OBJECTIVES: Real-world evidence (RWE) and other non-randomised evidence is increasingly being used for healthcare decision-making. However, these data are limited by missing observations and unmeasured confounders that can introduce uncertainty in and bias the results. Quantitative bias analysis (QBA) can test the impact of unmeasured confounders. Health technology assessment agencies in the United Kingdom and Canada recommend QBA in guidelines for using RWE as part of reimbursement submission packages. These guidelines outline best practices but do not provide an overview of QBA methods. This research aimed to summarise the current use of QBA methods for unmeasured confounding.

METHODS: A targeted literature review (TLR) was conducted in Embase, MEDLINE, and EconLit (May 2, 2023) for studies applying QBA for unmeasured confounders or discussing methods without application. Commentaries, letters, and systematic literature reviews were excluded, as well as studies with no description of QBA approaches, those focusing on probabilistic bias analysis, and those using QBA for misclassification or selection biases. Two reviewers screened and extracted results; a separate reviewer validated the results.

RESULTS: In total, 33 studies published from 2013 to 2023 were included. Twenty-six studies applied a QBA approach, while seven were methodology studies. More than half (n=14) of the QBA approaches were derived bias, of which eight employed the E-value formula. Methods were applied for many indications, most frequently non-small cell lung cancer (n=4). Fourteen studies using QBA did not report how the confounding factors were identified and six relied on previously published literature. The hazard ratio was the most common outcome of interest (n=13).

CONCLUSIONS: This TLR identified derived bias, particularly the E-value formula, as the most common QBA approach to evaluate unmeasured confounders. Information is lacking, however, on the clear methodology and application of QBA.