Using Artificial Neural Networks to Predict In-Hospital Cardiac Arrest Patients at a Medical Centre in Taipei

• Main Authors: Wei-Chih Huang, 黃威智
• Other Authors: Der-Ming Liou
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/fm9sjr

碩士 === 國立陽明大學 === 生物醫學資訊研究所 === 105 === In-Hospital Cardiac Arrest (IHCA) threaten life of the inpatients, cause serious effect to patient safety, quality of inpatients care and hospital service. Health providers must identify the signs of IHCA early to avoid the occurrence of IHCA. The aim of this study is to identify significant associations between the occurrence of IHCA and the essence of patient care provided by nurses and other professionals 8, 24, 48 hours before an IHCA occurs or during the admission, and develop a backpropagation neural network-based prediction model for IHCA in patients hospitalized in the general ward. Materials and Methods: This study utilized a cohort of adult patients hospitalized from July 1st, 2014 to June 30th, 2016 from a medical center in Taipei, Taiwan. Using five phase to define case and control group, and applied t-test to assess the statistical differences between inpatient’s age in each group by using MS SQL and R programming language. This study used fisher’s exact test to compute associations between nursing interventions and the occurrence of IHCA. The important features were also discussed with medical expertise. A backpropagation neural network was used to build a model to predict IHCA events by using R programming language. The performance of the model was validated by used 5-fold cross validation, ROC curve, confusion matrix, and AUC. Results: Total number of case group were 72, and 288 for control group. The results in the fourth different timing comparison of fisher’s exact test showed that 48 hours before IHCA events could present most significant association which was also considered by medical expertise. Whether performing the detection central venous pressure, degree of peripheral blood oxygen saturation, changes in neural function, seniors group have significant association with high risk of IHCA. After using 5-fold cross validation to train backpropagation neural network-based prediction model for IHCA events, overall sensitivity of the model was 85%, with a threshold of 0.25 and AUC of 0.72. Conclusion: The cut point of admission can be used as 48 hours before IHCA events that have 4 of significant associations between nursing interventions and occurrence of IHCA (P<0.05). Overall, the backpropagation neural network is worth to predict IHCA event. Furthermore, this result can be out-performed an existing model to discover IHCA risk factors and be implemented in a real-time system in order to readily predict IHCA events.