Capturing Heterogeneous Patient Needs in Hospitals: A Machine Learning Approach to Predict Staffing Needs

The global nursing shortage makes the efficient allocation of limited nursing resources increasingly critical. Current staff planning often relies on homogeneous care-demand assumptions, treating all patients as having identical care demand, although accounting for heterogeneity in patient care demand may enable more accurate and efficient staff planning.

This study aims to improve nurse scheduling by forecasting heterogeneous care demand, accounting for variability across patient categories, and evaluating its potential benefits compared with homogeneous care demand assumptions.

We use patient-level data from a German hospital and classify individuals into heterogeneous patient categories with distinct care demand. To forecast category-specific care demand, we apply 2 state-of-the-art machine learning models: LightGBM and TSMixer. Forecasts are generated for a 1-month horizon and subsequently translated into staffing needs.

Our findings show that forecasting heterogeneous care demand substantially improves the accuracy of staffing need estimation. Compared with a naive yet realistic baseline, heterogeneous forecasts enhance nurse scheduling efficiency by 34%. Moreover, this approach outperforms the best homogeneous demand estimation method by 14%. In addition, the heterogeneous approach provides a closer approximation to the historical monthly reference nursing staff needs by reducing both overstaffing and understaffing. The results highlight the value of integrating patient heterogeneity into staffing prediction processes.

Accurately forecasting heterogeneous care demand can increase the efficiency and reliability of nurse scheduling and outperform approaches based on homogeneous care demand.