應用知識模型技術與專家系統方法來加強臨床藥品資訊知識庫的查詢功能

對於所有醫護人員來說，舉凡相關臨床處理的問題，皆必須在短時間內得到最好的解答，才可以爲病人的健康帶來最大的福祉。在過去，當我們要查詢一個病人所用的一個藥品或是多個藥品的資訊，我們必須大費周章的查出單一個藥品的資料，再去除不要的資訊，留下想要的資訊，再整合所有資訊，並下最後的結論。而在分秒必爭的急性處理之下，這樣的查詢方法似乎在花費許多醫療人員的決策時間。爲了改善舊有資訊系統查詢不良的問題，我們著手研究改進舊有資訊系統的不足，並進一步從我們豐富的知識中推衍出隱藏在領域中知識。在本研究中，我們將以本體論的概念，對於複雜的臨床藥品資訊進行有效的知識管理。目前我們先以抗微生物用藥及藥品交互作用為例。首先，我們先用Protégé-2000知識模型技術及TOVE本體論方法論分別建立兩個知識系統DKB (Drug Knowledge Base)和DDI (Drug Drug Interaction)。在DKB中，主要把藥品的重點放在抗感染性疾病的抗微生物用藥 (antimicrobial agents)。例如：呼吸系統感染、中樞神經感染、尿道感染等疾病的治療用藥，而開發一個藥品知識庫的雛型 (DKB prototype)。由於目前臨床藥品的使用上，最常碰到的問題就是藥品跟藥品或是藥品跟食物之間有交互作用的情形。為了加速用藥查詢速度以減少非必要的不良反應發生，我們開始發展藥品交互作用的本體論雛型 (DDI prototype)。接下來，我們以CLIPS專家系統技術進行更進一步知識推理的部份。此部分主要是以知識庫為核心而來進一步建立法則式的專家系統 (Rule Based Expert System，簡稱RBES)，進行藥品知識的推論。而目前我們所要發展的目標為一個腦膜炎治療藥品決策的專家系統，並且針對台灣所使用的抗生素，提供完整的藥品產品資訊。目前DKB研究結果包含265個classes、100個slots、12個facets、2770個instances及3147個frames。DDI目前總共有298個classes、51個slots、12個facets、278個instances及639個frames。所可以回答的問題型態包括藥物層級的問題、數字問題的問題以及雙向詢問方式等等。而RBES的法則庫中總共有117條法則。所遵循的決策原則首先是，懷疑病人是否有感染腦膜炎；第二，是詢問病人是屬於哪一個年齡層；第三，根據參考書籍 (Koda-Kimble, et al., 2002) 等，我們找出各個年齡層病人最容易感染的菌種；第四，系統回到知識庫中推理出這類病人可以用哪些首選抗生素以及次選抗生素來治療；第五，使用者可以觀看關於用藥更詳細的資料，包括用藥的所有商品資訊；最後，相關於所有用藥時的小心事項及禁忌症等等都可以藉由本系統回到知識庫中推理出相關結果。本研究的結論可從整體研究結果當中發現DKB及DDI可以有效的改善以往資訊系統查詢的缺點，也可以促使我們進一步應用臨床藥學中所有知識。我們所開發的系統不再是像以往的資訊系統，而是一個知識系統，藉由過去知識的累積，進一步推理出新的知識。在醫療的領域中，很多知識是處於無法清楚描述的情況。而藉由RBES的協助，我們可以試著利用規則的推理來解決問題，進一步協助使用者解決所遇問題。

關鍵字

臨床藥學；藥品治療學；藥品資訊；本體論；專家系統； Protege-2000 ； CLIPS

並列摘要

It is important for health professionals (HPs) to make medical decision in a short time. In the past, if we wanted to search the information about a drug or many drugs that the patient used, we had to spend lots of time in searching, filtering, and integrating the collected information. Such searching method seems to consume a lot of time. In order to improve the querying problems of those present clinical drug information databases, we try to solve problems. Furthermore, we inferred the hidden knowledge from our complicated drug treatment domain. In this study, we used the form of conceptualization of ontology to manage the knowledge of complicated clinical drug information efficiently. At present, we took the antimicrobial agents and drug-drug interaction as examples. First, we applied Protégé-2000, a knowledge modeling technique, and TOVE methodology to build drug knowledge base (DKB) and drug drug interaction (DDI). In DKB, the domain knowledge was antimicrobial agents, such as the remedy for respiratory tract infections, central nervous system infections, urinary tract infections and so on. And then we developed the DKB prototype. Moreover, there were many problems happening in drug-drug interactions or drug-food interactions. In order to enhance the speed of querying, we attempted to develop the DDI prototype. Finally, we used CLIPS, an expert system shell, to extend the utilities of built DKB and assisted us to infer the results from complicated drug therapy knowledge. DKB was the core knowledge in the rule base expert system (RBES). We built the RBES of meningitis treatment and also provided the additional drug product information in Taiwan. This prototype DKB includes 265 classes, 100 slots, 12 facets, 2770 instances and 3147 frames and there are 298 classes, 51 slots, 12 facets, 278 instances and 639 frames in the prototype DDI. The ontology can answer all preset questions such as the drug hierarchy, numerical, inverse querying questions and so forth. The rule base of RBES includes 117 rules. The inference steps of RBES is described as follows. First, we ask whether the patient is infected. Then, we need to know how old the patient is. Third, according to the textbook (Koda-Kimble, et al., 2002), we can know what kind of bacteria the patient is infected. Fourth, RBES can infer which primary and alternative treatments (antibiotics) are for the suitable patient. The RBES provides detailed drug information about each antibiotic, such as its dosage, cost and so on. Finally, RBES can display all precaution and contraindication related to the treatment. In conclusion, built DKB and DDI can improve present clinical drug information databases, and facilitate the application of clinical drug knowledge. The system is not only an information system but also a knowledge base system. It enables us to make use of the knowledge accumulated in the past, and infers the new knowledge. Because the drug knowledge domain is difficult to describe in the medical domain, there are still many questions remained to be solved. RBES depicts our knowledge using formulated rules and make correct inference to solve problems.

並列關鍵字

Clinical pharmacy ； Protege-2000 ； Expert system ； Drug information ； Drug therapeutics ； Ontology ； CLIPS

參考文獻

曾憲雄、黃國禎、江孟峰、蔡昌均及林耀聰 (2002)，「專家系統：導論、工具、應用」，文魁資訊股份有限公司，台北市民生東路一段43號9樓

Google Scholar

戴汝為 (2003) 「人工智慧」，五南圖書出版股份有限公司，楊榮川，臺北市大安區106和平東路二段339號4樓，http://www.wunan.com.tw.

Google Scholar

廖榮貴工作室－黃聖豐、陳豪人(2003) 「資料庫系統理論與實務」，文魁資訊股份有限公司，台北市重慶南路一段69號，http://www.kingsinfo.com.tw/shop.asp.

Google Scholar

Alani, H. (2003), “TGViz V1.2 tab”, Intelligence Agents, Multimedia Electronics and Computer Science Department., University of Southampton, http://protege.stanford.edu.