What is it about?
In this paper we present a methodology to forecast the demand for intensive care unit (ICU) beds during the COVID-19 pandemic. The basic idea behind the forecast was to describe how many of the patients that were being diagnosed in a given day would require critical care in the following days and to describe how the beds that were being used would become available in the next few days. To use this simple approach, we required precise estimations of the relevant epidemiological and clinical parameters which, for the specific case of the COVID-19 could change rapidly over time. Similarly, because SARS-CoV-2 was a new virus, it involved continuous learning by medical teams. To accommodate all these short-term variations in the process, we combine the basic clinical-inventory model with a variety of Autoregressive and Machine Learning models that have the flexibility to capture these complex dynamics. The development of this model was carried out during the most critical times of the pandemic and the resulting forecasts were used by the officials in charge of decision-making, helping them to implement a progressive increase in the number of beds. Overall, ICU capacity was more than doubled in the most critical regions.
Photo by Piron Guillaume on Unsplash
Why is it important?
Unlike the typical forecasting model which is heavily trained before entering in production, in our case important costly capacity decisions were needed urgently. Furthermore, there were important data limitations to calibrate the parameters of the system. Thus, in this work we describe the challenges of implementing a forecasting system with incomplete information, little resources and almost no time to provide a viable product. Our modeling approach proved to be accurate in providing reliable estimates of the demand in times of high uncertainty, while being flexible enough to accommodate the continuous changes in the evolution of the pandemic.
Read the Original
This page is a summary of: COVID-19: Short-term forecast of ICU beds in times of crisis, PLoS ONE, January 2021, PLOS, DOI: 10.1371/journal.pone.0245272.
You can read the full text:
The following have contributed to this page