144 Time Series Analysis and Forecasting Daily Emergency Department Visits Utilizing Facebook's Prophet Method

Abstract

Emergency department (ED) boarding and crowding have significant implications for patient care. The capability of being able to model ED volumes to optimize staff and hospital resource allocation may lead to significant positive operational and patient care outcomes. The number of daily visits in 24 hours can often be challenging to predict, but over the last few decades, several studies have attempted to develop and deploy several forecasting models including exponential smoothing, and autoregressive integrated moving average models with varying levels of success. The analysis and identification of patterns and trends in large scale data are vital in other industries, such as social media, where several innovative methodologies have been employed to assess pattern recognition and trends. Facebook® uses a novel Bayesian forecasting model known as Prophet to decompose time series factors identifying patterns. This study leveraged this innovative methodology to identify patterns of ED daily visits with the hope of predicting future number of daily ED visits. Data were collected from a single urban community-based hospital in the New York Metropolitan area from January 2016 to December 2018. A time series analysis and decomposition were performed on daily visits using Prophet, which is an open source forecasting time series model based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality as well as holiday effects. The model trends were revealed and used for predicting daily ED visits for the first two weeks of January 2019. The Facebook Prophet method identified two distinct seasonal patterns. The resulting dataframe contained fourteen days of forecasted data with corresponding 95% confidence intervals. The model appears to provide insight into the general trend and seasonality in the time series data. Not surprisingly, the busiest day of the week is a Monday and the busiest time of the year is during the late winter season, such as February and March. A quick comparison with the actual visits from the first two weeks in 2019 shows that its predictions are relatively reasonable. The result of our study suggests that the Facebook Prophet model can be used to provide reasonably accurate forecasts and predictions in daily ED visits. However, the limitation was that the data was from a single site with only three years of data. Building on this single site study, future multi-site prospective longitudinal work can assess if such a methodological approach can provide forecasting information useful for departmental administration as well as validation in other hospital sites.View Large Image Figure ViewerDownload Hi-res image Download (PPT)