For many patients – especially those facing spinal surgery – the specter of an unknown outcome can be frightening. Knowing their odds for a successful procedure versus a failed operation can help many individuals decide whether and what type of surgery they’d like to undergo.
And, having the data to determine those success percentages can also help surgeons and other providers make the right decisions when it comes to expenditures and recouping reimbursement.
Using this data is called predictive modeling. To understand how its affected spinal surgeries and payments associated with it, Rheumatology Network spoke with Joseph Osorio, M.D., Ph.D., a resident with the Department of Neurological Surgery at the University of California at San Francisco, about his study, "Predictive Modeling of Complications," published recently in Current Reviews in Musculoskeletal Medicine.
Rheumatology Network: Why did you decide to study predictive modeling? What is the importance of this tactic?
Dr. Osorio: The importance really lies with the patients we operate on most frequently. It’s complex spine surgery. Patients are prone to having complications because their surgeries involve many levels, and there’s a high risk for complications. Patient satisfaction is something that we want to do well, and we wanted a tool where we could use all the data at our disposal in our practice in the clinic to give them a number they could understand. For example, they’ll have an 85 percent chance of success rather than telling them patients like them generally do well with this operation.
One of the benefits we have is the professional societies are collecting data the way we’ve been doing here at UCSF and at other centers that focus on adult spinal deformities. We have a repository of data and collect large volumes, prospectively, on patients that fall within the adult degenerative and scoliosis parameters set out in the literature. We put that information into a database and go back, retrospectively, and analyze it. There’s a huge wealth of information there to benefit us. Part of the nuts and bolts of predictive modeling is having quality data and having large volumes of it. If you don’t have quality data, then it is a limitation to the predictive model. If you do, you can apply these techniques to other conditions and settings, and achieve the same success that we have.
Rheumatology Network: How impactful has the modeling been on spinal surgery?
Dr. Osorio: I think it’s been extremely useful with aging individuals and those undergoing revision surgery. Some of them want to know because they have many options for surgery. Do they go with the larger surgery that may address the bigger problem of global spinal balance, or do they choose a less invasive one that might have a likelihood for needing another spine surgery because the bigger problem was not addressed? Are they someone who doesn’t want to take the risk of undergoing such a big operation? With this data, we can relay possible outcomes to the patient.
Rheumatology Network: What were your main findings? Why are they important?
Dr. Osorio: Some of the main findings started with one of our initial pieces of work. The first one looked at surgical complications in the adult spinal deformity patients. Those were patients we did large segment fusions on, and they returned with adjacent level disease. They come back with new pain or new neurological indications, and they need another operation. In some sense, they’ve failed the surgery and are in need of a revision operation. We were able to take 500 or so patients and look at those we had two-year follow-up data on. We look at X-rays and clinical information and we’re able to say which of all those patients — if we throw in a series of variables — are going to come back with a failed surgery. We added in variables that were strong predictors — some were based on radiographic parameters that are commonly used in the clinic to assess adult spinal deformity patients. From previous studies, we predicted that these parameters, including being over age 64, being sagittally imbalanced, and having flat back syndrome where patients develop much of their back pain by overuse of compensatory muscles, are important ones to monitor.
Rheumatology Network: Did you find anything unexpected?
Dr. Osorio: It wasn’t unexpected, but it is something that gets discussed in the literature a great deal. It’s where do we stop — at what level do we end? Do we go to the pelvis or the sacrum? Does it make a difference or not? Surgeons want to know. It seems that it does make a difference, but we need to do more studies to get a better sense of it all. At our institution, we commonly end our large spine surgeries at the pelvis. That’s something that seems to be coming out in the predictive model.
Rheumatology Network: Are there any challenges to spinal surgeries that were revealed?
Dr. Osorio: I think the biggest challenge with studies like this is that when we see patients in our clinic, we really do have an algorithm for the way we analyze each patient. Everyone receives a standing 36-inch long X-ray. We’ve essentially designed our clinic that way – we don’t see patients until they get that X-ray, so that might be a challenge for other providers who are not using this film as one of their tools in understanding a patient’s global alignment. If you start reading the literature now, you get the sense that all of us are using 36-inch X-rays to identify parameters that are important for choosing whom to operate on. In many cases, when we receive referrals from the community to see patients who have had a failed surgery, there often isn’t a record of having had a 36-inch X-ray acquired. Providers often obtain a CT scan, or an MRI of the lower back, but overlook the global alignment. We’re looking at global alignments, but the challenge in moving forward is that providers aren’t looking at the problem in a global way. We’re trying to move the field in that direction — to get providers to understand that patients often have problems that could impact more than one region. They need to think about the patient and the spine as a whole.
Rheumatology Network: How does this work fit in with MACRA?
Dr. Osorio: I would say as we move forward with this wealth of information, we’re moving toward having an abundance of information from our use of electronic heath records. Having access to this data, we’re going to have a better understanding as to what parameters most impact surgical outcomes. We can better justify an expected outcome. Are providers, when they’re offering surgeries, looking at other providers’ outcomes to get a sense as to how the field is adapting/improving? Are they actually at the mean or are they below average? Are more of the patients they’re providing surgeries to, patients they would’ve know might have failed if they’d looked at a set of high risk targets? We can provide them with these targets. We can educate providers that are going to be doing do this level of spine surgery, to pay close attention to these targets. Counsel them that it will impact their success. I think overall, MACRA is going to really try to differentiate providers that are given that kind of data and incentivize them to provide high value care. Those that are below the mean are going to suffer. This will provide them with a tool and give them a better understanding of what success is in this field.
Rheumatology Network: How can this work be used to promote better care and outcomes?
Dr. Osorio: One of the challenges we fall into, is justifying and providing large scale operations at tertiary and quarternary care centers. Operations are expensive. Smaller spine surgeries can offer immediate improvement, but if you don’t think about addressing the larger scale problem, the patient will suffer and they might be back within a year with a failed operation. This will result in the patient requiring another operation, which is something we see very often. So, in order to justify doing a larger cost operation up front, it helps to be able to look out several years in advance and predict outcomes. If you tell someone justifying cost that reimbursement for an operation is X dollars more than a smaller-scale operation, they will likely want evidence through data about why spending that much more can be justified. If they are looking at larger operations that are chosen based on the correct indications, they’ll realize over the long term that most patients aren’t returning for an initial failed surgery. In this ideal scenario, the cost is better justified for the larger operations because you need fewer surgeries overall. Ultimately, through these kinds of examples you can get an understanding how these large scale operations are justified.
I think, overall, predictive modeling and analyzing is something that’s used in the business and government worldwide. We’re simply applying it to healthcare now. We’ve been stuck in this more traditional statistical model, that is hindered because of the fundamentals that require hypotheses and assumptions. They answer a single question, but really now that we’re having an abundance of data, we’re able to better answer any questions that are patient specific and individualized to a particular problem. We can provide a number that’s easier for a patient to interpret. It’s hard to interpret an odds ratio when making a decision as to what surgery to have, but if I say you have a 96 percent chance of doing well from this surgery, that number is something that makes sense.