Talk to the Veterans Crisis Line now
U.S. flag
An official website of the United States government

VA Health Systems Research

Go to the VA ORD website
Go to the QUERI website
2015 Conference Logo



2015 HSR&D/QUERI National Conference Abstract


3007 — Risk Adjustment Tools for Learning Health Systems: the development of the Nosos risk score

Wagner TH, VA Palo Alto; Almenoff P, Operational Analytics and Reporting; Stefos T, Office of Productivity, Efficiency and Staffing; Upadhyay A, VA Palo Alto; Cowgill E, VA Palo Alto; Moran E, Office of Productivity, Efficiency and Staffing; Asch S, VA Palo Alto; Cashy J, Pittsburgh VA; Shen M, Office of Productivity, Efficiency and Staffing;

Objectives:
To compare risk scores computed by DxCG (Verisk) and Centers for Medicare and Medicaid Services (CMS) V21, we analyzed administrative data from the Department of Veterans Affairs (VA) for fiscal years 2010 and 2011. We then developed a new risk adjustment model for cost data using V21 along with additional mental health diagnostic information and pharmacy. This presentation highlights our comparison of DxCG to V21, and the development of the model, known as Nosos. We also highlight access to these new risk adjustment data on the Corporate Data Warehouse.

Methods:
We created six analytical files representing more than two million Veterans (samples of general users, high cost users, mental health and substance use users, multi-morbid users and healthy users). We regressed total annual VA costs on predicted risk scores in ordinary least squares, general linear models, log-transformed and square-room transformed models, controlling for age and sex. Model fit was judged by R2, root mean squared error, mean absolute error, and Hosmer and Lemeshow tests.

Results:
Square root transformed regression models consistently performed better than other regression specifications. No one risk model was universally preferred across the six subsamples. The DxCG risk score with pharmacy data yielded substantial gains in fit over the V21 model, but the Nosos model generated risk scores with similar fit statistics to the DxCG risk scores. Notable gains in fit were observed in the low risk users, as many of these patients use medications, but are otherwise infrequent users of inpatient or outpatient care.

Implications:
The DxCG system, which is proprietary, provides risk scores that offer improved fit over V21, developed for Medicare managed care plans. We created a new model, known as Nosos, that compared favorably to the DxCG with pharmacy data.

Impacts:
Instead of relying on an expensive risk adjustment system for cost data, we have developed a new system that can be used by operations and research. Prospective and concurrent Nosos risk scores have been calculated for all VA patients from 2006-2015 (Q1) and have been posted to the CDW in SAS and SQL tables. Access to these data and documentation is discussed.