Summary: | 碩士 === 國立高雄師範大學 === 環境教育研究所 === 94 === Background: Methicillin-resistant Staphylococcus aureus (MRSA) is the most frequent cause of hospital acquired bacteremia and of surgical wound infection in the healthcare facilities. Transmission of MRSA between patients can be prevented by early identification and isolation of patients carrying MRSA. An artificial neural networks (ANN) is a predictive model that mimics the mammalian brain of learning and continuing improvement in predictive ability. The objective is to develop an ANN that will accurately predict the likelihood of MRSA colonization of a patient at the time of admission to the hospital. Methods: Two hospitals participated in this study: Hospital A (Pittsburgh, PA, USA) and Hospital B (Kaohsiung, Taiwan). Nasal cultures were taken on asymptomatic patients who admitted to the designated wards. 46 and 86 risk factors were surveyed from each patient in Hospitals A and B, respectively. Cultures results and risk factors were collected, and 75% of data were used for training and the remaining 25% were used for validation in our ANN. Results: The culture results of MRSA+/MSSA+/No Colonization in Hospital A and B were 135/239/1,126 (n=1,500) and 117/209/1,139 (n=1,465), respectively. The accuracy of ANN model was 95.2% (False +/- = 1.9%/2.9%) and 94.2% (False +/- = 4%/1.8%) in Hospital A and B, respectively, when all risk factors were used. If the number of risk factors was minimized to achieve at least 90% accuracy, only 17 and 20 risk factors were needed in Hospital A (accuracy = 90.9%, False +/- = 8.3%/0.8%) and Hospital B (accuracy = 90.5%, False +/- = 7.9%/1.6%), respectively. Conclusion: our ANN can be used to predict the likelihood of patients who carry MRSA with less than 20 risk factors and accuracy rate of >90%. The false negative rate is significantly lower than false positive in ANN prediction which can serve as a safety buffer in case of patient mis-classification.
|