Dominique van Wonderen

Diet modeling is applied to support the design of diets that are healthy, environmentally sustainable, and acceptable. In this thesis, we evaluated different approaches to modeling iron absorption and food substitution. Iron deficiency is a major contributor to the global disease burden, particularly affecting children, premenopausal women, and individuals in low- and middle-income countries. People following vegetarian or vegan diets are also at risk, as nonheme iron from plant sources is less readily absorbed than heme iron from animal products. Although the literature provides several equations for estimating nonheme iron absorption from diets, many diet modeling studies still apply a constant absorption factor instead. In Chapter 2 of this thesis, we evaluated the differences in absorbable nonheme iron content when modeling with constant absorption factors versus diet-dependent absorption equations across omnivorous, vegetarian, and vegan diets. For all diet types, constant absorption factors tend to overestimate the absorbable nonheme iron content compared to diet-dependent equations. Unlike constant absorption factors, the equations enable the optimization of food components that either enhance or inhibit nonheme iron absorption, resulting in more realistic estimates of absorbable iron. These iron absorption equations are nonlinear and they can be integrated into diet modeling using either nonlinear programming or piecewise linear approximation. In Chapter 3, we compared both methods within a mixed-integer and a continuous diet model. For the mixed-integer model, piecewise linear approximation produced accurate results and outperformed nonlinear programming in terms of solution quality and computation time. For the continuous model, the performance of both methods was comparable. Food substitution involves replacing specific foods with alternatives to create healthier and more environmentally sustainable diets. One of the primary challenges in modeling food substitution is ensuring the acceptability of these changes. In diet modeling, strategies to maintain acceptability include minimizing the differences between observed and optimized diets - based on the assumption that smaller dietary changes are more acceptable - and imposing constraints on substitutions both between- and within food groups. However, many studies have modeled diets only at food group level, overlooking the benefits of substitutions within food groups, where variation in nutrient content and environmental impact can be substantial. In Chapter 4, we investigated how allowing substitutions within food groups can enhance the nutritional adequacy, environmental sustainability, and acceptability of modeled diets. We found that enabling substitutions within food groups led to notable improvements in nutritional adequacy and sustainability. Consequently, achieving the same nutrition and sustainability goals required only half as much dietary change when both between- and within-food-group substitutions were permitted, compared to using only between-food-group substitutions. In addition to food-group-based substitutions, machine learning offers an alternative approach for identifying acceptable substitutes. Since it is challenging to quantify individual preferences and translate them into explicit constraints, recent research has focused on learning these preferences directly from data. In Chapter 5, we introduced a novel data-driven method for food substitution by integrating the fields of diet modeling and recipe completion. Recipe completion algorithms aim to identify food items that are commonly consumed together and can complement each other within a meal. In our approach, we used such an algorithm to suggest acceptable food substitutes that pair well with the other food items within a meal and compared its performance to traditional food-group-based substitutions. Traditional substitution using broad food groups, which include many food items, tended to be less acceptable to consumers as evaluated by nutritionists. In contrast, substitution within narrow food groups, containing fewer items, resulted in higher acceptability but offered smaller nutritional improvements. Our recipe completion diet model balanced these trade-offs, achieving greater acceptability than substitutions based on broad food groups and improved nutritional adequacy compared to substitutions based on narrow food groups. In conclusion, the methods presented for modeling iron absorption and food substitution have proven effective in supporting the design of diets that are healthy, environmentally sustainable, and acceptable.