Improvement of surgery duration estimation using statistical methods and analysis of scheduling policies using discrete event simulation

摘要

The United States health care system currently faces many challenges, with the most notable one being rising costs. In an effort to decrease those costs, health providers are aiming to improve efficiency in their operations. A primary source of revenue for hospitals and some clinics is the surgery department, making it a key department for improvement in efficiency. Surgery schedules drive the department and affect the operations of many other departments. The most significant challenge to creating an efficient surgery schedule is estimating surgery durations and scheduling cases in a manner that will minimize the time a surgery is off schedule and maximize utilization of resources. To identify ways to better estimate surgery durations, an analysis of the surgery scheduling process at UnityPoint Health - Des Moines, in Des Moines, Iowa was completed. Estimated surgery durations were compared to actual durations using a t test. Multiple linear regression models were created for the most common surgeries including the input variables of age of the patient, anesthesiologist, operating room (OR), number of residents, and day of the week. To find optimal scheduling policies, simulation models were created, each representing a series of surgery cases in one operating room during one day. Four scheduling policies were investigated: shortest estimated time first, longest estimated time first, most common surgery first, and adding an extra twenty minutes to each case in the existing order. The performance of the policies was compared to those of the existing schedule.Using the historical data from a one-year period at UnityPoint Health - Des Moines, the estimated surgery durations for the top four surgeries by count and top surgeons were found to be statistically different in 75% of the data sets. After creating multiple linear regression models for each of the top four surgeries and surgeons performing those surgeries, the β values for each variable were compared across models. Age was found to have a minimal impact on surgery duration in all models. The binary variable indicating residents present, was found to have minimal impact as well. For the rest of the variables, consistencies were difficult to assess, making multiple linear regression an unideal method for identifying the impact of the variables investigated.On the other hand, the simulation model proved to be useful in identifying useful scheduling policies. Eight series based on real series were modeled individually. Each model was validated against reality, with 75% of durations simulated in the models not being statistically different than reality. Each of the four scheduling policies was modeled for each series and the average minutes off schedule and idle time between cases were compared across models. Adding an extra twenty minutes to each case in the existing order resulted in the lowest minutes off schedule, but significantly increased the idle time between cases. Most common surgery first did not have a consistent impact on the performance indicators. Longest estimated time first did not improve the performance indicators in the majority of the cases. Shortest estimated time first resulted in the best performance for minutes off schedule and idle time between cases in combination; therefore, we recommend this policy is employed when the scheduling process allows.