Construction and Validation of Artificial Neural Network Model Suggesting Nursing Diagnosis

There are challenges involving human resource management, as the selection and evaluation processes for nursing diagnostic labels are time-consuming, resulting in an excessive workload. This, in turn, can lead to insufficient attention being given to patients' medical issues. As a proof of concept, to solve challenges related to nursing diagnoses, we developed an artificial neural network model using progress records and evaluated its performance. Specifically, datasets were obtained from progress record data from the critical care department system in Japan between 2014 and 2019 and the corresponding nursing diagnosis data from electronic medical records. The model was trained, and its performance was evaluated. We compared several methods for vectorizing progress records and evaluated performance with and without oversampling for imbalanced data. We used a naive Bayes classifier for comparison. The model using term frequency–inverse document frequency achieved the highest values for both accuracy and the area under the precision-recall curve across all target nursing diagnoses (accuracy = 0.705–0.911; area under the precision-recall curve = 0.387–0.929). The artificial neural network model outperformed the naive Bayes classifier in both accuracy and area under the precision-recall curve, which indicated its superiority as a classifier.