A Diabetes Drug Cut Heart Attacks by 20% in a Landmark Trial. Then Observational Data Started Suggesting Something Stranger.

A drug developed to control blood sugar in type 2 diabetics just cut heart attacks by 20% in patients who don't have diabetes. That finding alone would have been remarkable. What happened next made it historic.

The SELECT trial, published in the New England Journal of Medicine in 2023, enrolled 17,604 adults with overweight or obesity and existing cardiovascular disease. None had diabetes. Half received semaglutide. After a mean follow-up of 39.8 months, the semaglutide group had 20% fewer major adverse cardiovascular events: heart attacks, strokes, cardiovascular deaths. A diabetes drug was protecting hearts through some mechanism other than glucose control.

Ask three specialists about the same patient on semaglutide, and you get three different reasons it seems to be working. The cardiologist sees cardiovascular risk declining. The gastroenterologist sees liver enzymes normalizing. The neurologist suspects something is protecting cognition. None of them prescribed it for their condition. What's accumulating across a range of diseases that share no obvious connection besides being diseases of aging requires a different kind of scrutiny.

The Evidence Hierarchy

Before listing the evidence, a framework. Randomized controlled trials (RCTs) can establish causation. Observational studies, which mine existing patient records, can only show association. When observational studies produce enormous effect sizes, the correct first instinct is suspicion, not excitement. Everything that follows is organized by this distinction.

GLP-1 receptor agonists were first approved for type 2 diabetes in 2005 (exenatide), with semaglutide and tirzepatide arriving in subsequent years. The causal evidence is now substantial. SELECT proved cardiovascular benefit independent of diabetes. The FLOW trial tested semaglutide in chronic kidney disease and was stopped early for efficacy: significant reduction in kidney failure progression (NEJM 2024). Phase III trials led to FDA approval of semaglutide for metabolic dysfunction-associated steatohepatitis (MASH), though it was not the first drug approved for the condition (resmetirom preceded it in March 2024). And tirzepatide (Zepbound) became the first FDA-approved drug for obstructive sleep apnea based on Phase III data.

Four conditions, four sets of controlled data. So far, solid ground.

Then there's the observational evidence, which is where the claims get dramatic and the caveats must get proportionally louder. A UC San Diego team analyzed 6,871 colon cancer patients across the UC Health system and found five-year mortality of 15.5% among GLP-1 users versus 37.1% among non-users, a relative reduction exceeding 50% after adjusting for age, BMI, and comorbidities (JAMA Oncology, 2025). An effect size that large in an observational study should provoke immediate questions about what wasn't adjusted for.

The dementia data comes from multiple sources with different effect sizes, a distinction that matters. A target trial emulation by Wang et al. using nationwide US electronic health records found semaglutide associated with approximately 70% lower Alzheimer's risk compared to insulin over 3 years (eClinicalMedicine, 2024). A separate 295,000-patient TriNetX cohort found a more modest 37% reduction in incident dementia across GLP-1 users versus non-users (Alzheimer's & Dementia, 2024). A Danish registry study showed 11% reduction per year of exposure (eClinicalMedicine, 2024). The range itself is telling: 11% to 70% depending on comparator, population, and follow-up window. When the same class of drug produces effect estimates that vary sixfold across studies, the honest read is that we're measuring confounding as much as pharmacology. When researchers test associations across dozens of conditions, some will reach statistical significance by chance. Individual findings need independent replication before they mean what they appear to mean.

See et al. (2025) reported 18% reduction in acute COPD exacerbations among GLP-1 users (BMJ, 2025). Preliminary case reports suggest reduced alcohol and substance use cravings, though this remains signal-level evidence without controlled trials.

The Weight-Loss Question

Before reaching for aging biology, the simplest explanation deserves full-strength engagement: weight loss alone might explain most of this.

Bariatric surgery, which causes substantial weight loss without activating GLP-1 receptors directly, produces a strikingly similar cross-disease benefit pattern. The Swedish Obese Subjects study found bariatric patients had lower rates of cardiovascular disease, cancer incidence, and mortality over 20 years of follow-up. Roux-en-Y gastric bypass has shown cognitive improvement. Post-surgery patients show reduced inflammation, improved insulin sensitivity, and lower cancer risk, the same profile attributed to GLP-1 receptor activation.

If bariatric surgery produces the same pattern without a GLP-1 drug, the parsimonious explanation is that weight loss drives the benefit, and receptor biology is secondary. Disentangling these requires head-to-head comparisons that don't yet exist.

The Aging Hypothesis

That said, a plausible biological framework does exist for why GLP-1 drugs might do more than help people lose weight.

Chronic low-grade inflammation, sometimes called inflammaging, is now understood to be a root driver of cardiovascular disease, cancer, neurodegeneration, kidney disease, and metabolic dysfunction. GLP-1 receptor agonists reduce systemic inflammation, improve insulin sensitivity, and promote autophagy, the cellular cleanup process that declines with age. Emerging evidence suggests they may influence cellular senescence, the accumulation of dysfunctional cells that drives tissue aging.

These are the same pathways targeted by every major longevity intervention currently under study: rapamycin (mTOR inhibition), metformin (insulin sensitivity), caloric restriction (autophagy), and senolytics (senescent cell clearance). If GLP-1 drugs genuinely operate on these pathways, it would mean the first commercially available drug that extends human healthspan arrived not through longevity research, but through an obesity pipeline.

That "if" is carrying enormous weight. The aging framework is a hypothesis, not a proven mechanism, and the bariatric surgery comparison suggests weight loss alone could be doing most of the work. Both explanations could be partially true.

The Healthy User Problem

People who take GLP-1 receptor agonists are not a random sample of the population. They have active healthcare relationships with regular doctor visits. They have insurance covering medications costing $800 to $1,600 per month, correlating with higher socioeconomic status. They've demonstrated motivation to manage their health by seeking and adhering to treatment. All of these factors correlate independently with lower rates of cancer, dementia, and cardiovascular disease.

The UC San Diego colon cancer study adjusted for age, BMI, severity, and comorbidities. It could not adjust for socioeconomic status, healthcare access, exercise habits, dietary quality, or the dozens of other confounders that separate GLP-1 patients from non-users. A 50% mortality reduction is enormous, and when an observational study produces an effect size that large, the prior probability of confounding as a partial or complete explanation goes up, not down.

The 70% dementia risk reduction is, if anything, more suspicious. Alzheimer's disease has resisted every targeted intervention for 30 years. A diabetes drug achieving what no purpose-built Alzheimer's drug has achieved should provoke deep skepticism before excitement. That number may ultimately prove directionally correct with a smaller true effect, or it may collapse entirely under confounding.

The $67 Billion Research Engine

Novo Nordisk reported $48.5 billion in 2025 revenue, driven by Ozempic ($19.2 billion) and Wegovy ($12 billion). Eli Lilly's tirzepatide franchise generated $36.5 billion (Mounjaro $23 billion, Zepbound $13.5 billion). Combined GLP-1 class drug revenue reached roughly $67 billion in 2025, making it the highest-revenue drug class in pharmaceutical history.

Lilly went from outside the top fifteen pharmaceutical companies by market capitalization to the most valuable pharma company on Earth in approximately eighteen months. Novo Nordisk became the most valuable company in Europe. Both trajectories were driven by drugs originally developed to manage blood sugar.

This scale matters beyond the balance sheet. With tens of millions of patients on GLP-1 drugs worldwide, every electronic health record system becomes a potential observational dataset. Signals that would take decades to emerge from small trials appear within years when the installed base is large enough. But this also creates a structural problem: Novo Nordisk and Eli Lilly fund a substantial share of GLP-1 research, and industry-sponsored studies with positive results are more likely to be published than negative ones. Publication bias is well-documented in pharmaceutical research generally, and a $67 billion drug class is not immune. When every positive signal gets a press release and every null finding gets filed away, the cross-disease evidence base may look more consistent than it actually is.

What's Missing from This Picture

The observational evidence has not been independently replicated across multiple datasets for most conditions. The colon cancer finding comes from one health system. The dementia data comes from two databases that may share population overlap. Single-dataset findings, regardless of cohort size, require independent replication before clinical conclusions.

Long-term safety data beyond five years is sparse. GLP-1 drugs carry known risks: pancreatitis, thyroid C-cell tumors (rodent models carry an FDA black box warning; human translation remains uncertain), and gastrointestinal side effects affecting 20-40% of clinical trial participants. Sarcopenia, the loss of lean muscle mass during rapid weight reduction, is a direct concern for a drug class being discussed in the context of longevity. Muscle mass loss accelerates aging. A drug that extends cardiovascular healthspan while degrading musculoskeletal health may be trading one form of decline for another. Decade-scale surveillance data on these tradeoffs does not yet exist.

Rebound weight gain after discontinuation is well-documented. If the cross-disease benefit depends on sustained weight loss, these drugs imply lifelong treatment at $800-1,600 per month. At that price, GLP-1 drugs are available primarily to patients in wealthy countries with employer-sponsored or government-subsidized healthcare. If they genuinely target aging biology, the equity implications are stark: approximately 10 million Americans currently take GLP-1 drugs, while 100 million have obesity. In low-income settings where the burden of age-related disease is highest, access is effectively nonexistent. Compounding pharmacies offer cheaper alternatives, but quality and efficacy data are limited.

No guidelines currently recommend GLP-1 drugs for cancer prevention, dementia risk reduction, or any indication outside their approved uses. Patients who bring observational headlines to their physicians deserve to know this. Physicians, in turn, face pressure from patients who have read headlines about 70% dementia reduction and 50% cancer mortality reduction without the methodological caveats. Articles like this one contribute to that dynamic. Reporting on the cross-disease pattern creates demand, and demand drives off-label prescribing. Acknowledging this tension doesn't resolve it, but ignoring it would be dishonest.

Bottom Line

Four conditions have RCT evidence. Six or more have observational signals that range from 11% risk reduction to 70%, a spread so wide it tells you more about methodology than medicine. We may be witnessing the largest healthy-user bias in pharmaceutical history, amplified by a $67 billion market incentive to interpret every positive signal as confirmation that semaglutide is a miracle drug.

Or GLP-1 receptor agonists may turn out to be the first broad-spectrum drug that operates on the biology of aging, a finding that arrived not through longevity research but through a diabetes pipeline, visible only because the commercial scale generated enough patient data for the pattern to emerge.

Randomized controlled trials for Alzheimer's and addiction are now enrolling. Within three to five years, we'll know whether the observational findings replicate under controlled conditions. Until then, the honest clinical answer is: four indications are proven, the rest are hypotheses, and the difference between those two categories matters more than the headlines suggest.

Methodology

This article compiles published evidence across GLP-1 receptor agonist indications as of March 2026, organizing findings by evidence strength (RCT vs. observational). Effect sizes are reported as published in original sources. No independent analysis was performed. The cross-disease compilation is this article's contribution; the underlying data belongs to the cited research teams. Sources include peer-reviewed publications in NEJM, JAMA Oncology, Alzheimer's & Dementia, eClinicalMedicine, and the BMJ. Corporate revenue figures are from Novo Nordisk and Eli Lilly annual reports. This article was produced by an AI system; for methodology, see AIPM.