Background: Observational studies estimating causal effects are often threatened by multiple sources of bias such as selection, measurement, and confounding bias that cannot be prevented utilising the observed data alone. In such cases, Quantitative bias analysis (QBA) approaches can be used to assess impact on findings, with quantitative multiple bias analysis (MBA) required to quantify the joint impact of multiple biases. However, MBA approaches can be methodologically complex, typically requiring specialised implementations, which has limited its routine application in practice.

 Methods: In this work, we aim to facilitate the practical application of MBA by proposing its conceptualisation as a missing data problem. This motivates the development of an MBA approach grounded on multiple imputation (MI), a popular approach for handling missing data that is widely accessible. Specifically, we derive exact-sequential imputation models implied by the joint bias structure and show how these can be approximated using fully conditional specification (FCS)–based MI implemented in standard software, flexibly propagating uncertainty arising from unobserved variables inducing bias, bias parameters governing the magnitude and direction of bias, and sampling variability. We conduct a simulation study to evaluate the performance of these approaches in estimating the causal risk ratio and difference and illustrate concepts and methods using a case study of the effects of breastfeeding on childhood asthma.  

Results: Simulation results showed that the exact-sequential and FCS-MI approaches performed similarly, with similar minimal bias and nominal 95% coverage. Both methods produced comparable estimates in the applied example.

Conclusions: These findings demonstrate that the proposed MI approaches can be used to implement MBA in a practical and accessible way, facilitating more routine and transparent assessment of robustness to multiple sources of bias in observational studies.