Stitching Ties: Physician Networks and Hospital's Surgical Care Performance

Speaker A Talk about today is research that I have going. It's fairly early stage work and it basically tries to get at this question, this broad question of how informal social structure in organizations might shape organizational outcomes. And by informal social structure I just mean things like communication and collaboration ties or information sharing ties basically networks inside of organizations, informal networks in organizations. So there's been a lot of research over the past couple of decades on networks and organizations. This isn't a new area and it's been very valuable, very useful work that's generated a lot of interest and useful insights. But what's a common theme about that work and the approach it's been taken is it tends to focus on individuals networks, individual person's position within a network, in an organization and how that influences the individual's performance or outcomes, right? So things like there's been research on how the strength of a person's connections might influence their ability to acquire new complex knowledge or research on, say, how being a broker and straddling disconnected contacts might affect a person's career advancement, right? Now, this empirical approach of focusing on the individual is somewhat though, in contrast with a lot of really classic theories about organizational design and just organization theory more broadly right. Which tended to focus on broader networks within organizations that connect all the members of an organization, kind of bring people together and in turn have consequences for the organization's performance.

Speaker B Right.

Speaker A Not an individual, but organization's performance. And so some classic theories ask questions like are communication patterns, are some communication patterns better than others for managing environmental uncertainty and turbulence? Or do bad communication patterns contribute to the failure of young firms? Or things like how do communication patterns within an organization shape overall decision making and its effectiveness? And so if I was going to try to visualize the difference between what's been a very common approach to studying networks and organizations and the kind of approach that I'm advocating, I could do it with this figure. So on the left hand side, you have what's been sort of the predominant approach and studies tend to focus on well, how does this person's position within this network or in this network neighborhood, the fact that they have three connections that are configured in some way affect their performance and same with this. And these might be two observations in our data set, right? And what I'm proposing is something that's a little bit more on the right hand side where we have information about this global network of informal network structure within an organization and how the fact that, well, say this is a firm, right? And it looks like there are three sub communities within this firm and there are a couple of boundary spanners that seem to be pretty important, how can we use that information to give us any insights about whether this organization is going to perform well and under what conditions and things like that right. So if we move to this global level of analysis now, one area where I've been looking at this, and I think it's pretty valuable, I think that this global network perspective can lend a lot of value. And I think also the context helps us learn a little bit more about how informal social structure matters for organizations more broadly. So it goes both ways is in surgical care delivery.

Speaker B Right?

Speaker A And so why focus on surgical care delivery? Well, there are a couple of reasons. And one is that there's this vexing problem within health policy. And it's the observation, the persistent observation, that the quality and efficiency of surgical care differs widely among hospitals.

Speaker B Okay?

Speaker A So just to give you a few numbers, mortality rates for vascular surgeries, the common surgery range anywhere from three and a half to about 7%, among the lowest and the highest quality quintile hospitals. That's sort of a known finding. What's probably even more striking is that within a single metropolitan area, it's been found that the average cost of a knee replacement surgery can range anywhere from about 17,000, just shy of $17,000 to nearly $62,000 within the same metro area. And these sorts of differences between quality and cost are persistent even after you adjust for things like patient health, local economic conditions, and a lot of other factors.

Speaker B Right.

Speaker A So there's wide variation among hospitals in their performance on these key dimensions of cost and quality.

Speaker B All right?

Speaker A And these differences matter. They add up to be very significant. There are over 50 million surgical procedures performed annually in the United States. That's about $500 billion spent on surgery every year. And surgery accounts for about 40% of all physician and hospital spending.

Speaker B Right.

Speaker A So cost and quality of health care has been a big part of the national dialogue recently. And so one place that we can focus if we're trying to make a difference is thinking about interventions into improving surgical care delivery.

Speaker B Right?

Speaker A Now, why? A network perspective is the next question. And I think there are also a few reasons here why this kind of approach could be pretty useful. The first is that surgical care is often described as a team sport, or at least in an ideal sense, it should be a team sport, right? And that's something that people strive for. One reason for that is that over the course of a surgical treatment, a patient will typically see multiple different providers of different specialties, and those providers are likely dispersed across different locations, and the patient's seeing them at different points in time, it could be a PCP, a surgeon, a medical specialist. And so a single physician is rarely overseeing all aspects of a patient's surgical care, especially in the US. Healthcare system, where insurance companies aren't really reimbursing for care coordination and other sorts of things. We rely on patients oftentimes to transmit this sort of information.

Speaker B Right.

Speaker A So if we're thinking about cost and quality and surgery as a team sport. It's really essential for having effective communication. It's really essential to have effective communication and collaboration. If we have that, it might do things like eliminate ambiguity around treatment, know, oh, is the PCP going to take care of this or is the surgeon going to take care of that? Who's going to prescribe this medication lapses and care transitions. So the surgeon thinks that some aspect of the recovery is taken care of, whereas the Pct was thinking that had already been done. Conflicting advice in areas where there aren't clear guidelines, one provider might feel one way, the other feels the other way. If they're not communicating, then that can create problems, particularly in the form of things like costly emergency department visits. That's something that's been a big priority for Medicare and Medicaid and intern insurance companies to try to avoid the benchmark of quality things like unnecessary readmissions that could be taken care of in a physician's office as opposed to in a hospital, and just to improve the overall quality of care. So within this context, we've been working on studying physician networks within hospitals and trying to map this global structure of how do different providers communicate with one another and what can we learn about how the structure of those networks might influence outcomes. Now, one reason why I think there hasn't been quite as much research on this topic until now is that it's been hard to get this kind of data right, because in order to study variation in these global networks, we need massive networks across lots of different organizations, potentially across different periods of time. So we have enough variation as opposed to before, to study individuals networks, we might just need to study one or two organizations to get enough variation. But now with all this research and all this interest in big data and administrative records becoming more and more available things on websites and so forth, we have lots and lots of data that let us get very rich pictures of this kind of thing that's happening inside organizations. So what we've been doing to map these networks is getting claims data, national claims data from Medicare through the center for Medicare and Medicaid Services, which is the organization that runs Medicare and Medicaid. And what we have are four years of data. So from 2008 to 2011, for four common major surgical procedures that are common among elderly patients, we have Medicare data. We're focusing on that. Those are collectomy prostatectomy, hip replacement, and then coronary artery bypass grafting, basically bypass surgery it's called also known as cabbage. And that's what I'm going to focus on today. And we have lots and lots of data too, which lets us get at this question. If you can imagine when a patient goes in and they're seeing all these providers, it generates a lot of claims, right? So that's what lets us get this rich picture. We have about 4000 hospitals just shy of 800,000 patients, about 700,000 physicians, about 15 million physician and patient encounters that we can use to make some kind of inferences and sort of describe what's happening in this setting. So I haven't been doing this by myself. This has been really a big team project because there's a lot of different aspects. And I've relied really heavily for understanding the medical context and the aspects of care delivery by leaning on physicians who do this day in and day out. So I have collaborators mostly at the University of Michigan, but also some people at other universities too, that have been able to contribute in a lot of ways to this and give lots of insight to many aspects of the project. So what I want to describe now is how we move from claims that are not created necessarily for research purposes, right. They're for billing and other financial records to network data.

Speaker B Right.

Speaker A And one way that I can help you try to see that is by describing a typical or somewhat simplified pathway through which a patient might go from feeling sick to getting coronary artery bypass grafting. So imagine this guy is an 80 year old patient and wakes up one morning not feeling too good, has some chest pain. He goes and contacts his primary care provider and goes in for an office visit. And she says, Run some tests and ask him a few questions and says, I think we need to look into this further. You might need surgery. And give him a referral to the guy, referral to a surgeon. The surgeon does some more tests and says, yes, you're a candidate for cabbage. Let's schedule a time in the operating room, get this done. So the patient goes into the hospital. While he's at the hospital, maybe for some other complications. He sees an endocrinologist, sees a psychiatrist because maybe there's some concern about depression. He gets out of the hospital, he's doing well, has a follow up with a PCP, follow up with a surgeon, maybe a few weeks later, a couple months later, checks in with the PCP again.

Speaker B Right?

Speaker A So every time that happens, a claim gets registered with Medicare, and from those claims, we get lots of information about who the patient is, who the physician is, what the nature of the visit was, other demographic information. And we also know about frequencies of interaction. So the patient saw the PCP three times, other providers just once. We can then go from these patient to physician interactions to start to look at interactions among physicians around shared patients. So we know that this is the patient's care team, and since they were all working on the same patient, they have some sort of relationship with one another.

Speaker B Right.

Speaker A There's also been work on this that's validated this approach and found that the more patients that paraphysicians share in common, as evidenced by Medicare claims, is actually a very strong predictor of a self reported tie. So it's very good data for getting at these sort of networks. Now, to map the hospitals and get the kind of data that we want to analyze, we repeat this procedure for all the patients within a hospital over the course of a year and then for each hospital in our data set. So there's 4000 hospitals. That's how we get these observations of this global network structure over time within a hospital. Now, one of the things we were most interested in when we first started this project was just getting some descriptive data on how much variation do you see in these networks and on what dimensions do they vary and what can we learn about health care delivery, surgical care delivery, by looking at these. So here's one example of the kind of wide variation that we see and how you can start to think about how these networks might be influential for organizational performance, right, and the ability to deliver high quality, low cost care. So what you have on the left is a hospital. These are both very similar hospitals. They're in a similar region of the US. They have sort of a similar number of patients, have a similar number of physicians. So they're sort of comparable. But you see, the structure of the network is very different. On the left hand side, we have a hospital that we say has high cohesiveness, right? So we use some network metrics that I could discuss, if you're interested in, to kind of quantify the extent to which physicians cluster around shared patients. They have many shared patients in common with one another, and the people that they share patients with also share patients with one another. So like a clustering measure, right? And you see this network has this dense core which is indicated by these red ties. So you can get the sense that there's a lot of information sharing. People know one another within this hospital. And so it seems less likely that information is going to fall through the cracks. If a patient has a problem, it's likely someone's going to be on call and they know who their Pct or who their surgeon is and can share information. That's not what we see in the hospital. On the right here, the network is much more sparse, all right? People are a little bit more isolated from one another. You see far fewer of these clothes or cohesive sorts of ties, which suggests that information sharing coordination might be a little more challenging. We've also looked at this in another way. We've tried to look at the extent to which there's information sharing or shared ties across different types of specialties among physicians that are involved with providing surgical care. So we have primary care physicians of the gold color, surgical specialists of the red color, medical specialists of the green color. And the idea is, again, going back to what I said earlier about these surgery being a team sport that requires input from surgeons, PCPs, medical specialists. The idea is if the PCPs within a hospital have more familiarity with a broader race of surgical specialists and surgical specialists have a broader familiarity with the PCPs and the medical specialists, it again creates more opportunity for information sharing. And in fact, this has been a big goal of healthcare reform, is to break some of these barriers and get people coordinating more as teams and not as silos of medical specialties. So unfortunately, it's not coming up quite as clearly as I hoped. On the screen, the lighting is a little weird, but according to our measure, what I'm trying to convey here is that this hospital is more integrated because you see more wider distribution of these blue ties, which are ties across specialties, right? Whereas in the hospital on the right, which is again comparable in a lot of dimensions, many of the cross specialty ties come from just a couple of different primary care providers and surgeons that refer heavily among one another, right? So you might expect that people concentrate their relationships a little bit more deeply within a particular set of providers and aren't quite as familiar with some of the other specialists within the hospital. Again suggesting some interesting things potentially about coordination. So the next thing we were interested in is sort of after spending a lot of time staring at these balls and stick diagrams and trying to figure out what was going on, is if we could see any connection to outcomes.

Speaker B Right?

Speaker A Really at this point in the project, as I mentioned this early stage, we're just trying to test for associations, so to get the best associations we can to see if there's anything interesting happening. So we started creating models and obviously there are a lot of other factors that likely get in the way and make this relationship sort of hard to observe, right? And so we've tried to control for a bunch of alternative possibilities, things like one of the things we spend a lot of time figuring out is how sick are the patients?

Speaker B Right?

Speaker A Because you could expect that would easily affect the network structure, could also affect the outcomes for the hospital. So we've included a lot of controls for that. We've included a lot of controls for the socioeconomic status of the patients and of the market in which the hospital is situated. Also other measures about the healthcare capacity and the healthcare market that the patients are trying to that the patients would be a part of. How many patients, how many physicians? What's the volume of activity being done at this hospital? We've also tried to account to some degree statistically for some of this, some of these possible sort of alternative things that could be affecting this relationship. What we've been interested in are three measures of quality and one measure of cost. So we've been exploring 60 day readmission. So after surgery, is the patient readmitted within 60 days. That's something that's interesting. Again, from a billing perspective, it's something insurance companies care a lot about. Ed visits, unplanned emergency department visits are much more expensive than if a patient goes to see a Pct. Mortality is obviously something we care a lot about. And then the total episode cost and what we find. So here's readmissions we find a similar pattern. Here's readmissions, here's adjusted EV visits and then our adjusted measure of mortality. And I just want to focus on mortality because the relationship is pretty similar. But one of the things we find is that a 10% increase in the cohesiveness of these networks above the mean is associated with something like a four and a quarter percent decrease in the adjusted mortality, right? And then you can even see differences across the high and low hospitals with high and low levels of cohesiveness or sometimes we've called it we're presenting this for a medical audience at teamwork levels. So the difference between the highest and lowest hospitals on teamwork is something like 28 and a half percent difference in mortality.

Speaker B Right?

Speaker A We also find some pretty interesting relationships with cost. Again, this is looking at that measure of integration. So do specialists coordinate well across different do they have good communication ties across different specialties PCTs medical specialists and so forth? And we find that from the hospitals with the lowest levels of integration to the highest levels, it's a difference of about $3,000 per procedure per cabbage, which, when the cost about fifty nine thousand dollars fifty seven to fifty nine thousand dollars might not seem like a lot. But one thing to remember is that each year there's something like 200,000 costs of these performed, right? So if you could save $3,000 on each one of those, then also keeping in mind that this is just one of dozens and dozens of very common surgical procedures, you could potentially talking about billions of dollars of savings for the healthcare system. This is something that could be implemented. So just as a way of kind of wrapping up, the things that we've found so far are that there is a lot of variability in the structure of these informal networks within hospitals and health systems around different surgical care episodes. We find that hospital systems where physicians have very cohesive ties that are conducive to information sharing and coordination have lower readmission rates, Ed visits and mortality and then cross specialty integration seems to have an association with lower cost care. That leaves a lot of kind of open questions. And one of the things that we're trying to figure out now is sort of like what are the mechanisms and how do we make sense of this relationship? I think one thing that needs a lot more exploration is trying to understand, well, what's the role of formal structure, right? So how are formal dimensions of the organization? Things like electronic health records or IP systems or other things like that influencing these networks? And how can they maybe complement or serve as substitutes for this kind of coordination? And then there's the big open question that something, again, we're trying to move forward on understanding. Also combining this with some qualitative work is figuring out the dynamics.

Speaker B Right.

Speaker A How can you change one of these networks if this could lead to something that would have substantial cost savings? How easy or how hard is it to get providers to coordinate across specialty lines? I'll go ahead and just stop there. Thank you.

Speaker C So what you're shown here actually, is that if there's more information processing across the transition points right. Then both the mortality goes down and the cost goes down. Is that the correct interpretation of your presentation?

Speaker A Yeah, the data seems to be suggesting that's what's happening. Yeah, absolutely. And that's consistent with other research that's been done more at a team based level.

Speaker B Right.

Speaker A And there's been some work that's been following particular patients and videotaping and seeing what happens as providers, even within a single unit, are passing on patients. A shift change, for example. And the quality of information transfer at those points in time can vary a whole lot and then can have significant implications for continuity of care, even when the patient's within the hospital.

Speaker B Right.

Speaker A Figuring out how information sharing can be more efficient is something that's really of interest to potentially having an impact on quality, cost, things like that.

Speaker C And what your research? Will that sense be generalized outside the hospital system? Because these formal structures in the hospital with emergency department actor created matrix structures. And if you find that there's both a cost reduction and performance improvement in other ways, that might be some resources could be taken to other types of matrix structures.

Speaker A So is it generalizable beyond the other?

Speaker C Yeah.

Speaker A What I will say is that we have seen consistent patterns across the different types of surgeries that we've been looking at, and they vary in terms of things like the complexity of the surgery. Some are much more routine, you're in the hospital for one day, overnight effects, and others are more complex. So I know that it generalizes, but beyond cabbage. But I think it's an open question. Certainly there's an assumption within the health policy literature, and there is some research as well too, that says fragmentation is a huge problem in the healthcare system. And that's just not within, but that's also across. So thinking about the hospital to the physician practice, to ambulatory surgery centers, and so the idea is to reduce that fragmentation should improve outcomes. And so I think that this could be viewed as some evidence in supporting that idea. I think it sort of complements that, but I can't really directly tell.

Speaker D Yes, I agree. Very interesting discussion. You have used the claims data that are sort of binary to build up the network structure. And then you've interpreted that structure as the informal structure but the claims data actually don't differentiate between informal, formal and so on. And it all kind of confounded in terms of you have a real structure. I understand, but I don't know how you can characterize it as an informal structure.

Speaker B Well.

Speaker A I've been thinking about this more as informal structure because it's something that is influenced. So physicians make referrals, right. But patients are the ones that need to actually implement those, and a physician might make a referral that's never followed up on. And so it's informal in the sense that I guess why I was using the term informal is that it's getting at, I think, patterns of communication around these shared patients. And we know that what's reflected in those ties, what's reflected in the claim, seems to also be reflected in informal patterns of communication. We know that hospitals vary in the ways that they're organized. They tend to be organized for the most part around specialties and primary care is separate from specialty care and so forth.

Speaker B Right.

Speaker A That's not the case everywhere, and there's a movement away from that by introducing things like patient centered medical homes where you're getting settings that are organized more around patient needs as opposed to specialties and things like that. But I still think the predominant way of organizing this is more of separation. And so I tend to think of these ties as things that are spanning some of those boundaries. And that's why I've been calling informal connections.

Speaker B Good.

Speaker E Just a quick question, and I'll preface this by saying I know nothing about healthcare, and I try and stay away from hospitals whenever I can. But I remember asking somebody once, what's so special about the Mayo Clinic? I know you're in Minnesota, so you might know, and somebody told me something about team based medicine. So I'm wondering if, when you created your structures, did it correlate with any self identified philosophies or missions of those different hospitals? I know you said these hospitals are similar. Maybe they're similar size, similar budget, but did they have different philosophies? And that explained why some of them had the different cohesiveness or the different integration.

Speaker A Yeah, well, so one of the things that we were trying to do to handle that is use mostly fixed effect models. So we're trying to make comparisons within hospitals. And so as long as the Mayo Clinic doesn't change its philosophy within this period and that's kind of a fixed thing, then that's the way we can it's not perfect, but that's the way we're trying to deal with it. I think places like the Mayo Clinic are more the exception than the rule, and I know that one of the things that makes them unique is the way that they pay physicians.

Speaker B Right.

Speaker A The physicians there are salary based they're not using fee for service. And I think the idea is that that allows them to spend more time with patients right. And not be they have to focus less on the volume of care that they're doing and can focus more on the quality and giving people the attention sort of things that they need, which lets collaboration happen a little more easily. But I think that even like replicating the male clinic model by male clinic in other settings has taken a long time. It's been very hard. And so although other places are trying to do that, I think it's not quite.

Speaker F Healthcare sees that the team model leads to successful outcomes or better outcomes. So I'm curious about when does that not apply or what are the limits to that? So from your study, do you have a sense of what are the detrimental or negative aspects of this? Because I want to be able to generalize this now to other studies. So is it the speed of which the outcome, as we make it more cohesive these networks? Is the surgery delayed? And what's the trade off between having these team work versus I haven't heard anything about the downside. Is there no downside?

Speaker A No, I think there definitely is a downside. This is one of the areas where I'm starting to move now, is to try to figure this out and to think about this more deeply. Two responses. One is just to give an example of what I think is a downside or risk with this and that has to do with some of the cross specialty sorts of collaboration, right? So coming from more of a sociological background and a huge field that I'm sure a lot of you are familiar with sociology of professions and there's a lot of work about how there's jurisdictional competition among professionals, people don't get along and it just makes this type of coordination collaboration very hard.

Speaker B Right?

Speaker A And so I think that there's a real risk that if this sort of thing is just implemented in a place where there are those very strong interprofessional rivalries and conflicts that it could just end up making things worse. Right, because people maybe don't want to work together and they feel like their specialty is superior and then now they're being told that they need to work together in more of a teamwork setting and it's just going to be a challenge. And so thinking about how can you change the institution or other things like that is something that I think needs a lot of consideration. That's just, for example, one of the ways that I've been thinking about it, but I certainly don't think that it's something that would always work. And then there's also the fact that within health policy they talk about this quadruple aim which is sort of four goals that a lot of people think need to be accomplished in order to improve the health care system. And one of those is care for care for patients, reducing costs. But another big one of those is care for providers, right? Teamwork takes work, right? And if providers are already. Just really strapped as far as time, and they're just struggling to keep up. And we have a shortage of physicians. And then asking them to maintain all these other relationships as well could also be something that could be problematic.

Speaker B Thank you.