Purpose: The objective of this session is to dig deeper than the common high-level presentations of quantitative bias analysis and explore the nuts and bolts of how they are conducted in practice.

Description: Quantitative bias analysis (QBA) comprises a collection of approaches for modeling the magnitude and direction of systematic errors in data which cannot be addressed with conventional statistical adjustment. QBA is useful when working with imperfect real-world data where uncertainty about bias often arises due to substantial missing covariate data, limited synthetic control arm data, or incomplete outcome data. While QBA is gaining in popularity, there has been little guidance on the nuts and bolts of applying QBA to real-world data.

The session will start with a brief introduction to QBA and its potential role for medical decision-making (Thorlund, 5 minutes). The nuts and bolts of QBA applied to unknown population-level confounders will be illustrated using two examples in advanced non-small cell lung cancer with substantial missing covariate data that is ‘missing not at random’, and synthetic control arms with sample sizes too small for statistical adjustments. (15 minutes, Hsu). The nuts and bolts of applying QBA for extrapolation of long-term benefits from surrogate outcomes will be illustrated across oncology examples in a Bayesian framework, further discussing how to best incorporate information from external sources such as expert opinion or related literature (Thorlund, 15 minutes). Finally, reporting requirements for applied QBA as well as perspectives on which approaches would work in practice for HTA and medical decision-making will be covered. (Duffield, 15 minutes). Throughout, the audience will be prompted with questions about the likely impact of bias that QBA will reveal as well as the pros and cons of applying simple versus complex approaches. The session will conclude with a Q&A (10 minutes).