Revisiting the Association Between Diabetes Mellitus and Hearing Loss: Methodological Heterogeneity, Shared Risk Factors, and Collider Bias

Hearing loss and diabetes mellitus impose substantial health and socioeconomic burdens, and their relationship has been investigated since Jordao’s case report in 1857. However, considerable heterogeneity persists across studies, raising questions regarding whether diabetes exerts a direct causal effect on hearing function. This review examines the potential reasons for these inconsistent findings from three perspectives. The first is inherent methodological heterogeneity, including variations in study populations, variable definitions, and measurement techniques. The second is unmeasured or residual confounding arising from inadequate data collection and insufficient adjustment for shared risk factors, such as age, sex, obesity, genetic factors, cardiovascular diseases, lifestyle factors, diabetes treatments, noise exposure, and ototoxic medication use. The third is collider bias resulting from inappropriate conditioning on common consequences of both diabetes and hearing loss, such as cognitive decline. Directed acyclic graphs can be employed to illustrate these biases and guide appropriate adjustment strategies. Sensitivity analyses, including E‐values and quantitative bias analysis, may help evaluate the robustness of findings against unmeasured confounding. Transparency in research reporting is also necessary to advance causal inference in diabetes‐related hearing loss research.