Background: Studies have shown geographically co-locating patients of the same physician (also known as regionalization) have improved provider satisfaction and increased team efficiency and collaboration. At most institutions, including ours, assigning physicians to patients and units is done manually. Furthermore, interventions to regionalize hospitals in the literature focus on grouping conditions, specialties, or teams (i.e., resident vs. nonresident teams). Still, no interventions specifically focus on minimizing the distance traveled by individual physicians. Additionally, the manual assignment process is very tedious, time-consuming, and ultimately unsustainable.
Purpose: We leverage technology and mathematical modeling to generate patient assignments that achieve the purpose of regionalizing each physician. In other words, we create an automated way to assign patients to physicians that satisfies all necessary criteria while ensuring that physicians’ patients are located as close to each other as possible so they spend minimal time traveling. This patient assignment technology/tool would replace the need for physicians to spend time manually making patient assignments that geographically regionalize each physician, which in the past took over 3 hours for hospitalists to do.
Description: We use a mathematical optimization model with integer decision variables to generate the assignment of patients to physicians. We formulate the problem of deciding which patients to assign to each physician mathematically, with constraints including load-balancing across physicians, continuity of care etc., and with the objective of minimizing distance traveled across all physicians. We coded the model in Python and used the OR-Tools solver to find an optimal solution. We then send the model-generated assignment to physicians for approval.
Conclusions: We saw the possibility of a 300% reduction in distance traveled when using the optimization model to generate patient assignments compared to the actual assignments that were used.Furthermore, the model generates patient assignments in significantly less time than previously observed, decreasing the amount of time physicians spend manually making assignments. This model shows promise in sustaining a complex project like provider regionalization and decreasing manual tasks currently performed daily by physicians.