Jian Gao1, Eileen Moran2, Peter L Almenoff3. 1. Department of Veterans Affairs, Office of Productivity, Efficiency, and Staffing, Albany, NY. 2. Department of Veterans Affairs, Office of Productivity, Efficiency, and Staffing, West Haven, CT. 3. Department of Veterans Affairs - Reporting, Analytics, Performance Improvement and Deployment (RAPID), University of Missouri Kansas City School of Medicine, Kansas City, MO.
Abstract
BACKGROUND: Accurate risk adjustment is the key to a reliable comparison of cost and quality performance among providers and hospitals. However, the existing case-mix algorithms based on age, sex, and diagnoses can only explain up to 50% of the cost variation. More accurate risk adjustment is desired for provider performance assessment and improvement. OBJECTIVE: To develop a case-mix algorithm that hospitals and payers can use to measure and compare cost and quality performance of their providers. METHODS: All 6,048,895 patients with valid diagnoses and cost recorded in the US Veterans health care system in fiscal year 2016 were included in this study. The dependent variable was total cost at the patient level, and the explanatory variables were age, sex, and comorbidities represented by 762 clinically homogeneous groups, which were created by expanding the 283 categories from Clinical Classifications Software based on ICD-10-CM codes. The split-sample method was used to assess model overfitting and coefficient stability. The predictive power of the algorithms was ascertained by comparing the R, mean absolute percentage error, root mean square error, predictive ratios, and c-statistics. RESULTS: The expansion of the Clinical Classifications Software categories resulted in higher predictive power. The R reached 0.72 and 0.52 for the transformed and raw scale cost, respectively. CONCLUSIONS: The case-mix algorithm we developed based on age, sex, and diagnoses outperformed the existing case-mix models reported in the literature. The method developed in this study can be used by other health systems to produce tailored risk models for their specific purpose.
BACKGROUND: Accurate risk adjustment is the key to a reliable comparison of cost and quality performance among providers and hospitals. However, the existing case-mix algorithms based on age, sex, and diagnoses can only explain up to 50% of the cost variation. More accurate risk adjustment is desired for provider performance assessment and improvement. OBJECTIVE: To develop a case-mix algorithm that hospitals and payers can use to measure and compare cost and quality performance of their providers. METHODS: All 6,048,895 patients with valid diagnoses and cost recorded in the US Veterans health care system in fiscal year 2016 were included in this study. The dependent variable was total cost at the patient level, and the explanatory variables were age, sex, and comorbidities represented by 762 clinically homogeneous groups, which were created by expanding the 283 categories from Clinical Classifications Software based on ICD-10-CM codes. The split-sample method was used to assess model overfitting and coefficient stability. The predictive power of the algorithms was ascertained by comparing the R, mean absolute percentage error, root mean square error, predictive ratios, and c-statistics. RESULTS: The expansion of the Clinical Classifications Software categories resulted in higher predictive power. The R reached 0.72 and 0.52 for the transformed and raw scale cost, respectively. CONCLUSIONS: The case-mix algorithm we developed based on age, sex, and diagnoses outperformed the existing case-mix models reported in the literature. The method developed in this study can be used by other health systems to produce tailored risk models for their specific purpose.
Authors: Jian Gao; Eileen Moran; Steffie Woolhandler; Andrew Toporek; Andrew P Wilper; David U Himmelstein Journal: J Gen Intern Med Date: 2021-10-04 Impact factor: 6.473