Here is a reasonably real-time feed of Twitter comments (“tweets) containing the word ‘overdose’, with a query performed about every 5 seconds.
Usernames have been stripped off (some usernames contain the word ‘overdose’ so it’s possible you won’t see the word ‘overdose’ in a given tweet). To experiment you can filter for a couple of other terms: cocaine, meth, kratom, purple drank.
The vast majority of these tweets are of course innocuous. However, a couple of interesting questions arise from data like this.
Could tweets reflect or warn of an epidemic of overdoses in a community or region? The widespread prescription, use, and diversion of opioids has contributed to an enormous problem, as noted in the popular press:
Drug deaths now outnumber U.S. traffic fatalities (September, 2011).
Drugs exceeded motor vehicle accidents as a cause of death in 2009, killing at least 37,485 people nationwide, according to preliminary data from the U.S. Centers for Disease Control and Prevention.
While most major causes of preventable death are declining, drugs are an exception. The death toll has doubled in the last decade, now claiming a life every 14 minutes. By contrast, traffic accidents have been dropping for decades because of huge investments in auto safety.
Public health experts have used the comparison to draw attention to the nation’s growing prescription drug problem, which they characterize as an epidemic. This is the first time that drugs have accounted for more fatalities than traffic accidents since the government started tracking drug-induced deaths in 1979.
Many of these deaths may be localized to certain areas. The U.S. Drug Enforcement Administration, for example, considers Los Angeles to be one of three “pill mills” where there may have been even easier availability of potentially dangerous prescription medications. Would data mined from the social web reflect that geographic availability?
Some Twitter feeds are geo-tagged (I have not displayed any geo-tagging in the demonstration feed).
It’s not unheard of for a badly synthesized chemical to find it’s way into a particular region, sometimes with disastrous effects. In other instances, increased exposure to non-prescription substances are problematic.
- What might other widely available data sources reveal (e.g. Google Trends (search data) and Ngrams (historical use of phrases in writing)?
- Could sentiment analysis of social data discover trends in experiences with various substances?