HbA1c, glucose and lipids: where the cohorts and the trials disagree
Observational studies make testosterone look like a cardiometabolic tonic. The randomised trials say it moves body composition and very little else.
The part of the panel where the observational literature and the randomised literature tell two different stories, and where it matters enormously which one you were told.
Why these tests are on a testosterone workup
Because low testosterone travels with obesity and type 2 diabetes, and because obesity and diabetes are treatable in ways that low testosterone is not.
The direction of causation is not symmetrical, and this is the single most useful fact on the page. Genetic evidence settles it: a one standard deviation genetically determined increase in BMI produces a 0.25 standard deviation lower testosterone (95% CI −0.42 to −0.09), with no effect in the reverse direction. Adiposity lowers testosterone. Low testosterone does not measurably drive adiposity.
The practical corollary follows immediately. Weight loss raises testosterone, and the amount of weight lost is the best predictor of how much it rises: total testosterone rose 8.73 nmol/L after bariatric surgery and 2.87 nmol/L on a low-calorie diet in a pooled analysis, with degree of weight loss the strongest determinant.
So the metabolic panel is not there to decide whether a man needs testosterone. It is there because the thing driving his testosterone down may be the thing worth treating.
What the trials found about glycemia
Two large randomised trials dominate this literature and they appear to disagree. They do not, quite — but the reconciliation matters.
T4DM: the positive trial
1,007 Australian men aged 50 to 74, waist circumference 95 cm or more, testosterone 14.0 nmol/L or below without pathological hypogonadism, and either impaired glucose tolerance or newly diagnosed diabetes. All received intramuscular testosterone undecanoate 1,000 mg or placebo for two years — and every participant was enrolled in a structured community weight-management programme. That co-intervention is central and is routinely dropped when the trial is cited.
The result: a two-hour oral glucose tolerance test above 11.1 mmol/L at two years in 12% on testosterone versus 21% on placebo, relative risk 0.59 (95% CI 0.43 to 0.80). Mean two-hour glucose fell by an extra 0.75 mmol/L (95% CI −1.10 to −0.40). The effect was independent of baseline testosterone, so this was not deficiency correction — it looked like a pharmacological effect.
And the number that gets left out: hematocrit exceeded the 54% safety trigger in 22% of the testosterone arm versus 1% of placebo. The investigators flagged it themselves as potentially treatment-limiting.
TRAVERSE: the null trial
The diabetes substudy of TRAVERSE followed 1,175 men with prediabetes and 3,880 with diabetes for up to four years on transdermal gel or placebo. Progression from prediabetes to diabetes, testosterone versus placebo: 0.7% versus 1.4% at six months, 7.8% versus 10.7% at twelve, 10.1% versus 14.6% at twenty-four, 12.8% versus 15.8% at thirty-six, and 13.4% versus 15.7% at forty-eight. The pre-specified omnibus test gave p = 0.49. Glycemic remission in the diabetes group, and the changes in glucose and HbA1c, were similar in both arms.
The authors’ conclusion is blunt: testosterone alone should not be used to prevent or treat diabetes in men with hypogonadism.
Two notes on reporting. The point estimates favour testosterone at every single time point, and it is tempting to read a real effect into that — but the analysis was designed to test exactly this and did not find one. And the substudy used repeated-measures log-binomial regression, not a Cox model, so it produces no hazard ratio. Any hazard ratio attributed to the TRAVERSE diabetes substudy — including two we published on this site before August 2026 — is not from the paper.
How the two fit together
| T4DM (positive) | TRAVERSE (null) | |
|---|---|---|
| Endpoint | Two-hour OGTT glucose — a provoked glucose challenge | Clinical diabetes and HbA1c; glycemic remission |
| Population | Central obesity and dysglycemia; not selected for hypogonadism | Symptomatic hypogonadism, high cardiovascular risk |
| Lifestyle co-intervention | All participants in a weight-management programme | None mandated |
| Formulation | 1,000 mg injectable undecanoate — supraphysiological peaks | Titrated transdermal gel — physiological replacement |
| Duration | 2 years | Followed to 4 years |
The endpoint is the biggest single difference. An oral glucose tolerance test is highly sensitive to muscle mass and glucose disposal — which is precisely what testosterone changes. HbA1c is much less sensitive to those, and can be confounded in the opposite direction by testosterone-driven red cell production shortening red cell survival. A drug that adds muscle can move an OGTT without moving HbA1c.
So T4DM should be described as reducing the proportion of men crossing an OGTT diabetes threshold, not as preventing diabetes. And it tested testosterone as an add-on to weight loss, not as a substitute for it. Anyone citing T4DM to justify testosterone instead of weight management is inverting the trial’s own design.
Evidence: Strong (null) that testosterone is not a diabetes treatment or prevention drug. Moderate that a supraphysiological injectable regimen on top of structured weight loss moves a provoked-glucose endpoint.
The meta-analyses openly disagree, and the disagreement is instructive
Two meta-analyses of overlapping populations published a year apart reached different answers on HbA1c. One reported −0.87% (95% CI −1.32 to −0.42) — an effect roughly the size of adding a sulfonylurea, from 5 trials in 351 men. The other reported −0.15% (95% CI −0.39 to 0.10), non-significant, from 7 double-blind placebo-controlled trials in 833 men. The smaller pooled sample produced the much larger effect, which is the classic signature of small-study bias.
The same paper contains the methodological heart of the matter. Insulin resistance by the simple HOMA1 equation improved substantially, −1.58 (95% CI −2.25 to −0.91). By the better-specified HOMA2 model, the effect vanished: −0.19 (95% CI −0.86 to 0.49).
There is genuine mechanistic evidence on the other side and it deserves stating. In 44 men studied with the hyperinsulinemic-euglycemic clamp — the reference method, not a calculated index — glucose infusion rate rose 32% after 24 weeks of testosterone with no change on placebo, alongside 3.3 kg less subcutaneous fat and 3.4 kg more lean mass. Visceral and hepatic fat did not change.
That is the honest shape of it: testosterone does appear to improve insulin sensitivity by the best available physiological measure, in a 44-man study, and that improvement did not translate into HbA1c benefit in trials an order of magnitude larger. Both facts belong in the same paragraph.
What testosterone does to the lipid panel
HDL falls. The size of the fall depends entirely on which literature you read, and this is the cleanest example on the site of the observational-versus-trial gap.
In the largest pooled randomised safety analysis, across 51 studies, HDL cholesterol fell by 0.49 mg/dL (95% CI −0.85 to −0.13). That is half a milligram per decilitre. The same analysis found no significant effect on mortality, prostate outcomes or cardiovascular outcomes, and its authors described the lipid and hematological findings as “of unknown clinical significance” in a body of evidence that was “of low quality.”
In uncontrolled before-and-after data from 272 men on intramuscular esters, HDL fell 4 mg/dL (95% CI −5 to −2), total cholesterol fell 14 mg/dL and LDL fell 5 mg/dL. Eight times the HDL effect of the randomised estimate — and these are before-and-after differences with no placebo contrast.
A further meta-analysis of 29 trials found total cholesterol down 0.23 mmol/L (95% CI −0.37 to −0.10), no change in LDL, and an HDL reduction only in studies with higher baseline testosterone. And one meta-analysis in men with diabetes found HDL going up in a subgroup, contradicting the other three. We are reporting that rather than omitting it; a literature that disagrees with itself should look that way on the page.
Does the HDL fall matter?
Nobody has shown that it does, and the premise underneath the worry is itself shaky.
TRAVERSE, in 5,246 men at elevated cardiovascular risk, found major adverse cardiac events at hazard ratio 0.96 (95% CI 0.78 to 1.17), meeting non-inferiority. Whatever testosterone does to HDL, it did not produce excess atherothrombotic events. TRAVERSE’s real safety signals were atrial fibrillation, acute kidney injury and pulmonary embolism — a different set of problems entirely.
More fundamentally, the assumption that lowering HDL raises risk is not established. In a Mendelian randomisation study, carriers of a variant giving 0.14 mmol/L higher HDL should have had an odds ratio of 0.87 (0.84 to 0.91) for myocardial infarction on observational grounds. The observed odds ratio was 0.99 (0.88 to 1.11). A 14-variant HDL score gave 0.93 (0.68 to 1.26) against an observational expectation of 0.62. The positive control worked exactly as predicted: the LDL genetic score gave 2.13 (1.69 to 2.69).
HDL cholesterol is a marker, not a reliable causal lever. A drug-induced change in it cannot be assumed to carry the risk implications of a naturally occurring difference.
Evidence: Established that HDL falls, by a small amount in randomised data. Unsupported that the fall causes cardiovascular events — and equally unsupported that it is definitely harmless. It has not been shown to matter.
The distinction that gets botched most often
When someone says “steroids destroy your lipids,” they are almost always citing data generated with oral 17-alpha-alkylated androgens — stanozolol, oxandrolone, methyltestosterone, danazol — which are a different drug class at a different route and dose.
In a six-week crossover in 11 weightlifters comparing oral stanozolol 6 mg daily against supraphysiological intramuscular testosterone enanthate 200 mg weekly:
| Oral stanozolol | Injectable testosterone | |
|---|---|---|
| HDL cholesterol | −33% | −9% |
| Apolipoprotein A-I | −40% | −8% |
| LDL cholesterol | +29% | −16% |
Note the sign on LDL. These are not the same class behaving with different intensity — on the fraction with the best causal evidence for atherosclerosis, they behave in opposite directions. The trial’s authors concluded that parenteral testosterone “may be preferable in many clinical situations.”
One further finding is directly relevant to men taking an aromatase inhibitor alongside testosterone. In a randomised crossover, testosterone alone lowered HDL by 16%; testosterone with aromatisation blocked lowered it by 20%. Blocking aromatisation makes the HDL fall slightly worse, not better — some of testosterone’s HDL-lowering is offset by its conversion to estradiol.
Evidence: Established that oral 17-alpha-alkylated androgens have far larger and partly opposite lipid effects. Moderate that blocking aromatisation worsens the HDL effect.
Lipoprotein(a)
Old, small, mechanistically consistent, and entirely outcome-free. Testosterone lowered Lp(a) by 37% in a randomised crossover, and by 20% in nine men on weekly injections — while orchiectomy raised it 20% by three months. Blocking aromatisation barely changed the effect (28% versus 37%), so it appears androgenic rather than estrogenic.
Nobody has shown this matters clinically, no trial has used Lp(a) as an endpoint, and there are null studies. It is a real and reproducible biochemical effect in search of a clinical meaning.
Evidence: Limited.
What the guidelines say — and where they contradict each other
This is a genuine disagreement between current, evidence-graded documents, and it is worth seeing laid out.
| Question | Endocrine Society 2018 | AUA 2018 | EAU | ADA |
|---|---|---|---|---|
| Measure testosterone in an asymptomatic man with diabetes? | Against routine screening in the general population | Yes — consider it “even in the absence of symptoms or signs” (Statement 4, Grade B) | Assess and treat comorbidities first | No — symptom-triggered only (Recommendation 4.19, Grade B) |
| Use testosterone to improve glycemic control? | Recommend against | Evidence “inconclusive” (Statement 15) | “Do not use TTh to reduce weight and enhance cardio-metabolic status” | Not recommended for this purpose |
| Routine metabolic monitoring on therapy? | Not specified | Not specified | “Baseline and, at least, annually… may be a reasonable consideration” | Not specified |
A urology guideline says consider measuring testosterone in an asymptomatic diabetic man; a diabetes guideline says wait for symptoms. Both are current and both carry Grade B evidence. The disagreement reflects different specialty priors about the value of case-finding, not a difference in the underlying data — and knowing that is more useful than being told one of them is correct.
Where all four agree: testosterone should not be used to treat dysglycemia. That is as close to unanimity as this field gets.
The Endocrine Society’s monitoring plan — symptoms, adverse effects, adherence, testosterone, hematocrit, prostate cancer risk — does not include glucose, HbA1c or lipids. The EAU is the only body that endorses periodic metabolic monitoring, and it does so in the softest language available while simultaneously telling clinicians not to use testosterone for cardiometabolic purposes.
GLP-1 receptor agonists
Worth a section because it is the question men now ask, and because the answer is unusually clean about mechanism.
GLP-1 receptor agonists raise total testosterone in men with obesity or type 2 diabetes: pooled across 7 studies in 680 men, a standardised mean difference of 1.39 ng/mL (95% CI 0.70 to 2.09), with a meta-regression showing the more weight lost, the greater the rise.
The mechanism is settled by the negative control. In 26 healthy, lean, eugonadal men given dulaglutide or placebo in a randomised crossover, total testosterone changed by 0.9 nmol/L (95% CI −1.5 to 3.3) — nothing. Sperm parameters and sexual desire were unchanged. So the testosterone rise in obese men is a weight-loss effect, not a direct drug effect on the testis.
Two nuances usually lost in summaries. Free testosterone gains are inconsistent, because SHBG rises in parallel. And unlike testosterone therapy, GLP-1 agonists preserve or increase LH and FSH rather than suppressing them — which is the fertility-relevant difference.
Exactly one head-to-head trial exists: 25 men with type 2 diabetes and functional hypogonadism randomised, open-label, to semaglutide or injectable testosterone undecanoate for 24 weeks. Both arms raised total testosterone and improved symptom scores. Sperm concentration and total sperm number fell significantly on testosterone; morphologically normal sperm rose from 2% to 4% on semaglutide. Erectile function improved significantly only in the testosterone arm.
That trial supports one specific conclusion and nothing broader: for a man with obesity-related functional hypogonadism who wants to preserve fertility, a GLP-1 agonist raises testosterone without suppressing sperm production, while testosterone therapy suppresses it.
And one contradicting signal, which we are not going to omit. A database study of 3,094 non-diabetic obese men prescribed semaglutide, propensity-matched, found new erectile dysfunction diagnoses in 1.47% versus 0.32% (relative risk 4.5, 95% CI 2.3 to 9.0) and new testosterone deficiency diagnoses in 1.53% versus 0.80%. The most plausible explanation is detection bias — men on semaglutide have far more clinical contact, and both outcomes require a clinician to ask a question — and the absolute rates are very low. But it is a published finding in over 6,000 men and it points the other way.
Evidence: Moderate that GLP-1 agonists raise total testosterone via weight loss. Limited on everything comparative.
Related
Total testosterone · SHBG · Hematocrit · Testosterone, the prostate and the heart · When testosterone isn’t the answer · Raising testosterone without a prescription · The Lab Library
Questions patients ask
Diabetes and glucose
Testosterone prevents diabetes — there was a big trial.
There were two, and the larger and longer one was null.
What the evidence showsT4DM randomised 1,007 men and found 12% versus 21% crossing a two-hour OGTT threshold at two years, relative risk 0.59 (95% CI 0.43 to 0.80). The TRAVERSE substudy followed 1,175 men with prediabetes for four years and found progression rates of 10.1% versus 14.6% at two years and 13.4% versus 15.7% at four, with a pre-specified omnibus test giving p=0.49, and no difference in glucose or HbA1c. Every participant in T4DM was enrolled in a structured weight-loss programme; testosterone was tested as an add-on to it.
What remains uncertainWhether the difference is the endpoint, the dose, the population or the co-intervention. An OGTT is sensitive to muscle mass and glucose disposal in a way HbA1c is not, so a drug that adds muscle can move one without moving the other.
Bottom lineT4DM reduced the proportion of men crossing an OGTT threshold. That is not the same claim as preventing diabetes, and it was achieved on top of weight loss, not instead of it.
Mixed — Moderate for the OGTT finding, Strong (null) for clinical glycemic outcomes
My HbA1c will come down on testosterone.
In the largest and best-controlled data, it did not.
What the evidence showsTwo meta-analyses disagree — one found −0.87% (95% CI −1.32 to −0.42) from 5 trials in 351 men; the other found −0.15% (−0.39 to 0.10) from 7 double-blind trials in 833 men. The smaller sample produced the larger effect. The TRAVERSE substudy, roughly six times the combined size of both in a single blinded trial, found no HbA1c or fasting glucose difference over four years.
What remains uncertainTestosterone does appear to improve insulin sensitivity by the clamp method — glucose infusion rate rose 32% over 24 weeks in one 44-man study — without that translating into HbA1c benefit at scale.
Bottom lineSomething real is happening to insulin sensitivity and body composition. It is not showing up in the number your diabetes is managed by.
Strong (null) for HbA1c
My diabetes means I should have my testosterone checked.
Two current guidelines disagree with each other about this, and you should know that rather than be told one answer.
What the evidence showsThe AUA's Statement 4 says clinicians should consider measuring total testosterone in men with diabetes "even in the absence of symptoms or signs," at Grade B. The ADA's Standards of Care take a symptom-triggered approach: inquire about sexual health, and measure a morning total testosterone only if symptoms or signs of hypogonadism are present, also at Grade B. The Endocrine Society recommends against routine screening of the general population.
What remains uncertainNothing about the documents. The disagreement is about the value of case-finding, not about the data.
Bottom lineIf you have symptoms, everyone agrees you should be tested. If you have none, a urologist and a diabetologist would give you different advice, and both would be following their own guideline.
Strong
Lipids
Testosterone will wreck my cholesterol.
This claim is more complicated than it sounds — and the data behind it usually come from a different drug.
What the evidence showsIn pooled randomised data across 51 studies, HDL fell by 0.49 mg/dL (95% CI −0.85 to −0.13). In uncontrolled before-and-after injectable data, it fell 4 mg/dL — eight times as much, without a placebo contrast. Meanwhile, in a direct comparison, oral stanozolol lowered HDL by 33% and raised LDL by 29%, while injectable testosterone lowered HDL by 9% and lowered LDL by 16%.
What remains uncertainWhether a small HDL fall matters at all. The relevant clinical question has never been answered directly.
Bottom linePrescription testosterone lowers HDL slightly and does not raise LDL. Oral 17-alpha-alkylated androgens do something considerably worse and in the opposite direction on LDL. Do not generalise from one to the other.
Strong
My HDL dropped on testosterone, so my heart risk has gone up.
Nobody has shown that, and the assumption underneath it is itself contested.
What the evidence showsTRAVERSE, in 5,246 men at elevated cardiovascular risk, found major adverse cardiac events at hazard ratio 0.96 (95% CI 0.78 to 1.17). The pooled randomised safety analysis found no significant effect on mortality or cardiovascular outcomes. And genetically, a variant conferring 0.14 mmol/L higher HDL should have given an odds ratio of 0.87 for myocardial infarction on observational grounds — the observed value was 0.99 (0.88 to 1.11). The same study's LDL positive control behaved exactly as predicted.
What remains uncertainWhether any drug-induced HDL change carries clinical meaning. HDL appears to be a marker rather than a lever.
Bottom lineAnyone telling you the HDL fall is dangerous is going beyond the evidence. So is anyone telling you it is definitely harmless.
Unsupported in both directions — it has not been shown to matter
Adding an aromatase inhibitor protects my lipids.
It does the opposite, by a small margin.
What the evidence showsIn a randomised crossover in 14 men, testosterone alone lowered HDL by 16%; testosterone with aromatisation blocked lowered it by 20%, with apolipoprotein A-I down 15% and hepatic triglyceride lipase up 38%. The aromatase inhibitor alone had little effect on any lipid. Separately, a meta-regression across 29 trials found that whether the androgen could be aromatised explained the between-study variation in the HDL effect.
What remains uncertainThe magnitude in men on ordinary replacement doses; these were weightlifters on 200 mg weekly.
Bottom lineSome of testosterone's HDL-lowering is offset by its conversion to estradiol. Blocking that conversion removes the offset.
Moderate
Weight, and what actually raises testosterone
Testosterone is a weight-loss drug.
It changes what your weight is made of. It does not reliably change the number.
What the evidence showsAcross 29 randomised trials, total body fat fell 1.6 kg and fat-free mass rose 1.6 kg — with no change in body weight. The observational literature looks very different: pooled across 32 observational studies in 4,513 men, weight fell 3.50 kg and waist circumference 6.23 cm at 24 months. Those authors themselves noted that placebo-controlled data are more conflicting and that their weight finding should be confirmed by a purpose-designed randomised trial. None has confirmed it.
What remains uncertainWhy the observational and randomised literatures diverge so sharply here. Selection, adherence and co-intervention are the usual suspects.
Bottom lineThe randomised literature says testosterone moves body composition and very little else. That is a real benefit. It is not weight loss.
Strong for body composition; Unsupported for weight loss
Losing weight won't fix my testosterone, so I may as well start treatment.
The causal arrow runs the way you would least prefer.
What the evidence showsGenetically, a one standard deviation increase in BMI produces 0.25 standard deviations lower testosterone (95% CI −0.42 to −0.09), with no effect in the reverse direction. Total testosterone rose 8.73 nmol/L after bariatric surgery and 2.87 nmol/L on a low-calorie diet, and the degree of weight loss was the best determinant of the rise. The EAU guideline is explicit: improve lifestyle, reduce weight, withdraw interfering drugs and treat comorbidities before starting testosterone.
What remains uncertainHow much weight loss is needed in an individual man, and how durable the testosterone gain is.
Bottom lineAdiposity lowers testosterone. Low testosterone has not been shown to drive adiposity. That asymmetry should shape the order in which things are tried.
Strong
A GLP-1 agonist would raise my testosterone directly.
It raises it, but through the weight, not through the testis.
What the evidence showsPooled across 7 studies in 680 overweight men, total testosterone rose with a standardised mean difference of 1.39 ng/mL (95% CI 0.70 to 2.09), and the more weight lost the greater the rise. In 26 lean, eugonadal men given dulaglutide in a randomised crossover, total testosterone changed by 0.9 nmol/L (95% CI −1.5 to 3.3) — nothing at all, with sperm parameters and sexual desire unchanged.
What remains uncertainFree testosterone gains are inconsistent because SHBG rises alongside. One database study of 3,094 men found more new erectile dysfunction diagnoses on semaglutide (1.47% versus 0.32%), which contradicts the hormonal data and is most plausibly detection bias — but it exists.
Bottom lineIn a lean man with normal testosterone, a GLP-1 agonist does nothing to the gonadal axis. In an obese man it helps, in proportion to the weight lost.
Moderate
I should take a GLP-1 agonist instead of testosterone.
For one specific man, there is a real argument. Beyond him, nobody knows.
What the evidence showsExactly one head-to-head trial exists: 25 men with type 2 diabetes and functional hypogonadism, randomised open-label to semaglutide or injectable testosterone for 24 weeks. Both raised total testosterone and improved symptom scores. Sperm concentration and total number fell significantly on testosterone, while morphologically normal sperm rose from 2% to 4% on semaglutide. Erectile function improved significantly only in the testosterone arm.
What remains uncertainEverything else. Twenty-five men, open-label, 24 weeks, surrogate endpoints. No adequately powered comparison has been done.
Bottom lineFor an obese man with functional hypogonadism who wants to preserve fertility, that trial is genuinely informative. Anyone stating more broadly that these drugs are or are not substitutes for testosterone is going beyond the evidence.
Limited
Ready for testosterone care built on a diagnosis?
The first step is a full endocrine evaluation, not a prescription. We see patients across San Diego County and welcome referrals from other physicians.
