Re-examining Facility-level Effects on Diabetes Care Quality for Veterans
Little is known about how organizational and structural factors mediate patterns of chronic disease care quality. In the Department of Veterans Affairs (VA), evidence suggests facility-level differences account for most variation in diabetes care quality. However, identifying specific facility-level determinants of diabetes care quality has proven difficult, in part because of cross-sectional study designs susceptible to bias due to reverse causality.
We re-analyze the effects of facilities’ average diabetes care quality measure performance and available resources on individual physicians’ diabetes care quality measure performance, employing an innovative longitudinal design that exploits the experience of physicians who “move,” treating veterans in different facilities over time, to overcome this concern.
Study population:
Our sample of moving physicians delivered 60% or more of their diabetes measure-eligible episodes in one VA facility one year and 60% or more of their episodes in another facility the following year, with at least ten episodes each year, during FY2008-FY2011.
Measures:
We measured diabetes care quality using intensity indicators for blood pressure control therapy (BPC) among patients with diabetes based on Kerr and colleagues’ (2012) clinical action measure. All diabetes episode and physician data were derived from the VA Clinical Data Warehouse (CDW).
Facility resource scales were constructed using data from CDW, the 2006-2007 VA Clinical Practice Organization Surveys, and the 2008-2009 VA Primary Care Survey.
Statistical analysis:
We estimated the effects of facilities’ rates of low-intensity, moderate-intensity, and high-intensity BPC on these same rates for moving physicians’ using physician fixed effects models. These models were identified using changes over time in physicians’ peers’ average performance: our key independent variable was the difference in average facility-year-level quality measure performance between the physician’s post-move facility (current year) and pre-move facility.
We also assessed whether facility resources mediated this relationship by introducing interaction terms. Lastly, we conducted parallel cross-sectional analyses—associating each physician’s performance with his/her peers’ in the same year—for comparison.
We controlled for numerous patient, facility, and region characteristics in each model and clustered standard errors at the facility level.
Results:
Using our sample of 314 moving physicians, we find large effects of facility-level quality on individual physicians’ quality across all models (p<0.05). For example, for each percentage-point increase in a physician’s current facility’s high-intensity BPC rate, relative to his/her pre-move facility’s rate, the physician’s high-intensity BPC rate rises 0.39 percentage points (p<0.001). Fixed effects model estimates were consistently larger in magnitude than cross-sectional model estimates. Facility resources did not significantly mediate these effects.
Conclusions:
VA facilities’ average diabetes care quality performance strongly affects the performance of individual physicians, more strongly than can be observed using cross-sectional study designs. Because we control for numerous patient-level and population-level characteristics and because the specifications of our BPC quality measures restrict heterogeneity among measured patients, our strong estimates reaffirm that there are important facility-level mechanisms leading physicians to perform like their peers. Additional evidence is needed to identify specific mechanisms—other than differences in facility resources—by which VA facilities influence their clinicians’ quality of diabetes care.