Posted by Ezra Klein
In “The Robot Will See You Now,” Jonathan Cohn took a long, reported look at the efforts to computerize and roboticize not just medical records but the practice of medicine itself. If IBM’s Watson can win a game of “Jeopardy,” why can’t it help a nurse practitioner diagnose even very tough-to-read symptoms? We spoke about the article over e-mail this week.
Ezra Klein: So let’s start with the hard question: After reporting this article, are you hoping, 10 years from today, that if you have to see a doctor, there’s a robot in the room?
Jonathan Cohn: Robot? No. Sophisticated computer? Yes.
Really there are two reasons for this. First, medicine is becoming a lot more complex. You see it in cancer care. We’re developing much more detailed information about tumors — about the specific mutation that’s causing a group of cells to divide uncontrollably. And we’re learning how to target treatments more specifically than ever before. But the amount of knowledge is already more than even academic oncologists — the ones with the most time to keep up with the latest research — can manage on their own. The challenge will only grow. And it’s not just cancer. The same sorts of information are becoming available for diabetes, high blood pressure — you name it. Making full use of this information is going to require technological assistance.
But the more important change may be something that seems a lot more mundane: correcting mistakes that even conscientious providers are bound to make. Marty Kohn, the E.R. physician who’s led the development of IBM Watson for health care, talks a lot about “anchor bias.” You’re a doctor, you’re listening to your patient, and you inevitably start to form a diagnosis after hearing a few facts. But that diagnosis may be wrong — or incomplete. It’s human nature to downplay information you get later, even if it suggests other possibilities. A computer isn’t subject to human nature, so it might point out that, gee, the diagnosis might not be exactly what you think.
Of course, computers can’t provide warmth. And we’re still a long ways from the day when they can pick up nuances from a medical record, in the way a trained professional can. Turns out that’s a little more complicated than producing conjuring up trivia for “Jeopardy.”
EK: I’m fully in the give-me-robots camp. I think the actual numbers on how our health-care system performs are shocking. The number of errors, of misdiagnoses, of patients who aren’t given the recommended care — it’s really scary, I think. That’s not to say there’s no good reason for it. As you say, medicine has grown inhumanly complex. But perhaps we need something inhuman to deal with it now. And that’s where I think this gets interesting.
Your article focused heavily on the economics of automating medicine. But I think this discussion cuts in a different direction. We ask constantly about how to bring costs down. But it would be just as good, in my opinion, to bring value way up. The problem right now is we’re spending too much and getting too little. It seems possible that computers won’t necessarily solve the spending-too-much part, particularly if we find lots of really great, really expensive new treatments. But I’m pretty confident they can help with the getting-too-little part. What are your thoughts on that possible future? Much better, but not necessarily much cheaper, care?
JC: Like you, I would happily take care that is better, even if it’s not cheaper. We spend too much on health care, but that’s not because spending 17 percent of GDP on health care — or 20, or 25 — is truly “unsustainable.” If we’re willing to put up the money, in principle, we can sustain it. The problem is that we get so little for what we spend. Like you say, the quality just isn’t that good. Increasing the value of health care, so that we’re actually getting our money’s worth, would be a clear improvement. And to some extent, that’s what’s bound to happen.
But I’m a bit more optimistic that we can also work it the other way — and bring cost down to value. I’m talking, of course, in relative terms. Health care spending would still go up, but it wouldn’t go up as quickly. We’d be bending the curve. The key is how. I don’t think it will happen primarily through reduction in misdiagnosis or duplication. I mean, IT and data have the potential to do that. But my gut says the savings will materialize for a different reason.
EK: So then the next question is obvious: What’s that reason?
JC: When we talk about health care, we talk a lot about doctors. But there are tons of other health professionals — nurses, medical assistants and so on — who provide care. With decision-making technology, they can do even more. It gives them access to more information and allows them to communicate with one another more fluidly, so that they can take a team approach to care. Of course, we’re moving to this model already. The best health care providers, like Group Health of Puget Sound, already have a team approach. But these IT innovations can make the change easier and faster. This saves money because doctors are really, really expensive to pay. The other professionals, not so much. In effect, this allows us to take some of the work doctors now do and give it to other professionals, whose labor isn’t as expensive. My bet is that this is where we’ll see the most savings. Of course, there are obstacles to this change too.
EK: One of the obstacles to change, I imagine, is regulatory. Doctors are very good at getting laws passed making it hard for there to be more doctors and making it hard for patients to be treated without them in the room. So when you were talking to the folks who are focusing on building these systems, how much did they worry about the regulatory side?
JC: A fair amount. This is, as you know, a very old fight in health care. And it’s not just doctors. Every group of professionals guards its turf very jealously. And every group has leverage in the state capitols. One such fight is playing out in California right now. But there’s reason to think the resistance may be weaker now. The big reason is that we’re facing a pretty severe shortage of primary care providers — the people who handle ongoing, frequently routine care. We don’t have enough doctors to provide that care, and the problem will only get worse with time. The only way to solve it is to spread the work around. And precisely because doctors will remain in high demand, doctors needn’t fear this shift.
In fact, I would make the argument (as many did to me) that doctors stand to benefit. They’ll get to concentrate on the most interesting cases and the most challenging parts of their jobs. Of course, it’s easy to say that, and I’m sure plenty of physicians (and other professionals) will resist the transition. And we obviously need to make sure this is being done in a smart, careful way — rather than simply handing off work to people not qualified to do it.
EK: Towards the end, your article turns towards the idea that the computerization of medicine could help the middle class, by creating a lot of high-paying positions that don’t require quite as much training as becoming a surgeon. Is the implication of that that there would be significantly fewer positions for high-paid doctors — that they will, in effect, become supervisors rather than the main care providers, and so their pay will disperse but so, too, will the number of positions for them to fill?
JC: Sort of. It’s quite likely other health professionals will be taking on greater responsibilities, as we were just discussing. Physicians, in turn, will take on that “team leader” role. This is already happening, in fact — technology just makes it a lot eaiser. But remember — there’s a looming shortage of doctors. That’s why this has the potential (emphasis on “potential”) to be win-win. The demand for doctors will remain sufficiently high, even with this team model, that nobody is going to lose their jobs. We might see the growth in physician positions — and maybe the growth in physician salaries — slow down. But both will still be getting bigger. I guess you could say I think the medical profession won’t be downsizing. It just won’t be upsizing as much as it might otherwise.
EK: Let me end by asking you about something a bit more low-tech. The “robots” of 2001-2009 were electronic health records, which seem like a blindingly obvious idea, were estimated to have huge benefits, and which turned out to be much more difficult to implement effectively than we thought. You talk a bit about them in your piece. So where do you think we are on electronic health records, and what do they tell us about medicine’s ability to absorb productivity-enhancing technologies?
JC: If you want a reason to pessimistic about the potential for innovation, the history of electronic health records is probably the best one. Like you say, we’ve been talking about EHRs for an awfully long time. The big problem isn’t so much developing the records themselves as coming up with a common standard for information and making sure the different systems can talk to each other. It’s daunting. The Recovery Act put a lot of money into pushing EHRs — Mike Grunwald writes about this a lot in his book — and the good news is that use of EHRs has increased dramatically. The bad news? We’re still behind other countries when it comes to the number of physicians that have systems. Your colleague Sarah Kilff wrote about this a while ago. And getting the systems to talk with each other is still very much a work in progress.
What’s holding things up? It’s complicated, and I’m not sure I fully know the answer. But one problem is that providers who have invested a lot of money in systems already don’t want to change them, and the companies that build this stuff don’t want to be using somebody else’s standards. It’s all very predictable and understandable, but it speaks to the inherent difficulty of realizing technology’s potential. These particular obstacles may not do much to affect the innovations I cover in my article — harnessing data to make better decisions, using IT to increase the skills of professionals and so on. Then again, a lot of these innovations depend on good electronic health records to work. What good is a wireless sensor for vital signs if it’s got nowhere to send the information? If the EHRs aren’t fully functional then these innovations can’t live up to their full potential.
Courtesy: The Washington Post