An indirect treatment comparison (ITC) involving lenvatinib plus everolimus (LEN+EVE) was conducted using networked data from HOPE 205, CHECKMATE-025, METEOR, AXIS, RECORD-1 and TARGET. The ITC incorporated adjustments for crossover to investigational treatment. Results showed superiority of LEN+EVE over EVE alone; and inferiority versus pazopanib (PAZ) or sunitinib (SUN) alone in overall survival (OS) for second-line treatment of advanced/metastatic renal cell carcinoma. No statistically significant differences in OS were found between LEN+EVE versus nivolumab (NIV), cabozantinib (CAB), axitinib (AXI), or placebo.
A subsequent analysis was conducted using intention to treat (ITT) to evaluate the impact of crossover correction on OS estimates and additionally to uncover any potential bias due to its absence. Three ITC scenarios were analyzed: A) all comparators plus placebo versus EVE; B) all comparators versus placebo; and C) LEN+EVE versus all comparators.
Scenario “A” showed consistent variance in survival benefit for ITT versus crossover by an average of 20%. Hazard ratios for AXI versus EVE shifted from below null (0.98) to above null (1.27); and mortality risk (placebo vs. EVE) moved 51% further from null (1.15 vs. 1.67). ITT estimates for Scenarios “B” and “C” showed on average 9% and 14% differences in OS benefits, respectively, versus crossover. In Scenario “C”, estimates for LEN+EVE versus PAZ or LEN+EVE versus SUN showed superiority with ITT data (0.82 or 0.75) but were inferior (1.2 or 1.09) with crossover.
Bias was observed in naive approaches to survival analysis in the presence of crossover. Failure to account for this in clinical trials may have implications on the comparative effectiveness profile and also on the cost-effectiveness results, and may lead to inconsistent resource allocation decisions.
Clinical trial identification
Legal entity responsible for the study
G. Meier, M. Guo: Employees of Eisai Inc. H. McElroy: Employee of Covance Market Access Services, which was paid by Eisai for literature review and statistical services. All other authors have declared no conflicts of interest.