Weather information improves a predictive model of emergency department arrivals

Variable emergency department (ED) volumes contribute to staffing challenges. Weather may impact ED arrival patterns. Enhancing prediction of arrivals based on weather conditions may be useful for operational planning, including adaptation to climate change. In this retrospective analysis of ED arrivals from an urban teaching hospital in the Northeast US from 2010 to 2019, two linear regression models were developed to predict daily arrival totals. The base model utilized calendar variables and a 28-day rolling average of daily arrival totals. The second model added weather variables. Models were tested for overfitting using training and test datasets, and residuals were plotted. Addition of weather variables to the model increased adjusted R-squared (0.418 to 0.483). Effect estimates and 95% confidence intervals for predictor variables in the final model were scaled to the mean daily arrival total (152.7) to facilitate interpretation. Monday was predictive of the most arrivals (+ 23.1%, CI + 22.0% to + 24.2%), relative to Sunday (reference). Holidays were associated with fewer arrivals (- 14.2%, CI - 15.9% to - 12.4%). Inclement weather was associated with fewer arrivals. Increases in the relative maximum daily temperature were associated with increases in arrivals. The effect of temperature varied across seasons. Inclusion of weather variables improved predictive value of a model of daily ED arrivals in this single-site, retrospective analysis. Rain, snow, and wind were associated with reduced arrivals, while warmer temperatures relative to historical averages were associated with increased arrivals. Inclusion of weather variables into models of ED utilization may support improved operational decision-making.