COVIDNearTerm: A simple method to forecast COVID-19 hospitalizations

Abstract

Introduction: COVID-19 has caused tremendous death and suffering since it first emerged in 2019. Soon after its emergence, models were developed to help predict the course of various disease metrics, and these models have been relied upon to help guide public health policy.

Methods: Here we present a method called COVIDNearTerm to "forecast" hospitalizations in the short term, two to four weeks from the time of prediction. COVIDNearTerm is based on an autoregressive model and utilizes a parametric bootstrap approach to make predictions. It is easy to use as it requires only previous hospitalization data, and there is an open-source R package that implements the algorithm. We evaluated COVIDNearTerm on San Francisco Bay Area hospitalizations and compared it to models from the California COVID Assessment Tool (CalCAT).

Results: We found that COVIDNearTerm predictions were more accurate than the CalCAT ensemble predictions for all comparisons and any CalCAT component for a majority of comparisons. For instance, at the county level our 14-day hospitalization median absolute percentage errors ranged from 16 to 36%. For those same comparisons, the CalCAT ensemble errors were between 30 and 59%.

Conclusion: COVIDNearTerm is a simple and useful tool for predicting near-term COVID-19 hospitalizations.

Keywords: COVID-19; SARS-CoV-2; forecasting; hospitalization; prediction.

Fig. 1.

Demonstration of how COVIDNearTerm works. The black line depicts hospitalizations in Santa Clara County starting on May 4, 2020. The red lines represent 100 possible paths as predicted by the COVIDNearTerm model starting from October 1, 2020. Each path is for 28 days, and the prediction for a particular day, the median of the paths, is shown in green.

Fig. 2.

Median absolute percentage error (MedAPE) by county as a function of days used in weighting for 14-day predictions. The symbol E is for equal (black), U is for unweighted (red) and T is for triangular (blue).

Fig. 3.

Median absolute percentage error by county for 14 days, 21 days and 28 days. Here 14 days is on the left, 21 days is in the middle, and 28 days is on the right. The models from bottom to top are COVIDNearterm (red), CalCAT Ensemble (brown), COVID Act Now (orange), Columbia (yellow), JHU IDDG (green), LEMMA (cyan), Simple Growth (gray), Stanford (blue), UCLA MLL (pink), UCSB (purple), and UCSD-COVIDReadi (black).

Fig. 4.

14-day predictions for multiple methods. The lines are for truth (black), COVIDNearTerm (red), CalCAT Ensemble (brown), LEMMA (cyan), and Simple Growth (gray). Note that Simple Growth was utilized by CalCAT starting only on December 8, 2020.