A new study shows that children of mothers who drank as
little as 1 to 6 units of alcohol per week had children whose IQs at age 8 were
statistically significantly lower than those of children who mothers abstained.
This finding was widely reported by news media yesterday.
Google “IQ alcohol” and you will find many articles which uncritically describe
the findings of the study. As is the case in many such articles, quotes from
the press release accompanying the paper’s publication appear to have been
liberally used.
The major problem with this study is one that I’ve written
about before. Results that are statistically significant may not necessarily be
clinically significant.
The difference in IQ between the two groups of children was
1.8 points. Do you really think that an IQ difference of 1.8 is going to be a
life-altering finding for a child? I don’t.
IQ tests are often unreliable and if taken more than once can
yield different results. A difference of 1.8 points is well within the margin
of error of such tests. One source I found
states that the margin of error of the IQ test used in this study is a minimum of 5 points. The children in this study were 8 years old when they took the test.
The study, done by a group in Bristol, England, was
published on PLoS One and the full text is available here.
Caution is advised if you plan to read it. There are many other problems to
consider.
Data collected for this paper were pulled from another study which was done from 1992 to 2000
about the genetics of alcohol metabolism. The
primary focus of the original study was not the hypothesis of the IQ study.
A unit of alcohol was defined as 8 grams by the authors. Since
I was not familiar with what 8 grams of alcohol really meant, I looked it up. A
“standard drink” was said to contain anywhere from 8 (in the UK) to 14 grams (in
the US) of alcohol.
Here’s a quote from the paper’s “Methods” section, “At
approximately 18 and 32 weeks of pregnancy women were also asked on how days
during the past month they had drank [sic]
2 pints of beer (or the equivalent amount of alcohol), any women who reported
doing this on at least one occasion was classified as a binge drinker in our
analysis of the association between genotype and binge drinking.”
Really? Two pints of beer on one occasion makes a
woman a binge drinker? Depending on the type of beer, a pint contains 2 to 3
units or 16 to 24 grams of alcohol.
To help you understand the paper better, here is a table:
I asked my wife what she thought of this study and she said,
“I think the women who drank were probably more fun to be with.”
13 comments:
I did a lot of FAS (fetal alcohol syndrome) research in med school, mostly on mice. Basically, we would inject pregnant mice (and not inject similar pregnant mice as controls) with different amounts of EtOH at different gestational days, rear their offspring, test the offspring in mazes/tasks etc., then finally look at the structure of the offspring's brains. Results were pretty consistent. I forget the exact metrics, but we could basically "binge drink" a mouse to a high BAC on just one particular day of the gestational cycle and produce rather slow-minded offspring. They mostly "looked" normal, but there would be varying "FAS-like" visible phenotypes even in the mice whose mothers only "got drunk" during one day of their pregnancy.
I mention all this to say, there are probably more sensitive times in pregnancy to alcohol than others. That time is probably, broadly, the first trimester. There's lots of neuron-building during that phase. I would not risk a drink of any alcohol during that time. I have begun to feel alcohol's teratogenicity may be somewhat akin to thalidomide. That is, it wasn't so much the dose of thalidomide, it was when the mom was exposed. The absolute most sensitive period was 20-40 days of gestation for thalidomide. I had a professor who looked at some data and said, "If a mother takes thalidomide on day 26 (or whatever) of pregnancy, the child will have no thumbs." I guess he was convinced the data was that precise. In a far off time, I remember reading some paper about looking at the IQs of children whose mothers received IV alcohol for pre-term labor. I remember that overall their IQs were normal.
The IQ differences were slight, not clinically significant, as per this study. But it was a huge population, and I would give the authors credit in trying to guard against shortcomings in their analyses. You could look at a population of San Franciscans and Little Rockians and probably find a mean difference in IQ (by chance? because one city is smarter than the other?) of 2 points or greater. So you'd make some hypotheses based on that. The hypotheses the authors make, that some alcohol exposure early-ish in pregnancy affects IQ and the IQ affectation may be linked to their EtOH metabolizing abilities, is reasonable in my opinion. It fits with the totality of data in this arena. Don't let the 1.8 mean IQ difference bother or comfort you too much. It's just one way to express "Hey, we see a difference here."
I'm a radiation oncologist. With radiation, we view exposure in a "linear no threshold" model, in other words, there is no safe radiation dose; there will be some measurable effect at any radiation level. Alcohol exposure in pregnancy and IQ may be like that too. If so, it would be important to know, just so we can counsel patients.
As a final personal point, my wife and I didn't drink any alcohol ever with our kids 'til her belly got very big. We then thought it was OK, but I encouraged her to keep it to less than a glass of wine every few days based on my own particular biases on the matter. Ultimately, because I probably made her so paranoid, she wound up never drinking anything while pregnant :)
Todd, thanks for the comments. That is interesting stuff about the timing of drinking alcohol.
I disagree with one of your points. The "difference" in the mean IQ scores is very likely not a real difference because 1.8 falls well within the 5 point margin of error of the IQ test itself.
I think to see a real difference, one would need a larger than 5 point spread between the two groups.
I'm not advocating that women get drunk during pregnancy. I'm saying this study does not convince me that women must completely abstain. And I always call out the media when they swallow the PR without questioning the findings.
Thanks for picking arkansas for your IQ comparison. Nothing better to get rid of stereotypes and debunk myths than by reinforcing others.
Oh it's "real." Depends on how you define real, because I don't know what "unreal" differences are; mathematically, there was a **mean** 1.8 IQ difference. In terms of a large population, you have to think mathematically and biostatistically here. You're arbitrarily stating there has to be >5 IQ **mean** average point differences between groups for the differences to be "real." In a huge population, that could be quite the discrepancy and suggest a LARGE effect; in FAS research, we already know a lot of alcohol decreases IQ a lot. So ongoing research continues to quantify how much if any a little alcohol affects IQ. The data may never convince you; I always already convinced before seeing this.
First, where are you getting the 5 cutoff. Second, are you talking plus or minus 2.5 points or 5 points in this "margin of error"? Third, repeat IQ tests in individuals fall within plus or minus 2 points of one another about 50% of the time, plus or minus three points of one another about 67% of the time, and plus or minus 5 points of one another about 95% of the time. So, again, the 5 seems arbitrary-ish. The authors couldn't mention this in their discussion because they didn't know anyone was interested in a >=5 point difference :)
The whole head circumference/brain size and IQ thing (http://analyseeconomique.files.wordpress.com/2011/09/brain-size-iq-and-racial-group-differences-evidence-from-musculoskeletal-traits.pdf)... if you look at some of those studies, the **mean** differences are slighter than 1.8 points, the correlation coefficients only perhaps 0.2 or 0.3, but the P-values... large. Yet many, as do I, feel that the issue is settled and that IQ is correlated with brain/head size and that the differences although slight (a very subjective word) are "real" (a very objective word).
Again... I don't focus on the mean differences and their absolute largeness or smallness. This study would never outright convince anyone, me included, that women should abstain (although the popular press would try to play it off as such perhaps). This study would only serve to suggest a trend or hypothesis. That hypothesis would be, alcohol affects IQ at low to medium doses. This hypothesis is supported by other studies. This hypothesis is not supported (but also not necessarily refuted) by other studies too, however.
Anonymous, I imputed no higher or lower IQ to either San Francisco or Little Rock. Perhaps your own biases made you think otherwise.
Apparently, I didn't make my main point clear enough in the post. IQ tests are BS. And a difference of 1.8 points, while statistically significant, is meaningless as far as an eventual job, relationships or anything else about life. Someone whose IQ is 105 is not really smarter than someone whose IQ is 103.
PS: It's also ludicrous to even mention IQ points into 1 decimal.
I don't mean to be so argumentative today, promise. But, I am a radiation oncologist, SS.
No, your point was very clear. Of course *two* people with an IQ difference of 1.8 points have equal intelligence, outlooks on life, etc. That's obvious. However, if you take an effect and look at its effect on a POPULATION, discounting a mean difference of 1.8 points in a POPULATION shouldn't be done so readily. As far as decimals and IQ points go, don't sweat it; remember, I think the average family has 2.8 kids or something like that. My point being, when we talk of means and statistics, we can talk of decimal points.
Now first off, I don't see a mean 1.8 IQ point difference between populations in this study. In the conclusions section they mention "per allele effect estimates were −1.80" but this is NOT implying mean 1.8 IQ point differences between a group.
The purpose of this study was really to look at ADH genes and see if there were any IQ differences amongst women who drank or didn't drink based on these ADH genes. In fact, they found that women who drank moderately had overally slightly higher IQs, were better educated, etc. So what were the "real" IQ differences?
Based on "risk alleles," non-drinkers had kids with mean IQs of 103.1, 103.5, and 103.2 if they had 1 or 2, 3, or 4 or more risk alleles; these IQ differences had a P-value of 0.8, indicating no significant differences. The women drinking "<1-6 units per week" had children with mean IQs of 107.5, 105.4, and 104 if they had 1 or 2, 3, or 4 or more risk alleles; this P-value was 0.00002, indicating differences likely not present due to mere chance.
So now I'll rephrase your angst over this study in a different way. Among women who drink moderately and frequently during pregnancy, offspring have mean IQs of 107.5 whose mothers easily metabolize alcohol, and mean IQs of 104 whose mothers don't. Is this mean 3.5 point difference "real"? Is it significant? Yes, at a P-value of 0.00002; a result which indicates that likely a "real" difference has been found.
Clinically significant? Of course we all fret over this question. I submit that JUST looking at the absolute difference in means between groups is a TERRIBLE way of gauging clinical significance.
Let's look at things another way. Let's say we do the "B.S." IQ tests on children from two groups: one group fed beans from birth, another group fed corn. We march them in as couplets, holding hands. We test 100 couples, 200 children total. We find that for 90 couples, each child has a 100 IQ. But for 10 of the couples, one child from the corn group had an IQ of 100 and one bean child had an IQ of 120. The mean of the corn group is 100, the mean of the bean group is 102. The P-value for the difference between the means per the Student's t-test? P<0.001. So is this 2-point *mean* difference "real"? Bad question! For ~10% of children, eating beans has an IQ effect which we all might consider substantial. Applying a (certain) test of statistical significance yields a P-value which enables us to answer a simple question: is the difference in the single metric which I have measured independently between these groups due to random chance?
I published one of the first IGRT comparison studies in radiation oncology (http://www.ncbi.nlm.nih.gov/pubmed/16563658). I found, for example, a mean ultrasound lateral shift of 1.92 mm, and Xray shift of 0.78 mm, a **mean population** difference of only about 1 mm. Of course a difference of 1 mm in IGRT shifts in radiation oncology is clinically inconsequential. The implication was much deeper: a difference was found, where there should have been none. (I will postpone discussing for now the "clinically significant" things I referred to in the paper...)
On the bright side, if it wasn't for Alcohol, many women wouldn't be pregnant.
In case you don't believe me, here's another blog which explains from a different perspective why the study is bogus.
http://www.carlheneghan.com/2012/271/should-i-drink-moderately-during-pregnancy/
He's making the same points, paraphrasitically, that you made. He said, "Overall, the problem with this paper is the sheer number of statistical tests being undertaken, which generally leads to spurious results." That's very true, and you have to look with caution at any significant P-values in a paper chocked full of P-values. But, you can at least use the significant findings to branch out into more research. He also said, "The research, published in Plos One, used data from over 4,000 mothers and their offspring, concludes that you shouldn't [drink]." I don't see that conclusion ANYWHERE in the paper; it would have been outlandish to conclude so. He also says, "It is misleading, and not good practice, to just look at reported statistical significance, irrespective of how small the p value is and then infer causality." Amen to that. That's why the authors state: "...causation cannot be proven by the current study design." Causation is a LARGE hill to climb.
Finally, the crux: "Although the differences reported between the groups in the paper are statistically significant they are clearly not significant in terms of the classification system. Not even close." This is a toughie for most folks, and speaking frankly, it reflects a certain lack of understanding.
There is much data that IQ scores are normally distributed. Yes, there will be intrapersonal variances and inter-administrator variances. But across a large sample, from many individuals, there will be a normal distribution to these scores; the central limit theorem can then apply, and many things follow from that. If we happen to find a normally-distributed sub-population whose mean differs either from another population subsample or the population-at-large, there is a suggestion that the samples are different. That's it. The magnitude of the mean difference is INCONSEQUENTIAL, by itself, to make “clinical significance” judgments.
It is mathematically and statistically fallacious to think that if you have a sample of 1000 people whose mean IQ is 100 and another 1000 whose mean IQ is 102 to think that, on average (whatever that would mean here) if you randomly select one person from the first group and one person from the second group, that the difference in their tested IQs would be 2 points; there may be group vs group “outliers.” Across many individual samples, a "real" difference--if found--is powerful evidence (depending on the magnitude of the P-value) that some effect is at work. With all the "noise" of an IQ test, when a difference in means betweens populations is found, that group difference is plausibly "real" (e.g. in non-drinkers, mean IQs were all 103-104 range across alleles, P>>0.05). The magnitude of the mean difference between groups informs little to no cogent discussion on the question at hand. My back of the envelope calcs tell me that based on the IQ data presented, a claim from the authors COULD have been, "Amongst women who drink, children with IQs of >135 occurred twice as often in the few risk allele group versus women who had many, and children with IQs below 80 occurred twice as often in the 4 or more risk allele group." I mean, they could have done a Chi-square test like this (2x2 table of "Hi IQ", "Lo IQ", "Group A", "Group B") and there would've been a significant P-value. Who knows. That might be more compelling than saying, "The mean(SD) IQs in children with mothers with few risk alleles were 107.5(16) versus 104(16) in those with four or more risk alleles."
I see the Heneghan fellow is director of the Centre for Evidence-based Medicine at Oxford. Well. For that he deserves a wide berth. But he should've read the paper a bit better and seen that the authors provided many reasonable caveats and made no causal links or crazy recommendations. He and you called the study "bogus." Seems harsh & uninformed, IMHO.
Worth noting that 1.8 IQ points is the estimated "average causal effect", which means that among the exposed, some people had no changes, or even higher than expected IQ scores, while others had much lower than expected IQ scores. It would be much more telling to know the variation in scores, and how many of the exposed experienced a large enough deficit to matter (e.g. did 25% of the exposed have an IQ that was 5 less than the average non-exposed IQ).
I haven't read the study so obviously won't comment on the validity of the results.
D L, you can read the whole paper. It's available free on line. The link is in the post.
Moderate Maternal alcohol use lowers childtens Iq. It has been reported here
Post a Comment
Note: Only a member of this blog may post a comment.