Development and validation of a backpropagation neural network model for predicting nursing unit staffing needs: A...

Nurse staffing requires a strategic approach that aligns staffing levels with actual patient needs. There is a need to explore staffing frameworks that both enhance patient safety and maximize workforce efficiency.

This study aimed to integrate efficiency evaluation with deep learning techniques to develop a prediction model for optimizing nurse staffing in clinical units.

A cross-sectional study.

The study was conducted in a large comprehensive tertiary public hospital in Zhejiang Province, China.

Data from fifty-five nursing units were collected to develop and validate the demand prediction model. Thirteen units with optimal nurse staffing efficiency formed the construction group for training the model, and forty-two units with suboptimal efficiency were used as the prediction group to assess the model's effectiveness in improving staffing.

The number of nurses on duty served as the predicted outcome, while predictors included the number of actual open beds, costs of nursing staff, number of nursing hours, number of diagnosis-related groups, case-mix index, total number of actual occupied bed days, bed utilization rate, nursing quality assessment results, and nursing adverse events. This study retrospectively analyzed data from 55 nursing units between January and December 2023. A backpropagation neural network model was developed using data from 13 units with optimal nurse staffing efficiency to predict nurse staffing demands. Model accuracy was evaluated via mean squared error, Pearson correlation coefficient, and Bland-Altman analysis, while the coefficient of determination (R-square) assessed goodness-of-fit. The validated model was then applied to predict staffing needs for units with suboptimal efficiency. Data envelopment analysis simulated pre- and post-prediction staffing efficiency comparisons to verify the model's practical effectiveness.

The demand prediction model for nursing units staffing achieved high predictive accuracy (R-square = 0.97, mean square error = 0.1674) with no systematic staffing bias (mean: +0.0736 nurses, 95 % Limits of Agreements: -0.7308 to +0.8779; P = 0.3693). Bootstrap-validated normal errors (skewness = -0.04, kurtosis = 3.18) supported parametric reliability. When simulated to implemented, the model improved technical efficiency in 91.67 % of months and optimized nurse-to-demand ratios (100 % scale efficiency), demonstrating its capacity to balance staffing precision with operational flexibility using existing resources.

This study proposes a hybrid model to predict nurse staffing by linking operational efficiency with staffing adjustments. Preliminary results show effective optimization through scale inputs, aiding managerial strategies like staff reallocation. Multi-center validation and integration of external factors are needed for broader application, advancing data-driven nursing workforce management. Social media abstract: Our research combines deep learning and efficiency evaluation to develop intelligent models that accurately predict nurse staffing demands. This may be able to reduce temporary labour costs and provide innovative solutions for patient safety and healthcare resource optimization.