When it comes to drug and alcohol abuse, the best interventions are those implemented as early as possible. That’s why researchers are increasingly focused on determining which risk factors spur behaviors like binge drinking and alcohol addiction, as well as how they interact. But integrating all that information accurately hasn’t always been easy. Now, a study published today in Nature describes how researchers "taught" a computer to weigh the risk factors associated with teenage binge drinking using data from 14-year-olds. And the results, says lead study author and cognitive neuroscientist Robert Whelan, are promising: the computer was able to predict future adolescent binge drinkers with "70 percent accuracy."
the model learns which features distinguish binge-drinkers
In the study, Whelan and his colleagues linked data such as brain function, personality, life history, and genetics from hundreds of 14-year-olds who had participated in the IMAGEN project to their future alcohol use behaviors at the age of 16. This allowed the computer model to "learn" how over 40 different variables can influence future alcohol misuse. For example, the model learned that smoking at age 14 merits a relatively high weighting, whereas pubertal development status at age 14 doesn’t. "It’s a form of supervised learning," Whelan says. "You tell the computer what are the two groups — either binge drinker or not — and it has to learn what features best distinguish them."
Once the computer model had determined which factors were most important, researchers applied it to a new group of 14-years-olds and compared its results to what actually happened once the teens turned 16. "Some people who aren’t binge drinkers were classified as binge drinkers, but overall, 70 percent were picked up correctly," says Whelan, who works at University College Dublin. "We learned that predicting teenage binge drinking is possible," and "we can be confident about that number."
"predicting teenage binge drinking is possible."
About 10 percent of American 13- to 14-year-olds report drinking alcohol regularly, and that number nearly triples by age 16. This is worrisome, Whelan says, because early alcohol use is an important risk factor for alcoholism as an adult. But there’s hope, he says, because according to a 2001 study, early interventions can reduce the odds of adult alcohol dependence by 10 percent for each year that drinking behaviors are delayed in adolescents. That’s why Whelan thinks future models should also include factors like peer pressure, as measured through social media interactions, for instance. "At the time the study was proposed — that’s probably going back to seven years ago — things like social media weren’t as prevalent," he says. "If we had that information now, I’m pretty sure it would bump up that 70 percent."
Whelan also thinks that furthering genetic research will lead to stronger predictions. "We know from familial studies that alcohol misuses and substance misuse has a strong genetic component but we don’t know much about the genes that drive that," he says. "So in however many years, we might be able to make a dent in that."
But even if those predictions become stronger, they might not have immediate real-world applications, given the $1,000-per-participant price tag that comes with the inclusion of brain imaging. "Brain imaging is very expensive to acquire and analyze," Whelan says. But it might still be possible to construct strong models without the brain imaging information, if researchers can use personality traits that overlap directly with what’s going on in the brain. "Identifying future alcohol misuse with perfect accuracy may never be possible," Whelan says, but identifying teens who are at risk already is. "Catching them so we can perform early interventions is really where this sort of research should be directed."