Tuesday, April 15, 2008

More on Rohrmann et al.

In a previous post on human variation in testosterone levels, I discussed the Rohrmann et al. study and its methodological flaws. This study tested black Americans, white Americans, and Mexican Americans in three age classes: 20-44; 45-69; and 70+. It found that even in the youngest age class the race differences were minor, with the Mexican Americans having the highest testosterone levels.

The main flaw here is that the black-white difference in testosterone levels shrinks after 24 years of age, is gone by the early 30s, and seems to reverse at older ages. So it should not show up in the average t levels of 20 to 44 year-olds. As for the Mexican Americans, their t levels were higher probably because they were younger on average. The median age of the 20-44 year-olds was 31.8 for the whites, 29.5 for the blacks, and 28.6 for the Mexican-Americans. This age disparity is probably understated by the use of medians rather than means, given the youthful age pyramid of the Hispanic-American population.

But another thing bothers me with this study. The 45-69 year old black men had higher t levels (5.62) than the 20-44 year old black men (5.35). This is more than a bit odd. Testosterone declines with age and the large number of participants rules out random sampling error.

Nor could recruitment bias be responsible. The authors of this study did not recruit their participants. In fact, they never even saw them. They used serum samples that the National Center for Health Statistics had earlier collected as part of its Third National Health and Nutrition Examination Survey (NHANES III). The authors state they used 1,479 samples that were still available out of an initial total of 1,998.

Uh, why were 519 samples no longer available? That comes to over 25% of the original total. What happened to them?

After some googling, I discovered that another study had used the NHANES-III serum bank for research on a sexually transmitted disease: Herpes Simplex virus type 2 (HSV-2).

The National Health and Nutrition Examination Surveys (1988–1994) (NHANES-III) reported that more than 25% of adults between 30 and 39 years of age were positive on serology for HSV-2 in those years. (Chorba et al., 2007, p. 655)

More than 25% ... Hmm, could these be the missing samples we’re looking for? If so, removing them would have disproportionately depleted the black American component of the study population:

Although HSV-2 infection is increasing among young Caucasians, who have a seroprevalence of approximately 17%, infection is more common among African-Americans, who have a seroprevalence of 45%. (Chorba et al., 2007, p. 656).

The study is written up in Fleming et al. (1997). The ‘Methods’ section describes why some serum samples were no longer available, including a “need to use serum for other tests”:

Of the persons originally selected for NHANES III, 82.5 percent were interviewed, and HSV-2 test results were available for 60.2 percent. The reasons that results were unavailable included the inability to locate the selected subject, refusal by that person to be interviewed or to have blood drawn, unsuccessful venipuncture, the need to use serum for other tests, and the loss of serum samples during transportation, storage, or processing.

When any serum samples tested positive for HSV-2, they were probably set aside for the duration of the study, with a view to further analysis. If, as seems likely, these samples were absent from the NHANES III serum bank when Rohrmann et al. undertook their study, their study population would have had disproportionately fewer participants with a polygynous sexual orientation and with correspondingly high testosterone levels. As Chorba et al. (2007, p. 687) point out, “The major known risk factors for acquiring genital HPV infection include having multiple sex partners and having sex partners who have had multiple partners.”

This data bias would explain Rohrmann et al.’s paradoxical finding of higher t levels in older black Americans. Young black Americans with high t levels were likely to test positive for HSV-2 and to have their samples removed from the NHANES III serum bank.


Chorba, T., Guoyu, T., and Irwin, K.L. (2007). Sexually transmitted diseases. In: Litwin MS, Saigal CS, editors. Urologic Diseases in America. US Department of Health and Human Services, Public Health Service, National Institutes of Health, National Institute of Diabetes and Digestive and Kidney Diseases. Washington, DC: US Government Printing Office, 2007; NIH Publication No. 07–5512 [pp. – 649-695]

Fleming DT, McQuillan GM, Johnson RE, Nahmias AJ, Aral SO, Lee FK, St Louis ME. (1997). Herpes simplex virus type 2 in the United States, 1976 to 1994. New England Journal of Medicine, 337, 1105-11.

Rohrmann, S., Nelson, W.G., Rifai, N., Brown, T.R., Dobs, A., Kanarek, N., Yager, J.D., Platz, E.A. (2007). Serum estrogen, but not testosterone levels differ between Black and White men in a nationally representative sample of Americans. The Journal of Clinical Endocrinology & Metabolism, 92, 2519-2525


Anonymous said...

1) The paper clearly indicates the values shown in the tables are adjusted for age. Age differences within age brackets should not be a factor, and T levels across age brackets likely do not have the same relationship before adjusting for age.

2) The main flaw here is that the black-white difference in testosterone levels shrinks after 24 years of age, is gone by the early 30s, and seems to reverse at older ages. So it should not show up in the average t levels of 20 to 44 year-olds.

Again: no, the second claim doesn't (necessarily) follow even if the first is true, and the data from this study suggest the truth is closer to the opposite of your first claim (the black-white differences is larger in the older age bracket than in the younger).

3) Instead of speculating wildly about the "missing samples", did you attempt to contact Rohrmann or the NHANES people or anyone else who might potentially give you a definitive answer?

Anonymous said...

1. You're referring to the linear regression model. I'm referring to the serum concentrations of each age group, as listed in Table 3 "Serum concentrations of sex steroid hormones". Those are simply geometric means.

Yes, the age-adjusted linear regression model failed to find a black-white difference. There are two reasons. First, testosterone levels do not decline linearly with age. The decline is much steeper in the mid to late twenties and is steeper for blacks than for whites.

Second, a significant number of high testosterone subjects were removed from the 20-44 year-old group of black Americans. This is not 'speculation.' Testosterone levels do not increase with age, as they purportedly have in this study.

Sorry n/a, but age differences within age groups are certainly a factor because the ethnic groups under study (blacks, whites, hispanics) have different age pyramids.

2. Yes, indeed, the black-white difference is wider in the 45-69 year olds. That is to be expected if t levels are higher in 45-69 year-old black men than in 20-44 year-old black men. Both of these findings indicate that a disproportionate number of high testosterone individuals were removed from the younger age group.

3. If and when I receive such information you'll be the first to know. Trust me.

Anonymous said...

Presented in Table 3 are multivariable-adjusted geometric
mean hormone concentrations.

TABLE 3. Serum concentrations of sex steroid hormones and SHBG by race/ethnicity within age strata, NHANES III (Phase 1), 1988–1991

The geometric mean concentration was adjusted for age, percent body fat, cigarette smoking, alcohol consumption, and moderate/vigorous
physical activity. MA, Mexican-American; NHB, non-Hispanic black; NHW, non-Hispanic white.

Anonymous said...

Hello. This post is likeable, and your blog is very interesting, congratulations :-). I will add in my blogroll =). If possible gives a last there on my blog, it is about the Home Theater, I hope you enjoy. The address is http://home-theater-brasil.blogspot.com. A hug.

Anonymous said...


I'm sorry but you're confused. The 'adjustment for age' in Table 3 is the breakdown of the data into three age groups. This point is explained in the Statistical Analysis section:

"We evaluated racial/ethnic differences in the hormones and SHBG (a) overall after adjusting for age and (b) within three age
categories reflecting hypothesized hormonal transitions through life: early adulthood (20-44 years old), mid-adulthood (45-69 years old), and late adulthood (70+ years old)."

Anonymous said...

With that response, Peter Frost loses any credibility he may have had.

A bad memory or laziness on the first read through might be forgiven. But if you can't correctly interpret simple English statements even after your attention is drawn to them repeatedly, then all your previous attempts at synthesizing large quantities of data must be called into question.

No neutral, intelligent English speaker could possibly interpret Table 3 the way you do.

I'm right. You're wrong. Period.

I'm done with this. Go email Rohrmann if you want someone to tell you you're wrong again.

Anonymous said...


I understand you have strong feelings on this topic. Fine. But please don't let strong feelings cloud your sense of judgment.

One last point. The Rohrmann et al. study is noteworthy for two findings:

1. There is no difference in testosterone level between white Americans and black Americans.

2. Testosterone levels are higher in middle-aged black Americans (45-69) than in young black Americans (20-44.

In my humble opinion, the second finding nullifies the credibility of the first. Perhaps you feel otherwise. Fine.

Anonymous said...

What I have strong feelings about is poor scholarship.

Your assertion that *I* am letting "strong feelings" "cloud my judgment" betrays an incredible lack of self-awareness.

Re-read the paper and pretend you're not trying to support any particular argument. Personally, I don't see any room for confusion, but at minimum you must acknowledge that my interpretation is plausible and the "explanation" you point to as counter-evidence is nothing of the kind (stating that the overall comparison was adjusted for age does not imply that age was not adjusted for when data were broken down into age brackets, particularly when it is clearly indicated elsewhere that the levels were adjusted for age; breaking data out into three age groups is not "adjusting for age", and such a notation would be unnecessary if the separation into three age brackets clearly labeled in the table were the only age "adjustment" made).

If you're still confused after that, email Rohrmann or the other authors. It reflects quite badly on you to misrepresent sources like this.

Anonymous said...

First, it’s of secondary importance whether or not Rohrmann and her colleagues age-adjusted their data within each age class of Table 3. I believe the younger age structure of Mexican Americans explains why they had the highest testosterone level of the three racial groups. Perhaps I am wrong. Fine.

This issue, however, does not bear on the two main findings of Rohrmann et al., i.e., 1) the level of testosterone did not differ between black American and white American participants; and 2) it was actually higher in the middle-aged black Americans (45-69 years old) than in the young black Americans (20-44 years old). The second finding is bizarre to say the least, all the more so because the study population was large.

How do you explain this finding? Or is it of no interest to you?

Second, Rohrmann et al. did discuss age adjustment of their data in the ‘Statistical analysis’ section of their paper, but only with regard to their linear regression models (which involved the entire data set). When they compared the three racial groups by age class, they used a simple ANOVA procedure.

Perhaps you feel I am wrong on this point. Fine. I don’t think so but perhaps I am. Please let me mention, however, that this point has no bearing on black/white differences in testosterone levels, yet that seems to be what gets you hot under the collar.

Or are you this way whenever you disagree with someone? If so, you must lead a lonely life.

Anonymous said...

How do you explain this finding?

I already did.

Anonymous said...

John Derbyshire said (and you have the proof in n/a' s attitude) that most people,even right leaning people, do not like the subject -not just certain conclusions- the subject of human variation.
Your post is superb,which is no doubt why it broght you ad hominum brickbats.