This article is more than 5 years old. Information may no longer be current.
Common predictive variables provided adequate risk adjustment for comparing clinical outcomes
Age, gender and ASA models were nearly as predictive as models using all covariates for THA and TKA.
Click Here to Manage Email Alerts
Recently published results showed by using only the most predictive variables commonly available within the clinical record, it is possible to provide adequate risk adjustment to compare outcomes for hip fracture repair and total hip and knee arthroplasty across providers.
“We looked at three common procedures in orthopedics [hip fracture repair, total hip and total knee arthroplasty] and found morbidity and mortality can be adequately risk-adjusted with a relatively limited number of the most predictive variables,” Peter L. Schilling, MD, MSc, an orthopedic surgeon at the San Mateo Medical Center and co-founder and chief medical officer at DataFascia Inc., told Orthopedics Today. “By moving to more limited risk-models, we can greatly reduce the burden of data collection necessary for fair comparisons of quality measures across providers.”
Risk-adjustment models
Utilizing data from the American College of Surgeons National Surgical Quality Improvement Program, Schilling and Kevin J. Bozic, MD, MBA, created derivation cohorts for hip fracture repair, total hip arthroplasty (THA) and total knee arthroplasty (TKA). Using age, gender, American Society of Anesthesiologists (ASA) physical status classification, comorbidities, laboratory values and vital signs-based comorbidities as covariates, logistic regression models were developed and validated for each procedure with use of data from 2012.
Results showed derivation models had C-statistics for mortality of 80% for hip fracture repair, 81% for THA, 75% for TKA and 92% for combined procedure cohorts. C-statistics for adverse events were 68% for hip fracture repair and THA, 60% for TKA and 70% for combined procedure cohorts. Age, gender and ASA classification were observed to be nearly as predictive as models utilizing all covariates for THA and TKA, while the addition of comorbidities and laboratory values improved hip fracture repair model discrimination. Functional status, low albumin, high creatinine, disseminated cancer, dyspnea and BMI were among the important covariates. When compared to validation cohorts, results showed model performance was similar.
“Our finding contradicts a widely held belief that extremely detailed risk-adjustment, which places additional burden on providers and their patients, is necessary for fair comparisons of quality across providers,” said Bozic. “We were surprised by how much of the variation in orthopedic outcomes can be predicted by demographics and ASA class. Age, gender and ASA class accounted for a large share of the explained variation for both mortality and adverse events —more than any of the other variables in the models.”
Data collection burden
Although Schilling and Bozic said “data collection burden has limited the growth of quality improvement initiatives and has led many to rely on administrative data to support their efforts,” they added the collection of large volumes of patient demographic and clinical data from a variety of sources at a low marginal cost will be enabled within the next 10 years. This will be done via data management platforms and will give physicians a “more nuanced understanding of patients’ health,” Bozic and Schilling said.
“Platforms like these are able to liberate data in real-time from a wide variety of data sources, including electronic health records, medical devices, biosensors and a full array of other health care software applications being used today,” Bozic said. “As these technologies spread beyond earlier adopters, we will be able to create a more complete patient information profile not only for the purposes of risk adjustment, but also to help providers make better informed and more timely decisions to improve outcomes and lower costs.”
Schilling added, “We are on the cusp of some exciting developments at the intersection of medicine and data science, but for now, the burden of data collection remains a problem. Much of our health data is locked up in data silos that are difficult to access — including the risk clinical data contained in electronic health records. This is about to change, but in the meantime, we can mitigate the problem by focusing on only the most critical pieces of information. This study shows we can still compare outcomes across providers fairly by using only a limited number of the most predictive variables in our risk-adjustment models.” – by Casey Tingle
- Reference:
- Schilling PL, et al. J Bone Joint Surg Am. 2016;doi:10.2106/JBJS.N.01330.
- For more information:
- Kevin J. Bozic, MD, MBA, can be reached at the Department of Surgery and Perioperative Care, Dell Medical School, The University of Texas at Austin, 1912 Speedway, Austin, TX 78712; email: kevin.bozic@austin.utexas.edu.
- Peter L. Schilling, MD, MSc, can be reached at the San Mateo Medical Center, 222 West 39th Ave., San Mateo, CA 94403; email: pschillin@gmail.com.
Disclosures: Bozic reports he is a consultant to the Yale Center for Outcomes Research and Evaluation for development of performance measures for the CMS. Schilling reports he is the co-founder and chief medical officer of DataFascia Inc.
Perspective
Back to Top
John M. Cuckler, MD
The study by Shilling and Bozic is an interesting first step in the effort to provide insight into attempts to quantify risk adjustment for the 30-day mortality and adverse events following hip fracture, total hip arthroplasty, or total knee arthroplasty. The study was based on data from the American College of Surgeons National Surgical Quality Improvement Program (ACS NSQIP).
Not surprising to a surgeon, but the data demonstrates increased age and ASA status had the strongest correlations with mortality and adverse events. Cormorbidities added little to the correlation, but were more significant for the hip fracture cohort.
This data needs to be put in front of the bureaucrats driving the changes in the health care system, particularly given the current emphasis on outcomes. As electronic medical records improve and increase in utilization, refinement in such models should emerge as the data collection burden is reduced. The effects of the healthcare setting (eg, teaching hospital vs. community hospital) need to be examined. The 90 day mortality and adverse events also would be of great interest, given the current reimbursement mechanisms in place. Validation of the current model with other data sources, such as Medicare claims, would further strengthen the value of this work.
Click Here to Manage Email Alerts