Submission 81

Submission 81

Drug indications knowledgebase can be achieved based on drug labels. But when facing complicated clinical scenarios with multiple clinical problems, there are lack of corresponding knowledge resources that can support which kind of drug should be used in this condition. The real-world clinical data which implies the clinical drug utilization knowledge can be used as a resource to mining such drug therapy knowledge. In this study, a statistical method was proposed to achieve the significant associations between drug and diagnosis in Electronic Medical Records. Then the statistically significant associations between drug and diagnosis were used to measure the distance between drugs. Then drugs can be unsupervised clustered based on this quantitative drug-drug distances. A bootstraps strategy was used to select the best cluster numbers in the dataset. The clinical problem sets that associated with each drug cluster can be enriched and define a clinical scenario. The drugs in the drug cluster will provide a drug therapy for such clinical scenario. An EMR dataset extracted from a 2000-bedded general hospital were used to test this method. Within total 36 drug clusters, there are 34 drug clusters associated with specific clinical scenarios. Such associations can be used to create a clinical drug therapy knowledgebase. In addition, when comparing with the up-to-date best practice knowledge, this resource could be used to evaluate the effectiveness of medication-related knowledge translation and identify drug use problem in practice. It also provided a quantitative way to measure the distance between different drug therapies.