Retrospective indirect comparison of alectinib phase II data vs ceritinib real-world data in ALK+ NSCLC after progression on crizotinib

Abstract

Background: Approvals of second-line ALK inhibitors (ALKi) ceritinib and alectinib are based on single-arm trials that lack comparative efficacy data to support health technology assessments. We assessed if real-world data (RWD) could provide this by acting as an external control for single-arm studies by comparing data generated from alectinib trials and ceritinib patient data. Methods: We retrospectively analysed patients (pts) with ALK+ advanced NSCLC receiving an ALKi after crizotinib failure. The alectinib arm (ALC; n=183) included pts from the phase II NP28673/NP28761 studies. To generate the ceritinib control arm (CER; n=67) eligibility criteria similar to the alectinib trials was applied to the Flatiron Health's electronic health record database. A propensity score based on prognostic factors was generated and applied by inverse probability treatment weighting. A multivariate Cox model was used to evaluate the association of ALC compared with CER on overall survival (OS) adjusting for age, sex, prior treatment, race and stage at diagnosis. Summary data from the CER trial ASCEND-2 were re-digitised to compare with CER RWD. Results: Prior to re-weighting, the armswere heavily imbalanced. Key differences between the arms included age, prior treatments and baselineCNS metastases (median: 53 vs 61 yrs, 36%vs 13%[3 lines], 61%vs 35%, ALC vsCER). After weighting, balance was achieved across all key prognostic factorswith a standardised mean difference<10%for all factors.A multivariate Coxmodel showed ALC was associated with lower risk of death (HR 0.65; 95%CI 0.48-0.88). Adjusted median OS was 24.2months for ALC (95%CI 17.5-NR) vs 15.6 months for CER (95%CI 15.5-18.6).MedianOS in the RWDCER groupwas similar to that reported in the CER ASCEND-2 trial (14.9months). Conclusions: The results show that ALC is associated with prolonged OS vs CER which was consistent through numerous sensitivity analyses. CERRWD median OS was similar to that observed in ASCEND 2, validating this analysis. For single-arm studies RWD can serve as an external control for producing comparative data.