Using Data Mining Methods to Support Antibiotic Selection for a Medical Center

• Main Authors: Ya-Yu Hsiao, 蕭雅尤
• Other Authors: Yung-Jin Lee
• Format: Others
• Language: en_US
• Published: 2006
• Online Access: http://ndltd.ncl.edu.tw/handle/30367268986429164564

碩士 === 高雄醫學大學 === 臨床藥學研究所碩士班 === 94 === This study was aimed to use data mining techniques to establish an antibiotic selection model. Based on the articles by Cunha, six issues were considered as the criteria of antibiotic selection, containing antimicrobial spectrum, pharmacokinetics and pharmacodynamics, antimicrobial resistance, side effects, drug cost, and miscellaneous item and thus creating 17 input attributes in total. The output attribute was the selection result (exclusion or inclusion). Cephalosporin was chosen as the target antibiotic. We set up a database comprising a group of simulated antibiotics as a web-based questionnaire for experts to login and make their own decision through internet. Four medical doctors and six clinical pharmacists participated in this study. Each of them needed to complete 300 cephalosporin selections. The nine data mining methods include J4.8, Id3, NNge, IBk, Ridor, SMO, Naïve bayes, multilayer perceptron, and JRip. In addition, 10-fold cross validation was used for data analysis. It demonstrates that these data mining methods can be used in antibiotic selection and maybe can be applied for a hospital to facilitate the efficiency of the hospital P&T committee. In the future, perhaps data mining methods can also be applied in the selection of other drugs.