Forecasting emergency department admissions as a tool for improved health service
Résumé
Emergency departments (ED) have difficulties coping with the rise of patients during seasonal epidemic crises. Particularly, during winter time, the influence of patients suffering from gastroenteritis, influenza and bronchiolitis lead to overcrowded emergency departments, and ultimately to patient rejection to other hospitals in the territory. The aim of the study is twofold, potential correlations between epidemic disease and environmental factors are sought and epidemics are then forecasted for classes of pathologies using the different correlation parameters that are found in the first part. The forecasting model serves as a mechanism to predict the incidence rate in ED and acts as the basis for a scheduling strategy to manage emergency departments. Historical data from various hospitals in the Rhone-Alpes region (France) as well as epidemic data collected from generalist physicians are used in the study. Environmental data of air pollutants and meteorological data are geographically matched with ED and used to extract potential correlation factors between them and the epidemics. These detailed historical meteorological data are obtained from the French national meteorological service. Thereafter, epidemic diseases of interest are modeled using a discretized compartmental epidemiological model. The epidemiological SEIR model allows us to track the evolution of the epidemics in four different categories of population (susceptible, exposed, infected and recovered). Additionally, a time series regression model is used to predict the time-varying infection rates of the epidemic model. Furthermore, a study on the number of individuals entering emergency departments following infections from the pathologies under analysis is performed. This last information allows us to study the effects of different epidemics on ED overcrowding and deploy a strategy to reduce waiting time of the patients in the ED. Waiting time reduction is done via the study of the interaction of the human resources of the ED. A scheduling strategy is then devised to organize the working hours of the different resources (physicians and nurses) according to the predictions made by the epidemic model under various constraints. Considering the multi-factor interaction of epidemics, we are able to devise an epidemic model that forecasts the infection peaks and use it to dynamically act changes in a health institution and thus reduce overcrowding during epidemic seasons. Under the hypotheses defined for the scheduling problem, the model gives us the optimal distribution of shifts for ED employees in front of stochastic demand patterns. The solution is obtained in only a few seconds for over 100 scenarios and the overall waiting time for the patients is significantly reduced compared to the solution obtained by fixing the shifts to what is currently practiced. Once the different factors affecting epidemics are determined, the epidemic model coupled with a statistical time series regression model allows us to get an accurate weekly forecast of the epidemics. As epidemics have a strong impact on overcrowding ED during the winter seasons, the forecast also forms the basis for an overcrowding reduction strategy that takes into account the key resources in the ED and reacts dynamically to changes in the estimated incidence.