(C-110) PEPRVision: A Data Insight Engine for Pharmacoepidemiology

Background: Pharmacoepidemiologic research relies on large-scale healthcare databases, such as administrative claims and electronic health records, to study medication safety and effectiveness. The Sentinel System’s Patient Episode Profile Retrieval (PEPR) provides tabular summaries of patient healthcare encounters but presents challenges in identifying patterns efficiently due to data volume and complexity.

Objectives: This study introduces PEPRVision, a novel data visualization tool that transforms tabular PEPR data into an intuitive graphical format to facilitate pharmacoepidemiologic research.

Methods: PEPRVision was developed using principles of pre-attentive processing, employing visual attributes such as color coding, spatial arrangement, and timeline-based displays to present chronological healthcare data. The visual attributes enable immediate pattern recognition to assist reviewers in quickly assessing a case, identifying patterns, and making decisions. The tool was evaluated through three case studies: (1) identifying mechanisms of prenatal exposure to teratogenic medications to inform risk mitigation strategies, (2) detecting protopathic bias in medication-outcome associations, and (3) assessing the validity of congenital malformation measures used in data mining approaches.

Results: PEPRVision effectively visualized longitudinal patient data, allowing assessment of temporal associations between medication exposure and outcomes. In a forensic study on failures of the iPLEDGE REMS program to prevent fetal exposure to isotretinoin, PEPRVision revealed gaps in contraceptive use and pregnancy testing adherence. The second case study illustrated its ability to identify reverse causation in the assessment of doxycycline safety on pregnancy loss, informing improved study design. Finally, the tool demonstrated its utility in validating congenital malformation signals by distinguishing maternal from infant conditions and highlighting relevant care procedures versus rule-out tests to support accurate measurement.

Conclusions: This study demonstrates the complementary role of disaggregated data visualization in pharmacoepidemiologic research. While summary statistics characterize population-level patterns, integrating visualization tools like PEPRVision with analytical approaches helps researchers explore patient-level variations, detect potential design-related biases, and evaluate measurement approaches.