The use of relative incidence ratios in self-controlled case series studies: an overview

The self-controlled case series (SCCS) is a useful design for investigating associations between outcomes and transient exposures. The SCCS design controls for all fixed covariates, but effect modification can still occur. This can be evaluated by including interaction terms in the model which, when exponentiated, can be interpreted as a relative incidence ratio (RIR): the change in relative incidence (RI) for a unit change in an effect modifier.

We conducted a scoping review to investigate the use of RIRs in published primary SCCS studies, and conducted a case-study in one of our own primary SCCS studies to illustrate the use of RIRs within an SCCS analysis to investigate subgroup effects in the context of comparing whole cell (wcp) and acellular (acp) pertussis vaccines. Using this case study, we also illustrated the potential utility of RIRs in addressing the healthy vaccinee effect (HVE) in vaccine safety surveillance studies.

Our scoping review identified 122 primary studies reporting an SCCS analysis. Of these, 24 described the use of interaction terms to test for effect modification. 21 of 24 studies reported stratum specific RIs, 22 of 24 reported the p-value for interaction, and less than half (10 of 24) reported the estimate of the interaction term/RIR, the stratum specific RIs and interaction p-values. Our case-study demonstrated that there was a nearly two-fold greater RI of ER visits and admissions following wcp vaccination relative to acp vaccination (RIR = 1.82, 95 % CI 1.64-2.01), where RI estimates in each subgroup were clearly impacted by a strong healthy vaccinee effect.

We demonstrated in our scoping review that calculating RIRs is not a widely utilized strategy. We showed that calculating RIRs across time periods is useful for the detection of relative changes in adverse event rates that might otherwise be missed due to the HVE. Many published studies of vaccine-associated adverse events could have missed/underestimated important safety signals masked by the HVE. With further development, our application of RIRs could be an important tool to address the HVE, particularly in the context of self-controlled study designs.