Predicting Patient Length of Stay in a Neurosurgical Intensive Care Unit of a Large Teaching Hospital

Background: The intensive care unit (ICU) has the highest mortality and admission rates compared to other wards. Therefore, to increase the performance of hospital services, it is very important to evaluate indicators such as mortality and length of stay of patients in ICU. The present study aimed to investigate the neural network analysis method and Particle Swarm Optimization - Support Vector Machine to predict the length of stay in the neurosurgical intensive care unit. Materials and Methods: This descriptive research deals with data mining and modeling of intensive care unit processes, leading to a practical example of the application of health systems engineering knowledge, using MATLAB software. Data of 1200 patients admitted during the years 2017 to 2019 in the intensive care unit of neurosurgery. Then we evaluated all data with SVM + PSA and NCA. Results: Identifying the important features and using them has gradually reduced the LOS prediction error from 40% to 7%. Using the NCA technique makes better results for predicting ICU LOS. Conclusion: PSO + SMV in addition to NCA is a good predictor of ICU LOS screening in patients after neurosurgery and can provide more accurate prognostic factors.