4030 — Method to Develop Facility-Specific Groups of VA Hospital Peers
Lead/Presenter: Patrick O'Mahen, COIN - Houston
All Authors: O'Mahen PN (COIN Houston)
Mehta PD (University of Houston)
Rajan SS (University of Texas -- School of Public Health)
Knox MK (COIN Houston)
Yang C (COIN Houston)
Kuebeler MK (COIN Houston)
Petersen LA (COIN Houston)
At the request of VA network leaders, develop a methodology to identify peers for acute-care facilities based on key VA facility characteristics for the purposes of administrative-, performance-, and research-motivated comparisons.
We incorporated feedback from executives, clinicians, economists, and statisticians to select measures of facility characteristics that influence variations in cost and scope of service as a potential alternative to current Office of Productivity Efficiency and Staffing (OPES) complexity groups. We used principal components analysis (PCA) to identify linear composites that discriminate among facilities based on our measures. These components were then used to compute Euclidian distances between all pairs of facilities. We identified a set of peers centered around each facility based on the differences in distance and the score on the first principal component between the target and other facilities.
We used data from 122 facilities providing acute care at the VHA. We identified 15 measures of facility characteristics that describe facility size, disease burden, reliance on VA, academic mission, care delivery structure, infrastructure, and community context. We computed the Euclidean distance between all facility-pairings based on the 15 principal components weighted by the proportion of variance they explain. Shorter Euclidean distances between facilities indicate greater similarity among the 15 measures. To determine peers of a given facility, we selected the 50 closest facilities by distance then eliminated those not falling within one standard deviation of the index facility's score on the first principal component. Average size of peer groups is 20.9 with a standard deviation of 5.5. Groups ranged in size from 3 to 31 facilities, with a median and mode of 22
PCA can be used to compute distance from correlated measures of facility characteristics. Using a multi-dimensional measure of distance and principal components differentiates acute-care facilities based on a wide range of characteristics and identifies unique sets of peers centered on individual facilities.
Developing methods to identify peers for acute care facilities aids numerous comparisons. Using PCA to compute distance between sites also provides a novel yet readily intuitive interpretation of important characteristics contributing to facility differences.