Standardized data were simulated using the “dagitty” (0.2-3) R package (18 (link)) to reflect the data generating process depicted in Supplemental Figure 1, where total energy was fully determined by the energy intake from 7 macronutrients: 1) sugars, 2) carbohydrates, 3) fiber, 4) saturated fat, 5) unsaturated fat, 6) protein, and 7) alcohol. Total energy and remaining energy intake were not directly simulated. Instead, they were calculated from the sum of all macronutrient energy variables, or the sum of all energy variables except sugar, respectively. Each macronutrient was assigned a unique effect on fasting plasma glucose. Specific path coefficient values were chosen to represent plausible causal effects, and simulated variables were rescaled with plausible mean ± SD values informed by the National Diet and Nutrition Survey (see Supplemental Table 1) (19 ). All simulations and models (see below) were repeated in the presence of a single variable (U) that affects the intake of all macronutrients, to demonstrate the influence of confounding by common causes of dietary composition. Each simulation included 1000 observations and was repeated over 100,000 iterations. We report the median effect estimate and 2.5th and 97.5th centiles [representing the 95% simulation interval (SI)] from the 100,000 iterations for each model. For ease of illustration, effect estimates are expressed in mg/dL per 100 calories (mg/dL/100 kcal).
Free full text: Click here