Editor's note: Dr. Meltzer is the Fanny L. Pritzker Professor of Medicine, Chief of the Section of Hospital Medicine, and Director of the Center for Health and the Social Sciences at the University of Chicago. His research aims to improve the cost and quality of hospital care. We spoke with him about the Comprehensive Care Physician Model, which he pioneered and was recently featured in an article in The New York Times Magazine.
Dr. Robert M. Wachter: Tell us about the genesis of your interest in the Comprehensive Care Physician model.
Dr. David O. Meltzer: I'd been studying hospitalists for a long time, trying to understand whether they improved costs and outcomes and how we could make the model even better. When I was an MD–PhD student getting interested in this question, on the one hand there was so much to know and on the other hand I could see even as a medical student the cost of fragmentation of care. At the time, the Medical Outcomes Study had just come out, and the results suggested that patients who were cared for by specialists had higher costs and worse outcomes than patients cared for by generalists.
When I came to Chicago as a new faculty member, I started collecting data on the inpatient services, and that began my studies and my interest in the value of specialization. We've been studying the outcomes of hospitalists and finding effects, but still have deeper questions about effectiveness. One of my heroes in graduate school was Bob Fogel, who won the Nobel Prize in economics. His big idea was that economic historians should use data from the past to learn fundamental things about behavior. I thought maybe I could do an economic history of hospitalists. I had the opportunity to apply for the Robert Wood Johnson Foundation Investigator Awards. During the project, I started developing ideas about why hospitalists grew. One idea was that they potentially were better at hospital care. But another was that hospitalists grew because it was too hard for primary care doctors to do the job they had previously done—seeing patients both in clinic and in the hospital.
As we did these analyses, we discovered that over time traditional general internists had had dramatic increases in ambulatory volume compared to inpatient volume, and that wasn't because inpatient volume had fallen but because ambulatory volumes had grown. The idea is that patients started going to the doctor not because they didn't feel well but because they wanted to stay well. You could fill your whole day as a primary care doctor with patients and almost never have any patients in the hospital. When you did have a patient in the hospital, it was a big disruption to your schedule. Because you couldn't just block your morning to hypothetically see a patient in the hospital, you had to fill it with clinic appointments. This ambulatory economics theory of hospitalist growth was something we developed then found some strong empirical support for by looking at which primary care doctors decided to give up their patients to hospitalists. We found a lot of evidence that this push was coming from primary care initially, rather than the pull from better hospital care.
In looking at the math behind this idea, I realized that if you were to change the likelihood that patients were admitted to the hospital so that a physician had more patients admitted out of their panel size, you could reset the incentives so that it made sense for a doctor to see patients both in clinic and in the hospital. Then the idea was to have a subset of doctors who focused their practice only on patients at high risk of hospitalization so that they could have a clinic panel that was small enough to provide them with the ambulatory care they needed while only seeing them in the clinic in the afternoon, and then have the morning free to provide hospital care the way internists had traditionally done—and in fact is the way that hospitalists then were doing. So out of this fundamental research trying to understand why hospitalists grew I had this idea for the Comprehensive Care Physician Model.
RW: How important were changes in the policy environment, particularly readmission penalties or moving toward value-based payments, as creating the economic and policy milieu in which this effort might be interesting and potentially feasible?
DM: When I originally came up with the idea, I wasn't thinking about that per se. I was thinking more about whether care would be improved, and if we might lower the total cost of care by doing this. As we began to move it into practice, of course we thought about that in the context of our own hospital. But it wasn't the fundamental driver. The fundamental driver was the belief that sooner or later we would move down a road where we had incentives to reduce utilization.
RW: You mentioned that the only way this works is that you reduce the primary care doctor or the comprehensive doctor's panel size to create bandwidth to do this. I assume the bandwidth is both to free up time to go to the hospital as well as to manage more comprehensively people in the outpatient settings.
DM: Certainly, it's decreasing clinic volume so you have the time to be in the hospital. We tend to have longer appointment times. But some forms of care coordination need go away in a model like this, because as a Comprehensive Care Physician (CCP) you don't have to communicate with a primary care doctor. And as the CCP you don't have to communicate with a hospitalist.
RW: How about the numbers? How did you take a guess at what the number would be and then what did it turn out to be?
DM: Basically, I took a guess as to how often patients would be hospitalized and for how many days. I guessed about how often you would see them in clinic, imagining every couple of months and some distribution of things. With that back-of-the-envelope calculation, I guessed you could handle around 200 patients. We managed to handle that and a bit more. One big lesson from doing this for a couple of years is the incredible heterogeneity in how much people are hospitalized across this population, and it varies a lot in the course of their illness. The key thing to make the model work is that you have enough inpatient volume to justify presence, and you don't have too much ambulatory volume to overwhelm people. There is much less variability across people and their needs for ambulatory care than there is in their needs for inpatient care. The longer people are in the program, the more their hospitalization tends to decline. I would love to tell you that's all because the program is so great and that we're reducing hospitalizations. But the truth is a big piece of that is simply because we typically bring people into the program at a time when they've recently been hospitalized or been in the emergency department, and they tend to have more active illness. We make them better, they get better on their own, or they die. Any of those tends to result in decreased hospitalization rates over time. One big worry we had is that the model would prove to be unsustainable in terms of the volume of inpatient and outpatient care simply because of this natural tendency. But it has worked out. We've managed to maintain 3 or 4 people on average in the hospital every day for each of our doctors with this panel size of about 200 people.
RW: Do you use any artificial intelligence–based algorithm or is the back of that same envelope to figure out high utilizers?
DM: When we started out, we thought about machine learning and did some predictive analytics looking at what the highest risk populations would be. But it quickly became clear, both looking at our own data and reading the literature, that the very best predictor of future hospitalization is past hospitalization. The simple question "Have you been hospitalized in the past year?" in our population was enough to get adequate inpatient volumes that this would work. The other important thing is just being practical—the biggest danger in a program like this is that you build it and you don't enroll anyone and you have unused capacity. We were doing an RCT [randomized controlled trial], so not only did we have to find patients who would potentially benefit from it, we had to convince them to enroll in an RCT. It took us a good 4 years to enroll 2000 patients in this study. If we had been incredibly selective, we would have been a small program for a long time. That would never have worked from a study perspective, but it also would have created real programmatic instability. Because if the program had been that small, we probably would have only been able to justify having two doctors do it or something like that. And stuff comes up in doctors' lives, they have to move and things happen. If you have a two-person program and you lose one of the doctors, you have a disaster. So, it was a good thing that we had somewhat broader inclusion criteria.
We've analyzed the data that we have, and we think the benefits of the program are much greater in people who are at higher risk of hospitalization than with people who are at lower risk. But the truth is that if you're not at very high risk of hospitalization, the model is not more costly than just the typical ambulatory care model. All you're doing is using up a spot that could be used by someone who would benefit more, and as long as there are enough people in the program who are hospitalized enough to justify you having blocked off a couple of hours in the morning to do this, it works out fine. Since we started doing this, we also realized that there was a complementary model where you serve as the rounder for a set of ambulatory-based physicians and then basically are the hospitalist who is always admitting for them. That is another way to get adequate inpatient volume.
Comprehensive Care Programs promote continuity in two ways. The obvious way is to think about physicians seeing patients in clinic and seeing them in the hospital. The other way is that they spend a couple of hours in the hospital every morning throughout the vast majority of the year. Whereas a typical hospitalist model where you work about half the year clinically implies that half the time you're in the hospital—as a patient you couldn't possibly have the same doctor—in this model you regularly can get the same doctor. By including that rounder model along with a CCP model, you can relax a lot of the potential restrictions on how often people are in the hospital and still have adequate inpatient volume to justify their presence. Of course, if you're enrolling a lot of people in the study who are never going to be in the hospital anyway, you're hardly going to produce many benefits. But you can make the model economically viable, you just can't produce benefits.
RW: Recognizing not all of your outcomes are published yet, what can you tell us about the key outcomes with a particular focus on anything you know about safety or quality?
DM: We organized our evaluation around the triple aim of better care, better health, lower costs. For the better care part of it, we used HCAHPS [Hospital Consumer Assessment of Healthcare Providers and Systems] scores to measure patient experience. It's clear that patients are very happy with this care. When patients enter the study, they're about at the 20th percentile nationally—they're so unhappy that they're willing to leave their current doctors. The control group—where we get them a new doctor, just not one that will see them in clinic and in the hospital—goes to the 80th percentile, and the CCP arm goes to the 95th percentile. We haven't presented these results publicly yet, but we're following things like knowledge, trust, communication, and interpersonal relationships. There are some very promising trends in all of the things. We haven't measured patient safety outcomes in the way you might imagine—like errors and things like that. But just observing the model, a lot of handoffs just don't occur. With mental health, we've seen statistically significant improvements for the intervention group versus the control group. Not so for general health outcomes, which are about the same. But they're not very sensitive measures. We found big decreases in the hospitalization rates and about a 15% reduction out to a year, and there are some hints that the benefits go beyond that.
RW: Given that this was supported and catalyzed by grant funding, once the grant funding goes away, how does the economic ledger play itself out?
DM: This was originally supported by CMMI. They paid the cost of hiring the doctors, research costs, and other things. As that funding started to end, we had to make our case to the hospital. It came down to the following: there was evidence it was producing benefits and the cost was very close to zero—because it's a lean approach to care coordination. Rather than hiring people whose job is to connect the inpatients and outpatient physicians, having one inpatient physician and one outpatient physician, we mix together those inpatient and outpatient jobs and have someone whose job is half inpatient and half outpatient. If anything, we eliminate the need for some of that care coordination effort. In that sense, it's not only free it's even cost-saving. We already had some evidence that length of stay was decreasing, readmissions were decreasing, and hospitalizations were decreasing. Those were all positive things from a health system perspective. At the same time, our health system was thinking about becoming an accountable care organization and recognized they would have a business case around this. We've also been fortunate to get more research funding. But that research funding is not supporting the clinical operational costs. Essentially, the clinical operation costs have been completely supported by the institution, and those costs are no different than they would be if we were doing this in the model where we have an ambulatory physician and an inpatient physician.
RW: Tell us about where the economic tradeoffs are. I can understand from a health system perspective, if you're decreasing hospitalizations or lengths of stay are shorter that may have value, but it sounds like you need to subsidize about five-sixths of the ambulatory physician's time to create a full salary. Where does that tradeoff happen?
DM: This is an important point. If you think about a panel size of 200 versus a panel size of 1200, it's very easy to imagine that that physician is doing fewer RVUs [relative value units]. But the reality is that those 200 patients are very sick. Let's say a typical primary care doctor has a panel size of 2000, and 100 of those patients are super sick. It takes half their time. Another doctor also has a panel of 2000, and 100 of their patients are really sick and take half their time. Instead, you give the 200 sick people to one doctor and that is their panel. So these doctors are not producing a comparable number of RVUs. The only difference is you've reallocated the high-risk patients to one doctor.
RW: Got it. And what kind of doctor wants that job?
DM: Someone who likes taking care of complicated patients, who likes caring for them in the hospital, and who likes the complexity that goes with that—which is partially medical but also psychosocial. Someone who is not afraid to roll up their sleeves and deal with these complicated people and complicated problems. They are also doctors who just like people. A lot of patients who get hospitalized frequently are some of the hardest patients for a lot of physicians to deal with. A lot of information has to be assimilated, and often the experience of being seriously ill and being pushed around the health care system doesn't leave patients with a great feeling about their care. While these patients can be tough for physicians who just care for a piece of them, for the physicians who get to spend time getting to know them, it can become incredibly satisfying. That's not to say there aren't days when it's hard or people with whom the issues are difficult. But I also truly believe there are incredible forms of satisfaction. One of the things we did in starting the program is we had a wonderful clinic coordinator. She would collect comments from the patients when they called in to schedule their clinic appointments. Every week she would get up and repeat back the comments, which named the individual doctors by name. And 20 to 1, they were positive and just glowing gratitude for the care that these doctors had been able to provide. A lot of it is just being there.
RW: What has been the experience in both recruiting and retaining those docs?
DM: We've always been able to recruit, which is great. But it is not a job for everyone. We've kept our eyes open for people who wanted to do this and that has also helped us. Sometimes even when we haven't had a job open, just at that moment we've brought someone in as a hospitalist and then transitioned them into this role when one became available. We've lost a few people over the years. Mostly just for personal reasons—family reasons to move or something like that. Occasionally someone who just felt this wasn't the right patient population for them and they wanted to be doing something different. But not too much of that. By and large, we've had a group of people who've gotten a lot of satisfaction. In terms of scaling the model, I don't think there is any shortage of doctors who got into medicine both wanting to have deep relationships with patients and valuing the complexity that is inherent in hospital care and these complex patients.
RW: In some ways, this is a complement to the hospitalist model and in some ways it's a contrast to it. Has this changed your feeling about the value of the hospitalist model? Assuming that you still believe there is some role for hospitalists, what's the mix of these two models?
DM: If anything, this has only added to my appreciation of why hospitalists are critical. I don't think we could have hospital care the old way. It's just not possible. Hospitalists at some level are just a necessity. On top of that, a ton of great things have come from having hospital medicine arise as it has as a discipline and a profession. All sorts of improvements have come from that, including improvements in handoffs and a greater body of people who have expertise in hospital care. Particularly with changes in academic medical centers and duty hours, it has been important too. I also would say that these doctors are hospitalists. They care for patients in the hospital with at least half their time. Our CCPs care for patients independently without housestaff, which very few academic general internists do. This is a form of hospitalists, and it's a reflection of the evolution of the model rather than a departure from it.
One of the challenges has been how does it handle handoffs? A lot of effort has been put into trying to recognize that handoffs are important and improve them. One of the areas that has been unappreciated is the set of opportunities to decrease handoffs, and this is an example of that. It's sort of a strategy that you can think about where you don't need to do an inpatient–outpatient handoff. But there are other ways to think about diminishing handoffs. If you go from 7 days on and 7 days off to 10 days off and 10 days on, you diminish handoffs. We've done some neat work here looking at things like having people do bridge shifts and admit patients in the evening before they come on and start their week on service so that you have continuity there. It illustrates an important challenge identified by hospital medicine as a field, which is discontinuities and fragmentation. This model offers one valuable strategy to try to diminish some of that. But I don't think of this as a substitute for hospital medicine. It is a form of hospital medicine that has evolved in the context of the evolution of the field.
RW: If a new hospital came to your health system and said, based on what you've come to understand, we want to create the optimal way to care for hospitalized patients, how many CCPs do we need? How many more traditional hospitalists? Should we confine the CCP just to the 5% of patients that are the most complex and utilize the hospitalists? Are there empirical data to help answer those questions?
DM: We don't have a lot of data right now and we need a lot more. I would start with being practical about it, which is that the model doesn't work unless you have adequate inpatient volume. I would try to figure out what strategies you have available to either identify patients at high risk of hospitalization who would switch fully into this model, or to partner with ambulatory practices using some form of the rounder model that I mentioned. I would try to build this up at a sustainable practice scale, then evaluate and see how it did. What's a sustainable practice scale? In my mind, it's probably a bare minimum of three, but more likely four to five CCPs. The nice thing about that number is that they can work as a tightly knit group. They can each do one in four or five weekends, which is pretty sustainable. They can rotate covering the day service on the afternoon. You can lose one of them and take a little while to replace them and not have it be the end of the world.
If I were going to grow our program, I wouldn't create a group of 10, I'd create 2 groups of 5. Because the intimacy in the relationship between this group is extraordinary. One great thing is they all cover for each other's patients. So even when they're gone for the weekend or on vacation or they've just gone to clinic for the afternoon, the patient is being seen by someone who these doctors know very well and the patient gets to know them.
RW: You had a major feature on you and the program in The New York Times Magazine, how did that change your life?
DM: Lots of wonderful emails and response. People interested in trying it. Doctors who this resonated with from their own personal practice. Sometimes payers interested in it, hospitals interested. Just a lot of people curious about how this fits in. It seems to be opening a bunch of doors. We'll see which ones stay open and turn into real things.