Using tree-based scan statistics to assess vaccines for possible associations with Guillain-Barré syndrome in the Vaccine Safety Datalink

Tree-based scan statistics are used to study the safety of specific drugs or vaccines with respect to a broad range of health outcomes. But instead of choosing an exposure of interest and scanning patients' post-exposure outcome data, one can choose an outcome of interest and scan case-patients' pre-diagnosis exposures to vaccines, checking for potential associations. We used tree-based scan statistics, a new hierarchical structure of vaccines, and a self-controlled risk interval design to detect possible associations between vaccines and Guillain-Barré syndrome (GBS) in adults in the U.S. Vaccine Safety Datalink in 2017-2023. Incident GBS cases were identified with ICD-10-CM code G61.0. We assessed all vaccines administered in the 84 days prior to the incident GBS code, using a vaccine "tree" containing 99 CVX codes organized into 7 branches and 5 levels. We conducted both fixed and varying risk-window analyses. P ≤ 0.05 signaled a possible association. There were 2584 incident GBS cases; 40% were ≥ 65 years old. Only 25% had any vaccination code in the preceding 0-84 days. Using a Bernoulli tree-based scan statistic and risk and control windows of Days 1-21 and 43-84 before diagnosis, respectively, statistical signals were detected for influenza vaccines; Ad26.COV2·S (Janssen) COVID-19 vaccine; AS01-adjuvanted vaccines, including recombinant zoster; pneumococcal vaccines, including conjugate; and tetanus-diphtheria-acellular pertussis vaccine. Varying risk-window analysis found signals for influenza and Janssen COVID-19 vaccines. Attributable risk point estimates for these vaccines were in general agreement with other studies. All the vaccine groups that signaled, or their components, have had GBS signals in prior work. Conversely, there are no vaccines with a previously known association with GBS for which no signal was found. Tree-based data-mining holds promise for studying other outcomes suspected of association with vaccination, if case-finding algorithms are relatively specific, and allows detection of effects common to groups of related vaccines.