The 83,010-Veteran Study: What an Observational Cohort Can and Cannot Show

Sharma and colleagues examined 83,010 male veterans with documented low testosterone and sorted them three ways: treated with levels that normalised, treated without normalisation, and never treated. Men whose testosterone normalised had substantially lower all-cause mortality (hazard ratio 0.44, 95% CI 0.42–0.46), myocardial infarction (HR 0.76, 0.63–0.93) and stroke (HR 0.64, 0.43–0.96) than untreated men.[1]

Those are large effects in a very large cohort. They are also observational, and the distance between "associated with" and "causes" is the entire point of this article.

What the study did

• Population: 83,010 male veterans with documented low total testosterone, no previous myocardial infarction or stroke.
• Groups: Group 1 — testosterone therapy with resulting normalisation (n = 43,931, median age 66, mean follow-up 6.2 years). Group 2 — therapy without normalisation (n = 25,701, median age 66, mean follow-up 4.6 years). Group 3 — no therapy (n = 13,378, median age 66, mean follow-up 4.7 years).
• Method: retrospective cohort with propensity score-weighted Cox proportional hazard models.
• Outcomes: all-cause mortality, myocardial infarction, stroke, and a composite endpoint.

What it found

Group 1 versus Group 3 (normalised versus untreated):[1]

Outcome	Hazard ratio	95% CI
All-cause mortality	0.44	0.42–0.46
Myocardial infarction	0.76	0.63–0.93
Stroke	0.64	0.43–0.96

Group 1 versus Group 2 (normalised versus treated-but-not-normalised):[1]

Outcome	Hazard ratio	95% CI
All-cause mortality	0.53	0.50–0.55
Myocardial infarction	0.82	0.71–0.95
Stroke	0.70	0.51–0.96

And the comparison that gives the paper its internal logic: there was no difference in myocardial infarction or stroke risk between Group 2 and Group 3 — between men treated without achieving normal levels and men not treated at all.

That gradient is the study's strongest argument. If treatment mattered only as a marker of being the kind of patient who gets treated, Group 2 should have looked like Group 1. It looked like Group 3 instead.

Why the mortality number should still give you pause

A hazard ratio of 0.44 for all-cause mortality means the normalised group died at less than half the rate of untreated men. For comparison, TRAVERSE — a randomized trial — found cardiovascular event rates statistically equivalent between testosterone and placebo, and the Hudson meta-analysis found an odds ratio of 1.07 for cardiovascular events.

When an observational study reports a benefit several times larger than randomized trials of the same intervention can find, the most likely explanation is not that the trials missed it.

Healthy adherer bias is the standard candidate. Achieving a normalised testosterone level requires attending appointments, filling prescriptions, tolerating the medication and returning for repeat labs. Men who do all of that reliably differ from men who do not, in ways that predict survival and that no propensity model built on claims data can fully capture. The outcome being predicted — normalisation — is partly a measure of engagement with healthcare.

Propensity weighting adjusts for what was measured. It cannot adjust for adherence behaviour, frailty, or health-seeking disposition that was never recorded.

The smoking sub-analysis

A later analysis from the same group examined 18,055 of these men by smoking status and found something that sharpens the picture: among non-smokers, normalisation was associated with lower mortality and lower MI. Among current smokers, normalisation was associated with lower mortality but no benefit for MI (HR 1.096, 95% CI 0.698–1.720, P = .69).[2]

Compared with non-smokers who normalised, current smokers who normalised had significantly higher mortality, MI and stroke. The authors concluded that active smoking may negate the protective association.

How to use this study

Not as evidence that testosterone prevents heart attacks — no observational cohort can carry that claim, and the randomized evidence does not support it.

What it does contribute is scale and consistency. Alongside Hudson and TRAVERSE, it is part of why the position on testosterone and cardiovascular risk shifted from suspected harm to no demonstrated harm. Its dose-response gradient — normalised better than not-normalised, not-normalised no better than untreated — is the kind of internal structure that makes a cohort finding more credible, without making it causal.

The full arc of this argument is told in what we actually knew about low-T and the heart.

Bottom line

83,010 veterans, six years of follow-up, and a consistent association between normalised testosterone and better cardiovascular and mortality outcomes — with a dose-response gradient that resists the simplest confounding explanations, and an effect size on mortality large enough that residual confounding remains the leading explanation. Strong supporting evidence; not proof of benefit.

References

2 sources

1. Sharma R, Oni OA, Gupta K, et al. Normalization of testosterone level is associated with reduced incidence of myocardial infarction and mortality in men. Eur Heart J. 2015;36(40):2706–15. doi:10.1093/eurheartj/ehv346 · PMID 26248567
2. Oni OA, Sharma R, Chen G, et al. Normalization of Testosterone Levels After Testosterone Replacement Therapy Is Not Associated With Reduced Myocardial Infarction in Smokers. Mayo Clin Proc Innov Qual Outcomes. 2017;1(1):57–66. doi:10.1016/j.mayocpiqo.2017.05.003 · PMID 30225402