Low Testosterone Is a Flag, Not a Prescription

The Short Version

• Across 24,510 men and 276,931 participant-years, the lowest fifth of total testosterone carried an 18 percent higher risk of cancer death than the highest fifth. That is an association measured in observational data. It is not a reason to write a testosterone prescription.
• The finding that sells nothing is the one that matters most. A man's own testosterone showed no association with his risk of being diagnosed with prostate cancer. A fear that has shaped fifty years of prescribing just lost more of its evidence base.
• The two hormones that did predict prostate cancer, SHBG and luteinising hormone, are the two almost nobody orders. Their relationship to cancer death is a U, not a line. Mid-range was safest. More is not better at either end.

On June 19 a paper landed in The Lancet Healthy Longevity. Within a week the coverage had compressed into five words: low testosterone raises cancer risk.

I have read the paper. That summary is close enough to be repeated and wrong enough to matter.

Here is what will happen next, and some of it already has. Every men's clinic with a marketing budget will put that sentence in an ad. Somewhere between the abstract and the landing page, "associated with" will quietly become "causes," and "we found a biomarker" will become "we found a reason to treat you." The authors have said plainly that the second thing does not follow from the first. That correction will travel about a tenth as far as the headline.

So let me be precise about what I think of this study, because my objection is not to the research. This is a better paper than most of what gets written about male hormones, and the parts of it that will never be quoted are the parts worth your time.

It is a biomarker paper. It tells you where to look. It does not tell you what to write.

This is a better study than the ones you are used to reading

The design is an individual participant data meta-analysis, which means the investigators pooled the raw records from ten prospective cohorts rather than averaging one another's published summary estimates. That distinction matters more than it sounds. Pooling summaries means inheriting every other group's modeling choices. Pooling participants means you get to make one consistent set.

The inclusion criteria are where the real quality control lives. Cohorts had to enroll at least 1,000 community-dwelling men, follow them at least five years, and, critically, measure testosterone by mass spectrometry. The protocol was registered on PROSPERO in 2019, before the results existed. Risk of bias was scored on the Newcastle-Ottawa scale. Models were adjusted for age, previous cancer, BMI, marital status, alcohol, smoking, physical activity, hypertension, and diabetes.

That mass spectrometry requirement is the one to sit with. A great deal of the testosterone literature rests on immunoassay, which is cheap, fast, and least reliable exactly where clinical decisions get made, at the low end of the range. Requiring mass spectrometry threw out a lot of studies. It also threw out a lot of noise.

I want to flag something about that, because I have criticized this exact move before. When I wrote about the NASEM report on compounded hormone therapy, my complaint was that a committee reviewed thirteen studies and set hundreds of submissions aside, then reported a "dearth of evidence" it had manufactured through exclusion. Here you have exclusion again, aggressive exclusion, and I think it is the right call. The difference is direction. NASEM excluded evidence and then drew a conclusion the exclusion had preselected. Marriott and colleagues excluded measurement methods and then reported a result that cuts against their own funder. Same tool. Opposite intent. That is what the standard is supposed to look like when it is applied honestly.

The number everyone will quote, and what it can carry

Men in the lowest quintile of total testosterone had a hazard ratio of 1.18 for cancer death compared with the highest quintile, 95% confidence interval 1.04 to 1.34. Dihydrotestosterone behaved almost identically at 1.21.

Note where the lower bound sits. It is 1.04. This result clears 1.00, and it clears it narrowly. That is not a criticism, it is a description, and it should shape how hard you lean on the finding.

Now convert it into something you can feel, because a relative risk on its own is the single most abused number in medicine. The paper reports 2,847 cancer deaths across 276,931 participant-years. That works out to roughly 10 cancer deaths per 1,000 men per year across the pooled cohorts. An 18 percent relative increase on a baseline like that is on the order of two additional cancer deaths per 1,000 men per year, which over a decade lands somewhere near one man in fifty.

The paper does not publish absolute risks, so treat that arithmetic as an order of magnitude rather than a figure to quote back. It is real. It is also nothing like what "low testosterone raises cancer risk" puts in a reader's head.

The finding that sells nothing

Look at the last row of that plot. Across 12,280 men, 151,373 participant-years, and 918 diagnosed prostate cancers, a man's own circulating testosterone showed no association with whether he developed prostate cancer.

Understand how long the opposite assumption has been load-bearing. Lowering testosterone treats advanced prostate cancer, and it has since the 1940s. From that clinical fact the field ran the inference backwards for decades: if removing androgen treats the disease, having more androgen must cause it. That inference has been eroding for years. This paper, built on gold-standard assays and pooled raw participant data, is one of the cleaner tests of it yet, and it finds nothing there.

A null result will never lead the coverage, because nothing happened and nothing sells. It is also, for a practicing clinician, the most immediately useful sentence in the paper, because it removes a reflex objection that has kept appropriate therapy off the table for men who needed it.

The two hormones almost nobody orders

What did predict prostate cancer was the pair sitting in the middle of that plot. Lower SHBG carried a hazard ratio of 1.28. Lower luteinising hormone carried 1.45, the largest single effect anywhere in the paper.

Think about what gets ordered in practice. A man reports fatigue and low libido, and a total testosterone comes back. Maybe a free testosterone. SHBG shows up occasionally as an afterthought to interpret the free value, and LH shows up when someone is specifically working up the cause. Neither is ordered as a risk signal. In this dataset they were the risk signal, and total testosterone, the one number everybody orders, was the one that carried nothing.

Then there is the shape. For cancer death, SHBG and LH were not linear. Both were U-shaped, with the lowest risk in the middle of the distribution: SHBG at the third quintile, LH at the second. Men at the top of the range and men at the bottom both did worse than men in the middle.

A U-curve breaks the rule that clinical software and clinical habit are both built on. You cannot write "below X is concerning" when above X is also concerning. The moment a relationship is U-shaped, a threshold stops being a summary of the evidence and starts being a replacement for it.

The lines do not line up

Here is the detail I have not seen anyone mention, and it takes about thirty seconds with a conversion factor to find.

The paper reports that risk of cancer death began rising below 8.6 nmol/L, and risk of incident cancer below 7.3 nmol/L. American labs report in ng/dL. Multiply by 28.84 and those inflection points land at roughly 248 and 211 ng/dL. The harmonized lower limit of normal for healthy young men, the CDC-standardized figure the Endocrine Society uses, is 264 ng/dL, or 9.2 nmol/L.

Read that graphic and the practical problem becomes obvious. The threshold your lab flags is not the threshold this paper identified. A man at 9.0 nmol/L gets an L next to his result and sits above both inflection points. A man at 8.0 nmol/L sits below the cancer-death inflection and above the incident-cancer one. The flag and the signal are neighbors. They are not the same thing, and treating the flag as though it carries the paper's finding is how a careful result turns into a careless decision.

Reverse causation is not a footnote here

Now the part that should temper everything above, and the reason I keep saying biomarker instead of cause.

Cancer that has not been diagnosed yet lowers testosterone. So does chronic illness, so does poor nutritional status, so does the accumulated systemic burden that tends to arrive before a cancer diagnosis does. Obesity lowers testosterone and independently raises cancer risk. The investigators adjusted for BMI and for a reasonable list of comorbidities, and requiring five years of follow-up pushes back against the most obvious version of the problem. Neither move eliminates it.

The most defensible reading is that low testosterone in an older man is often a readout of general physiological reserve. It may be the smoke rather than the fire. The senior author has been consistent about this in interviews, and his practical recommendation is the conservative one: a man with a genuinely low testosterone deserves a full workup, because the number is telling you something about him that is probably larger than his gonads.

The authors' own conclusion is careful in exactly the way the coverage is not. They write that sex hormones "could serve as biomarkers for cancer risk," and they ask for further investigation. That is the correct size of claim for this evidence.

Who funded it, and why that cuts the other way

My habit with any hormone paper is to read the funding line first. This one lists the Medical Research Future Fund, the Government of Western Australia, and Lawley Pharmaceuticals, which manufactures testosterone therapy.

State it plainly: a testosterone manufacturer helped pay for a study about testosterone and cancer. That is a real interest and you are entitled to weigh it. Be honest about which findings it touches, too. The prostate cancer null is commercially convenient for a company that sells testosterone, because it retires the oldest objection to the product. Nobody should pretend otherwise.

Then look at what the money did not buy. The commercially valuable version of this paper closes a loop: low testosterone predicts cancer death, therefore raise it. Everything was in place for that sentence. The paper refuses to write it, and its authors have gone out of their way in the press to say there is no evidence yet that giving men testosterone reduces this risk. That is the one sentence a marketing department would have asked them to cut.

The reason the convenient finding is still credible is that the method was fixed before the answer was known. The protocol went on PROSPERO in 2019. The assay requirement was a rule about measurement quality, not about which studies produced which result. You cannot reverse-engineer a null out of a design you registered years earlier.

I hold this paper to the same standard I held the NASEM report to, and it comes out the other way. Funding is a reason to read carefully. It is not a verdict. When a funded study declines to make the claim its funder would profit from, that is evidence of discipline, and it should raise your estimate of the work rather than lower it.

Why a single lab flag cannot carry any of this

Step back and count what a clinician would actually need to hold in mind to use this paper well. Four hormones. Two of them non-linear. Inflection points that do not coincide with the reference range. A prostate cancer signal that lives in the two analytes least likely to be on the panel. And a live possibility that the whole pattern is a marker of general health rather than a lever to pull.

Now count what the chart shows. One number, with an H or an L beside it.

That gap is not a knowledge problem, and no amount of continuing education closes it. It is an interface problem. A binary flag is a lossy compression of a continuous, multivariate, non-monotonic relationship, and the information it discards is precisely the information this paper produced.

I am not writing this from the bleachers. We are building clinical decision support at Nimbus, and this is the design constraint we keep running into: the useful output is almost never a recommendation. It is a well-drawn picture of where a patient sits, with the uncertainty left visible instead of rounded away. A system that says "consider testosterone therapy" has overstepped. A system that surfaces the SHBG and LH the clinician did not think to order, shows the shape of the association, and links the paper has done its whole job.

What I would actually do with this on Monday

1. Treat a low total testosterone as a reason to work the man up, not a reason to write a script. The paper's own senior author lands here. A number in the bottom quintile is a signal about a man's general physiology. Go find out what it is pointing at before you decide it is the thing to fix.

2. Order SHBG and luteinising hormone, and read the panel as a shape rather than a row of flags. They carried the prostate cancer signal in this dataset, where lower was worse. For cancer death the relationship was a U, and the middle of the range, not the top, was the safest place to be. If you are ordering a total testosterone at all, these two cost very little to add and tell you something the total cannot.

3. Ask your lab which assay it runs. This entire body of evidence was built on mass spectrometry. If your low-end results come from an immunoassay, you are comparing your patient against a literature your measurement does not match, and the disagreement will be worst exactly where you are making the decision.

4. Retire prostate cancer as the reflex objection, and do not let this paper become the reflex permission. Endogenous testosterone showed no association with incident prostate cancer, and that should change how the conversation opens. It says nothing about whether therapy is right for the man in front of you, and anyone who cites this study as evidence that it is has not read past the abstract.

The best thing about this paper is how carefully it refuses to overclaim. The worst thing that can happen to it is being quoted by people who will.

The number tells you where to look. It does not tell you what to write.

Key sources: Marriott RJ, Murray K, Antonio L, et al. "Associations of testosterone, sex hormone-binding globulin, and related hormones with risks of cancer death, incident cancer, and incident prostate cancer in men: individual participant data meta-analyses." Lancet Healthy Longev. 2026;7(6):100857. doi:10.1016/j.lanhl.2026.100857 (PMID 42320510, free full text; PROSPERO CRD42019139668). Reference range figures from Travison TG, et al. "Harmonized Reference Ranges for Circulating Testosterone Levels in Men of Four Cohort Studies in the United States and Europe." J Clin Endocrinol Metab. 2017;102(4):1161-1173. Author commentary from University of Western Australia press coverage, June 2026.

Disclosure: I am the founder and CEO of Nimbus Healthcare, which operates telehealth services and a compounding pharmacy. That is a commercial interest in men's health, and you should weigh this piece accordingly. This is a reading of the published evidence, not medical advice, and it is not a substitute for a clinician who knows your history.