Could the US Food and Drug Administration (FDA) soon use Google to track drug-related adverse events?
In a new research letter published this week in Clinical Pharmacology & Therapeutics, two FDA researchers argue that "Internet search logs" could be substantially useful to the agency's drug regulatory staff in determining which drugs might have unforeseen adverse events.
At present, FDA tracks those adverse events through a hybrid voluntary/mandatory reporting model which feeds data into its FDA Adverse Event Reporting System (FAERS). While pharmaceutical and biopharmaceutical companies are required to report all adverse events they are made aware of to FDA, doctors and patients only do so on a voluntary basis. The result of this system, FDA officials note, is that many side effects are under-reported, especially from patients who may not know how to report to FDA.
But even while FAERS sometimes falls short, it also has certain benefits. Each FAERS report contains extensive details about the adverse event, including an identified individual patient, the drug s/he took, his/her condition, whether they had any comorbid illness or were taking other medications, and a description of the event.
But over the last few years, FDA has expressed interest in moving past individual reports about patient experiences and moving toward "big data" approaches. Last year, FDA proposed a new surveillance program within its Office of Clinical Pharmacology (OCP) known as the Pharmacological Mechanism-Based Drug Safety Prediction (PMDSP) program.
The program, according to FDA, was to mine data and recognize patterns through the use of technology in order to find as-yet unrecognized safety signals. As reported in Focus, the program has experienced several setbacks and does not appear to have launched yet.
But while the PMDSP program was set to mine databases like the National Library of Medicine's MEDLINE database of medical articles and launch as many as a dozen pilot projects, until now FDA has said little about data it might obtain from search giants like Google and Yahoo, even as it has expressed interest in leveraging "big data" approaches and used its 160-million patient "Mini-Sentinel" database to probe for adverse events.
Can Regulators Use Search Engine Data?
According to the authors of the research letter—Sirarat Sarntivijai and Darrell Abernethy, both fellows at FDA—the point of using a search engine is to generate hypotheses about drug safety. In other words, to raise questions about drug safety, not to answer those questions.
"When a disproportionately strong association between an adverse event and a drug exposure is noted during ongoing analysis of FAERS data, it can be viewed as generation of a hypothesis," the duo wrote in their article, Use of Internet Search Logs to Evaluate Potential Drug Adverse Events. Similarly, "hypotheses generated by use of Internet search logs could be viewed as yet another data source," they said.
But Sarntivijai and Abernethy both conceded that search engine data is not quite as rigorous as existing approaches utilized by FDA, even if it does have its advantages. For starters, FDA just doesn't have the same amount of experience dealing with internet search data as it does with its FAERS system.
But more importantly, Sarntivijai and Abernethy noted that it can be extremely difficult to determine the true size and cause of an adverse event signal. For example, could a spike in people using Google to search for influenza vaccine side effects be the result of a new side effect, or might it be caused by the publication of an alarming news article containing decades-old information?
It's hard to say, and the example isn't hypothetical, either. The articles cites previous research in the field that found that Google's previous "Flu Trends" product, which sought to determine the scope of a flu outbreak by measuring searches for flu-related terms, had "over-reported the transmission" of the virus by being unable to distinguish between actual reports and other information.
Therefore, FDA—if it is to use search engine data—will have to figure out, paradoxically, how to use less data from its searches. Instead of treating every search query equally, it will have to determine which sources are worth listening to, and which to treat with more skepticism.
The Known Unknowns
And then there are simply the unknowns, the authors explain. "It is not at all clear how Internet search behavior would differ between instances of well-known drug–event pairs and potential pairs that have not been described in the news media and are not mentioned in available drug information sources," they wrote. "That analysis is essential and will require prospective study."
And perhaps the biggest problem of all may well be validating the data. The authors note that FDA would have to track down individual patients to establish if "what was inferred using the Internet search log methodology is what actually happened."
While that's routine practice for FDA, it also means that so-called "big data" approaches won't be a panacea for regulators, but simply a new starting point—a generator of drug safety hypotheses.
So will we one day see FDA using these types of data to track down adverse events? The authors say such a move would seem to be "a step in the right direction," even if some problems will need to be worked out prior to its use.
For now, though, if you're "feeling lucky," you're better off just communicating an adverse event to FDA the old fashioned way—through FAERS.
Use of Internet Search Logs to Evaluate Potential Drug Adverse Events