Semantic distance-based creation of clusters of pharmacovigilance terms and their evaluation.

BACKGROUND

Pharmacovigilance is the activity related to the collection, analysis and prevention of adverse drug reactions (ADRs) induced by drugs or biologics. The detection of adverse drug reactions is performed using statistical algorithms and groupings of ADR terms from the MedDRA (Medical Dictionary for Drug Regulatory Activities) terminology. Standardized MedDRA Queries (SMQs) are the groupings which become a standard for assisting the retrieval and evaluation of MedDRA-coded ADR reports worldwide. Currently 84 SMQs have been created, while several important safety topics are not yet covered. Creation of SMQs is a long and tedious process performed by the experts. It relies on manual analysis of MedDRA in order to find out all the relevant terms to be included in a SMQ. Our objective is to propose an automatic method for assisting the creation of SMQs using the clustering of terms which are semantically similar.

METHODS

The experimental method relies on a specific semantic resource, and also on the semantic distance algorithms and clustering approaches. We perform several experiments in order to define the optimal parameters.

RESULTS

Our results show that the proposed method can assist the creation of SMQs and make this process faster and systematic. The average performance of the method is precision 59% and recall 26%. The correlation of the results obtained is 0.72 against the medical doctors judgments and 0.78 against the medical coders judgments.

CONCLUSIONS

These results and additional evaluation indicate that the generated clusters can be efficiently used for the detection of pharmacovigilance signals, as they provide better signal detection than the existing SMQs.