Testosterone and Your Joints: The 51 Percent That Disappears

A large American dataset appears to show that adults in the top quarter for testosterone have half the arthritis of those in the bottom quarter. It is exactly the number a hormone clinic would put on a billboard. We are going to show you why it does not survive the rest of the paper, and why that matters more than the finding itself.

What Was Actually Measured

This is a cross-sectional analysis of the National Health and Nutrition Examination Survey, or NHANES, covering 2013 to 2016. NHANES combines a household interview with a physical examination and laboratory testing, sampled to represent the United States population rather than one clinic's patients. Cross-sectional means everything was measured once, on the same day, in the same people. No follow-up, no before and after.

From 20,146 eligible individuals the authors removed 5,380 with no testosterone value and 4,327 with no arthritis data, leaving 10,439 adults, mainly aged 20 and over. Of those, 48.11 percent were men, mean age was 47.25 years with a standard deviation of 16.80, and mean serum testosterone was 214.97 ng/dL with a standard deviation of 234.70. 26.97 percent reported arthritis: 2,734 people with it, 7,705 without.

Testosterone was measured by isotope dilution liquid chromatography tandem mass spectrometry, the reference method, and that is a real strength. Arthritis was self-reported, established by asking whether a doctor or other health professional had ever told them they had it. The authors cite earlier work reporting 81 percent agreement between self-reported and clinically confirmed osteoarthritis. Among those who said yes, 52.00 percent called it osteoarthritis or degenerative arthritis and 14.34 percent rheumatoid.

The Crude Picture, Which Looks Convincing

On the surface the association is strong and tidy. Mean testosterone in the arthritis group was 163.67 ng/dL against 233.91 ng/dL in the group without arthritis, with a p value below 0.0001.

The authors then split testosterone into quartiles, four equal groups from lowest to highest. In their fully adjusted model, with the lowest quartile as reference, the odds ratios ran 1.00, 0.85, 0.53 and 0.49. That last figure is the headline: the highest quartile had 51 percent lower odds of arthritis than the lowest. An odds ratio is the odds in one group divided by the odds in another, so below 1.00 means less, above means more, and 0.49 reads as roughly half.

Now the Number That Never Gets Quoted

The same table in the same paper also reports testosterone as a continuous variable, per 100 ng/dL, across three models of increasing adjustment.

• Model 1, no adjustment at all: odds ratio 0.87, 95 percent confidence interval 0.85 to 0.89.
• Model 2, adjusted for age, sex and race: 0.93, confidence interval 0.90 to 0.97.
• Model 3, fully adjusted: 1.02, confidence interval 0.97 to 1.07.

A 95 percent confidence interval is the range of values compatible with the data. When it crosses 1.00, as 0.97 to 1.07 plainly does, the honest reading is that no association was demonstrated. Not a weak one. None.

An odds ratio of 1.02 with a confidence interval from 0.97 to 1.07 is not a small effect. It is no effect.

Watch the sequence. Unadjusted, each 100 ng/dL of testosterone carried an odds ratio of 0.87. Adjust for age, sex and race alone and most of it has gone. Adjust for everything and it is gone entirely.

What Full Adjustment Actually Means

Model 3 accounted for age, sex, race, education level, marital status, body mass index, alcohol status, smoking status, hypertension, diabetes, cardiovascular disease, estradiol, sex hormone binding globulin and the income to poverty ratio.

Adjustment is not a trick. It asks a plain question: comparing people alike in age, sex, weight and the rest, does testosterone still tell you anything about their joints? Here the answer is no, and the baseline table shows why.

• Mean age was 59.90 years in the arthritis group against 43.36 years in the group without it.
• A body mass index of 30 or above: 55.24 percent of the arthritis group, 34.59 percent of the others.
• Hypertension 55.57 against 25.58 percent, diabetes 18.24 against 7.78 percent, cardiovascular disease 18.56 against 5.01 percent.
• Women were 61.69 percent of the arthritis group and 48.27 percent of the group without.

Older people have more arthritis and lower testosterone. Heavier people have more arthritis and lower testosterone. Women here had far more arthritis and, for ordinary physiological reasons, much lower testosterone. The crude association was carrying all of that on its back.

Why the Quartiles Still Look Significant

Two numbers from the same fully adjusted model point in opposite directions, and that deserves an explanation rather than a choice between them.

The quartile result does survive adjustment: Q3 at 0.53, confidence interval 0.35 to 0.79, and Q4 at 0.49, confidence interval 0.31 to 0.76. But those intervals are wide, meaning the data are compatible with anything from a very large effect to a modest one. And the p value for trend, below 0.0001 before adjustment, fell to 0.0327 after it, sitting just inside the conventional threshold rather than near it.

Then read the quartile boundaries, which the paper prints and the discussion does not dwell on. The lowest quartile runs from 0.52 to 18.77 ng/dL. The highest runs from 378 to 2000 ng/dL. A ceiling of 18.77 ng/dL is a concentration you essentially only see in women. A floor of 378 is a male one. Comparing the top quartile with the bottom is, to a large extent, comparing men with women, even though the model does adjust for sex.

The overall mean makes the same point arithmetically: 214.97 ng/dL with a standard deviation of 234.70. When the spread is larger than the average, you are not looking at one population. You are looking at two, pooled.

One last point, since we would want to be told it ourselves. The paper's abstract and discussion state that the association held after adjustment. Its own Table 2 shows the continuous estimate did not. We have reported the table.

Where the Association Was Strongest: Women and Obesity

The subgroup analysis ought to give any testosterone clinic pause. The negative association was more significant in women and in people with a body mass index of 30 kg/m2 or above, and the interaction tests for sex and for that BMI category were statistically significant. Age, hypertension and diabetes showed no significant interaction, with p values above 0.05.

So the strongest signal in a study about testosterone and joints sat in women and in people with obesity, close to the opposite of the marketing use this paper will be put to. It is also where confounding is worst. The authors note the observational evidence that obesity is associated with low testosterone and that the relationship may run in both directions. There, a hormone level and a joint problem can share a cause without either one touching the other.

The Word Risk Is Doing Work It Cannot Do

The paper describes a reduced risk of developing arthritis. A cross-sectional study cannot measure development. Blood was drawn and the arthritis question answered on the same day, so nobody knows which came first, and the authors state it plainly: because of the design, no causal relationship could be obtained.

The reverse explanation fits just as well. Painful joints reduce activity, reduced activity costs muscle and adds fat, and persistent pain disturbs sleep. Arthritis lowering testosterone is as consistent with this dataset as low testosterone causing arthritis, and a snapshot cannot separate the two.

The authors list their limitations honestly: arthritis was established by interview, so recall bias was inevitable; the findings may not transfer to other populations; and residual confounding, unmeasured confounding and measurement error may all bias the analysis. Add the 5,380 and 4,327 individuals dropped from the 20,146 eligible, close to half the starting sample, with no multiple imputation, which they acknowledge may affect accuracy.

What to Do With This If Your Joints Hurt

Be suspicious of any clinic that offers you a hormone for a joint problem. This is one of the strongest recent datasets on the question and it does not support the offer. No trial, no treatment, no before and after, and the fully adjusted continuous estimate says testosterone did not explain arthritis on its own.

That does not make the paper useless. It makes it an excellent lesson in how to read one. A headline percentage tells you what a comparison looked like. A fully adjusted model tells you whether the comparison meant anything. When the two disagree, the second one wins.

What is worth doing is separating the two questions instead of merging them. Joint pain deserves a proper mechanical and rheumatological assessment. Genuine symptoms of low testosterone deserve their own, which means more than one morning measurement and the surrounding pituitary and binding protein picture rather than a single figure. We set out what that involves in how low testosterone is diagnosed and monitored and in what a proper men's health blood panel measures.

The one modifiable factor in both halves of this study is body weight, and the authors make the point themselves: weight reduction through physical exercise and controlled diet may have a role in relieving osteoarthritis symptoms. That requires no hormone. If you would like your own results read in full context rather than as a single number, our specialist consults by video and at private partner settings across the Costa del Sol.