Forecasting Microbiology Laboratory Test Volumes using Time Series Models

Hospitals find themselves in an increasingly uncertain financial situation. A key reason for this being the increase in patient related costs. Specifically laboratory costs stand out after having increased at a compound annual growth rate of 8.54% in the last decade. This study addresses the predictive capabilities and comparative performance of various time series forecasting models (SARIMAX and Prophet), within a major Dutch hospital in order to aid in the forecasting ability of microbiology laboratory volumes. The findings of the study reveal that SARIMAX and Prophet both exhibit comparable predictive efficacy in forecasting microbiology laboratory test volumes. both models demonstrate a cluster with satisfactory performance, as the error metrics fall below the designated thresholds. Satisfactory being phrased as a Symmetric Mean Absolute Percentage Error (SMAPE) under 30%, and a Mean Absolute Scaled Error (MASE) below 1.00. Nevertheless, the presence of several outliers suggests that SARIMAX and Prophet may not be optimal fits for certain datasets.