Human interaction is probably going to change because of Google Glass and/or derivative technology. I think it's easy to overlook the speech recognition baked into Google Glass; it's really quite stunning (and to be fair I've been impressed by Apple's Siri as well, along with the Mac OS X continuous speech recognition capabilities for dictation). Terms like "pheochromocytoma" (a type of tumor that precipitously secretes adrenaline into your bloodstream) are handled with aplomb. We think of this as a nice-to-have, but really this is a critical aspect for clinical medicine -- how frail are we, as human doctors, in recognizing everything the patient has said? If the patient mutters under their breath that they've had ankle pain when they're on a fluoroquinolone, does the physician brain even recognize they said that and tie it into the issue at hand as a possible side effect?
Scribing, it turns out, is an important function. Katie Hafner writes about it in yesterday's New York Times, in an article titled "A Busy Doctor's Right Hand, Ever Ready to Type". An ENT surgeon and close friend of mine has often yearned for some sort of system that could automatically scribe your entire interaction with a patient -- or the very least the highlights. When you have a scope jammed up someone's nose it's not really a great time to be scribbling something. Charting, it turns out, is a pain.
Writing physician notes -- or typing them while your patient is talking -- can be a good process in and of itself though. For the non-doctors out there, it's important to understand the difference between the type of medical chart note written by a medical student or trainee, and one written by a full-fledged attending doc (or more senior resident).
When medical students first start out, they learn a structure, to be complete and avoid missing something. This can be agonizing to listen to because they're working their way through a very standard structure, which almost literally goes head to toe. As a result, their medical documentation also is very complete to the point of containing a lot of stuff that really doesn't weigh into our diagnostic decision making -- like a census surveyor trying to catalog every detail of your illness.
By comparison, when you listen to an attending physician take a history from a patient (i.e. interviewing them), they're problem-focused. In addition to already having a good sense of the patient's history, they'll zoom in on hypotheses going through their head -- we call this differential diagnosis -- and ask questions to refine those hypotheses. One example of this process is "Guess the Dictator/Sit-com character: http://www.smalltime.com/Dictator, although that's more of a decision tree than a probabilistic Bayesian process.
Anyhow, the point is that the note the attending scribbles down appears brief -- one of my favorites from a legendary cardiologist at NY Presbyterian - Cornell read, simply, "LASIX!". But it reflects a thought process, a sort of filtration and re-prioritization of what matters.
Google Glass, therefore, could be a very interesting scribe. But a computer could be a scribe, too, right? Well, this is where the display comes in. Like a human scribe in the exam room, who knows just enough to be clever, Glass could point out simple things -- like, say, inconsistent stories in a pain medication-seeking patient. Or search against common mispronounciations of drugs to speed up the game of "I take this pill, I think it's called, umm...." These sorts of things are handy for a doc to be able to glance at as they come up.
In a more advanced form, you'd tie this sort of natural language processing to start parsing out and tokenizing the text. The reason for this is that you could start applying Clinical Decision Support rules. Clinical Decision Support is what separates quality measures reporting from action and improving outcomes. As Dr. Jacob Reider at HHS/ONC might put it, it's like the difference between knowing how to give kids a C- grade vs. giving teachers a way to help those kids improve their grades.
So in addition to scribing, Glass should eventually be a nice mechanism to point out "wow they just described Lyme disease... you want to ask this other question to clinch it?" I think of systems like Dxplain for that, to improve diagnostic certainty. Other systems have aimed at diagnosis as well, and certainly there's a fascination with artificially-intelligent diagnosticians. IBM Watson is a trendy example of this.
But it turns out that another arena of Clinical Decision Support is just common stuff that docs forget. Like remembering to prescribe a medication called an ACE-inhibitor in a diabetic with early kidney damage called microalbuminuria. I once led a chart review study of patient records looking at the cause of gaps in care, like this example, to see whether it was the doctor who had failed to prescribe the drug, or if the patient had failed to fill the prescription. While we think of patient non-adherence as a huge issue, it turns out that if the doctor forgets or neglects to write the prescription, there's no chance for the patient to be non-adherent. And, as the study turned out, doctors just not remembering to refill medications was all too common -- the medication was nowhere to be found in the chart, rather than the medication written in the chart but the patient not picking up the prescription.
This begets a whole user interface and usability issue -- what's the best way to offer a tip or correction to a doc? Too many and they get alert fatigue. Too few and they'll just keep making the mistakes they're destined to make. Glass offers a great way to slip a message in unobtrusively in the heads-up display, to the point of even remembering just to ask another question. Likewise, Glass could listen for the appropriate action as the result of the prompt/alert; e.g. once you said "Mrs. [Patient], I'd like to start you on a drug called lisinopril" Glass would recognize that you had fulfilled the clinical decision support prompt and dismiss it for you.
So speech recognitions is, I believe, not only a critical part of Glass, but an incredible enabler of a potentially much more fluid experience between doctors and patients -- and, if driven by Clinical Decision Support, a whole lot better for quality as well.
(Next up: medical student James X. Wang. Fear not: we'll get to the neurologist a little later!)
No comments:
Post a Comment