In a public release, the American Association for the Advancement of Science (AAAS) reported:
A group of 11 genes can successfully predict whether an individual is at increased risk of alcoholism, a research team from the United States and Germany reported Tuesday.
"This powerful panel of just 11 genes successfully identified who has problems with alcohol abuse and who does not in tests in three patient populations on two continents, in two ethnicities and in both genders," said Alexander B. Niculescu III, M.D., Ph.D., principal investigators and associate professor of psychiatry and medical neuroscience at the Indiana University School of Medicine.
This press release highlights the publication of the article "Genetic risk prediction and neurobiological understanding of alcoholism" in the journal Translational Psychiatry, published online on May 20. The lead author of the study claims that, as HealthDay reporter Randy Dotinga put it, "it may be possible someday for young people to take a blood test and learn if they're susceptible to alcoholism."
This is a surprising release and an unexpected claim, because the study itself shows that the genetic test did not predict alcoholism at all; the researchers demonstrated that the predictive ability of their test was not better than tossing a coin.
This incorrect interpretation of genetic prediction studies is common, unfortunately.
Researchers tend to base conclusions about the predictive ability of a genetic test on the observation that the genes are associated with the risk of the disease of interest, particularly when the genes are "biologically plausible." In this study, certain copies (alleles) in the 11 genes were seen more frequently among people with alcoholism than among those without.
Yet the mere fact that gene variations were seen more frequently in people with alcoholism does not imply that the genes can be useful to predict the disease. What determines predictive performance is the answer to the question of how much more frequently.
To answer this question, the researchers examined the gene variations together by calculating a so-called genetic risk score for each person in the study. This genetic risk score was the total number of risk copies on 66 gene variations in the 11 genes. Given that individuals can inherit either zero, one, or two risk copies from their parents, the genetic risk scores could range from 0 to 132 (copies). A higher genetic risk score indicates a stronger genetic predisposition to alcoholism.
The researchers investigated whether the group of people with alcoholism on average had higher genetic risks scores than those without alcoholism. They found that the average risk score was 64.0 among people with alcoholism and 62.7 among those without.* On average, the people with alcoholism had only one risk copy more than those without alcoholism.
When the average scores are so similar between the groups with and without alcoholism, the predictive ability of a genetic test based on these 11 genes is expected to be low. This is because people do not exactly have the average risk scores; rather, they have risk scores that vary around these averages.
Because the risk scores vary and the difference between the averages was tiny, the genetic risk scores between the two groups will be very similar. As a result, the genetic test will be unable to distinguish between individuals with and without alcoholism.
The authors had formally assessed the predictive ability. They performed an analysis that rates the predictive ability on a scale from 0.5 to 1. A value of 0.5 equals the predictive ability of a coin toss, and a value of 1 indicates perfect prediction. They performed several analyses and found that the value for their genetic risk score was 0.54 at best -- hardly better than tossing a coin.
The study unmistakably shows that the genetic test based on 11 genes could not predict alcoholism.
When researchers are overly optimistic about the predictive ability of a genetic test, we usually find that they had performed only the association analysis and not the prediction analysis. These researchers correctly did both, but they were mistaken about the interpretation.
In their article, the authors conclude that the genetic test "can differentiate between alcoholics and controls at a population level, although at an individual level the margin may be small." They refer to the small differences of the average scores between the two groups for the first conclusion, and to the assessment of the predictive ability for the second.
However, the prediction analysis does not indicate the performance at the individual level; it is a population-level assessment as well. The small difference in the average risk scores and the low value for the predictive ability are two sides of the same coin.
In the coming days, media around the world will copy and distribute this public release from AAAS. People will read that a test "could identify people who are at higher or lower risk for the disease," and that "knowing one has a genetic predisposition to alcohol abuse could encourage behavioral and lifestyle changes."
That is pretty bad news.
*The average genetic risk scores were recalculated for this blog post. The researchers had transformed the scale of 0 to 132 into a scale of 0 to 100. The averages of 64.0 and 62.7 are 48.5 and 47.5 in the article. The transformation has no impact on the interpretation; the difference between averages remains tiny.