A Temporal Fusion Transformer for Long-Term Explainable Prediction of Emergency Department Overcrowding

Emergency Departments (EDs) are a fundamental element of the Portuguese National Health Service, serving as an entry point for users with diverse and very serious medical problems. Due to the inherent characteristics of the ED, forecasting the number of patients using the services is particularly challenging. And a mismatch between the affluence and the number of medical professionals can lead to a decrease in the quality of the services provided and create problems that have repercussions for the entire hospital, with the requisition of health care workers from other departments and the postponement of surgeries. ED overcrowding is driven, in part, by non-urgent patients, that resort to emergency services despite not having a medical emergency and which represent almost half of the total number of daily patients. This paper describes a novel deep learning architecture, the Temporal Fusion Transformer, that uses calendar and time-series covariates to forecast prediction intervals and point predictions for a 4 week period. We have concluded that patient volume can be forecasted with a Mean Absolute Percentage Error (MAPE) of 5.91% for Portugal's Health Regional Areas (HRA) and a Root Mean Squared Error (RMSE) of 84.4102 people/day. The paper shows empirical evidence supporting the use of a multivariate approach with static and time-series covariates while surpassing other models commonly found in the literature.

Access paper here

Bibtex:

@inbook{Caldas_2023,
    author = "Caldas, Francisco M. and Soares, Cláudia",
    editor = {Koprinska, Irena and Mignone, Paolo and Guidotti, Riccardo and Jaroszewicz, Szymon and Fr{\"o}ning, Holger and Gullo, Francesco and Ferreira, Pedro M. and Roqueiro, Damian and Ceddia, Gaia and Nowaczyk, Slawomir and Gama, Jo{\\textasciitilde a}o and Ribeiro, Rita and Gavald{\`a}, Ricard and Masciari, Elio and Ras, Zbigniew and Ritacco, Ettore and Naretto, Francesca and Theissler, Andreas and Biecek, Przemyslaw and Verbeke, Wouter and Schiele, Gregor and Pernkopf, Franz and Blott, Michaela and Bordino, Ilaria and Danesi, Ivan Luciano and Ponti, Giovanni and Severini, Lorenzo and Appice, Annalisa and Andresini, Giuseppina and Medeiros, Ib{\'e}ria and Gra{\c{c}}a, Guilherme and Cooper, Lee and Ghazaleh, Naghmeh and Richiardi, Jonas and Saldana, Diego and Sechidis, Konstantinos and Canakoglu, Arif and Pido, Sara and Pinoli, Pietro and Bifet, Albert and Pashami, Sepideh},
    title = "A Temporal Fusion Transformer for Long-Term Explainable Prediction of Emergency Department Overcrowding",
    ISBN = "9783031236181",
    ISSN = "1865-0937",
    url = "http://dx.doi.org/10.1007/978-3-031-23618-1\\_5",
    DOI = "10.1007/978-3-031-23618-1\\_5",
    booktitle = "Machine Learning and Principles and Practice of Knowledge Discovery in Databases",
    publisher = "Springer Nature Switzerland",
    year = "2023",
    pages = "71-88",
    address = "Cham",
    abstract = "Emergency Departments (EDs) are a fundamental element of the Portuguese National Health Service, serving as an entry point for users with diverse and very serious medical problems. Due to the inherent characteristics of the ED, forecasting the number of patients using the services is particularly challenging. And a mismatch between the affluence and the number of medical professionals can lead to a decrease in the quality of the services provided and create problems that have repercussions for the entire hospital, with the requisition of health care workers from other departments and the postponement of surgeries. ED overcrowding is driven, in part, by non-urgent patients, that resort to emergency services despite not having a medical emergency and which represent almost half of the total number of daily patients. This paper describes a novel deep learning architecture, the Temporal Fusion Transformer, that uses calendar and time-series covariates to forecast prediction intervals and point predictions for a 4 week period. We have concluded that patient volume can be forecasted with a Mean Absolute Percentage Error (MAPE) of 5.91{\\%} for Portugal's Health Regional Areas (HRA) and a Root Mean Squared Error (RMSE) of 84.4102 people/day. The paper shows empirical evidence supporting the use of a multivariate approach with static and time-series covariates while surpassing other models commonly found in the literature."
}