Observations on Surgical Demand Time Series

Abstract

Background Surgical scheduling is complicated by both naturally occurring and human-induced variability in the demand for surgical services. Surgical demand time series are decomposed into periodic, lagged, and linear trends with frequent occurrences of nonconstant variations in mean and variance. The authors used time series methods to model surgical demand time series in order to improve the scheduling of scarce surgical resources. Methods With institutional approval, the authors studied 47,752 surgeries undertaken at a large academic medical center. They initially extracted periodic information from the time series using two frequency domain techniques: the harmonic F test and the multitaper test. They subsequently extracted lagged (correlated) behavior using a seasonal autoregressive integrated moving average model. Finally, they used moving variance filters on the residuals to identify variance in the time series that coincided with major US holidays. Results Linear terms such as periodic cycles, trends, and daily and weekly lags explained 80% of the variance in the raw time series. In the residuals, the authors used moving variance filters to detect nonlinear variance artifacts that correlated with surgical activities on specific US holidays. Conclusions After extracting linear terms, the remaining variance was attributable to a combination of nonlinear and unexplained random events. The authors used the term holiday variance to describe a specific nonlinear disturbance in surgical demand attributable to statutory US holidays. Resolving these holiday variances may assist in management and scheduling of scarce surgical personnel and resources.