Forecasting COVID-19 Cases Based on Social Distancing in Maryland, USA: A Time-Series Approach.

Abstract

Our objective is to forecast the number of coronavirus disease 2019 (COVID-19) cases in the state of Maryland, United States, using transfer function time series (TS) models based on a Social Distancing Index (SDI) and determine how their parameters relate to the pandemic mechanics. A moving window of 2 mo was used to train the transfer function TS model that was then tested on the next week data. After accounting for a secular trend and weekly cycle of the SDI, a high correlation was documented between it and the daily caseload 9 days later. Similar patterns were also observed on the daily COVID-19 cases and incorporated in our models. In most cases, the proposed models provide a reasonable performance that was, on average, moderately better than that delivered by TS models based only on previous observations. The model coefficients associated with the SDI were statistically significant for most of the training/test sets. Our proposed models that incorporate SDI can forecast the number of COVID-19 cases in a region. Their parameters have real-life interpretations and, hence, can help understand the inner workings of the epidemic. The methods detailed here can help local health governments and other agencies adjust their response to the epidemic.