OBJECTIVES: The rank preserving structural failure time model (RPSFTM) can be used to adjust time-to-event efficacy estimates for treatment switching. The RPSFTM relies on two key assumptions (1) common treatment effect (CTE) assumption, which assumes that the effect of treatment was equal regardless of when it is received and (2) randomisation assumption, which can be violated if non-random drop out occurs during follow-up. The aim of this analysis was to assess the sensitivity of the RPFSTM results to these assumptions when applied to time to 6-month confirmed disability progression in the CLARITY and CLARITY Extension study.

METHODS: We applied the rank preserving structural failure time model (RPSFTM) to adjust for treatment switching from placebo to low-dose cladribine. A propensity score matching (PSM) method was used to test the sensitivity of the RPSFTM to the CTE assumption. The PSM method does not rely on the CTE, however estimation of an unbiased HR still requires that the randomization assumption holds. To overcome this issue, the PSM method was combined with inverse probability of censoring weights (IPCW) to adjust for potential selection bias from non-enrolment into the extension study. The PSM method and IPCW require all relevant confounders are included in the estimation procedure.

RESULTS: During CLARITY, the cladribine tablets (3.5 mg/kg) vs placebo HR was 0.58 (95% CI 0.40-0.83). During CLARITY+ CLARITY Extension, the unadjusted HR was 0.67 (95% CI 0.50-0.90), the RPSFTM HR was 0.62 (95% CI 0.44-0.88), the PSM was 0.62 (95% CI 0.40-0.84), and the PSM+IPCW HR was 0.63 (95% CI 0.40-0.87).

CONCLUSIONS: The adjustment methods produced consistent results. The addition of IPCW to the PSM made little difference. Provided the assumption of no unmeasured confounders holds, these results indicate no significant bias in the RPSFTM cladribine efficacy outcomes due to participant non-enrolment into the extension study or violation of the CTE assumption.