(004) Mapping the Information Pathways of Adverse Reactions Through Clinical Event Sequences, Patient Interactions with the Healthcare System and Generation of Different Data Streams – a Qualitative Pilot Study

Publication Number: 1000

Background: Multiple data streams are used in Pharmacovigilance (PV), providing evidence for signal management and risk quantification. With an increasing volume and diversity of data, the question arises as how to better optimize the use of different data streams, by assessing which data would be of most value for specific safety problems. It is hypothesized that modelling how a patient interfaces with the healthcare system may give insights into where different data of relevance for PV resides and the data reliability.

Objectives: To analyze information pathways across heterogeneous cases of adverse reactions (ARs) according to the clinical event sequence, patient-healthcare interactions and data streams generated. To consider the value of data streams in providing useful evidence for PV.

Methods: We constructed a matrix with 6 rows representing the clinical event sequence from drug to AR (establishing indication, drug initiation, early symptoms or signs of the AR, treatment of early symptoms or signs, AR diagnosis and AR treatment) and 8 columns representing the points where data streams are generated in the system (patient, pharmacy, physiotherapist, GP, specialist, hospital emergency, inpatient and outpatient clinics). The three cases selected were NSAIDs and GI bleeding; nonsedating antihistamines and ventricular arrhythmias; and statins and myopathy. The matrix was filled according to the events and patient-healthcare system interactions. Based on this mapping, the relative value of each data stream was evaluated.

Results: Each healthcare data stream contributed with data on 1-4 (median 2) clinical events per case. Thus, for all cases, no single healthcare encounter would allow a complete collection of all data of importance to PV. The highest contribution came from GP clinics (4 events per case). All data streams provided useful information, although the relevance varied. We noted differences in information pathways even within a specific exemplar, depending on the development of symptoms and signs, and the final outcome diagnosis. For example, the relevance of early symptoms (dyspepsia, palpitations, muscle pain) as a proxy for the AR depends on the probability of developing a more serious outcome and the temporality of the event sequence. Data completeness and validity were also important.

Conclusions: This qualitative analysis offers a systematic approach to map information pathways and data streams that could be useful for optimizing information collection and evidence generation in PV. The model could be further formalised by assigning probabilities to each path, allowing for quantitative modelling. Limitations of the analysis are inclusion of only few cases, and the generalizability across different healthcare systems.