Predicting use of intensive care units during the COVID-19 pandemic

With the prevalence of the SARS-CoV-2 pandemic, sudden planning needs emerged in intensive care units (ICUs) in many countries, particularly Chile. Chile was chosen for this study due to its diverse geographical regions, which presented unique challenges in managing ICU capacity during the pandemic. The researchers’ understanding of the local healthcare system provided a significant advantage in accurately analyzing these challenges. In ICUs, the most severe COVID-19 patients require specialized treatment, stressing operational-level decision-making. Understanding patient arrival dynamics became essential to predicting the additional ICU beds needed. We propose ten approaches using machine learning and classical time series models to estimate the required beds, setting upper and lower bounds. Evaluating the predictions with 2020 and 2021 data from three representative regions produced lower errors in the largest region. The low errors produced by the Holt-Winters model suggest that the data have seasonal and trend characteristics. Specifically, Holt-Winters achieved a mean absolute error of 0.00 in the smallest region and 9.26 in the largest region, demonstrating its effectiveness in forecasting ICU demand. Although the models were evaluated in only three regions, extending them to other situations would require training with local data.