Researchers Utilise AI Technology to Scan Twitter for Signs of Opioid Abuse

In the US, a much-publicised opioid epidemic sadly continues to ravage pockets of the population. The issue has been plastered across major news networks and even piqued the interest of some accomplished documentary filmmakers; yet despite all the attention being placed upon the issue, citizens from all walks of life continue to fall under the sway of these oft-damaging drugs.

The scope of this issue has now reached a truly-frightening level, with drug overdoses (the majority of which involved heroin and/or prescription opioids) claiming the lives of more than 64,000 people in the US throughout 2016 alone, according to the US Centres for Disease Control and Prevention; this constitutes a rise of 21% from the previous year’s figures. Deaths attributed to the use of fentanyl specifically, the drug which is reported to have claimed the life of music industry legend Prince just last year, more than doubled from 2015-2016, and these figures are only expected to rise further once the appropriate health organisations have the time to gather additional data. It is this delay in receiving the relevant data which is thought to be most negatively affecting efforts to address the widening problem.

In light of this fact, researchers in the US are continually looking for more advanced and, importantly, faster methods of collecting data regarding the epicentres of this now-major public health problem in an effort to get ahead of the trend and hopefully stop it in its tracks. Their latest effort involves a combination of two fairly recent technologies; namely artificial intelligence technology and prominent social media platform Twitter.

Approximately 500 million messages are posted to micro-blogging platform Twitter on a daily basis, and this makes the publicly-searchable platform a veritable goldmine of information for those with the inclination to delve into its darker corners. Combine this with the very nature of the platform itself, which encourages short-yet-frequent posts incorporating a variety of topics which often include location data and other demographical information, as well as offering a level of anonymity which encourages a more candid tone, and Twitter soon becomes a particularly reliable source of data.

“There’s a confessional effect,” says study lead Michael Chary, a resident physician in emergency medicine at New York-Presbyterian/Queens Hospital. “People may discuss or reveal things on social media that, when directly asked, they may not. There may be a level of candor there that’s not present in the emergency room or internist’s office.”

Manually combing through millions of messages each day is not exactly a feasible proposition, so the researchers utilised AI technology to analyse publicly-visible tweets in the hope of gathering data which could help them to estimate the location and relative prevalence of prescription opioid misuse just as accurately as established epidemiologic studies in a mere fraction of the time. Traditionally, major medical research projects such as the National Survey on Drug Usage and Health (NSDUH) take years to complete, and so the team hoped that by careful analysis of freely available information on Twitter, they could create a form of early warning system which could then be used to better coordinate immediate action such as localised public health campaigns. This would also help to ensure that available resources are spent when and where they are most needed.

“We found that our estimates agreed with [NSDUH] data, suggesting that social media can be a reliable additional source of epidemiological data regarding substance use,” asserts Chary. “We can analyse social media to canvass larger segments of the general population and potentially yield timely insights.”

The method employed by the researchers involved the development of a custom AI software which then  set to work analysing more than 3.6 million tweets, identifying words and phrases thought to refer to opioids or the consumption of, including “dope,” “percs,” “white,” “TNT” and “Captain Cody”. The research also identified relevant slang terms previously unknown to the team, such as the fact that fentanyl can also be referred to as “dummies”, and that codeine also goes by the names of “syrup” or “Tango and Cash”.

Of course many of the above terms can also refer to perfectly innocent or unrelated substances and/or actions – take syrup as an example – so the team’s next task was to distinguish to those tweets which were relevant to their research from the innocuous content via the analysis of previously identified word-use patterns. The AI performed this task diligently, providing results that lined up similarly to NSDUH state-by-state estimates, thereby proving the validity of the data. This was found to be especially true amongst 18-25 year olds, which the Pew Research Center attributed to the fact that 36% of Twitter’s user base are between the ages of 18 and 29.

For those worried that their data may have been used in the study without their consent and that information may have been gathered regarding them specifically, it should be stated that the team followed established medical protocols and keep the identities of individual tweeters anonymous. The team did however acknowledge that it would not be all that difficult to later trace a tweet back to an individual user if a government or law enforcement agency should want to conduct a study of a similar nature.

“Twitter data is high-volume and the content is short-form, brief statements [that] are easier to classify than very long and complex statements,” says Michael Gilbert, a Portland, Ore.–based epidemiologist and social media researcher not involved in Chary’s research. “The combination of the volume of data and the format of the data makes Twitter suitable for machine-learning tools. Are people talking about getting high, controlling pain or some other motivation that is underlying a common behaviour? People are more likely to share certain types of information with their peers than they will with their health care providers.”

Sam Bonson

Sam is an aspiring novelist with a passion for fantasy and crime thrillers. Currently working as Editor of Social Songbird, he hopes to one day drop that 'aspiring' prefix. Follow him @Songbird_Sam

Contact us on Twitter, on Facebook, or leave your comments below. To find out about social media training or management why not take a look at our website for more info: TheSMFGroup.com