Relating Subjective and Objective Pharmacovigilance Association Measures - Clustering Challenges in Biological Network - page 300

Mean Sab Cor(Sab,Uab) Cor(Uab,Psiab)

Cor(Sab,Rab) Cor(Rab,Psiab)

C1 C2 C3 C4 C1 C2 C3 C4 C1 C2 C3 C4 C1 C2 C3 C4 C1 C2 C3 C4

Cluster

Fig. 15.10. Boxplot summaries of the variables V 1 , V 2 , V 3 , V 7 and V 9 by cluster for the partitioning
with k =4obtained from these five variables.

the two drugs forming this cluster are two of the three drugs with highest mean
S ab values from the second partition of the basic clustering. Thus, while the level
of subjective blame associated with these two drugs is high, it is well supported
by the objective evidence.

15.8. Summary

The fundamental problem of pharmacovigilance is to accurately assess the rela-
tionship between drugs and adverse medical reactions that may or may not be
related to these drugs. This paper has presented a clustering study, aimed at
elucidating the relationships among different objective and subjective association
measures between drugs and adverse events in the FDA's Adverse Event Report-
ing System (AERS) database. As noted previously, the ultimate objective of this
work is to define a quantitative index of blame that characterizes the tendency
for drugs to be subjectively associated with adverse events to a greater or lesser
extent than is warranted by the objective evidence. The results presented here
consider a subjective association measure based on the classification of drugs as
“suspect” or “concomitant” in the AERS Individual Safety Reports, along with