在台灣,於鼻中膈鼻道成型術使用預防性抗生素已是相當普遍。抗生素雖能成功地治療細菌感染,但細菌本身也會發展出各種對抗抗生素的機制。造成無藥可醫的困境。在美國抗生素的花費就占了20%以上(張峰義等,2005),而預防性抗生素在醫院又占約30%用量。在台灣各地的醫院之外科與骨科於清淨傷口手術預防性抗生素使用,有發現很多的預防性抗生素並無依照實證臨床指引來執行。因此,如何在術前發揮抗生素的最大效用成了非常重要的議題。 本研究以全民健保資料庫2010年接受鼻中隔鼻道成型術之病患為研究對象,彙整影響該手術預防性抗生素使用合適性之相關因子,運用倒傳遞類神經網路、支援向量機、粒子群演算法、基因邏輯斯模式以及相互結合之人工智慧方法,建構六個抗生素使用合適性之預測模型,再以倒傳遞類神經網路、粒子群演算法及基因邏輯斯模式之權重做為預設權重導入案例式推理系統,設計一抗生素使用合適性之三個評估系統。 本研究之研究結果顯示,在預測模型方面,粒子群演算法結合支援向量機有較高之預測準確度,達99.93%。而評估系統方面,倒傳遞類神經網路結合案例式推理系統之評估績效為最佳,其系統準確度為85.30%,平均相似度為96.98%,ROC曲線下面積為0.8865。本研究之結果將可提供醫師及相關醫療單位在臨床使用抗生素時之輔助評估參考依據。
The setomeatoplasty use preventive antibiotic prescriptions are general in Taiwan. Although antibiotic can success to treatment disease of bacterial infection, bacterium can also growth any mechanisms to fight the antibiotic, made no medicine to treat this kind of bacterium predicament. The cost of antibiotics in the United States accounted for more than 20% (Zhang Fengyi et al, 2005).The preventive antibiotics in the hospital account for approximately 30% of consumption. So, how to use antibiotic appropriate is a serious issue. This study is major to investigate setomeatoplasty of preventive antibiotic prescriptions appropriate use. From the claim data of the Bureau of National Health Insurance in 2010 were analyzed. This study construct the Back Propagation Neural Network, Support Vector Machine combine, Particle Swarm Optimization, Genetic Logistic Algorithms and combined to six forecast models. Also use Back Propagation Neural Network, Particle Swarm Optimization, and Genetic Logistic Algorithms combine with Case-Based Reasoning, design three evaluate models to evaluate antibiotic appropriate. This study’s results show that Particle Swarm Optimization combine with Support Vector Machine has 99.93% high accuracy in forecast model. In evaluate models, Back Propagation Neural Network combine with Case-Based Reasoning has 85.30% high accuracy, 96.98% average similarity and ROC area at 0.8865. So, this study’s conclusion can provide for doctor to evaluate in the clinical.