Are U-shaped relationships between risk factors and outcomes artifactual?

The distribution of a risk factor in the population is an important determinant of its nadir value. Populations with high or low values may have high and low nadirs, respectively. Identification of a nadir for a modifiable risk factor from epidemiologic relationships may therefore arise from this distribution bias and is therefore unrelated to therapeutic targets.