Abstract:
|
Background: Comparing hospital performance in a health system is traditionally done with hierarchical models that adjust for case-mix (direct standardization). In contrast, "template matching" compares outcomes of similar patients at different hospitals (indirect standardization). We evaluated these methods in simulated situations when there were no hospital level effects. Methods: Using one year of nationwide VA electronic health record data, we randomly assigned 460,213 patients to 111 pseudo hospitals, thereby eliminating hospital level effects. We randomly selected a representative template of 240 patients and matched 240 patients on demographic and physiological factors from each pseudo hospital to the template. We assessed 30-day mortality at each pseudo hospital and identified significant outliers. We compared the false outlier detection rates between the two methods. Results: Preliminary results suggest that template matching performs similarly to hierarchical models with a false outlier detection rate of 7-8%. Conclusion: Template matching may be a credible alternative method to compare hospitals, with the added benefits of interpretability and actionability by end-users.
|