Forecasting daily emergency department arrivals using high-dimensional multivariate data: a feature selection approach

Abstract

Background and objective Emergency Department (ED) overcrowding is a chronic international issue that is associated with adverse treatment outcomes. Accurate forecasts of future service demand would enable intelligent resource allocation that could alleviate the problem. There has been continued academic interest in ED forecasting but the number of used explanatory variables has been low, limited mainly to calendar and weather variables. In this study we investigate whether predictive accuracy of next day arrivals could be enhanced using high number of potentially relevant explanatory variables and document two feature selection processes that aim to identify which subset of variables is associated with number of next day arrivals. Performance of such predictions over longer horizons is also shown. Methods We extracted numbers of total daily arrivals from Tampere University Hospital ED between the time period of June 1, 2015 and June 19, 2019. 158 potential explanatory variables were collected from multiple data sources consisting not only of weather and calendar variables but also an extensive list of local public events, numbers of website visits to two hospital domains, numbers of available hospital beds in 33 local hospitals or health centres and Google trends searches for the ED. We used two feature selection processes: Simulated Annealing (SA) and Floating Search (FS) with Recursive Least Squares (RLS) and Least Mean Squares (LMS). Performance of these approaches was compared against autoregressive integrated moving average (ARIMA), regression with ARIMA errors (ARIMAX) and Random Forest (RF). Mean Absolute Percentage Error (MAPE) was used as the main error metric. Results Calendar variables, load of secondary care facilities and local public events were dominant in the identified predictive features. RLS-SA and RLS-FA provided slightly better accuracy compared ARIMA. ARIMAX was the most accurate model but the difference between RLS-SA and RLS-FA was not statistically significant. Conclusions Our study provides new insight into potential underlying factors associated with number of next day presentations. It also suggests that predictive accuracy of next day arrivals can be increased using high-dimensional feature selection approach when compared to both univariate and nonfiltered high-dimensional approach. Performance over multiple horizons was similar with a gradual decline for longer horizons. However, outperforming ARIMAX remains a challenge when working with daily data. Future work should focus on enhancing the feature selection mechanism, investigating its applicability to other domains and in identifying other potentially relevant explanatory variables.

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

“…Most focus is on the use of classical statistical tools for temporal linear regression such as moving averages [28], and their many extensions, namely ARIMA, SARIMA or VARIMA [3,34,40,7]. In recent years, with the advent of machine learning, newer studies have been conducted that use neural networks [21,43], or otherwise other machine learning techniques to tackle the same problem [33,29,37]. From the use of Feed-forward Neural Networks [21,29], to Recurrent Neural Networks [22,15], 1-D Convolution Neural Networks [35], and later to Long Short-Term Unit (LSTM) [36,15], there has been a constant advance in the field, from linear models to deep neural network models.…”

“…In the literature, we found only one example that used a Temporal Fusion Transformer model to predict Emergency Department (ED) volume in one hospital for one day ahead [31]. While not being the only work that performed only daily predictions [33,37], we find that a longer forecasting window produces increased value for hospital management and poses a different challenge from a machine learning perspective, as seasonal fluctuation needs to be fully represented, and common forecasting models tend to decrease in predictive quality as the forecast period becomes wider.…”

Enhancing emergency department patient arrival forecasting: a study using feature engineering and advanced machine learning algorithms

“…can be classified into four groups: time series models, regression models, machine learning algorithms (ML), and hybrid methods. Among the time series models, ARIMA [1], [18] and its variants ARIMAX [19], [41], SARIMA [5], [22], and SARIMAX [3], [42] have been widely used. Naive models, Snaive [41], [43], ETS [21], [44], and ES models [5], [42] are also reported, with emphasis on the seasonal HW model [1], [17].…”

A multi-granular stacked regression for forecasting long-term demand in Emergency Departments

“…Previous studies have mostly focussed on predicting short-term ED attendances [ 4 – 12 ] using past attendances [ 4 , 5 , 7 , 8 , 10 – 14 ], calendar [ 5 , 7 , 9 , 10 , 14 ] and meteorological variables [ 5 , 9 , 14 ]. The most common techniques employed are Multiple Linear Regression [ 7 , 9 , 11 , 15 ], Autoregressive Integrated Moving Average (ARMIA) and variants [ 4 , 5 , 7 , 8 , 12 , 13 ], Exponential Smoothing [ 7 , 8 , 11 ] and, more recently, Machine Learning (ML) algorithms [ 5 , 8 , 10 , 12 , 13 , 16 ]. Cohort studies have found that measures of social deprivation and co-morbidities are predictive of ED attendances [ 15 , 17 , 18 ] and low General Practice (GP) attendance is associated with low ED attendance [ 18 ].…”