Using Active Comparators in Self-controlled Case Series: An Application to Two Studies in the Clinical Practice Research Datalink

Background: Active comparators have historically rarely been applied in self-controlled studies. Hallas et al (2021) proposed methods for their use in this setting.

Objectives: To evaluate and apply the proposed methods in two self-controlled case series (SCCS): 1) thiazolidinediones and fracture risk using sulphonylureas as a comparator to address potential residual time-varying confounding by age, and 2) fluoroquinolones (FQ) and uveitis using cephalosporin as a comparator to address potential time-varying confounding by indication.

Methods: We used data from two CPRD studies. We fitted conditional Poisson regression models deriving incidence rate ratios (IRRs) using risk periods of 1-28 days for thiazolidinediones, and 1-29, 30-59 days and >=60 until end of treatment for fluoroquinolones. The models were adjusted for age, and additionally for calendar time for uveitis. We adapted the parameterisation of the model of Hallas et al. to Wilkinson and Rogers' (1973) framework for factorial models for the analysis of variance, identifying nested model terms, and extended this to non-binary exposure variables. We then used the IRR regression coefficients from these models to obtain a ratio of ratios, and its confidence interval, using Wald methods (the simple ratio method). We also fitted a single nested regression model in each SCCS to derive the same estimand (the regression method).

Results: We included 1,095 individuals from the fracture case series, and 9,639 from the uveitis case series. Use of thiazolidinediones was not strongly associated with an increased risk of fracture (adjusted IRR (aIRR) = 1.50, 95%CI = 0.74 — 1.58). Incorporating sulphonylureas as an active comparator led to aIRRs of 2.14 (1.37 — 3.33) using the simple ratio method and 1.88 (1.23 — 2.87) using the regression method, with the difference between the two methods due to age adjustment. The aIRR for each time-period after fluoroquinolone exposure and uveitis was 1.17 (0.90-1.52), 1.20 (0.93 — 1.55) and 0.89 (0.52 — 1.52). Incorporating cephalosporin as a comparator reduced the aIRRs slightly to 0.96 (0.69 — 1.34), 1.18 (0.84 — 1.65), 0.83 (0.43 — 1.60) using the simple ratio method and 0.96 (0.69 — 1.34), 1.18 (0.84 — 1.66) and 0.85 (0.44 — 1.63) using the regression method.

Conclusions: We have clarified recently proposed methods using active comparators in self-controlled studies by expressing this in standard factorial-model notation, and demonstrated how it can be applied to SCCS with multiple risk periods. Our findings illustrate that active comparators, representing a comparable standard of care for the population of interest, can be a valuable tool for controlling unmeasured time-varying confounding in SCCS.