Harnessing Clinical Data Streams for Nursing Workload Prediction Using Artificial Intelligence

The increasing workload in inpatient care and the ongoing shortage of skilled workers require new approaches to demand-oriented personnel planning. A machine learning approach is presented that predicts the nursing workload in minutes on a daily basis based on routine nursing and medical data. Different data sets were examined for suitable predictors. A selection of predictors consisting of administrative, nursing, diagnostic, and procedural data was determined. Two tree-based machine learning models were trained, adjusted, and evaluated. Depending on the task area, the models achieve an R2adj of 0.58 to 0.78 with an NRMSEstd between 47 and 64% of the standard deviation of the target variables. Adding clinical data sources enhances the results and enables prediction of not only the independent scope of nursing care practice, but also different task areas.