Ensemble-based methods for forecasting census in hospital units.

Area of Science:

• Healthcare Operations Research
• Biostatistics
• Medical Informatics

Background:

• Accurate hospital census forecasting is crucial for effective resource allocation.
• Existing forecasting methods often fail to leverage patient-specific data.
• This limitation impacts the precision of hospital capacity planning.

Purpose of the Study:

• To develop and evaluate an ensemble-based methodology for hospital census forecasting.
• To integrate both temporal trends and patient-specific information into the forecasting model.
• To improve the accuracy and precision of short-term census predictions.

Main Methods:

• An ensemble model combining a Poisson Autoregressive model for arrivals and logistic regression for departures.
• Incorporation of patient-specific baseline and time-varying covariates.
• Application to neonatal intensive care unit (NICU) data for validation.

Main Results:

• The proposed model demonstrated statistically significant improvements in prediction accuracy for 3, 5, and 7-day census forecasts.
• Increased precision of forecasts was observed compared to models ignoring patient-specific information.
• The ensemble approach effectively utilized available patient data.

Conclusions:

• Forecasting models incorporating patient-specific data offer substantial improvements in census prediction accuracy.
• Leveraging typically available patient information enhances the utility of forecasting models.
• This methodology provides a robust framework for optimizing hospital resource management.