Predicting Hospital Length of Stay with Neural Networks

Critical careprovidersare facedwithresourceshortagesandmust find waysto effectively plan their resourceutilization. Neural networksprovide a newmethodfor evaluating traumapatient (and other medicalpatient) level of illness and accurately predictinga patient'slengthof stayat the critical carefacility. Backpropagation, radial-basis-function, and fuzzy ARTMAP neuralnetworksare implementedto determinethe applicability of neuralnetworksfor predictingeither injury severityor lengthof stay (or both). Neuralnetworksperformwell on this medical domainproblem. Thebackpropagationnetworksachieved the best performancefor predictinga patient's lengthof stay, butthe fuzzy ARTMAPproduced superior performancein evaluating patient's level of injury (especiallyfor the moreseverelyinjured patients). Thusa combinationof backpropagationand fuzzy ARTMAPneural networks is recommendedto produce the optimalcombined(injury severityandlengthof stay) results.

• Type: Conference
• Year: 1998
• Where: FLAIRS
• Authors: Steven Walczak, Walter E. Pofahl, Ronald J. Scorpio
• Keywords: Artificial Intelligence, FLAIRS 1998, Fuzzy Artmap Neuralnetworksare, Lengthof Stay, Severityor Lengthof Stay