Evaluation of time-series models for forecasting demand for emergency health care services

Objective: To evaluate a combined set of 6 time-series models for improving the

management of short-term calls for emergency health services.

Methods: The demand for emergency health services in the province of Malaga was

analyzed between 2004 and 2008. Using standard software, we constructed and

evaluated 3 decomposition models and 3 econometric models. The models considered

summer months and atypical values, influenza cases, and number of overnight

admissions as the exogenous inputs. We compared the models using the usual

econometric tests, such as the root mean square error (RMSE), the mean absolute

percentage error (MAPE), and the maximum absolute percentage error (MaxAPE)

among others.

Results: The models had MAPEs under 5%. Autoregressive integrated moving average

(ARIMA) modeling with intervention had the lowest RMSE. Harmonic analysis had the

smallest difference between the MAPE and MaxAPE. In the validation phase, ARIMA with

intervention had the poorest fit, and harmonic analysis and ARIMA with exogenous

input had the best fits.

Conclusions: A forecast of the demand for emergency calls can be generated using 2

models simultaneously to improve short-term planning. Decomposition models and

ARIMA with intervention warn of unexpected changes, whereas ARIMA or other models

with exogenous inputs and harmony component analysis can introduce alternative

planning scenarios, improve our understanding of demand, and facilitate decisionmaking.

Implementing these models with standard software decreases the cost of this

approach in emergency services.