(C-203) Incorporation of Tokenized Real-World Data into a Phase 4 Vaccine Effectiveness Study: Assessment of Feasibility and Utility

Background: Tokenization is a process that transforms identifiable patient information into random, irreversible character strings used to link study participants to de-identified real-world data (RWD). Linking tokens to vaccination records in RWD may present a new opportunity to improve the accuracy of vaccination data collected from site Case Report Forms (CRF) and reduce misclassification of vaccine exposure status in vaccine effectiveness (VE) studies.

Objectives: To evaluate the feasibility and utility of using tokenized RWD (tRWD) to improve the accuracy of vaccination status information in a test-negative design (TnD), Phase 4 PCV20 VE study in the US.

Methods: Hospitalized adults age ≥65 years diagnosed with community acquired pneumonia were enrolled in the study and given the option to consent to tokenization. Study sites created tokens using 5 demographic variables, which were linked to open and closed claims from the Komodo Healthcare Map. National Drug Codes (NDC) and Common Procedure Codes (CPT) were used to identify influenza and pneumococcal vaccinations within 1 and 5 years prior to the enrollment date, respectively. After linked data were integrated into the clinical database, CRF-reported data and tRWD were compared, then reconciled. Final vaccination status was determined using a combination of CRF-reported data and tRWD. Feasibility was assessed by percentages of site participation, participant consent, tokens linked to RWD and the validity of tRWD linkages, which were measured by splits (multiple tokens represent the same patient) and collisions (one token represents multiple patients). Utility was evaluated by examining how well tRWD aligned with CRF-reported data (percent agreement), and whether tRWD improved identification of PCV20 exposure status.

Results: All sites agreed to participate and >80% created tokens. More than 50% of participants consented to tokenization, >90% of tokenized participants were successfully linked to RWD and < 5% resulted in a collision or split. There was a high level of agreement between CRF-reported and tRWD pneumococcal vaccination status (81.2% [PPSV23], 86.0% [PCV13], 89.1% [PCV20], 99.6% [PCV15]). Agreement was lower for influenza vaccination status (63.6%). Among participants classified as vaccinated, 5.7% of influenza, 5.9% of PPSV23, 8.4% of PCV13, 0.0% of PCV15 and 7.1% of PCV20 vaccines were identified by tRWD only. Incorporating tRWD improved the percent identified as vaccinated with PCV20 from 15% to 16%.

Conclusions: Additional vaccine administrations were identified after incorporating tRWD, which reduced PCV20 exposure misclassification. tRWD should be used to supplement CRF-reported data. Minimizing exposure misclassification is critical to accurately estimating VE in TnD VE studies.