Optimization of Multidisciplinary Staffing Improves Patient Experiences at the Mayo Clinic

Common approaches to emergency department (ED) staffing are to optimize shifts based on historical patient volume or arrival patterns. The former is problematic because historical patient volumes are based on the volumes during existing shifts. Therefore, optimizing shifts based on these volumes can replicate the inefficiencies in these shifts. The latter approach ignores queueing effects. To address the shortcomings of these commonly used approaches, we use classification and regression trees to identify thresholds for patient-to-staff ratios, which split the patient subpopulations into two groups that have different empirical cumulative distribution functions (ecdfs) for patients’ lengths of stay in the ED; one has an extended length and the other has a shorter length. We apply these thresholds and ecdfs to historical patient volumes to calculate an ideal patient volume. After accounting for arrival patterns of ED patients, ideal patient volumes represent the load on the entire ED if patient-to-staff ratios are always kept under the identified thresholds. We then use a mixed-integer programming model to minimize understaffing with respect to the ideal patient volumes. The ED at Mayo Clinic Saint Marys Hospital in Rochester, Minnesota, a trauma center for both adults and pediatrics, implemented the new shift templates in the fourth quarter of 2015. The templates resulted in statistically significant improvements in several patient-centered metrics. In particular, the median length of stay, door-to-provider time, and door-to-bed time decreased by 11, 2.7, and 3 minutes, respectively, despite a six percent increase in patient volume.