Abstract
Ratio variables, such as body mass index (BMI) and left ventricular ejection fraction (LVEF), are deeply ingrained in cardiovascular research and practice, yet the use of ratios in regression models can confuse associations and obscure underlying mechanisms. Drawing on examples from genome-wide association studies of adiposity traits and echocardiographic studies of cardiotoxicity, we illustrate how spurious correlation, mathematical coupling, and collider bias can arise when ratios are analyzed without attention to their component variables. Associations with ratio traits like BMI or LVEF are not unique to the ratio; instead, they intermix signals for the numerator and denominator. Regressing ratios on their components can generate tautological findings that are difficult to translate or interpret clinically. We propose a practical decision framework to guide analysis that begins with separate regressions of the numerator and denominator on the exposure, uses adjusted models when the exposure is independent of the denominator, and otherwise favors multivariate approaches, adopting a ratio outcome as a fallback. We encourage cardiovascular investigators to view ratios as one of several competing representations rather than default endpoints, and to choose modeling strategies that promote interpretability and clinical utility. In highlighting limitations of ratios for mechanistic and etiologic research, we do not challenge their use in evidence-based and guideline-supported clinical practice.
Keywords
Cardiovascular research, Body mass index (BMI), Left ventricular ejection fraction (LVEF), Heart failure