1D-04 - Current approaches to observational study designs and analytical options to evaluate the safety of multi-dose vaccines: A systematic review

Background: Observational designs are crucial for evaluating vaccine safety post-licensure. Biases from time zero specification, time-varying confounding, and assumption violations threaten the validity of results in the presence of multiple doses.

Objectives: To review current practices of observational study designs and analytical options in evaluating the safety of multidose vaccines.

Methods: Following PRISMA guidelines, we pre-registered our protocol in the Open Science Framework. We searched EMBASE, MEDLINE, Web of Science, and Scopus (2018–2022) for studies evaluating vaccines given in more than one dose for complete immunization. Eligible articles tested a hypothesis on vaccine safety, were published in peer-reviewed journals, and had full-text availability. We extracted key design and analytical features for the most frequent study designs.

Results: From 5,953 records, 123 studies on multi-dose vaccine safety were included, with 101 assessing multiple doses. Study designs included 59 cohort, 53 self-controlled case series (SCCS), 11 self-controlled risk interval (SCRI), 17 case-control, and 19 using multiple designs.
Among cohort studies that evaluated the safety of multiple doses, 20 (40.0%) set time zero at a specific calendar time and time-updated exposure status, 16 (32.0%) set time zero at vaccination, 2 (4.0%) emulated a target trial and 16 (32.0%) did not explicitly define time zero. Out of those that set time zero at vaccination, 11 (68.7%) matched either unvaccinated or vaccinated comparator to the vaccinated at vaccination time. Time zero for subsequent doses was defined in 12 (85.7%) studies by censoring and restarting follow-up at the time of second dose. Time-varying confounding was considered in 8 (16.0%) cohort studies evaluating multiple doses primarily using time-dependent Cox models 7 (87.5%). Of the 17 case-control studies, 9 (52.9%) were nested within a cohort, with one using incidence-density sampling of controls. All defined time zero at diagnosis for cases, and 16 (94.1%) used matched calendar time for controls. Among 16 studies evaluating multiple doses, 10 (62.5%) estimated odds of any dose from time zero, while 6 (37.5%) focused on specific doses either by analysing the latest dose or each dose separately.
Of SCCS studies, 42 (79.2%) addressed the event-dependent exposure assumption, mostly using pre-vaccination periods, 23 (54.7%) followed by the modified SCCS models, 19 (45.2%). Twelve (28.6%) applied multiple correction methods.
Among 19 studies employing multiple designs, 13 (68.4%) reached consistent clinical conclusions.

Conclusions: Cohort and SCCS were frequently used for multidose vaccine safety evaluation. Key areas for improvement include clearer reporting of design choices, especially time zero specification and control of time-varying confounding in cohort studies. Target trial emulation, when feasible, could be employed to improve transparency in reporting at both design and implementation. This review also highlights the need to understand appropriate approaches to use to account for event-dependent exposure in SCCS design and guidance for interpreting results when multiple designs are used.