Op/Ed by Christopher Ames, MD
In the debate over affordable health care in the United States, there is an increasing expectation for physicians to prove the benefit of their interventions and accurately estimate costs. As we move away from fee-for-service models and towards value-based models, driven largely by Medicare, defining value remains a hotly contested battleground.
The bundled payment model—in which a hospital is given a fixed payment for a procedure over the entire care trajectory—has especially gained traction among policy makers who reason that it incentivizes providers to deliver high-quality care at the lowest possible cost.
As part of the Center for Medicare & Medicaid Innovation’s Bundled Payments for Care Innovation project, UCSF was one of the first institutions to test bundled payments in a pilot program for total joint athroplasty (TJA). The investigators of that program found that outcomes heavily depended on factors unrelated to the procedure itself, and thirty-five percent of episode-of-care payments were related to post-discharge care.1 This study clearly demonstrated that we needed to better understand the breakdown of costs and services across the entire care trajectory and that we needed to do a better job at ensuring appropriate post-discharge care for patients.
In the TJA study, there was considerable variation found among patients and procedures that affected cost, but that complexity is minimal compared with the variation found within complex spine surgery. There are currently no validated predictive models of improvement that could be used to generate reliable bundle payment options for complex spine surgery. However, this is an area we have been working on here at UCSF, in collaboration with the International Spine Study Group (ISSG), by applying predictive analytics.
For surgical procedures, estimating risk is usually done using odds ratios and regression analyses that factor in variables like age, degree of disability, history of smoking, and body mass index. These familiar statistical methods rely on a set of control data to confirm or refute a specific hypothesis. Predictive modeling, on the other hand, uses advanced computational methods to identify patterns in large datasets that can then be applied to individuals.
Modern datasets for spine surgery patients have yielded some surprising information. Elderly patients, for example, have long been advised against complex spine surgeries over fears related to comorbidities and frailty, but we found that they often do better than their younger counterparts. When we examined the outcomes of patients who had severe structural deformity but little disability with the outcomes of patients with minimal deformity but severe disability, there was no difference in Oswestry Disability Index (ODI) scores or neck disability after surgery. What are the factors contributing to good or bad outcomes at the individual patient level?
In 2014, we began developing models that analyzed variables related to demographics, surgical data, quality of life, and imaging. We started out with a model that incorporated 43 clinical variables and applied it to a prospective dataset collected by the ISSG. It was able to predict with 86% accuracy whether a patient would develop proximal junction kyphosis or reach a minimally important difference on the ODI scale after surgery. In another study, we found 20 variables that could predict with 87.6% accuracy the risk of intraoperative and perioperative complications.2-3 In 2016, an adult spinal deformity frailty index (ASD-FI) incorporating 40 variables was shown to be predictive of length of hospital stay and major complications.4
Psychosocial factors are also an important consideration. In spine surgery, we are primarily hoping to relieve pain. The complex nature of pain, especially pain that has been long-standing over a period of years or even decades, is not always easily solved, even when the surgery is considered successful from an anatomical perspective. This is a challenge, but validated clinical outcome assessments to measure overall mental health and well-being need to be standardized and incorporated into payment schemes. Similarly, it is thought that patients with better social support systems have better long-term outcomes. In hospitals that take care of sectors of the population that historically do not have good support systems, this needs to be taken into account.
Building accurate predictive models is critical to prevent incentives for providers to withhold care in value-based care models and for us to be able to better counsel our patients on the risk-benefit of surgery and expectations for recovery. These models also have powerful implications for a wide range of public health issues, ranging from health insurance to informed consent.
References:
(1) Bozic KJ, Ward L, Vail TP, Maze M. Bundled payments in total joint arthroplasty: targeting opportunities for quality improvement and cost reduction. Clin Orthop Relat Res 2014;472:188-93.
(2) Scheer JK, Osorio JA, Smith JS, Schwab F, Lafage V, Hart RA, Bess S, Line B, Diebo BG, Protopsaltis TS, Jain A, Ailon T, Burton DC, Shaffrey CI, Klineberg E, Ames CP, International Spine Study Group. Spine (Phila Pa 1976) 2016;41(22):E1328-35.
(3) Scheer JK, Smith JS, Schwab F, Lafage V, Shaffrey CI, Bess S, Daniels AH, Hart RA, Protopsaltis TS, Mundis GM Jr, Sciubba DM, Ailon T, Burton DC, Klineberg E, Ames CP; International Spine Study Group.. Development of a preoperative predictive model for major complications following adult spinal deformity surgery. J Neurosurg Spine. 2017 Mar 24:1-8. doi: 10.3171/2016.10.SPINE16197. [Epub ahead of print].
(4) Miller E, Sciubba DM, Neuman BJ, Smith JS, Kebaish KM, Kleinstuck F, Obeid I, Perez-Grueso FJ, Pellise F, Ames CP, European Spine Study Group. Development and external validation of the adult spine deformity (ASD) frailty index (ASD-FI) [abstract 208]. Spine J 2016;16:S213.