In his talk, two biases are mentioned as biases specific to case-crossover studies which are particularly important when the exposure is chronic or successive. One of those biases is “bias due to within-subject exposure dependency” which is bias produced by the conditional logistic regression for matched case-control studies when the exposure status is NOT independent between periods. For binary exposure, this bias can be detected when the Mantel-Haenszel odds ratio (OR) is substantially different from the OR by the conditional logistic regression for matched case-control studies and can be removed by the Mantel-Haenszel method. The second type of bias is “bias due to lack of pairwise exchangeability” which is equivalent to “bias due to the exposure time trends” in many cases but can occur even without the exposure time trends on a population level. This bias can be detected when the OR of control subjects is substantially different from 1.0 and can be removed by the case-time-control approach. Those two biases can occur separately and simultaneously, and when they occur simultaneously, the simultaneous use of the remedies for these two biases (i.e., the Mantel-Haenszel method and case-time-control approach) is needed to remove them. Several relevant studies including those using real-world data and simulation studies are also explained.

Self-controlled case series (SCCS) has been used to investigate the association between drug or vaccine exposure and acute outcomes for post-market drug safety surveillance. As SCCS can eliminate time-invariant confounding by comparing risks of outcome within an individual, some studies used SCCS to complement cohort study design to obtain robust results.
In this plenary session, Angel will present her previous work on using active comparators in SCCS to reduce time-varying confounding and address protopathic bias using routine clinical data from Hong Kong. She will also discuss her recent work implementing SCCS in environmental pharmacoepidemiology using Japanese claims data linked to meteorological records. With these new applications of SCCS, is SCCS the best thing since sliced bread?