GLP-1 Drugs Have Positive Evidence Against 7 of the 10 Leading Causes of Death in America. Only 4 Have Randomized Trial Support.

Seven out of ten.

That is the number of leading causes of death in America for which GLP-1 receptor agonists now have at least one positive clinical or observational signal. Heart disease. Cancer. Stroke. Alzheimer's. Diabetes. Kidney disease. Liver disease. Seven causes that together kill 1,819,685 Americans per year, according to the CDC's 2022 mortality data. That is 53% of the 3.4 million annual deaths in the United States.

No drug class since antibiotics has touched this many disease categories simultaneously. But the word "positive" does a lot of work in that sentence, and failing to distinguish between a randomized controlled trial and a retrospective cohort study is how pharmaceutical hype cycles spin up. So here is the accounting, sorted by evidence strength, with the math showing exactly what each signal is worth and what it is not.

The RCT Floor: 209,000 Deaths

Four of the seven indications rest on randomized, controlled, double-blind trial data. These are the signals worth taking seriously.

Heart disease (702,880 deaths/year): The SELECT trial, published in the New England Journal of Medicine in November 2023, enrolled 17,604 patients with BMI ≥27 and pre-existing cardiovascular disease, without diabetes. Semaglutide 2.4 mg reduced major adverse cardiovascular events (MACE) by 20% over a mean follow-up of 39.8 months (HR 0.80; 95% CI 0.72-0.90; P<.001). A critical caveat: MACE is a composite of cardiovascular death, non-fatal myocardial infarction, and non-fatal stroke. It is not a mortality-only endpoint. SELECT's all-cause mortality showed a favorable trend (HR 0.81) but did not reach statistical significance within the trial's hierarchical testing framework. Applying the 20% MACE reduction to 702,880 deaths is therefore an upper-bound estimate, not a direct calculation. It assumes event reduction translates proportionally to mortality reduction, which overstates the case. I use it as a ceiling: 140,576.

Stroke (162,890 deaths/year): Stroke falls within the MACE composite endpoint of SELECT. The same caveat applies: MACE includes non-fatal stroke. Applying the composite 20% reduction to stroke deaths gives a ceiling of 32,578.

Kidney disease (67,473 deaths/year): The FLOW trial was stopped early by its Data Monitoring Committee after semaglutide demonstrated a 24% reduction in kidney disease progression. Note: this measures progression to end-stage renal disease or sustained eGFR decline, not kidney disease mortality directly. Stopped-early trials also carry overestimation bias. Applying 24% to 67,473 deaths gives a ceiling of 16,194, with the real number likely smaller.

Diabetes (102,188 deaths/year): This is the original indication. GLP-1 receptor agonists have been FDA-approved for type 2 diabetes since 2005 (exenatide) with well-established A1c reduction and weight loss data across dozens of trials. A conservative mortality reduction estimate of 20%, consistent with cardiovascular risk reduction in diabetic populations, yields 20,438 fewer deaths.

Total from RCT-backed indications: 209,786 fewer deaths per year (ceiling estimate; see caveats above).

That number is not a forecast. It is a theoretical maximum that assumes universal treatment, proportional translation from event reduction to mortality reduction, and no competing risks. None of those assumptions hold in practice. Realistic penetration rates of 10-30%, combined with the MACE-to-mortality gap, likely cut the number to 15,000-50,000 actual lives saved. Still, in a year when the FDA typically approves drugs that affect thousands of patients at a time, a single drug class with plausible reach across 200,000+ deaths is worth the arithmetic.

The Observational Middle: Promising, Unproven

Three additional indications have positive data, but none of it comes from randomized trials.

Cancer (608,371 deaths/year): A retrospective cohort study published in JAMA Oncology used target trial emulation on the OneFlorida+ electronic health record system, matching 86,632 adults. GLP-1 receptor agonist use was associated with a 17% reduction in overall cancer incidence (HR 0.83; 95% CI 0.76-0.91; P=.002). Subgroup effects were dramatic: 47% reduction in ovarian cancer (HR 0.53), 31% in meningioma (HR 0.69), 25% in endometrial cancer (HR 0.75). One red flag: a possible increase in kidney cancer (HR 1.38; 95% CI 0.99-1.93), which missed statistical significance but sat right on the border.

Applied naively, 17% of 608,371 equals 103,423 fewer cancer deaths. I do not believe that number. Obesity is a known risk factor for at least 13 types of cancer. A cohort of people on semaglutide is a cohort losing 15% of body weight. Distinguishing the drug effect from the weight-loss effect requires a randomized trial. None exists.

Alzheimer's disease (119,399 deaths/year): Observational data from target trial emulations involving more than one million patients with type 2 diabetes showed semaglutide associated with a 40-70% reduced risk of Alzheimer's disease. In women, the hazard ratio hit 0.22. These are enormous effect sizes.

Then came EVOKE. The Phase III EVOKE and EVOKE+ trials enrolled 3,800+ patients with early symptomatic Alzheimer's and randomized them to semaglutide or placebo. Result: no superiority on the primary cognitive endpoint. Semaglutide did reduce phosphorylated tau, an Alzheimer's biomarker, and showed 18% slower cognitive decline on secondary endpoints. But the trial was not powered for cognition, and "failed primary endpoint" is the phrase that matters in drug development. EVOKE is the strongest evidence that large observational signals in GLP-1 research may be confounded, not causal.

Liver disease (56,585 deaths/year): GLP-1 agonists have shown improvement in non-alcoholic steatohepatitis (NASH) markers in early studies, with resmetirom (not a GLP-1, but often studied alongside) recently receiving FDA approval for NASH. Semaglutide's Phase II NASH data showed histological improvement, but Phase III results for liver endpoints are still pending. Early-stage signal only.

The Mortality Coverage Table

No one, to my knowledge, has published this calculation: mapping every top-10 cause of death against GLP-1 evidence quality simultaneously. Here it is.

Rank	Cause of Death	Annual Deaths	GLP-1 Signal	Evidence Type	Ceiling Estimate†
1	Heart disease	702,880	20% MACE ↓	RCT (SELECT)	140,576
2	Cancer	608,371	17% incidence ↓	Observational	103,423*
3	COVID-19	186,702	None	-	-
4	Accidents	227,039	None	-	-
5	Stroke	162,890	MACE composite	RCT (SELECT)	32,578
6	Chronic lower resp.	142,342	None	-	-
7	Alzheimer's	119,399	40-70% risk ↓	Obs. (RCT failed)	47,760*
8	Diabetes	102,188	Original indication	RCT (multiple)	20,438
9	Kidney disease	67,473	24% progression ↓	RCT (FLOW)	16,194
10	Liver disease	56,585	NASH signal	Early phase	TBD

*Observational only. No RCT confirmation. The Alzheimer's estimate is particularly suspect given EVOKE's failure.

†Ceiling estimates assume universal treatment and proportional MACE-to-mortality translation. Actual mortality reductions would be substantially lower. See methodology caveats in each section.

Total if all signals held: ~361,000 (ceiling). RCT-supported ceiling: ~209,000. Realistic penetration- and methodology-adjusted estimate: 15,000-50,000.

The Strongest Counterargument

Every observational GLP-1 signal may be a weight-loss confound dressed in pharmaceutical clothing.

Consider the mechanism. A patient on semaglutide 2.4 mg loses, on average, 15% of body weight over 68 weeks. That patient eats fewer calories. Metabolic syndrome markers improve: fasting glucose drops, triglycerides fall, blood pressure decreases, systemic inflammation declines. Each of these downstream changes independently reduces the risk of heart disease, stroke, cancer, kidney disease, and liver disease. Obesity is a recognized risk factor for 13 cancers. It accelerates kidney disease. It worsens NASH. It drives cardiovascular events.

So when a retrospective study finds that GLP-1 users get less cancer, is that the drug or is that the 40 pounds they lost? Target trial emulation attempts to control for this, but no statistical technique fully substitutes for randomization. EVOKE proved the point: a massive observational signal (HR 0.33 for Alzheimer's vs. insulin users) collapsed when subjected to a proper randomized trial.

If the cancer signal follows the same pattern as Alzheimer's, the true mortality coverage of GLP-1 drugs shrinks from 7-of-10 to 4-of-10. That is still historically significant for a single drug class, but it is a different story than the one the observational hype suggests.

What This Analysis Cannot Show

Several important limitations bound this calculation.

First, the mortality reduction estimates assume universal treatment. In practice, GLP-1 drugs cost approximately $1,000 per month without insurance in the United States. Even with expanded coverage under Medicare Part D and private insurers, penetration rates in eligible populations remain below 10%. Cost is the largest single barrier between the theoretical ceiling and actual lives saved.

Second, most trial populations are obese or diabetic. Whether GLP-1 agonists produce cardiovascular or renal benefits in normal-weight, non-diabetic patients is unknown. Extrapolating from the SELECT trial (BMI ≥27 required) to the general population introduces unquantifiable uncertainty.

Third, long-term safety data beyond 3-4 years is sparse. Rodent studies have shown thyroid C-cell tumors at supratherapeutic doses. The kidney cancer signal in the JAMA Oncology paper (HR 1.38) did not reach statistical significance but warrants monitoring as larger datasets accumulate.

Fourth, Novo Nordisk funded SELECT. Novo Nordisk funded FLOW. Novo Nordisk funded EVOKE. When the same company that sells the drug funds most of the trials studying the drug, conflicts of interest deserve explicit acknowledgment, even when the trials are well-designed and independently monitored.

Fifth, the central calculation in this article applies MACE event reductions and disease progression endpoints to mortality figures. This overstates the case. A 20% reduction in cardiovascular events does not equal a 20% reduction in cardiovascular deaths. A 24% reduction in kidney disease progression does not mean 24% fewer kidney disease deaths. All numbers in the table above are ceilings, not predictions.

The Bottom Line

GLP-1 receptor agonists have RCT-backed mortality reduction signals across four of the top 10 causes of death in America and observational signals across three more. That 4-of-10 floor is already unprecedented for a single drug class. But the gap between 4 and 7 is the gap between randomized evidence and retrospective hope, and EVOKE just demonstrated how wide that gap can be. Track the cancer RCTs when they come. Track the NASH Phase IIIs. And be deeply skeptical of any headline that conflates "associated with" and "caused."