Prescription-Event Monitoring

Abstract

Prescription-event monitoring (PEM) is a non-interventional intensive method for post-marketing drug safety monitoring of newly licensed medicines. PEM studies are cohort studies where exposure is obtained from a centralised service and outcomes from simple questionnaires completed by general practitioners. Follow-up forms are sent for selected events. Because PEM captures all events and not only the suspected adverse drug reactions, PEM cohorts potentially differ in respect to the distribution of number of events per person depending on the nature of the drug under study. This variance can be related either with the condition for which the drug is prescribed (e.g. a condition causing high morbidity will have, in average, a higher number of events per person compared with a condition with lower morbidity) or with the drug effect itself. This paper describes an exploratory investigation of the distortion caused by product-related variations of the number of events to the interpretation of the proportional reporting ratio (PRR) values ("the higher the PRR, the greater the strength of the signal") computed using drug-cohort data. We studied this effect by assessing the agreement between the PRR based on events (event of interest vs all other events) and PRR based on cases (cases with the event of interest vs cases with any other events). PRR were calculated for all combinations reported to ten selected drugs against a comparator of 81 other drugs. Three of the ten drugs had a cohort with an apparent higher proportion of patients with lower number of events. The PRRs based on events were systematically higher than the PRR based on cases for the combinations reported to these three drugs. Additionally, when applying the threshold criteria for signal screening (n > or =3, PRR > or =1.5 and Chi-squared > or =4), the binary agreement was generally high but apparently lower for these three drugs. In conclusion, the distribution of events per patient in drug cohorts shall be examined when comparing the 'strength of the signals' across drugs using PRR values. Further research will be required to address the sensitivity and specificity of the two ways of calculating PRR using data derived from drug cohorts.