Rough-Set-based ADR Signaling from Spontaneous Reporting Data with Missing Values

Spontaneous reporting systems of adverse drug events have been widely established in many countries to collect as could as possible all adverse drug events to facilitate the detection of suspected ADR signals via some statistical or data mining methods. Unfortunately, due to privacy concern or other reasons, the reporters sometimes may omit consciously some attributes, causing many missing values existing in the reporting database. Most of research work on ADR detection or methods applied in practice simply adopted listwise deletion to eliminate all data with missing values. Very little work has noticed the possibility and examined the effect of including the missing data in the process of ADR detection. This paper represents our endeavor towards the exploration of this question. We aim at inspecting the feasibility of applying rough set theory to the ADR detection problem. Based on the concept of utilizing characteristic set based approximation to measure the strength of ADR signals, we propose twelve different rough set based measuring methods and show only six of them are feasible for the purpose. Experimental results conducted on the FARES database show that our rough set based approach exhibits similar capability in timeline warning of suspicious ADR signals as traditional method with listwise deletion, and sometimes can yield noteworthy measures earlier than the traditional method.