Ask a cardiologist whether testosterone threatens the heart, and the answer depends on which study they read last.
A Mendelian randomization analysis says yes — genetically higher testosterone raises coronary artery disease risk by 17%. A prospective cohort study says the opposite — lower testosterone predicts earlier death. The largest randomized trial says neither — treatment is cardiovascular-neutral. A long-term retrospective study says treatment increases risk by 55%. An electronic health record study says treatment is safe.
Five studies. Five answers. The field treats this as a controversy to be resolved — eventually, a sufficiently large trial or a sufficiently clever meta-analysis will tell us which answer is right. But the studies don't contradict each other. They aren't even asking the same question.
What Each Study Actually Measures
The table below strips away the framing. Each row shows a different study design, what it actually exposes the body to (or correlates with), the population, and what it found. Read the third column carefully — that's where the construct lives.
| Design | What It Measures | Population | Finding |
|---|---|---|---|
| Mendelian randomization Morbey et al., JCEM 2025 |
Lifetime genetic exposure — decades of slightly higher endogenous T, encoded in SNPs, never chosen | General male population (UK Biobank) | OR 1.17 for CAD mediated by blood pressure |
| Prospective cohort UK Biobank / Yeap & Anawalt, JCEM 2026 |
Cross-sectional endogenous level — a single blood draw, reflecting current health, adiposity, inflammation | ~200K men, baseline survey | Low T → all-cause mortality but NOT CV events or CV death |
| RCT (TRAVERSE) Lincoff et al., NEJM 2023 |
Medium-term pharmacological exposure — 1.6% testosterone gel vs placebo, ~22 months median | n=5,246, ≥45y, hypogonadal, ≥1 CV risk factor | HR 0.96 for MACE noninferiority met |
| Long-term retrospective Glasgow / JES 2025 |
Chronic prescription exposure — ≥2yr prescription window in NHS records, mean 8.3yr follow-up | n=440 exposed vs 136,051 unexposed, ≥51y | HR 1.55 (1.19–2.01) severe confounding by indication |
| EHR real-world Lin et al., JES 2025 |
Matched treatment cohort — propensity-scored T users vs non-users across 3 health systems, median ~3yr | n=928 vs 928, from 2,663 TD men | HR 0.93 (0.68–1.26) neutral |
The Construct Does the Work
The MR study measures what it means to be a man who produces slightly more testosterone for an entire lifetime. That's not the same exposure as a hypogonadal 58-year-old applying gel for two years. The prospective cohort measures what low testosterone reflects — obesity, inflammation, chronic illness — and finds it predicts death the way fever predicts infection: as a signal, not a cause. TRAVERSE measures the pharmacological safety of topical testosterone gel in carefully selected, monitored patients. The Glasgow study measures what it means to be the kind of person who gets a long-term testosterone prescription in the NHS — which says as much about the patient as the drug. And the EHR study tries to match away those differences, arriving back at neutral.
None of these is wrong. Each is measuring something real. But they are not measuring the same thing, and treating their results as comparable — as votes to be counted for or against testosterone — is the error.
"A saga that refuses to die."
— Franck Mauvais-Jarvis, JCEM editorial on testosterone and cardiovascular disease, 2026
It refuses to die because we keep asking "Does testosterone protect or harm the heart?" as if it were one question. It isn't. The MR study asks about a genetic endowment. The cohort study asks about a biomarker. The RCT asks about a medication. The retrospective asks about a treatment pattern. The EHR study asks about a matched clinical decision. Five questions, five instruments, five constructs — each faithfully returning an answer to what it was actually asked.
What This Means Clinically
For a man considering testosterone therapy: the study that most directly addresses your situation is TRAVERSE, and it found no increased cardiovascular risk over ~22 months in men with pre-existing CV risk factors. The Yeap and Anawalt review adds that low endogenous testosterone predicts overall mortality but not cardiovascular events specifically — meaning the low number on your lab result is a health marker, not necessarily a heart risk. The MR finding of 17% CAD risk applies to lifelong genetic exposure, not to a decision you're making at 50.
For a physician: the Glasgow HR of 1.55 comes from 440 exposed men in an NHS cohort with no adjustment for diagnosis severity, symptom burden, or the reason T was prescribed — the exposed group is not the general population receiving a random drug. Lin et al. tried to address this with propensity matching and found the signal disappeared. Confounding by indication is not a minor footnote here. It is the finding.
For the field: the cardiovascular question will not be resolved by another study, because resolution requires agreement on the construct being measured. Until we distinguish between genetic endowment, endogenous biomarker, pharmacological intervention, and long-term treatment pattern as fundamentally different exposures, every new study will simply add a sixth answer. Then a seventh.
The saga refuses to die because the question was never singular.
Sources: Morbey et al., JCEM 2025 (MR, OR 1.17, DOI: 10.1210/clinem/dgaf582) · Yeap & Anawalt, JCEM 2026 (review, DOI: 10.1210/clinem/dgaf622) · Lincoff et al., NEJM 2023 (TRAVERSE, HR 0.96) · Glasgow retrospective, JES 2025 (HR 1.55, PMID 41163812) · Lin et al., JES 2025 (EHR, HR 0.93, bvaf074) · Mauvais-Jarvis, JCEM 2026 (editorial, DOI: 10.1210/clinem/dgaf629)