Technology in Cancer Research & Treatment (Oct 2022)

Forecasting Institutional LINAC Utilization in Response to Varying Workload

  • Srinivas Raman MD,
  • Fan Jia BASc,
  • Zhihui Liu PhD,
  • Julie Wenz MRT(T),
  • Michael Carter PhD,
  • Colleen Dickie MRT(T)(MR), MSc,
  • Fei-Fei Liu MD,
  • Daniel Letourneau PhD

DOI
https://doi.org/10.1177/15330338221123108
Journal volume & issue
Vol. 21

Abstract

Read online

Objectives Pandemics, natural disasters, and other unforeseen circumstances can cause short-term variation in radiotherapy utilization. In this study, we aim to develop a model to forecast linear accelerator (LINAC) utilization during periods of varying workloads. Methods: Using computed tomography (CT)-simulation data and the rate of new LINAC appointment bookings in the preceding week as input parameters, a multiple linear regression model to forecast LINAC utilization over a 15-working day horizon was developed and tested on institutional data. Results: Future LINAC utilization was estimated in our training dataset with a forecasting error of 3.3%, 5.9%, and 7.2% on days 5, 10, and 15, respectively. The model identified significant variations (≥5% absolute differences) in LINAC utilization with an accuracy of 69%, 62%, and 60% on days 5, 10, and 15, respectively. The results were similar in the validation dataset with forecasting errors of 3.4%, 5.3%, and 6.2% and accuracy of 67%, 60%, and 58% on days 5, 10, and 15, respectively. These results compared favorably to moving average and exponential smoothing forecasting techniques. Conclusions: The developed linear regression model was able to accurately forecast future LINAC utilization based on LINAC booking rate and CT simulation data, and has been incorporated into our institutional dashboard for broad distribution. Advances in knowledge: Our proposed linear regression model is a practical and intuitive approach to forecasting short-term LINAC utilization, which can be used for resource planning and allocation during periods with varying LINAC workloads.