Use of Decision Trees and Neural Networks in Predicting Patient Disposition using Clinical Decision Support Systems

Triage Applications have failed to provide reliable decision support on proper disposition of the patient for appropriate patient care. This lack of a reliable triage application in Emergency Department (ED) has significantly affected the patient care. A feasible solution to this issue is to utilize predictive techniques such as Artificial Neural Networks to create support process for triage that could be operated by a nurse. The output of this decision support process is going to be reliable disposition of the patient for appropriate patient care. Although medical specialists have developed several clinical decision support systems (CDSSs) for triage assistance by using their experience and knowledge, there is not yet an appropriate and positive evaluation about a significant impact of CDSSs directly built by specialists. An alternative approach to create CDSSs for triage assistance consists of utilizing algorithms that are able to directly learn the model from knowledge. Feed Forward and Back Propagation algorithms are an ideal match for these kinds of studies. Some of the important knowledge variables such as Chief Complaint, ED Acuity Assessment, ED Responsiveness Assessment, Systolic Blood Pressure, Diastolic Blood Pressure etc., will play a vital role in the scoring of classification and contribute to the accuracy of triage disposition. In the sample study, a sample set of test patients who were suffering from chest pain were taken into consideration. The input variables are chest pain categories (1–7) scored based on knowledge variables and pain score (1–10) scored based on patient physical condition examined when arrived at ED. In this study, we have attempted to implement the feed forward-back propagation algorithm. After examining various training duration periods and regularly retaining with most recent patient information, the study has achieved 80% accuracy of the triage disposition. The accuracy is being continuously improving after updating and incorporating most recent data. Finally, the study has achieved better performance results than similar kind of researches carried out in the area of triage disposition using Artificial Neural Networks.