Wharton's Hummy Song discusses her research on hospital capacity pooling and its impact on patients.

Hospitals can seem like confusing, chaotic places for patients, even when the employees are following well-established protocols and resource management techniques. But there is always room for improvement. That’s the goal behind the research of Hummy Song, Wharton professor of operations, information and decisions, who focuses on improving the performance of health care delivery systems. Her latest study looks at capacity pooling, which solves the problem of hospital bed shortages by admitting patients to wards that belong to different services where there are empty beds. This practice, known as off-service placement, requires an extra level of coordination to ensure the patient receives the proper care. But does it always work? Is there a better, more efficient way? Song sat down with Knowledge at Wharton to talk about the findings in her paper, “Capacity Pooling in Hospitals: The Hidden Consequences of Off-Service Placement.” Her co-authors on the study are Anita Tucker of Boston University, and Ryan Graue, Sarah Moravick and Julius Yang from the Beth Israel Deaconess Medical Center. The paper was recently published in the journal Management Science.

An edited transcript of the conversation follows.

Knowledge at Wharton: This paper focuses on something called off-service placement of hospital patients. Can you explain what that is and how it fits into the way that hospitals categorize patients for care?

Hummy Song: In this paper, we focus on the impact of off-service placement on hospitals’ patient outcomes. Just to start with exactly what I mean by off-service placement: Imagine you’re a patient. Maybe you have a cardiac condition and need to get admitted to the cardiology service. But let’s say the beds that have been assigned to the cardiology service are completely full. Or maybe there are a couple left, but they really want to save them for someone who might be much sicker than you.

The bed managers, who typically are nurses whose job it is to assign patients into beds, will look around the hospital to see where there is an available bed for this patient, even if it’s not a bed in the cardiology ward. I’ll use “ward” and “service” separately. Service is the specialty, like cardiology; the ward is the physical place where those beds are located.

If the cardiology ward is full or there aren’t a lot of beds left, this bed manager might decide to place you, the cardiology patient, in a bed that belongs to another service. Let’s say it’s a bed that belongs to the orthopedics specialty. This is what we call off-service placement. The cardiology clinical team is still responsible for your care, and they’re the ones who are going to be coming and rounding on you. However, it’s going to be the orthopedic nurses in that ward who are going to be delivering the day-to-day care at your bedside.

Knowledge at Wharton: For the hospital, what are the benefits or drawbacks of off-service placement?

Song: The big benefit is that it allows hospitals to address this mismatch problem between the supply of hospital beds and the demand of patients. Ideally, we would have an exact match between the two at all times. Oftentimes, this doesn’t happen. The number of beds you have in the hospital are typically fixed, whereas the number of patients, especially if you think across different specialties, is always variable.

It’s a way to do what we call capacity pooling. You’re pooling the capacity of beds across all these different wards. This really helps you utilize your beds much more effectively by making sure you’re not keeping a bunch of beds open and unused when you might have patients who could utilize them. From the patient’s perspective, if your alternative is to wait for hours and hours for an on-service bed, your wait times would be much reduced if you just get placed off-service.

In our paper, we’re taking at face value this potential benefit of doing so. What we’re trying to dig into instead is understanding some of the potential drawbacks. What are some of the unintended consequences of placing patients off-service? Because at the moment, we and hospitals tend to think just about the positive implications of the capacity-pooling strategy, not so much what the potential drawbacks might be.

“The bed manager is working under the assumption that placing a patient on service is going to be better than having them off service.”

Knowledge at Wharton: How did you go about studying this?

Song: We were really fortunate to have a partner hospital with which we were able to talk through the problem, collect data, etc. They kept incredibly detailed records on exactly which beds are occupied, open, reserved, or closed. This really helped us be able to causally identify this effect of off-service placement. Essentially, it’s a totally data-driven project. We’re looking at not only things like what the patient had done to them during their stay, but also let’s take a look at the hospital’s capacity. Let’s take the perspective of what is happening to each bed in the hospital and try to isolate this effect of off-service placement.

To give you the big picture, the main things we asked in terms of these potential drawbacks were: What’s the impact of off-service placement on the patient’s length of stay? Obviously, it matters for the patient, but it also matters for the hospital in terms of how long your beds are going to be occupied by each patient. Second, what’s the impact on quality of care? Things like 30-day hospital readmission, likelihood of dying in the hospital during your stay. We looked at some other things as secondary quality measures, but those were the main ones. And third, we focused on what are some of the drivers: What are the mechanisms that could explain why we’re finding an effect on length of stay, mortality, readmission, etc.?

Knowledge at Wharton: This is a little different from previous research on this topic, which looked at patients who should have been in a more serious level of service but were put in a less serious level because of capacity issues. In this case, the patients are being cared for by the team that treats their condition, and they’re just being housed in a different department.

Song: Exactly. This is what we’ll call medical-surgical patients. We’re distinguishing them from ICU patients. Most of us know that if you end up in an ICU, that means you really need a high level of care. There has been prior work asking that question that you just mentioned of what happens if a patient who really needs ICU-level care ends up in a regular medical-surgical ward, or vice versa? In this paper, we’re going to set aside all those ICU patients. We’re focusing on the rest of the hospital, so what you think of as your standard, regular, in-patient ward. That’s what we call just regular medical-surgical patients.

Knowledge at Wharton: I assume even within that population of patients, there’s still a big variation in their conditions. How did you control for that?

Song: That gets at a really important identification issue that we had to address, and that’s one of the key contributions of our paper. Just to take a step back, it’s not like hospitals haven’t thought about this question before. They might wonder, “We’re assigning all these patients, we’re putting all these patients off-service. What might happen to them?”

Typically, the most straightforward way that this question has been asked and addressed is by comparing the means of their outcomes. Let’s say I’m going to compare the average length of stay of all the patients who are on-service to the average length of stay of the patients who are off-service. If I find that there’s not much of a difference, it looks like it’s OK. You can imagine doing this for all the other outcomes. Even when we adjust for the observable patient characteristics — like differences in severity of condition, gender, demographics, age, co-morbidities, all these things — you find that actually there’s not much of a difference.

That might make you think off-service placement is no big deal, no problem. But what we really need to pay attention to is that those bed managers that I mentioned to you earlier have a lot more insight into differences in the severity of these patients that they’re assigning than we can gather from just the recorded observable data. What we really need to do, and what we were able to do in our paper, was essentially address those endogeneity issues.

The bed manager is working under the assumption that, if you can, placing a patient on-service is going to be better than having them off-service. So, what do they do? They look at the patients that they have to assign, and they select the healthiest patients, comparatively speaking, and place those patients off-service. That means if there are going to be any negative effects of being placed off-service, that’s going to be somewhat hidden just because these patients are healthier to begin with. So, it’s perhaps not a fair comparison to just compare their means, even if you’re adjusting for all the observable covariates.

“We found that being placed off service is associated with a 23% increase in your length of stay. That means you’re staying at the hospital about an extra day.”

To address this issue, we use an approach called instrumental variables, where we’re trying to leverage some other exogenous source of variability to try to make this comparison as equal as possible. What our incredible data allows us to do is try to see: what’s the variation in the utilization levels or the availability of beds in the service the patient is trying to get admitted to at the point when this bed manager is making those decisions? And how does the patient’s likelihood of being placed on or off service depend on that? Because it has nothing to do with the patient or their condition, we can rely on the current utilization level of the hospital or that service as an exogenous source of variability that is going to impact how likely this patient is going to end up on versus off service. Then, we’ll utilize just that source of variation to identify the causal impact of off-service placement.

Knowledge at Wharton: What were some of the key takeaways that you found when you studied this data?

Song: Three big things. I’ll say even something that we didn’t explicitly hypothesize: Just being able to identify how often this happens, how many patients are being placed off service, was of interest to the hospitals and other people we’ve talked to. In our study hospital, we found that about one out of every five patients is being placed off service, so about 20%. Since working on this paper, I’ve talked to several other hospitals around the world, and I’ve heard figures up to 40%. Just think what that means in terms of how many patients, how many people this is impacting. That’s one big thing that came out of this paper.

Secondly, to get at our patient outcomes in terms of length of stay, we found that being placed off service is associated with a 23% increase in your length of stay. That means you’re staying at the hospital, on average in our data, about an extra day, even if your clinical condition doesn’t really call for it. That was a huge, huge number. I’ll get back to that in a second.

In terms of readmissions, mortality, those quality measures, we found a 13% increase in your likelihood of being readmitted to the hospital. Not good. But fortunately, we did not find any impact on your mortality in the hospital, so that’s a good thing.

One thing that was really surprising was this 23% percent longer length of stay. As researchers, when we first saw that number, I thought, “This must be wrong. There’s no way the location of your bed makes such a big difference in your length of stay.” We had a bunch of conversations with our clinical collaborators, and they did not find it very unbelievable. Once they thought about it, they were like, “This makes total sense.” I asked them, tell me why this is happening and why it makes sense to you? Essentially, it came down to how the physicians and nurses coordinate care in the hospital.

What this hospital has is something called patient progression rounds. These are rounds that are led by the nursing team. If you recall from our example, the cardiologists are still responsible for rounding on the cardiology patients. The orthopedic nurses on that orthopedics ward are the ones who are actually taking care of that patient. If you need any labs drawn or medications administered, the orthopedic nurses are the ones doing that.

Knowledge at Wharton: They’re carrying out what the cardiologist managing your case is asking them to do, correct?

Song: Exactly. One thing to keep in mind is that when these patient progression rounds happen, what they’re doing is discussing, “For today, what do each of the patients on our ward need?” They go around discussing each of their patients. When the next patient is up for discussion, they’ll pull in the physician who is responsible for their care so they can have a team discussion.

Obviously, this is going to be much more difficult to do for the patients who are off service because those physicians are not co-located on that floor. They might be in a totally different part of the hospital. When the physician comes to round on that patient, they pop in for a couple of minutes or however long it takes, but then they go back to their home base. So, the discussion for these off-service patients usually gets deferred until the end of the patient progression rounds, after they’ve had a chance to discuss all the patients who are physically there and whose physician is also physically there. These providers have so many things going on simultaneously, those conversations about the off-service patients might get delayed quite a bit, into the afternoon or later in the day. By that time, even if that patient was meant to be discharged that day, if you haven’t had that conversation early enough so that you can coordinate all the logistics of getting the patient out of there, that’s going to get bumped until the next day.

That’s typically what happens, and that’s why our clinical collaborators thought it makes total sense that if there’s going to be a delay, it’s not going to be just a couple of hours. It’s going to be on average about a whole day’s extra length of stay.

Knowledge at Wharton: It’s more of a logistical issue or even a team dynamics issue than a medical issue. I also wondered whether the orthopedic nurses, for example, know the orthopedic doctors really well, but maybe not so much the cardiologists. They might not have that shorthand that they would have with the people they work with every day.

Song: Exactly. That familiarity and the coordination abilities really seem to matter. That makes me think about another part of our paper that I also wanted to talk about. What’s driving these effects? Many people thought if you’re going to place patients off-service, you should try to place them in a clinical specialty that’s as closely related as possible. I’ll call that an alignment in your clinical needs, or a match between the specialty of your home service and the ward you’ve been placed on. The assumption is that that matters a lot more than the physical distance between the ward you ended up getting placed in and the ward you ideally should have been placed in. Let’s call that your home ward.

We tested that. We wanted to understand which one seems to be really driving these effects. To our surprise, and a lot of our clinical collaborators’ surprise, we find that it’s really the physical distance that is driving this effect, not so much the proximity of the clinical specialization. It matters less that you’re placing, let’s say, a patient of a medical subspecialty into a ward that also belongs to a medical subspecialty. The alternative would be placing a medical subspecialty patient into a surgical subspecialty ward. That doesn’t seem to matter at all. According to our results, what really seems to matter is the physical distance between the two units.

If you think back to the patient progression rounds and how that all works out, I think it makes a lot of sense in hindsight. Because presumably, it’s a lot easier for you to ask the physician from the home service to pop up if you’re just a floor away or down the hall from one another, as opposed to if you’re in a totally different building than where your patient is located. So, it seems to be the physical distance that’s really driving the effect, less so the closeness and the clinical specialization between the two.

Knowledge at Wharton: What does this mean for hospitals? Should they stop doing this completely? What else can they do?

Song: We’ve made a couple of key recommendations to the hospital we’ve been working with. One, even if it’s not possible to completely get rid of it, they should really try to minimize how often they’re using off-service placement. That means, first of all, get rid of the practice where you’re proactively reserving beds with the expectation that you might have a future arrival who is sicker than the current patient. You’re taking a gamble. Chances are the time savings is not quite going to be worth the extra bed capacity that you’re ending up using because this off-service patient is going to stay longer. So, try to minimize the use of off-service placement as much as possible.

“Try to minimize the use of off-service placement as much as possible.”

Two, if you have to use it, prioritize placing patients in a bed that’s going to be as close as possible distance-wise to where they would have ideally ended up. Clinical match between the two doesn’t matter as much.

Third, and I think this is especially relevant for a lot of hospitals that are reorganizing or building new buildings and opening new wards, try to a priori allocate your number of beds to match the capacity needs of the services as much as possible. Historically, there’s been a tendency to try to have more extra beds for surgical services relative to medical services, just because surgical patients tend to bring in a lot more revenue than your medical patients. That’s something that we would recommend they not do as much, and really try to optimize the allocation from the get-go so that you have less of a need to engage in off-service placement anyway.

Knowledge at Wharton: Should they also look at proximity when reorganizing or building new wards?

Song: Should you have a couple of wards that are explicitly flexible wards, where you can take patients from a couple of different specialties and they would all be co-located, and also done in a way that it’s close to the main non-flexible wards for that specialty, so that you a priori determine this set of beds to be flexible beds? That’s one possibility.

Another thing that we’ve been thinking about working on is, how can you leverage predictive analytics to better predict when the next bed is going to become available on the ward that you want this patient to go into? If I can do a better job predicting when exactly the currently admitted patients are going to get discharged, then that means I’m going to know when the next bed is going to open up. I can have a better sense of, “If I hold on to this incoming patient and not place them in an off-service bed, how long am I going to have to wait to place them in an on-service bed?” A lot of the decisions that are being made right now without that kind of information, they’re trying to use their best guess, but they obviously aren’t able to go back and look at all the historical data that might have been recorded about all the admitted patients. There’s simply way too many beds to be thinking about.

Knowledge at Wharton: I also wondered if there’s a patient-communication part of this, too. As a patient, you hate to wait. But would the experience be better if it were explained to you, “Well, if you stay here for another hour, then we can get you on service as opposed to off service.”

Song: I hadn’t really thought about that, but I think that’s a really good point. It goes back to the fact that as patients, we typically don’t notice that we’re in an off-service bed when that happens. Maybe if you’ve been made aware of these kinds of things, you’ll start noticing it more. But I bet you most people who have recently been admitted to the hospital wouldn’t be able to tell you whether they were in an on- or off-service bed. And, as we can see from our data, this happens quite a bit. Thinking about communicating this to the patient upfront so that they can better manage their expectations around waiting — I think that’s actually a really good idea.

Knowledge at Wharton: What’s next for this research?

Song: A couple of things that we already talked about, like the predictive analytics. We’re thinking about how to do the bed allocation across services to be a better matched number from the get-go. Those are a couple of things. But more immediately, what I’m working on right now is thinking about, what is the second-order effect? The effect of off-service placement on those patients that have been placed off-service, you can think of that as a first-order effect that we’ve identified. But in our conversations with the clinical teams, what’s become apparent to me is that there might be a spillover effect onto the patients who might be placed on service but might be in a ward with a lot of off-service patients.

If I am another cardiology patient placed on service, but so many of the cardiology patients are placed off service that my physician keeps having to run around the hospital chasing after all those other patients, you can imagine that could have an implication for the care I get, even if I’ve been placed in the right kind of bed. So, we’re trying to find clean ways to identify that spillover effect, both for patients who might be in the same service as those off-service patients — that’s an immediate next step that I’m interested in looking into further — and then trying to think about different admission policies you can design to make the situation better overall.