Variation in the Quality of Diabetes Care for Veterans

Diabetes is a significant health problem in the U.S. and particularly for veterans: nearly one fourth have the condition (VHA, 2013).  The Veteran Health Administration (VHA) has performed excellently on standard measures of diabetes care quality (e.g., Health Plan and Employer Data Information Set (HEDIS)), but concerns have been raised that these measures may be insufficiently evidence-based and patient-centered.  Two recently developed, “tightly linked” (i.e., to evidence) clinical action measures (CAMs) offer a unique opportunity to robustly assess patterns of quality in the VHA’s diabetes care with little concern of Hawthorne effect bias because they had not been used to evaluate performance in the VHA until recently and are not yet widely known.

Kerr and colleagues (2012) and Beard and colleagues (2013), who developed the CAMs, noted wide variation in CAM performance across the VHA.  In this paper, I rely on the VHA’s largely homogeneous financial incentives and two theoretical frameworks from the geographic variation literature to explore this variation’s underlying structural causes.  My hypotheses include, first, based on a framework of resource sufficiency and coordination (Baicker & Chandra, 2004), that VHA facilities with greater access to care resources have higher CAM measure results.  However, the marginal benefits to patients of additional resources should diminish (e.g., due to spillovers costs of information overload) and may become negative without effective, facility-level coordination.  Second, based on a framework of peer effects and physician learning (Phelps & Mooney, 1993; Epstein & Nicholson, 2009), I predict that, after being introduced into a new facility, a physician’s diabetes care quality will converge toward that facility’s average quality over time.  Lastly, the two frameworks jointly suggest the physician’s rate of convergence toward facility average levels should be higher when the facility’s resource constraints are more binding—measured as lesser resource sufficiency.

These predictions’ empirical tests—which use VHA clinical data for nearly all veterans receiving care for diabetes between July 2007 and June 2011 (nearly 4 million veteran-year records), linked to three facility-level surveys conducted between 2007 and 2008, Vital Status records, and the Area Health Resource File—are ongoing.  My principal independent variables are indices of resource sufficiency and coordination constructed using survey data and facility-level measures aggregated from VHA clinical records.  I first assess the associations between these indices and CAM results, controlling for numerous facility-level and market-level factors, and test for any mediating effects of coordination using least squares regression with interaction effects. I also test these relationships using a panel data analysis framework and a more limited set of time-variant, facility-level measures.  Next, I use quantile regression methods to test for the declining marginal benefits of resource sufficiency and differences in the mediating effects of coordination at different resource levels.  Finally, I use a longitudinal sample of physicians moving between VHA facilities to test for convergence in care quality patterns and differences in rates of convergence in high- versus low-resource facilities.

These analyses will inform facilities’ expectations regarding the improvements in diabetes care quality attributable to greater access to and coordination of care resources.