W-03: Development and Operationalization of a Method for Determining Adverse Drug Reactions from a Clinical Study Safety Data Set
Poster Presenter
Fred Jerva
Pharmacovigilance
Janssen United States
Objectives
Large drug-development programs receive thousands of adverse events. Determining which are true adverse drug reactions (ADR) can be a daunting task for pharmacovigilance teams. We sought to develop and operationalize a novel method for determining ADR from a clinical study data set.
Method
We combined a Bayesian and risk-based quantitative analysis, MedDRA High-Level Group Term clustering of adverse events, individual case-level and focused medical reviews, and global introspection to create a robust, repeatable, and defendable methodology for ADR determination.
Results
Adverse event (AE) data were pooled from similarly designed late-phase clinical studies. Reported events (MedDRA PT) from patients enrolled in three experimental study arms were compared with events from the control arm. More than 13,000 AE (1,300 unique PTs) were reported from ~4,000 patients across all study arms.
A probabilistic statistical model of risk identified 14 AEs (MedDRA PT) reported more frequently for treated patients compared with those receiving placebo. One additional PT was identified when data from all experimental study arms were pooled before comparison with the control arm. These 15 PTs were clustered with other reported PTs using MedDRA HLGT to develop AE concepts (containing 5?25 PTs) for further review. An initial inspection focused on the consistency of the PTs within the AE concept and a medical review filtered PTs common to the disease indication and background population. At this stage, nine AE concepts were selected for a deeper case-level inspection.
Individual AE case-level details were scrutinized to identify events confounded by patients’ pre-existing relevant medical histories and concomitant medications. Approximately 70% of cases were confounded by medical or medication history. After subtraction of confounded cases, imbalances between treated and placebo were recalculated.
A structured evaluation framework was developed to scrutinize the remaining AE concepts consistently by answering the following four questions: 1) Was the time to onset of the reported AE realistic given the timing of drug administration? 2) Was the AE biologically plausible given the known mechanism of action of the investigational product? 3) Was the AE consistent with what is already known about the investigational product? 4) Was the AE consistent with a known drug class effect? A simple one-page scorecard captured the results of the case-level inspection and medical review. Following this review, one AE was nominated as a candidate ADR for governance.
Conclusion
We developed a novel, systematic, comprehensive, repeatable and less-subjective methodology for ADR determination, and our cross-functional team triaged a large clinical trial data set with more than 13,000 AE (1,300 unique PTs) down to a single candidate ADR for governance.
Within AstraZeneca, this was the first example of a pharmacovigilance team operationalizing a standardized sequence of steps (including statistical analysis, MedDRA-based clustering, selected case-level review, and a structured medical assessment) to consistently and transparently determine ADR.
With this methodology, safety reporting teams can present a simplified, time-efficient, and defensible strategy for ADR determination; improve communications with internal stakeholders about the emerging safety profile of an investigational drug product; and help inform key patient safety issues.
