Prediction of Hospital Length of Stay Using Artificial Neural Networks

With the increasing demands on hospitals and the financial constraints they face, the management of hospital beds has become a complex issue. This study proposes and designs a decision support system for predicting the length of stay (LOS) of specific patients using an artificial neural network (ANN) data mining technique. This system can serve as an effective tool for measuring hospital resource utilization. The target population of this study consists of patients diagnosed with acute myocardial infarction at Seyed al-Shohada Hospital in Urmia, Iran, over a four-year period. A total of 997 records, comprising 32,934 fields, were extracted from medical records and analyzed using MATLAB 2013. After training and testing the network with various hidden layers and learning rates, a three-layer neural network with 10 neurons was constructed. The network was trained using the optimized TRAINBR function, resulting in an acceptable mean error of 5.1 and a correlation coefficient of 0.83, leading to the development of an optimal model. Following successful testing, the findings indicated that the artificial neural network has a strong capability for predicting patient length of stay. It is worth noting that the randomness of the data used for training the network enhances the model’s accuracy and robustness. Such a model enables more efficient and effective utilization of human resources and hospital facilities.