*Geek Box: Statistical Methods

*Geek Box: Substitution Models

There are two ways to think about potential confounders in nutritional epidemiology: confounding from other non-dietary lifestyle factors, like smoking, and confounding from other nutrients which may be correlated with the nutrient exposure of interest. The former can be accounted for in an adjustment model; the latter can be addressed through careful substitution modelling.

So what is substitution modelling, and how does it work? In nutritional epidemiology, it is standard practice to adjust for total energy: this is because all nutrients positively correlate with total energy intake, and adjusting for energy provides a means to assess the effects of the nutrient, independent of total energy.

From that, researchers may want to investigate the effects of replacing one nutrient with another. Because total energy is adjusted for, this can be done assuming isocaloric substitution, for example, what is the effect of replacing 5% of dietary energy from sugar with 5% of energy from wholegrain carbohydrates. Let’s take an example of the effects of isocaloric replacement of saturated fats with unsaturated fats [without distinguishing between MUFA and PUFA, for simplicities sake]. In this model, you would have total energy + total fat, adjusted for total energy + saturated fat, adjusted for total fat.

Because all fat subtypes are under the umbrella of total fat, this means all that is excluded is unsaturated fats. So by energy-adjusting total fat and saturated fat, the effect of unsaturated fats on the outcome of interest is the effect of these fats replacing saturated fat. So when you read about substitution analysis, this is [albeit in a very simplified explanation!] what is going on – examining the effects of isocalorically substituting one nutrient with another while holding other nutrients and total energy constant.