|
Comparing hospital performance in a health system is traditionally done with regression models that adjust for case-mix. In contrast, "template matching" compares outcomes of similar patients at different hospitals. However, in template matching, hospitals for which it was not possible to achieve a high-quality match to the template are excluded, which limits usefulness in real-world benchmarking.
To address this limitation, we developed a tiered template matching procedure for the nationwide Veterans Affairs healthcare system, in which we used separate (N=300 hospitalization) templates for each of 5 tiers of hospitals defined by critical care and procedural capabilities. We match on patient demographics, diagnoses, and acute physiology using an optimal matching algorithm with fine balance and near-exact constraints.
Preliminary results suggest this approach yields good balance on key covariates at most hospitals, and far better balance than was achieved with a single template. This approach retains the potential benefits of template matching over a regression model without greatly restricting the dataset and yielding similar benchmarking results.
|
