8th Eurasian Congress on Emergency Medicine & 18th Turkish Congress on Emergency Medicine & 17th WINFOCUS World Congress, Antalya, Turkey, 01 December 2022, pp.420-421
OBJECTIVE: Increasing patient density causes delays in the emergency patient management process in Emergency
Departments (ED), negatively affecting patient satisfaction and service quality. Long stay in ED is an important problem
especially in crowded EDs. Machine learning and artificial intelligence applications are used to solve this problem. These
applications enable to reach the correct diagnosis quickly, to manage the patient circulation per unit area and unit time
effectively and objectively.The aim of our study is to predict the hospitalization and discharge requirements and management
process through machine learning modeling in the evaluation of patients admitted to the ED.
MATERIAL and METHODS:The patients aged 18 years and older classified as category 1 admitted to the ED were included in
the study. From machine learning techniques to create predictive models with clinical data; A prediction model was created
using Logistic Regression, Artificial Neural Network (ANN), XGBoost, Support-vector clustering (SVC), Random Forest and
Gaussian Naive Bayes. In the analysis of the data set, 20 independent variables affecting the result were determined by
applying Normality Test, Correlation Analysis and Binary Logistic Regression Analysis. As a result of the correlation analysis,
those above threshold=0.6 were eliminated and input variables were determined.
RESULT: The study was conducted on 1086 patients. Of the patients, 527 were discharged (49%), and 559 were hospitalized
(51%). For machine learning, 70% (n=760) of this dataset was used for training and the remaining 30% (n=326) for testing.
Evaluation was made according to the heat map showing the correlation of the best variables. From machine learning
techniques; ROC indexes of Logistic Regression, Artificial Neural Network (ANN), XGBoost, Support-vector clustering (SVC),
Random Forest and Gaussian Naive Bayes algorithms, respectively; 0.77, 0.73, 0.75, 0.73, 0.69 and 0.59 (Figure 1). The results
related to the algorithms used are given in Table 1. According to the Confusion Matrix, 129 of the 158 actually discharged
patients were predicted correctly. In this case, the success of correctly estimating the patients who were actually discharged
CONCLUSION:DISCUSSION: The primary purpose of machine learning is to create models that can train themselves to
improve and detect complex patterns and produce solutions to new problems based on historical data. Due to the high
patient load and the chaotic environment, errors in hospitalization and discharge should be minimized in EDs. In order to
contribute to this situation, machine learning algorithms should be used in addition to clinical evaluations