Patient stratification and identification of adverse event correlations in the space of 1190 drug related adverse events.

Purpose: New pharmacovigilance methods are needed as a consequence of the morbidity caused by drugs. We exploit fine-grained drug related adverse event information extracted by text mining from electronic medical records (EMRs) to stratify patients based on their adverse events and to determine adverse event co-occurrences.

Methods: We analyzed the similarity of adverse event profiles of 2347 patients extracted from EMRs from a mental health center in Denmark. The patients were clustered based on their adverse event profiles and the similarities were presented as a network. The set of adverse events in each main patient cluster was evaluated. Co-occurrences of adverse events in patients (p-value < 0.01) were identified and presented as well.

Results: We found that each cluster of patients typically had a most distinguishing adverse event. Examination of the co-occurrences of adverse events in patients led to the identification of potentially interesting adverse event correlations that may be further investigated as well as provide further patient stratification opportunities.

Conclusions: We have demonstrated the feasibility of a novel approach in pharmacovigilance to stratify patients based on fine-grained adverse event profiles, which also makes it possible to identify adverse event correlations. Used on larger data sets, this data-driven method has the potential to reveal unknown patterns concerning adverse event occurrences.