Can diagnostic decisions be automated?

June 8, 2007

Jerome Groopman's new book How Doctors Think has revived the debate over whether physicians are too quick to choose diagnoses that fit their prior experience or their preconceptions about a patient. But can computer programs help physicians improve their diagnostic skills?

Jerome Groopman's new book How Doctors Think has revived the debate over whether physicians are too quick to choose diagnoses that fit their prior experience or their preconceptions about a patient. But can computer programs help physicians improve their diagnostic skills?

Joseph Britto, MD, CEO and founder of Isabel Healthcare, argues that his company's web-based product—which is interfaced with some leading EHRs—can help jog doctors' memories about the range of possible diagnoses at the point of care. "It gives you a checklist of diagnoses based on a patient's symptoms and signs." When you click on a particular diagnosis, he says, you're linked to the latest testing and treatment information in textbooks and medical journals.

How does this differ from evidence-based medicine resources like Uptodate, DynaMed and Inforetriever? Unlike these tools, which are only useful when you have a diagnosis in mind, he says, Isabel guides physicians through a range of possibilities that start with the patient's symptoms.

Britto admits that similar decision support systems, including D-Explain, QMR, and Iliad, haven't caught on widely among physicians. Part of what makes Isabel different, he says, is that it uses a kind of "natural language processing" to pose questions, and it responds in far less time than these other systems do. He cites studies showing that in 90 percent of cases, Isabel will include the correct diagnosis in the list of possible diagnoses it provides. Nevertheless, it's still up to the physician to choose the right diagnosis, he points out.

Internist Peter Basch, director of e-health for MedStar Health, a hospital system in the Washington, DC, area, takes a skeptical view of diagnostic support programs. "Clinical decision support is best for prompts and reminders about services due, and has never been that effective for practicing clinicians in terms of suggesting a diagnosis," he says. "Eventually, it may get there, but we're not there yet."