Applying the self-controlled case series to analyse safety endpoints in single-arm, open label extension studies: A case study of Belimumab for the treatment of systemic lupus erythematosus and risk of infection

Background: Open Label Extensions (OLE) of randomized controlled trials (RCT) can generate additional long-term data on the safety and efficacy of a drug, but their interpretation is complicated by the lack of a control arm. The self-controlled case series (SCCS) could offer advantages when quantifying potential safety signals arising during an OLE study, as external, real-world control arms are likely to suffer from differences in data collection and coding compared to the RCT.

Objectives: To evaluate whether the SCCS could be used to analyse safety signals in an OLE study.

Methods: We used data from an extension study of the BLISS-76 RCT of belimumab, a treatment for systemic lupus erythematosus, which recruited patients in North America between 08/2008 and 09/2014. Planned study durations were 76 weeks (RCT) and 5 years (extension). The safety endpoint of interest was any acute infection event, assumed to independently recur. We included individuals originally randomized to placebo that continued into the OLE with >1 infection during either the RCT or OLE phases. Belimumab exposure was assumed to last 42 days following an injection. We constructed an SCCS, evaluated whether core assumptions were likely to hold, and applied conditional Poisson regression to quantify the association between belimumab and the risk of any infection. Infection served as a positive control, as an increased risk of infection with belimumab has been established.

Results: In total 87 individuals, out of 139 originally randomized to placebo, met the eligibility criteria. Most (44/52 = 85%) of the excluded patients withdrew during the RCT, with the most common reason being subject request (34%), followed by adverse event (30%). Due to small numbers, it was challenging to model the association between infection and inclusion in the OLE in order to assess whether the assumption of event-dependent exposures held. We constructed an SCCS of 628 infections over a median of 6.3 years of follow-up (5.0 – 6.6). After adjustment for age in 2-year age bands, there was no evidence that belimumab exposure was associated with risk of infection (RR = 0.86, 95%CI = 0.69 – 1.09). We assessed the assumption of independent event recurrences graphically by plotting the cumulative hazard of the distribution of time between successive events. Although this did this not indicate major violations, the number of individuals with a single event was very low (N = 7).

Conclusions: SCCS could offer an efficient and robust means of evaluating safety signals arising in single arm OLE compared to using external control arms. However, researchers need to carefully consider whether the assumptions of the design hold, and if applied, age and calendar-time adjustment requires careful consideration when treatment exposures are monodirectional.