Pattern-based system to detect the adverse drug effect sentences in medical case reports

Background The detection of adverse drug effect sentences in medical text reduces the efforts required for the manual task of drug safety monitoring by decreasing the number of reports which need to be investigated by drug safety experts. Moreover it helps in compiling a highly accurate and machine-understandable drug-related adverse effects knowledge bases which can support pharmacovigilance research and aids computational approaches for drugs repurposing. In this study, we proposed a pattern-based method to detect the sentences containing drug-adverse effect causal relation from medical case reports.

Materials and methods A set of 500 full abstracts from Medline medical case reports containing 988 adverse drug effect sentences from ADE corpus were used to evaluate the sentences detection task. Our method combined an outcome of a concept recognition system with a method for automatic generation of numerous patterns.

Results Our method achieved recall of 72.8, precision of 93.6 and F-Score of 81.7 % in the adverse drug effect sentences detection task.

Conclusion The results of this study can help database curators in compiling medical databases and researchers to digest the huge amount of textual information which is growing rapidly. Moreover, the mining algorithms developed in this study can be employed to detect sentences contain new associations between other medical entities in medical text.