Time Series forecasting of daily ICU admissions. SARIMA model development and validation.

Background Accurate forecasting of intensive care unit (ICU) admissions is critical to optimize resource allocation, improve patients’ outcomes, and maintain hospital readiness. Time series models, particularly Auto-Regressive Integrated Moving-Average (ARIMA), are increasingly used in healthcare settings for short-term demand forecasting but remain underutilized for ICU admissions. Objectives The primary objective was to identify the most suitable ARIMA or SARIMA model for forecasting daily ICU admissions. Secondary objectives included generating forecasts for a one-month period and comparing the predicted values to actual ICU admission counts to evaluate model performance. Methods This was a retrospective observational time series analysis of ICU admissions. The study used 28 months (January 2023–April 2025) of retrospective daily ICU admission data to fit the model, and admissions from May 2025 for validation. Model selection was performed using the auto.arima function in R, optimizing for the lowest Akaike Information Criterion (AIC), with several model diagnostics. Actual admissions were inspected within the 80% and 95% prediction intervals. Results The best-fitting model was ARIMA (0, 1, 1) (2, 0, 1)7, indicating weekly seasonality, with AIC (3340.4) and showed significant improvement in predictive accuracy over the naïve model, with RMSE and MAE ratios of 0.8 and Diebold-Mariano test p value < 0.0001. The empirical 95% and 80% prediction interval coverage was 1 and 0.77 respectively. Conclusion ARIMA models, particularly those accounting for seasonal patterns, can reliably forecast daily ICU admissions with good short-term accuracy. Integrating such models into hospital operations may improve readiness, reduce overcrowding, and support proactive critical care planning. Slight overestimation observed in predictions may be preferable to underestimation, which carries greater clinical risk.