|
|
Can Google make you a better doctor? | ||||
| ||||
We're talking, of course, about Google. This gargantuan answer machine in cyberspace may strike you as a poor resource, particularly when you can turn to authoritative healthcare websites—some, like Google, free of charge—to research a perplexing clinical problem. But, Google has more medical credentials than you might imagine. Google Scholar (scholar.google.com), for instance, searches only scholarly literature, weeding out online pseudoscience. And with a Google tool bar embedded in your web browser, it's just as tempting to type in "epistaxis" as it is "American Idol." Two Australian doctors put Google to the test recently and came away impressed. They selected 26 diagnostic cases published in The New England Journal of Medicine during 2005 and, without reading the differential diagnoses and conclusions, queried Google using search terms derived from each study. They reported in BMJ last December that Google guided them to the correct diagnosis in 15 of the 26 cases. For example, searching on "bullous skin conditions," "respiratory failure," and "carbamazepine" yielded the correct answer of "toxic epidermal necrolysis syndrome." The authors' conclusion: In difficult cases, it's often useful to "google for a diagnosis." Mixed reviews from the EBM world Not everyone thinks Google is such a sharp diagnostic tool. As one hospital librarian notes, the 58-percent success rate cited by the Australian physicians is just "slightly better than flipping a coin." Another bad review comes from Eamon Armstrong, an FP at Northern Arizona University who teaches other physicians how to find evidence-based medicine, or EBM, on the Internet. EBM is all about treating individual patients in light of the best available research, says Armstrong, and Google isn't precise enough to locate it in most situations. "The cases in the BMJ article deal with diagnoses so rare that most physicians will never make them in their lifetime," Armstrong says. "Google is useful for them, but not for common conditions where presenting complaints are vague, like 'nausea, fatigue, generalized muscle pain.' Google would yield a morass of disorganized information with no built-in quality filter. I can't think of a single person in the EBM field who would use Google on a regular basis." Armstrong says the BMJ study would have been more convincing had it compared Google with EBM search engines such as the TRIP Database (for more information, see "The best treatment? It's at your fingertips," in the March 4, 2005 issue). Interestingly enough, Jon Brassey, a co-founder and co-director of TRIP (www.tripdatabase.com), holds Google in higher regard. "Google isn't anywhere near as good as TRIP," Brassey wrote in a blog. "But it's pretty fine." Google was one reason why Brassey stopped charging people to use TRIP last September. "Why should people pay for TRIP when they could use Google?" wrote Brassey, an information specialist with the National Public Health Service for Wales in the United Kingdom. "If I was hard-pressed financially, I wouldn't bother subscribing to TRIP." How the CDC helps focus Google searches Brassey does fault Google for not being transparent about its search and rank methodology, which, he says, often ranks older journal articles higher than recent ones. Although the company's search-engine algorithms are proprietary, at least part of what Brassey complains about may be a reflection of the value "label" placed on any given piece of information by recognized medical sources. To see how that works, type in "hypertension" on the main Google search engine. In addition to ranked results, you'll see links to eight categories: Treatment, symptoms, test/diagnosis, cause/risk factors, for patients, for health professionals, from medical authorities, and alternative medicine. Click on "for health professionals," and more categories pop up, such as "practice guidelines" and "patient handouts." Most of the top-10 results in the various categories indicate that they have been "labeled" by heavyweight medical organizations like the Centers for Disease Control and Prevention, the National Library of Medicine, Kaiser Permanente, and the Mayo Clinic. These labels were created through a program called Google Co-op, which allows both organizations and individuals to recommend either entire websites or individual pages in selected subjects like health. Labeling can push web content higher in search-result rankings (more on that later). Take the top 10 results for "hypertension" in the practice-guideline category, for example. Heading the list is a page from the federal National Heart Lung and Blood Institute containing guidelines for the condition. The page was labeled by the CDC. Enlisting organizations like the CDC to separate web wheat from web chaff is reassuring, but Eamon Armstrong says Google Co-op still falls short of EBM standards. After all, endorsements of particular web pages amount to expert opinion, the lowest level of clinical evidence. "It's not good enough if the Mayo Clinic likes a website," says Armstrong, noting that a group's endorsement could be tied to services it provides. The ability for any individual to label clinical websites through Google Co-op raises another question mark about reliability. A chiropractor can tag web pages on health topics and then invite patients to "subscribe" to his picks at Google Co-op. When a participating patient types "brain cancer" in Google, a pertinent website that the chiropractor earmarked would be higher in the rankings than it ordinarily would be. Eventually, the ranking in any Google search could be influenced by the work of individual labelers as well as institutional ones deemed trustworthy by the company, says internist Roni Zeiger, a product manager with Google. Evaluating these folks turns out to be an automated process. "In an ideal world, we could carefully interview a labeler, but we are unable to do that," says Zeiger. "It's Google's business to learn how to make such a determination algorithmically about a labeler's trustworthiness. It's not perfect, but it's pretty darn good." Zeiger notes that one criterion for favoring a given labeler is the number of people who subscribe to his recommended websites. To assuage fears about a black-box system, Zeiger says that the labeling activity of a single individual won't nudge search results up or down very much, and that labelers like the CDC tend to carry more weight. If Google is too democratic for some doctors' tastes, it still never ceases to amaze. FP Robert Flaherty at Montana State University says he recently encountered a patient who developed a runny nose and a skin rash after grinding and polishing an object made with epoxy cement. Using Google clinically for the first time, he keyed in "epoxy" and "urticaria." Result number seven was a scholarly article from Japan confirming Flaherty's hunch—exposure to airborne epoxy particles, as opposed to skin contact, can trigger such symptoms. Flaherty, who teaches medical students about EBM, could have found the article with PubMed, but the Google tool bar on his web browser was handier. "Absolutely fascinating," says Flaherty. "The take-home message for me is that Google is another tool in the EBM toolbox." |
