Variables were chosen for inclusion in the statistical models based on their reported or suspected associations as independent risk factors or potential confounders.
As mentioned, total energy & dietary fats were forced to be constant in the models.
A variable was retained in the final model, as all of these were, if it was statistically significant or, if it was not statistically significant and its removal from the model resulted in 10% or greater change in the effect estimates of protein, in which case it was considered a confounder and retained.

- Although it may seem counter-intuitive to include components of dietary protein, such as dietary methionine, into the model with total protein, we did so with the attempt to isolate the effects of the protein component on outcome independent of known risk factors. For example, methionine has been reported to be the dietary precursor of homocysteine, a potential risk factor for CVD.