Quantile regression for hypothesis testing and hypothesis screening at the dawn of big data

摘要

One implication of the application of generalized linear models in epidemiology is that only mean differences matter. Their exclusive use implies that investigators believe other parts of the distribution do not contribute information to our understanding of the relation between exposure and outcome (other than for an assessment of variance). In this issue of Epidemiology, Liu and colleagues1 examine the distributional differences in the association of education with a cardiovascular risk score and with body mass index (BMI). The novel contribution of this work is the use of quantile regression to examine how the associations of education with cardiovascular risk score and BMI vary at different points in the distribution of these outcomes. The authors add to a small but growing literature in epidemiology using this approach.