Forecasting local surges in COVID-19 hospitalizations through adaptive decision tree classifiers

Abstract

During the COVID-19 pandemic, many communities across the US experienced surges in hospitalizations, which strained the local hospital capacity and affected the overall quality of care. Even when effective vaccines became available, many communities remained at high risk of surges in COVID-19-related hospitalizations due to waning immunity, low uptake of booster vaccinations, and the continual emergence of new variations of SARS-CoV-2. Some risk metrics, such as the CDC's Community Levels, were developed to predict the impact of COVID-19 on the community-level healthcare system based on routine surveillance data. However, they had limited utility as they were not routinely updated based on accumulating data and were not directly linked to specific outcomes, such as surges in COVID-19 hospitalizations beyond local capacities. Regression models could resolve these limitations, but they have limited interpretability and do not convey the reasoning behind their predictions. In this paper, we evaluated decision tree classifiers that were developed in "real-time" to predict surges in local hospitalizations due to COVID-19 between July 2020 and November 2022. These classifiers would have provided visually intuitive and interpretable decision rules for local decision-makers to understand and act upon, and by being updated weekly, would have responded to changes in the epidemic. We showed that these classifiers exhibit reasonable predictive ability with the area under the receiver operating characteristic curve (auROC) >80%. These classifiers maintained their performance temporally (i.e, over the duration of the pandemic) and spatially (i.e., across US counties). We also showed that these classifiers outperformed the CDC's Community Levels for predicting high hospital occupancy.