them, and the halo comes off.
Start with the VA target-trial emulation, because it is the cleanest falsification. The team built 8,869 matched sets—GLP-1RA initiators (76.6% semaglutide), SGLT-2i initiators (99.7% empagliflozin), and DPP-4i initiators (95.9% alogliptin)—and tracked insulin discontinuation over three years.
The numbers: 16.7% of GLP-1RA users discontinued insulin, versus 17.9% for SGLT-2i and 17.1% for DPP-4i. Risk ratio 0.93 (0.86–1.01) against the SGLT-2i arm—not superior, if anything numerically behind.
"None of the subgroups showed a comparative advantage of GLP-1RAs," the authors wrote. Alogliptin costs a fraction of semaglutide. On this specific outcome, the fraction bought you the same result.
That is the pattern this piece is about. The GLP-1 evidence base has split into two tiers, and the split runs along a single fault line: where hard randomized outcomes exist in the exact indication, the drugs earn their price. Step outside it, and a $4 alternative catches up or wins.
Where hard randomized outcomes exist in the exact indication, the drugs earn their price. Step outside it, and a $4 alternative catches up or wins.
Look at where the price is earned. The 25% cardiovascular-death signal is real—but read the subgroups. In the meta-analysis, the benefit concentrated in obese HFpEF patients at the 2.4 mg dose: HF rehospitalization RR 0.41, cardiovascular mortality RR 0.75.
At 1.0 mg weekly, no significant benefit on either endpoint. The dose that fuels the marketing is the dose that fails. The effect is not a class property floating over all of heart failure; it is indication-locked and dose-locked.
Now watch what happens when you leave HFpEF and enter HFrEF—reduced ejection fraction—where the signal doesn't just weaken, it reverses.
The FIGHT and EXSCEL trials enrolled or subgrouped HFrEF patients on liraglutide and exenatide. Pooled (n=769), these GLP-1RAs increased heart-failure hospitalizations: OR 1.49, 95% CI 1.05–2.10, P=0.02. That is harm, not neutrality.
SELECT is where the optimists point, because its HFrEF subgroup still showed a cardiovascular mortality benefit (HR 0.77). But the JACC analysis dissects the fine print: on HF hospitalization specifically, the HFrEF point estimate sat above 1.0—the same directional pattern seen with exenatide and liraglutide.
There's a deeper problem with reading SELECT as HFrEF evidence at all. The placebo event rate for HF hospitalization or cardiovascular death in that subgroup was, per the JACC analysis, roughly 4% per year—four to five times lower than symptomatic HFrEF in EMPEROR-Reduced (21.0%) or DAPA-HF (15.6%). Those patients likely didn't have symptomatic HFrEF; obesity, not heart failure, may have driven their exertional symptoms.
So the "semaglutide works in HFrEF" reading rests on a subgroup that probably wasn't HFrEF, while the trials that actually enrolled HFrEF show harm on hospitalizations. That is not a drug you reach for in reduced ejection fraction.
Contrast the comparator. Across the HF ejection-fraction spectrum, SGLT-2 inhibitors cut HF hospitalization and cardiovascular death by roughly 30–37% (HR 0.63–0.70), including in HFrEF where GLP-1s falter. GLP-1RAs reportedly manage about an 11% reduction in HF hospitalization (HR 0.89), primarily by preventing atherosclerotic events and driving weight loss.
The head-to-head net-benefit synthesis makes the division explicit. Across 17 cardiovascular outcome trials (N=132,038), both classes cut MACE equally (HR 0.87, no class difference, p=0.73).
But SGLT-2is reportedly beat GLP-1RAs on HF hospitalization (HR 0.69 vs 0.88, p<0.0001) and on renal outcomes (HR 0.68 vs 0.79, p=0.026). GLP-1RAs won net benefit for MACE (+2.5/1000). Each class owns a lane: GLP-1s for atherosclerotic risk, SGLT-2is for heart failure and kidneys.
The authors put it plainly—"the overall advantage depends on which adverse event is prioritized as sentinel harm." There is no universally superior drug here. There is a right tool for a specified job.
Which brings us to the adherence fix that was supposed to universalize GLP-1 benefit: oral semaglutide. If injections are the barrier, a pill should democratize the whole package—cardiac and renal alike.
Half the package showed up. In SOUL, oral semaglutide met superiority for MACE versus placebo in patients with established cardiovascular disease, the benefit driven largely by fewer nonfatal myocardial infarctions.
But it did not significantly reduce major kidney outcomes. The renal protection people assume travels with the molecule did not make the jump to the oral form. Injectable GLP-1RAs still hold the strongest cardiorenal evidence; the oral agent closed the cardiovascular gap and left the renal gap open.
Stack the failures and a rule emerges. Insulin discontinuation: matched by alogliptin. HFrEF: harm, not benefit. HF hospitalization and renal endpoints: SGLT-2is win outright. Oral renal outcomes: null.
Every one of those is a claim people make to justify GLP-1s off the strength of a different, narrower trial. The atherosclerotic-MACE win and the obese-HFpEF win are genuine and worth the money. The rest is halo.
The Alzheimer's literature belongs in this frame, not as a neuroprotection review but as the purest example of enthusiasm outrunning comparison. Preclinical models look striking—reduced amyloid, lower BACE1, preserved synapses. The human trials are "mixed or modest": liraglutide preserved cerebral glucose metabolism over 26 weeks but did not significantly change amyloid or cognition. Use to prevent or treat AD remains, in the literature's own word, off-label and investigational.
That is the shape of every over-claim in this class: a mechanistic story and a surrogate signal, mistaken for an outcome that hasn't been demonstrated against anything, let alone a cheaper alternative.
So the honest prescribing question has changed. It is no longer "GLP-1 or not." It is "which indication, and is a $4 alternative already as good."
For obese HFpEF at 2.4 mg, and for atherosclerotic MACE, prescribe with confidence—the randomized outcomes are there. For getting a patient off insulin, alogliptin matched it. For heart-failure hospitalization and kidney protection, empagliflozin beat it. For HFrEF, avoid it.
Price each claim to its evidence level. The drugs are extraordinary inside their proven indications and unremarkable—sometimes worse—outside them. Buying the halo means paying premium prices for outcomes a generic delivers, or reaching for a drug that raises hospitalizations in exactly the patients it's least suited to help.