Developing Risk Adjustment Models in Veterans' Health Status
Lewis E Kazis ScD
VA New England Health Care System
Bedford, MA
Funding Period: April 1999 - March 2001

BACKGROUND/RATIONALE:
Changes in PCS and MCS as measured by the SF-36V are recognized by the VHA as outcomes of care. However, other factors including patient’s clinical characteristics or case mix can influence changes in PCS and MCS. These patient factors are particularly important for the VHA because there are significant case mix differences across VISNs.

OBJECTIVE(S):
To compare the performance of three diagnosis-based case mix measures, the Adjusted Clinical Groups case-mix adjustment (ACGs) system (Weiner et al. 1991), the Deyo adaptation of the Charlson Index (Deyo et al. 1992), and the Extended Comorbidity Index (Selim et al. 1998) in predicting change in veterans’ functional status. To modify these measures so as to improve their performances in predicting changes in veterans' functional status. After having identified the most predictive diagnosis-based measure, to build risk adjustment models by adding other patient characteristics including sociodemographic and baseline functional status information. Finally, to make comparisons of the observed and adjusted changes in functional status scores at the level of the networks.

METHODS:
This study used data from the 1998 National Survey of Ambulatory Care Patients, a prospective monitoring system of outcomes of patients receiving ambulatory care in the Veterans Affairs (VA) integrated service networks. The physical health and mental health status were measured using the Veterans SF-36, a modified version from the Medical Outcome Study (MOS) SF-36. Of the 31,823 patients that completed the Veterans SF-36 in 1998, 70% (22,116) completed a follow-up questionnaire 18 months later. The eight important concepts of health that are included in the Veterans SF-36 (physical functioning, role physical, social functioning , general health perceptions, vitality, bodily pain, role emotional, and mental health were summarized into physical (PCS) and mental component (MCS) scales. The main study outcome measures were the observed and risk-adjusted PCS-D (defined as patients with a decline in their PCS scores greater than 5.66 points between baseline and 18-month follow-up and/or death) and MCS-I (defined as patients with an increment in their MCS scores greater than 6.72 between baseline and 18-months follow-up) rates.

FINDINGS/RESULTS:
Of the 31,823 patients, 5 percent died during the 18-months follow-up. The observed mortality rates across the 22 integrated service networks in the VA system ranged from 3.3 percent to 6.7 percent. The variation in the observed mortality rates across the 22 integrated service networks was statistically significant (p<0.001). Of the cohort of 22,116 patients with complete follow-up survey information, 3,028 patients had better physical health (an increase by more than 5.66 points in their PCS), 13,488 had unchanged physical health (PCS scores that remained within 5.66 points at baseline and 18 months of follow-up), and 5,593 had worse physical health (a decline by more than 5.66 points in their PCS) during the 18-months follow-up. The MCS scores were better (an increase by more than 6.72 points in their MCS) in 4,131, unchanged (MCS scores that remained within 6.72 points at baseline and 18 months of follow-up) in 12,420 patients, and worse (a decline by more than 6.72 points in their MCS) in 5,558 patients. On logistic models including age, gender, a diagnosis based case mix measure (the extended Comorbidity Index, the Charlson Index, the modified Charlson Index, ADGs, and ACG case-mix adjustment system), all disease-based case mix measures showed weak c-statistics for predicting both PCS-D and MCS-I. We developed final risk adjusted models using age, gender, and the extended CI in addition to marital status, means test, service connected disability, baseline PCS and MCS. The resulting risk-adjusted models for both PCS-D and MCS-I performed well in cross-validated tests of discrimination (c-statistic = 0.72; 95% CI, 0.71 to 0.73 for PCS-D and 0.65; 95% CI, 0.62 to 0.68 for MCS-I) and calibration. Analysis of variance confirmed that the 22 integrated service networks differed in their average level of expected risk for both PCS-D and MCS-I (p<0.001). Risk-adjusted rates and ranks of the networks differed considerably from unadjusted ratings for both PCS-D and MCS-I.

IMPACT:
Risk adjusted models that are validated and can predict changes in functional status at the network level are important evidence of outcomes. Thus, this proposal expands and greatly strengthens risk adjustment models for comparisons of outcomes at the network level. The VHA performances measurement system can incorporate this methodology into their process of evaluation. While not the focus of this proposal, future efforts can begin to identify the processes of care that may affect patient centered outcomes.

PUBLICATIONS:

Journal Articles

1. Weeks WB, Kazis LE, Shen Y, Cong Z, Ren XS, Miller D, Lee A, Perlin JB. Differences in health-related quality of life in rural and urban veterans. American Journal of Public Health. 2004; 94(10): 1762-7.
2. Selim AJ, Berlowitz DR, Fincke G, Cong Z, Rogers W, Haffer SC, Ren XS, Lee A, Qian SX, Miller DR, Spiro A, Selim BJ, Kazis LE. The health status of elderly veteran enrollees in the Veterans Health Administration. Journal of The American Geriatrics Society. 2004; 52(8): 1271-6.
3. Selim AJ, Berlowitz DR, Fincke G, Rosen AK, Ren XS, Christiansen CL, Cong Z, Lee A, Kazis L. Risk-adjusted mortality rates as a potential outcome indicator for outpatient quality assessments. Medical Care. 2002; 40(3): 237-45.

DRA: Health Services and Systems
DRE: Technology Development and Assessment
Keywords: Functional status, Research measure, Risk adjustment
MeSH Terms: none