OBJECTIVES: The acceptability of indirect treatment comparisons (ITCs) for health technology assessment (HTA) is very challenging because many biases can alter treatment effect estimates, including selection bias, confounding bias, and data quality. There are many recommendations for either correcting or assessing potential biases arising from ITCs. Residual bias due to unmeasured confounders or missing not at random values can be addressed by quantitative bias analysis (QBA). Bias plots, E-value and tipping point analyses are increasingly used to support the primary results of ITCs, but not in the context of MAIC. This project aims to explore sensitivity and bias analyses to better support the primary results of MAICs applied to ROS1-positive first-line metastatic NSCLC patients, comparing aggregate data from Entrectinib clinical trials and the French national ESME lung cancer cohort.

METHODS: As post-hoc sensitivity analyses, residual bias from potential unmeasured confounders was explored using QBA methods with bias plot and E-value.

To assess the robustness of the results to the assumption of missing at random data, sensitivity analyses were performed on complete cases. To simulate worse than expected ECOG status in the tipping point analysis, we included δ-shifts with multiple imputation.

RESULTS: The primary results showed a progression-free survival benefit for entrectinib compared with standard French treatments (HR: 0.49, p-value<0.01). An E-value of 2.673 was estimated, which is higher than the strongest observed association with outcome and treatment (risk ratios of 1.29 and 1.44, respectively). The primary result was robust to complete case analysis and no tipping point was reached.

CONCLUSIONS: This study presents the first comprehensive applications of quantitative bias analysis to MAICs. It demonstrates the usefulness of these approaches in supporting the robustness of the efficacy results of MAICs to residual bias and missingness assumptions, even in the presence of a limited sample size.