1041 — Risk Adjustment in VA Populations - A Method for Partial Recalibration
Loveland SA (CHQOER at Bedford VAMC & BU School of Public Health) , Christiansen CL
(CHQOER at Bedford VAMC & BU School of Public Health), Zhaos S
(CHQOER at Bedford VAMC ), Rivard PE
(COLMR at Boston VA), Rosen AK
(CHQOER at Bedford VAMC & BU School of Public Health)
Objectives:
Risk-adjustment (RA) is essential for comparing outcomes across providers and allocating resources for care. However, RA coefficients are often tied to populations that are dissimilar to the VA and recalibration is needed. Resource and/or data constraints may preclude developing a complete VA-specific RA. Our objective was to develop a methodology for applying RA to the VA using the benchmark population data yet still re-estimating coefficients for age/sex categories and comorbidities that are more prevalent or severe in the VA population.
Methods:
The AHRQ Patient Safety Indicator (PSI) software provides RA based on HCUP’s Nationwide Inpatient Sample (NIS). For each of 16 PSIs we calculated the logit of the “expected” outcome for each hospitalization, using AHRQ’s RA coefficients. This was used as an offset in the logistic regression model along with indicator variables for age/sex categories and 15 comorbidities, e.g., Depression, Diabetes, Alcohol-Use, Hypertension. Estimates were assessed for statistical significance and effect size. Keeping only significant variables, the model was rerun and coefficients re-examined. If the comorbidity was already part of AHRQ’s RA, the coefficient indicated a differential effect on VA- based expected compared to NIS; if the comorbidity was not part of AHRQ’s RA and its coefficient significant, it was added to the VA RA.
Results:
Adjustments were needed in 10 PSIs. Coefficients ranged from -0.7 to 1.3. Among comorbidities with additional impact were Depression on PSIs 3 and 13 (Decubitus Ulcer and Postoperative Sepsis, coefficients 0.337, 1.180) and Alcohol-Use on PSI13 (0.875). Among comorbidities added (not part of NIS-based RA) were Alcohol-Use in PSI9 (Postoperative Hemorrhage/Hematoma, 0.486), and Hypertension in 7 PSIs.
Implications:
Certain comorbidities, not accounted for in the NIS-based RA, are important to VA RA and need either recalibration or inclusion. This method of calculating RA coefficients specific to the VA had two results: recalibration to VA case-mix and comparison of VA case-mix to that of the benchmark population.
Impacts:
RA is important in accounting for VA patient case-mix and allocating resources for care; however non-VA-based RA can be misleading. This method of calculating VA RA coefficients is widely-applicable for VA research where resources or data prohibit complete recalibration.
