• Feature

    The Power of Prediction

    While the use of predictive analytics in physical therapy is still in its infancy, early adopters see great potential for it to transform the profession and society.

    Feature - Predictive

    Like most physical therapists (PTs), Stephen Hunter, PT, DPT, OCS, welcomes any edge he can get if it ultimately means he can provide better care. So when his smartphone buzzes as he's taking a patient's history, he doesn't offer his client an apology—he pulls it from his pocket and takes a look.

    On the screen, explains Hunter, who is director at Intermountain Physical Therapy in Salt Lake City, Utah, he'll typically find that the front desk has just sent him an email. "I'll open it up it and read the message. Depending on the patient, it will say something like, 'Mrs Jones has a 67% likelihood of success'" in achieving a minimal clinically important difference (MCID) on the Oswestry Disability Index.

    That prediction, Hunter says, did not just appear out of thin air. Rather, it's the bottom-line appraisal of his patient's condition as interpreted by a software program that Intermountain Healthcare calls ROMS, or the Rehabilitation Outcomes Management System. The platform, developed by Hunter and others at Intermountain (but now commercially available), collects self-reported measures of disability and pain, compares those numbers with previous patients who reported similar problems, and then funnels everything into an equation.

    "What it tells me," Hunter says, "is that with Mrs Jones, this patient right in front of me, this is probably going to be pretty straightforward. I can continue with my history and physical and initiate treatment. But if it's Mr Smith, and the email says he has a 40% chance of getting better, I know I need to do something different. I need to engage the patient, to use motivational interviewing techniques, and I may even tell Mr Smith that—based on our predictive model—there really isn't a great chance that he's going to get better unless we really work at it."

    Hunter will stress to his patient how important it is that he, as the PT, determines exactly which exercises are most critical to his rehabilitation. He'll also explain that the patient's own job is even more important—that it's up to him, as the patient, to take those exercises and do them at home exactly as directed. "It's about shared accountability. 'I want you to succeed. You want to succeed. We've got to work together at this. And if we don't, then you'll be wasting your money and I'll be wasting my time.'"

    Better Predictions, Better Patients

    The concept of making these outcome forecasts is called "predictive analytics." According to Charlie Miraglia, MD, chief medical officer at the company hc1.com (with the self-described mission of personalizing the health care experience), predictive analytics "refers to using data-mining techniques and statistical analysis of current and historical events to predict future outcomes. It can be used in any industry, from insurance to telecommunications, but in the health care arena it allows us to examine large amounts of clinical data and understand the outcomes achieved by following certain paths of care." 1

    You may recall ROMS from an earlier article in this magazine about data analytics and the ways in which physical therapists are collecting and using data to practice more efficiently, achieve better outcomes, reduce costs, and boost reimbursement.2

    In that story, Gerard Brennan, PT, PhD, Intermountain's director of clinical quality and outcomes research, explained how pain-and-disability data collected through ROMS can be layered over other patient-related health care information to identify opportunities for clinical improvement. Brennan described how outcomes measurement and data analytics are critical to Intermountain Healthcare's ability "to know if we're delivering quality care" and to its ongoing efforts to reduce the number of patients who use their services but fail to get better.

    What he didn't mention in that article was Intermountain's recent foray into predictive modeling, which in 2014 helped the organization earn a 2.0 Innovation Award from the American Physical Therapy Association.3 (The award comes with funding and other assistance from APTA, and with the expectation that Intermountain's winning proposal—a "Pay-for-Quality Program to Improve Value-Based Care for Patients With Low Back Pain"—will be refined over the coming year into a replicable template that other physical therapists can use.)

    The modeling program, as the proposal title suggests, is specifically intended for patients with low back pain, Brennan says. And its predictive capabilities depend on data—related to demographics, comorbidities, outcomes, and a variety of other factors—that Intermountain has amassed on its outpatient-orthopedic population for more than 15 years.

    "What really enhances the power of the ROMS database," Brennan says, "is that we collect information on every patient who comes in the door, and we try to do that every single time they visit. It's the only way to get a real picture of how things evolved," which in turn is the piece that makes those predictions possible.

    The other key aspect of the Intermountain database is its reliance on what Brennan calls "a treatment-based classification system."

    "When you come in and you have back pain," he explains, "we're not just looking at an ICD-9 code that puts everything that's reported into 1 big bucket of back pain. We're using a system that subcategorizes that back pain, and those subcategories are based on the best evidence in the literature."

    That is, the smaller and more homogenous the subgroup, the easier it is—in theory, at least—to predict how an individual patient within that group will respond to treatment. "This is all brand new," Brennan notes. At Intermountain Healthcare predictive modeling "is in place and is being used," but it's still too early to say if it will lead to better outcomes. "Our hope is that it will."

    Brennan, of course, is not alone in this regard, as health care providers—and PTs in particular—more than ever must focus on improving patient outcomes, and not just providing more services.

    "We're in a value-based health care market," notes Carmen Elliott, APTA's vice president of payment and practice management. "In the past, clinicians were incentivized to provide more visits and services, as reimbursement was directly tied to volume."

    In the current health care environment, and in accordance with the "triple aim" of the Patient Protection and Affordable Care Act, "providers now are held accountable for the quality of their services. Payers in all market segments are increasingly requiring measurable outcomes that include the patient's experience," Elliott says.

    Mining that data to make predictions, which in turn might be used to direct clinical decision-making and boost outcomes even more, or to demonstrate efficiency, or to improve practice management—well, Elliott says, that's the logical next step. "If the average physical therapist takes a patient with a certain diagnosis through an episode of care in 8 visits, and you're seeing that same type of patient for 12 visits but your outcomes are the same," that's a good indicator that there's room to do better, she says. And if the opposite is true, and the patients you're seeing are defying all expectations and improving faster than predicted and seeing better outcomes, the sign there also is clear: You're doing great work.

    Predictive modeling even could lead a PT to refer her patient to another provider if, for example, the data show that by doing so they're more likely to achieve that triple aim: best quality at a lower cost with a better patient experience. And that's good practice, Elliott says. "Providers may have to change their business model to reflect what is happening in the health care environment," she notes. "What's important today is patient-centric care. It's about doing what's best for the patient."

    Putting Theory Into Practice

    The question, then, is how such models might look in practice, and how predictive analytics might actually work in a real clinical environment. "The hard part," says Tannus Quatre, PT, MBA, president and CEO of Vantage Clinical Solutions Inc in Bend, Oregon, "especially for smaller practices, is that there's really no end to the data that's available," and the path to aggregating that data to make it manageable and usable is not always easy to find.

    The EMR (electronic medical record), for example, is often a practice's greatest potential data source, notes Quatre, but not all EMRs are created equal. Some, for example, will make documentation "a breeze," but "then on the back end, where you want it to turn all those data inputs into intelligence that allows you to run your practice and achieve better outcomes, that part may be missing."

    A typical practice, meanwhile, employs a variety of data-rich systems (EMR, claims management, and so on) that may not talk to one another. "If you have 17 disparate sources of information, the challenge becomes: How are we going to link these things? How can we make it so that when we're trying to run numbers and come up with intelligence, it's all pulled together" and easily accessible "in 1 data warehouse?" The issue, Quatre adds, is "not whether the data exist, or even where. It's: How do you actually synthesize it into something meaningful?"

    A Guide for Care

    One way to do it is through an analytics program. Al Amato, PT, MBA, is president of Focus on Therapeutic Outcomes Inc (FOTO). The program, explains Amato, who also practices at his clinic, Amato Physical Therapy in St Louis, Missouri, is a subscription-based service that collects, compares, and interprets data in ways that are useful to physical therapists. Part of that (as discussed in the March article) involves practice management and demonstrating the value of physical therapist services to payers.

    But the data analysis is also about "managing patient care at the bedside level" and about leveraging the company's growing repository of outcomes-related data—which, at last count, included information collected by more than 15,000 therapists across 6 million episodes of care.

    FOTO's predictive capability, Amato says, depends entirely on making risk-adjustments to that data, so that the care provided for a "16-year-old basketball player who just twisted his ankle in practice" can be compared with that given a "75-year-old man with arthritis who sprained his ankle stepping off a curb." By identifying others in the database who are similar to the patient in a PT's office, Amato explains, the system can "tell you with a very high level of certainty how many visits that patient should need and the functional change they will have at discharge."

    Insurers take this kind of information and use it in their pay-for-performance reimbursement models, Amato says, while PTs use it to work more efficiently (the system features automatic G-code reporting, for example) and improve their skills. "If you're consistently beating the predictions, when we send out our quarterly reports you're going to be ranked near the top of our database."

    And if you're consistently falling short? "It's a way to objectively identify your strengths and weaknesses. It highlights those areas where you might consider continuing education and getting some extra work."

    When it comes to their ability to make predictions, FOTO and similar programs "are nice tools that people can use to help them decide what to do with their patients," says Derek Vraa, PT, DPT, OCS, CSCS, FAAOMPT. "They're not," as some fear, "telling you what to do, or micromanaging your care; they're there as a kind of guide."

    In his work with Physical Therapy Orthopaedic Specialists Inc (PTOSI) in Plymouth, Minnesota, Vraa says, he uses FOTO every day, and with each patient each time they come in, to get a sense of how they "stack up against others with similar risk factors." It's "comparing apples to apples. If I just used the straight Oswestry or the pencil-and-paper method," making accurate comparisons would be all but impossible, he says.

    The program tells him "roughly how my patients should be progressing," so if they're off track in a significant way he knows to adjust his care. And critically, he says, reports—calculated immediately after a patient finishes answering the system's questions—are available to him instantly there in the clinic. "They come into the treatment booth, and I have all that information right in front of me on my computer. I can pull it up, I can read it, and we can discuss it."

    Those results, Vraa says, create a numbers-based portrait of an episode of care along with visit-specific snapshots of the patient's progression that he and his client can consult together. "Some patients see it as a kind of challenge—'I'm better than the national average'—so they try to beat what's predicted, and they do it."

    Other patients, though, are either not as motivated or as lucky. The program's reports make it clear where they stand. "That's when I need to take a look at myself and ask, 'Is there something that I need to change in this patient's plan of care? Or is physical therapy maybe not appropriate at this time?' Maybe they need further intervention—like an epidural steroid, or maybe surgery. Or does this patient just need to change the way he or she is going about things? If that's the case, then we probably need to have a conversation and figure out why."

    In his opinion, Vraa adds, "having more data is always better than having less data. It's up to the physical therapist to decide if that information is useful or not."

    Big Data, Big Future

    That kind of thinking is entirely in sync with what the global management consulting firm McKinsey & Company has called "an era of open information" in health care, where unknown terabytes of digitized data have grown increasingly "usable, searchable, and actionable" by the industry as a whole.4 It's an era of "big data," McKinsey notes, and for anyone in health care, its implications are enormous.

    "All practices are collecting data in some form or another," notes Matt Elrod, PT, DPT, MEd, NCS, APTA senior clinical practice specialist. "Whether they're actually using that data is a different question."

    So as EMRs proliferate and datasets merge, and as companies such as FOTO work to organize that data and massage it into reports that clinicians can actually use, how might that change the way PTs practice? Will predictive analytics one day become commonplace? Or will the concerns that Vraa describes ultimately hold the profession back?

    Stephen Hunter, for his part, is optimistic about that future—and especially the piece that is within the reach of ROMS. Intermountain, he says, has measured patient outcomes for many years and has the data, and the results, to show for it. "But this predictive model is more robust. When a PT is using prediction models to make decisions about treatment to improve patient outcomes, that's powerful, and that's where we're hoping to get."

    When they first started ROMS, Hunter says, he, Brennan, and others at Intermountain "couldn't have imagined we would be developing predictive models. All we wanted to do was set up a system to determine if we were delivering good care." Having achieved that, he notes, they turned to the next question: Now that they were collecting data on almost every patient who walked through their door, "what else can we do with this?"

    ROMS, Hunter admits, is by no means a perfect system. But the data it has amassed "has really helped us to self-evaluate to try to get better and better. And that's the most important thing. To do just a little bit better this year than we did last year." Toward that end, he says, Intermountain is teaching its PTs in Salt Lake City to become experts at motivational interviewing, "so when they get that email that says, 'Hey, you've got a 20% likelihood of success,' they'll have tools they can use to really do something about it."

    Although predictive analytics within the physical therapy sphere is being applied to existing patients and already-diagnosed conditions, its greatest value may lie in helping PTs achieve APTA's vision of "Transforming society by optimizing movement to improve the human experience."

    Miraglia says, "The inherent goal [of predictive analytics] is to identify people earlier on the path (or, ideally, before they get on the path), and prevent the undesirable outcomes experienced by the people who have gone before them. In the simplest of terms, this is what population health is all about.

    "Although predictions can be made for a variety of clinical activities and outcomes, including length of stay in the hospital, readmission rates to the hospital, and a patient's likely response (or lack thereof) to a certain medication, it's important to remember that these predictions are made primarily at the individual level. However, altering clinical practice to improve the care of a single individual, if done similarly under the same circumstances for each patient, leads to benefits for entire populations," Miraglia says.

    That's certainly in line with the innovation component of APTA's vision: "The physical therapy profession will offer creative and proactive solutions to enhance health services delivery and to increase the value of physical therapy to society."

    Ultimately, though, as Derek Vraa of PTOSI says, predictive analytics, despite all the attention it's receiving, is really only as powerful as the clinician who decides to use it. "When you get that prediction, it's not like anything is set in stone," he says. Like any PT, he says, you still have to use your best clinical judgment and consult the best available evidence. In the end, Vraa notes, "it's still just you and the patient. That hasn't changed."

    Chris Hayhurst is a freelance writer.


    1. Miraglia C. "Minority Report." Hc1.com. http://hc1.com/minority-report/. Accessed March 29, 2015.
    2. Hayhurst C. Putting data to work. PT in Motion. 2015;7(2), 34-42.
    3. Innovation 2.0. American Physical Therapy Association. http://www.apta.org/Innovation2/
    4. Kayyali B, Knott, D, Van Kuiken S. The big-data revolution in US health care: accelerating value and innovation. Insights & Publications. McKinsey & Company. April 2013. http://www.mckinsey.com/insights/health_systems_and_services/the_big-data_revolution_in_us_health_care. Accessed March 29, 2015.

    APTA's Innovation 2.0

    To build on the dialogue of Innovation Summit 2013: Collaborative Care Models, APTA launched Innovation 2.0—an initiative aimed at bolstering the impact of physical therapy in innovative and emerging models of health care such as accountable care organizations, bundled payment, direct access, patient-centered medical homes, prevention and health promotion, and value-based purchasing. APTA will aid the selected models by providing financial support, as well as technically sound and practical advice about health care issues, health care policy, evidence-based practice, data management, and dissemination of outcomes.

    APTA named 4 awardees. Two others received honorable recognition, which resulted in a financial award.

    The 4 awardees are:

    • Pay for Quality Program to Improve Value-Based Care for Patients with Low Back Pain—Gerard Brennan, PT, PhD
    • Facilitating Access Improving Care—PTs as Integral ACO Members—Timothy W. Flynn, PT, PhD, OCS, FAAOMPT
    • Adding Value to Postacute Care Settings Through Evidence-Based Physical Therapy Services—Robyn Marcus, PT, PhD, OCS
    • Patient-Centered Medical Home: An Innovative Model for Childhood Obesity Prevention With the Physical Therapist as a Key Player to Improve Quality of Care and Reduce Costs—Brian Wrotniak, PT, PhD.

    The 2 models for honorable recognition are:

    • A New Model of Care in Workers Compensation: Direct Access to Physical Therapist Services by Workers With Low Back Pain—Craig Johnson, PT, MBA
    • Integration of Physical Therapy in 90-Day Postacute Episodes of Care—Allison Orofino, PT

    More information: http://www.apta.org/Innovation2/.

    Leave a comment: