Study Design
Has the “obesity paradox” been resolved by semaglutide trials？ Zhu Liduzi Jiesisibieke* Zhu Liduzi Jiesisibieke Zhu Liduzi Jiesisibieke Mary Beth Terry C Mary Schooling
Background: Semaglutide reduces weight and cardiovascular disease in older people with obesity, diabetes, or heart failure. Although, some benefits could be pleiotropic effects, these findings, consistent with previous small trials in older people, question the ‘obesity paradox’, which posits that higher body mass index (BMI) may be beneficial for older people or patients with heart disease. Here, we consider the role of selection bias, specifically survival bias, in the ‘obesity paradox’.
Methods: We clarified the role of different instances of survival bias using directed acyclic graphs (DAGs), which are often used to identify confounding. Specifically, we focused on different ways survival to recruitment, in general or amongst patients, could generate bias alone or as the result of the interplay of survival with confounders and competing risk.
Results: We developed a prototype DAG comprehensively identifying when survival bias could potentially occur, including due to selection on potential confounders. Typical situations where unrecoverable survival bias might occur, include selection on exposure and competing risk of the outcome, for example, the association of BMI with stroke cannot be observed when death from heart disease precludes occurrence of a stroke, making heart disease a competing risk before recruitment for stroke. The best way to mitigate selection bias is to address it at the study design stage, and to consider birth cohort studies, studies in younger people and target trial designs, apart from randomized trials which obviates both confounding and selection bias.
Conclusion: Selection bias due to survival, like confounding, can occur in many ways and may be difficult to recover from, especially if not addressed during study design. Survival bias may be inevitable in observational studies when selecting exposure and outcome, or a competing risk of the outcome, on survival to recruitment, such as in studies of patients.
