We assessed the relation between dietary protein and each outcome with multivariable-adjusted nutrient density models.
These models allow estimation of the effect on each outcome of an increase in the percentage of energy from protein intake.
By forcing total energy and other intake (such as, dietary fats) to be constant, and by excluding carbohydrate from the model, modeling the effects of an increase in protein intake, by definition, statistically results in a decrease in carbohydrate intake.
Thus, the effect estimates of protein assume a substitution interpretation.
This study was not an intervention; we did not ask women to consume higher protein in place of carbohydrate; rather we used a statistical modeling technique with already-collected dietary data to estimate the effect of a protein for carbohydrate substitution.
The % of energy from protein that is “substituted” for carbohydrate is the difference between the median intake in the highest and lowest quintiles.
The models were also adjusted for other risk factors.

(We also evaluated the effect of an isoenergetic substitution of various intakes of high-protein foods in place of carbohydrate-dense foods, and expressed as servings per 1,000 kcalories per day)