基于数据挖掘的药物不良反应因果关系研究

Research on the Causal Relationship of Adverse Drug Reactions Based on Data Mining

目的:针对药物相互作用导致广大患者的发病率和死亡率提高,利用现有的药物不良反应自发报告系统的资源,研究如何高效快速地发现药物不良反应的发生,降低医疗意外的发生率。方法:药物不良反应的数据挖掘和因果发现是一个非常有挑战性的课题,需要对药物信息和不良事件之间的所有组合进行分析,同时准确估计药物组合与不良事件之间的因果关系。为有效识别药物与不良事件之间的因果关系,通过研究发现,可以将因果概念引入关联规则的数据挖掘中,通过关联规则找出关联程度高的药物和不良事件的频繁项集,再通过贝叶斯网络中V结构的性质识别出药物与不良事件之间的因果关系。结果与结论:通过比较由关联规则和因果发现所得出的100个高度关联结果,与关联规则相比,因果关系发现的结果准确度更高,并且通过因果关系发现的结果中存在更低的未知药物不良反应。

Objective: In view of the harm that drug interaction leads to the increase of morbidity and mortality of patients, this paper studies how to efficiently and quickly find the occurrence of adverse drug reactions and reduce the incidence of medical accidents by using the resources of the spontaneous reporting system of existing adverse drug reactions. Methods: Data mining and causal discovery of adverse drug reactions is a very challenging subject. It is necessary to analyze all combinations of drug information and adverse events, and accurately estimate the causal relationship between drug combinations and adverse events. To effectively identify the causal relationship between drugs and adverse events, through the study, we have found that we can introduce the causality concept into data mining of association rules, find frequent itemsets of drugs and adverse events with a high degree of correlation through the association rules, and identify the causal relationship between drugs and adverse events through the properties of V structure in the Bayesian network. Results and conclusions: Among the 100 highly correlated results obtained by association rules and causal findings through comparison, the accuracy of results found by causal relationship detection is higher than that of results found by association rules, and there are fewer unknown adverse drug reactions in the results discovered by causal relationship.