OR WAIT null SECS
From 620,000 tweets, machine learning identified nearly 1,800 sales pitches for the deadly drugs.
Credit: Flickr user Grumpy Puddin. Original image resized for this story.
Few disagree that opioids’ widespread grip on the nation is primarily a health issue. President Donald Trump, for instance, is expected today to order the US Department of Health and Human Services to declare the matter a “public health emergency,” providing federal dollars and loosening regulations for the fight against opioid addiction.
And artificial intelligence might also prove to be a valuable weapon.
Researchers from the University of California San Diego’s medical and engineering schools have used machine learning to create technology that combed Twitter for people illegally marketing prescription opioids on the platform, according to an announcement. Their efforts uncovered 1,778 tweets advertising the drugs, 90% of which included links to virtual cash registers, according to the study, published this month in the American Journal of Public Health.
The innovation comes at a time when opioid abuse and death rates are mushrooming across the country. In 2015, 12.5 million Americans misused prescription opioids, 2 million were found to have prescription opioid use disorder, and 33,091 died from opioid overdoses, according to the federal government. Tim K. Mackey, PhD, a university med school associate professor of anesthesiology and global public health who led the project, said the use of prescription and non-prescription opioids continues to rise.
“Public policy and law enforcement efforts are trying to address this crisis, but closer attention to the potential negative influence of digital technologies is needed,” he said in a statement. “Our study demonstrates the utility of a technology to aid in these efforts that searches social media for behavior that poses a public threat, such as the illegal sale of controlled substances.”
His team scanned Twitter between June and November 2015, amassing 619,937 tweets containing the words codeine, Percocet, fentanyl, Vicodin, Oxycontin, oxycodone, or hydrocone. Of that data set, AI pinpointed nearly 1,800 tweets that aimed to sell opioids, according to the study.
Mackey and his colleagues employed a 3-step process to get the job done. First, they used cloud-based computing to compile the hundreds of thousands of tweets. Then, they used machine learning to identify those ones marketing opioids. Finally, they underwent a web forensic examination to analyze tweets containing links to outside websites, according to the announcement.
What they found, according to the study, was that 1% of the targeted tweets actually marketed opioids over the 5-month span. Of that number, just 46 hyperlinks remained live when researchers analyzed the data 8 months later, according to the university.
Many directed users to forums, classified ads, and other blogs, most of which were based in other countries, including Pakistan, a major drug counterfeiting area, according to the study. (The research team published an interactive map of its findings.)
Mackey said if the tool were used for active surveillance, it could identify additional live links and other illicit health-related activities online. Selling or advertising drugs online, researchers noted, is prohibited by the Ryan Haight Online Pharmacy Consumer Protection Act, which went into law in 2008, following the death of an 18-year-old who died after buying Vicodin over the internet.
“This technology could help improve enforcement of the Ryan Haight Act,” Mackey said, noting that social media platforms could also use the tool to identify and censor illegal content.
This work jumped off findings published earlier this year in the journal Addictive Behaviors, according to the school. In that case, Mackey and colleagues used machine learning to “identify nonmedical uses of prescription drugs and behavioral trends” on Twitter, the university wrote. Of 11 million tweets, they flagged 2.3 million for the discussion of nonmedical drug use. Another key finding? Their methodology was able to sift through the mass of data “with minimal human intervention” to spotlight themes and trends.