TRAVERSE Cleared Testosterone on the Heart. Almost Nobody Read the Rest of the Trial.
For about a decade the argument over whether testosterone therapy hurts the heart was conducted almost entirely with weak evidence. Two observational analyses in 2013 and 2014 suggested harm, the FDA added a warning to testosterone labels in 2015, and prescribing continued to climb anyway because nobody had run the trial that would settle it. In 2018 that trial started enrolling. TRAVERSE randomized 5,246 men with symptomatic low testosterone and serious cardiovascular risk to daily testosterone gel or placebo gel, followed them for a mean of twenty seven months, and adjudicated every event. When it reported in 2023 the answer was clean: no increase in major adverse cardiac events. That result deserved the attention it got. What happened next is the reason to write about it now. The cardiovascular verdict became the entire public memory of the trial, and the other things TRAVERSE measured, several of which are less comfortable and one of which came out in the exact opposite direction from what the field expected, mostly stayed inside the papers. This article walks through what the trial and its substudies reported. It is educational background on published data, not medical advice, a protocol, or a recommendation to start or stop anything.
The Question That Took a Decade to Get a Real Answer
The case against testosterone was never strong, but it was loud. A 2013 analysis of veterans and a 2014 claims-database study both reported more cardiovascular events among men who had been prescribed testosterone. Neither was randomized. Men who get prescribed testosterone differ from men who do not in ways that are difficult to fully adjust for, and the studies drew heavy methodological criticism almost immediately. The FDA still added a cardiovascular warning to testosterone product labels in 2015 and, more consequentially, required manufacturers to run a properly powered safety trial. That requirement is what produced TRAVERSE. The design was built to be difficult to argue with. It was randomized, double blind, placebo controlled, and deliberately enrolled the men most likely to have an event, because a safety trial in low risk participants would take forever to accumulate enough endpoints to say anything. Participants were 45 to 80 years old, had two fasting morning testosterone measurements below roughly 300 nanograms per decilitre, reported symptoms consistent with hypogonadism, and had either established cardiovascular disease or a stack of risk factors that put them close to it. They received transdermal testosterone gel titrated to a normal range, or an identical placebo gel, every day.
The Number the Headlines Were Written From
The primary endpoint was the standard composite used across cardiovascular outcome trials: death from cardiovascular causes, non-fatal heart attack, or non-fatal stroke, whichever came first. It occurred in 182 men on testosterone, which is 7.0 percent of that arm, and 190 men on placebo, 7.3 percent. The hazard ratio was 0.96 with a confidence interval running from 0.78 to 1.17, and the trial comfortably met its prespecified noninferiority margin. Read plainly, this says that in a large population of men who were both genuinely hypogonadal and genuinely at cardiovascular risk, replacing testosterone to a normal range for a little over two years did not raise the rate of the events everyone had been worried about. That is a real finding produced by a well-run trial, and it substantially weakened the observational case that started the argument. It is also, on its own, a narrower statement than the coverage made it sound. Noninferiority on one composite endpoint over twenty seven months is not a general finding of safety, and the trial did not claim to be one.
The Endpoint the Trial Was Built to Answer
Incidence of the primary composite of cardiovascular death, non-fatal heart attack and non-fatal stroke over a mean 27.1 months, in 5,246 men aged 45 to 80 with symptomatic hypogonadism and either established cardiovascular disease or a high risk of it. The primary event occurred in 182 men on testosterone and 190 on placebo. The hazard ratio was 0.96, and the trial met its noninferiority margin. This is the number every headline was written from, and it is a real result.
Noninferior Is a Narrow Word
Cardiovascular outcome trials collect far more than their primary endpoint, and the tables further down the TRAVERSE papers are where the trial stops being purely reassuring. Atrial fibrillation was reported in 91 men on testosterone against 63 on placebo. Acute kidney injury ran 60 against 40. Pulmonary embolism was 24 against 12, which is a doubling in relative terms even though the absolute numbers stay small. The two arms were nearly identical in size, so these counts can be compared directly. Every one of these was a secondary or exploratory outcome rather than something the trial was powered to detect, which matters and should temper how hard anyone leans on them. In a trial reporting dozens of outcomes, some will come out lopsided through chance alone. The counterargument is that these particular ones are not random draws from a hat. Testosterone raises haematocrit, which thickens blood, and a thicker circulation is a mechanistically plausible route to a clot lodging in the lung. Atrial fibrillation and thromboembolism have appeared in other testosterone datasets. Reviewers have also pointed out that a large number of participants entered the trial with stage 3 kidney disease already, which muddies the kidney signal considerably. None of these are verdicts. Three signals leaning the same way in the largest trial ever run on the question are still worth knowing about.
The Rows Further Down the Safety Table
Adjudicated event counts in the two arms, which were of near-identical size. Atrial fibrillation ran 91 against 63, acute kidney injury 60 against 40, and pulmonary embolism 24 against 12. These were secondary and exploratory outcomes rather than the endpoint the trial was powered for, so each is a signal rather than a verdict. Three signals pointing the same direction is still worth reading.
- Testosterone (n)
- Placebo (n)
The Fracture Result Nobody Ordered
The most interesting thing TRAVERSE produced was not in the main paper at all. A prespecified substudy tracked clinical fractures in 5,204 of the participants over a median of 3.19 years, with every reported fracture adjudicated rather than taken from a checkbox. The expectation going in was straightforward. Testosterone raises bone mineral density in men, by several percent in trabecular bone, and higher bone density is associated with fewer fractures. Most people involved assumed the substudy would show a modest protective effect and close the question. It did not. After adjudication, 3.50 percent of the testosterone group had sustained a clinical fracture against 2.46 percent of the placebo group, a hazard ratio of 1.43 with a confidence interval of 1.04 to 1.97. That interval does not cross 1.0. The trial did not merely fail to show the expected benefit. It pointed the other way, in the direction of more fractures on treatment, in the only trial large enough and long enough to count them properly.
Adjudicated Clinical Fracture, Against the Expected Direction
Clinical fracture in the 5,204 men of the fracture substudy over a median 3.19 years of follow-up. After adjudication, 3.50 percent of the testosterone group had a fracture against 2.46 percent on placebo, a hazard ratio of 1.43 with a confidence interval running from 1.04 to 1.97. Testosterone reliably raises bone mineral density, so the trial was widely expected to show fewer fractures. It showed numerically more.
Why Denser Bone Can Still Break More Often
A result like that is worth sitting with rather than explaining away, because it is a clean demonstration of something that trips up a great deal of health reasoning. Bone mineral density is a surrogate endpoint. It is a number that stands in for the thing anyone actually cares about, which is whether a bone breaks. Surrogates are used because they are cheap and fast to measure, and most of the time they track the real outcome well enough to be useful. Sometimes they do not, and a fracture is not purely a function of how much mineral is in the bone. It is a function of mineral, of bone architecture and quality, and of whether the person fell over in the first place. A treatment that makes men feel more energetic and more physically capable may plausibly change how much they do, how much they lift, and how much risk they take on their feet, all of which affect the fall side of that equation rather than the bone side. The substudy authors were careful not to claim they had identified the mechanism, and it would be wrong to state one here as though it were established. What can be said is narrower and still useful. Testosterone did what it was supposed to do to the density number and did not deliver the fracture reduction that number predicted, so anyone reasoning from a scan result to a real-world outcome is making an inference this trial specifically failed to support.
The Gap That Looks Convincing and Is Not
A second substudy followed the 1,175 participants who had prediabetes at baseline, 607 on testosterone and 568 on placebo, to see whether restoring testosterone slowed the slide into type 2 diabetes. The rationale was reasonable, since low testosterone travels with insulin resistance and visceral fat. Plot the two curves and they look like a drug working. At six months, 0.7 percent of the testosterone group had progressed against 1.4 percent on placebo. At twelve months it was 7.8 against 10.7, at twenty four months 10.1 against 14.6, at thirty six months 12.8 against 15.8, and at forty eight months 13.4 against 15.7. The placebo line sits above the testosterone line at every single timepoint, which is exactly the shape a genuine effect makes. None of those differences reached statistical significance, and the substudy authors concluded that testosterone therapy did not reduce progression from prediabetes to diabetes. This is worth showing precisely because the chart is more persuasive than the statistics. A consistent-looking gap across five timepoints in a study that was not sized to detect one is not evidence of a benefit, and a review that reads that same substudy as a twenty two percent reduction in new-onset diabetes has read the picture rather than the result.
A Gap That Looks Convincing and Did Not Hold Up
Cumulative progression from prediabetes to diabetes among the 1,175 men in the glycaemic substudy, 607 on testosterone and 568 on placebo. The placebo line sits above the testosterone line at every timepoint, which is exactly the shape a real benefit would make. None of the differences reached statistical significance, and the substudy authors concluded that testosterone did not slow progression to diabetes. This is a useful picture of how a consistent-looking gap can still be nothing.
- Testosterone (%)
- Placebo (%)
What Actually Improved
The trial was not a string of disappointments, and skipping the genuine positives would misrepresent it as badly as skipping the safety signals. Anemia responded clearly. Among participants who were anemic at baseline, roughly 41 percent of the testosterone group had corrected by six months against roughly 28 percent on placebo, which is the kind of separation that does not need careful reading to see. Sexual function improved too, but the shape of the improvement is instructive. Sexual desire and sexual activity both rose significantly on testosterone. Erectile function did not. That distinction lines up with what testosterone is understood to do, which is act centrally on drive rather than on the vascular machinery of an erection, and it is exactly the nuance that gets flattened in advertising aimed at men who are worried about a specific problem. Depressive symptoms and energy showed little to no benefit in the relevant substudies. Prostate events were low and similar between arms, with five high-grade cancers on testosterone and three on placebo, which is reassuring against one of the older fears about the therapy even though the numbers are too small to settle it.
Who These 5,246 Men Were, and Who They Were Not
Every number above describes a specific population, and the gap between that population and the men who most often read about testosterone online is large enough to matter. TRAVERSE participants had to fail two separate fasting morning testosterone measurements, report actual symptoms, and be either 45 to 80 with established cardiovascular disease or carrying enough risk factors to be close to it. They had a mean age around 63. This is a group with real, measured, symptomatic hypogonadism and substantial comorbidity. It is not a 38-year-old with a mid-range testosterone result and a general sense that he could be sharper. The trial cannot say what daily gel does to that man over twenty years, because it did not enroll him and ran for under three years. It also studied one specific formulation, a transdermal 1.62 percent gel titrated into the normal range, which is not interchangeable with injectable esters that produce very different peaks and troughs, and not remotely comparable to supraphysiological dosing. Extending a result past the population, the dose, the delivery route, and the duration that produced it is the most common way good trial data gets misused, and it happens in both directions, by people selling testosterone and by people warning about it.
A Reasonable Reading Without Overreaching
Put every adjudicated outcome on one axis and the trial reads as neither the exoneration nor the indictment it gets used as. The primary cardiac endpoint came in at 0.96 and did not move. Progression to diabetes came in around 0.85 and was not significant. Clinical fracture landed at 1.43, atrial fibrillation near 1.44, acute kidney injury near 1.50, and pulmonary embolism at roughly double, and those four are the ones that reached significance or came close enough to warrant attention. Ratios presented this way hide the confidence intervals and hide the fact that only one of these was a powered primary endpoint, so the picture is a map of where to look rather than a scoreboard. What is established is that testosterone therapy in symptomatic hypogonadal men at cardiovascular risk did not increase major adverse cardiac events over roughly two years, that it corrects anemia and raises sexual desire without moving erectile function, and that it did not reduce fracture despite raising bone density. What is not established is what happens beyond three years, what any of this means for men outside the enrolled population, whether the thromboembolic and arrhythmia signals are causal, or why the fracture result went the way it did. Those are open questions, and a trial that answers one thing well while opening four more is a normal outcome in clinical research rather than a scandal. The practical version has nothing to do with buying anything: the cardiovascular fear that dominated this conversation for a decade has been substantially addressed, the conversation that replaces it should be about haematocrit, clotting risk, and the limits of reasoning from a bone scan, and decisions about testosterone therapy belong with a clinician who can see the labs and the history. This article is educational and describes published trial data. It is not a protocol and not a recommendation to start or stop any treatment.
The Whole Trial on One Axis
Each adjudicated outcome as a single ratio of testosterone against placebo, arranged so that lower is always better and the vertical line at 1.0 marks no difference between the arms. The two outcomes to the left of that line, progression to diabetes and the primary cardiac endpoint, did not reach statistical significance and are drawn in grey. The four to the right of it did, and are drawn in red. Ratios alone hide the confidence intervals and the fact that only the cardiac endpoint was a powered primary outcome, so this is a map of where to look rather than a scoreboard.
Frequently Asked Questions
Have more questions? Contact us or explore our full peptide intelligence platform.
Related Articles
Tribulus Is Sold as a Testosterone Booster. Eight of Ten Trials in Men Found No Change in Male Hormones, and the Erection Benefit It Does Show Arrives Without a Testosterone Rise.
Does tribulus raise testosterone? Eight of ten trials in men found no change. A 180-man trial did show better erections. What the research found, plus safety.
Three Beers a Night for Three Weeks Cut Testosterone by About 7 Percent. The Bigger Cost to Men Who Train Shows Up in the Muscle.
Does alcohol lower testosterone? In a diet-controlled trial, about 40 grams of alcohol a day for three weeks lowered testosterone in middle-aged men by 6.8 percent. Heavy drinking lowers it far more, and alcohol after training cut muscle protein synthesis by 24 to 37 percent. An educational review of the dose, the timing and what a month off actually changes.
One in Five Men Carries an Inherited Cholesterol Particle Their Standard Lipid Panel Never Measures. The First Drug Built to Lower It Just Missed.
Lipoprotein(a) is a mostly inherited cholesterol particle that raises heart attack and aortic stenosis risk, and about 1 in 5 people have elevated levels. The 2026 ACC/AHA guideline now advises testing it once in a lifetime, yet the pelacarsen Lp(a)HORIZON outcomes trial missed its primary endpoint in September 2026. An educational review of what the Lp(a) number means for men.