Side effects of drugs that are rare or those that happen when combinations of drugs are taken may take years to turn up and be recognized by the medical cognoscenti, yet they can also be severe or even life-threatening. Rick and I agree on PodMed this week that the emerging field of 'webscale pharmacovigilance,' as described in a study in a journal we've never talked about before, the Journal of American Medical Informatics, is very persuasive about the utility of this field in helping identify and hopefully ameliorate many of these drug reactions.
The article is entitled "Web-scale pharmacovigilance: listening to signals from the crowd," and it's an analysis of Web search data from 2010 from literally millions of users who opted in to allow their searches to be captured anonymously by Microsoft, using a browser add-on. 82 million drug, symptom and condition queries from 6 million searchers were included in the analysis. Researchers specifically searched for two drugs whose interaction was known to produce high blood sugar, or hyperglycemia: pravastatin, one of the class of cholesterol-lowering drugs known as statins, and paroxetine, an antidepressant. For comparison, the study also examined searches for drug pairs also known to be associated with hyperglycemia and those known not to be associated with the condition.
In short, the results of extensive data analysis of searches did reveal the concern of hyperglycemia among those who also searched for paroxetine and pravastatin. The authors speculate that anonymized (I LOVE this as a verb!) data collection from internet searches, in conjunction with electronic medical record data analysis and AERS, the adverse event reporting system, can provide early signals of rare adverse drug reactions as well as drug-drug interactions and would comprise an important contributor to post-marketing surveillance activity mandated by the FDA, and Rick and I heartily agree. Another bene is that such searches are very inexpensive to conduct, and my suspicion is that with cloud computing will become even less expensive as time goes on.
Any discussion of this type would be remiss in not mentioning the burgeoning field of research specifically directed at mining data from gigantic data sources. Recent research from Johns Hopkins by Mark Dredze and colleagues has fine-tuned tweet analysis to pinpoint when the tweeter actually has the flu or is merely chatting about it, with an eye toward using such data to follow in real time the progress of infections like influenza as they make their way around the globe. Mark points out that tweets are much more contemporaneous with actual infection spread than internet searches of things like flu symptoms or 24 hour pharmacies, which follow the spread of the virus and subsequent development of infection days later. In the hands of public health types the hope is that when a pandemic strain arises we can get our arms around it quickly and reduce disease burden. And the flu, of course, is just the model for almost any infectious disease., with other researchers already on outbreaks of malaria or TB with regard to social media. Pretty cool, methinks.
Also this week, three studies in NEJM on endovascular strategies for stroke management, the intersection of HIV infection and risk of MI in JAMA Internal Medicine, and in MMWR, yet another enteric pathogen group demonstrating antibiotic resistance. Until next week, y'all live well.
Internet Searches and Drug Reactions,No Comments