目的：心臟血管內科門診處方向來有高的潛在藥物交互作用發生率，一旦發生藥物交互作用，將使預期療效無法達到，甚或引發其他疾病導致死亡。本研究期以決策樹技術發掘影響藥潛在物交互作用有意義的資訊，俾利資訊科技建置有效的檢核提示系統，以提升病人的用藥安全。方法：2004 年全民健保研究資料庫系統抽樣檔門診處方及 2004  年衛生署藥物交互作用資料庫為研究資料。以 Clementine7.2 建 立  C5.0  及  CART 決策樹演算法分類規則。結果：以心臟血管內科藥物潛在交互作用的發生率 19%為最高，主要族群年齡為 65~74 歲(26.8%)及 75 歲以上(24.8%)。就心臟血管內科發生率而言，1 至 5 級分別為 12.9%、51.4%、4.7%、14.9%及 16.1%， 可能具致死性或造成永久性傷害及導致病情惡化嚴重度為 1 及 2 級共占了 64.3%。 潛在藥物交互作用組合中以  digitalis  glycosides  與   loop   diuretics  的組合發生 率28.16%為最多。資料探勘決策樹演算法 C5.0  的藥物交互作用預測正確率較 CART 為高，C5.0 分類模式的最重要預測變數為主診斷，反應出主診斷病況在潛在藥物交互作用上所扮演的重要角色。建議醫院成立臨床專家小組，建立客製化的藥物交互作用資料庫，以共同促進病人用藥的安全。
 

OBJECTIVE: Cardiological prescriptions are with high risk of potential drug-drug interactions (DDIs). DDIs will enable pharmacotherapeutic failure, and be associated with morbidity and mortality. The objective of the  present study was to build a decision tree to determine the important factors for the effective DDIs screening software development. METHODS: The 2004 systemic sampling claims data  of  the  Bureau of National Health Insurance (BNHI) from hospitals throughout Taiwan which contains ambulatory prescriptions, therapeutic and  registration files were the data source. The 2004 DDIs  Database developed by the Department of Health was used to select potentially harmful drug  combinations in the outpatient setting. Clementine 7.2 was used for the classification task of data mining in order to determine DDIs predictors for drug safety. RESULTS: In ambulatory prescriptions, 19%  (2,307/12,350) cardiological prescriptions were  detected with  potential DDIs; the  incidence was  the highest; patients whose age  over  65  y/o  were   the  major population of  potential DDIs. Of  these 2,307 potential DDIs prescriptions, the  severity of the  potential adverse effect was  rated  as 1st grade in 12.9%, 2nd grade in 51.4%, 3rd grade in 4.7%, 4th grade in 14.9% and 5th grade in 16.1% respectively. A“serious” interaction which was defined as potentially life or organ threatening (1st grade) or disease aggravation (2nd grade) were estimated in a total of 64.3% potential DDIs. We  found that  digitalis Glycosides and  loop  Diuretics were  the most   common (28.16%) combination involving potential DDIs. From the results of decision tree,  C5.0 algorithm was  better than CART algorithm in terms of  predictive accuracy;  the first  judgment  of C5.0 model was principal diagnosis indicated  its importance to which levels of DDIs was influenced. This  research suggests that  hospitals should set  up  clinical expert group to  build a customized DDIs database for  improving DDIs screening software systems.