Developing a novel optimisation approach for keeping heterogeneous diets healthy and within planetary boundaries for climate change

Current diets in high- and middle-income countries significantly contribute to chronic diseases and the climate crisis. The global food system needs a transformation to align with the Paris Agreement and Sustainable Development Goals. Consumer eating habits are crucial in this transformation, due to the strong link between food demand and supply in market economies. The EAT-Lancet Commission proposed a healthy reference diet within planetary boundaries, emphasizing plant-based foods. However, operationalizing this diet and achieving consumer acceptance across diverse cultural groups remains challenging. Adopting the EAT-Lancet diet would require significant dietary changes for many high-income populations. Optimization analyses, particularly linear programming (LP), offer holistic approaches by considering nutritional and environmental factors alongside affordability. However, optimizing for a single population average might overlook the heterogeneity of diets within populations, potentially leading to unrealistic dietary changes for certain subgroups. This study aimed to optimize diets for groups with different eating patterns, investigating whether this yields a more realistic approach than optimizing for the national average, focusing on nutritional requirements, food-based dietary guidelines (FBDGs), and a greenhouse gas emissions (GHGE) limit of 1.57 kg/day as suggested by the IPCC. The study compared optimized diets to the EAT-Lancet diet.

The literature review highlights the significant negative impacts of current dietary habits on both human health and the environment. It cites the EAT-Lancet Commission's recommendations for a sustainable, plant-forward diet, emphasizing the challenges of translating these recommendations into practical, culturally acceptable dietary guidelines for diverse populations. Existing research using linear programming (LP) for diet optimization is discussed, pointing out limitations of focusing solely on population averages and neglecting dietary heterogeneity. The review underscores the need for approaches that account for the variability in dietary patterns within a population to create more realistic and acceptable dietary recommendations.

This modelling study combined hierarchical clustering analysis with linear programming to design nutritionally adequate, health-promoting, climate-friendly, and culturally acceptable diets using data from the Swedish national dietary survey, Riksmaten Vuxna 2010–11 (n=1797). Hierarchical clustering, identified as the optimal method, grouped participants into three dietary clusters based on their consumption of various food groups. Linear programming (LP) was used to optimize diets for each cluster and the total population. The objective function minimized the total relative deviation (TRD) from baseline diets to maximize similarity and cultural acceptability. Constraints included dietary reference values (DRVs) based on Nordic Nutrition Recommendations 2012, Swedish FBDGs, and in a second set of models, a CO₂eq limit of 1.57 kg/day. The CO₂eq values of foods were obtained from the RISE Climate Database. The cost of foods was estimated using data from "Matpriskollen". Post-hoc comparisons of clusters were conducted using Kruskal-Wallis test, ANOVA, and Pearson’s chi-squared test. The SHEIA15 index was used to assess the healthiness of each cluster. All optimizations were performed using the CBC solver in OpenSolver.

Three dietary clusters were identified: "Classic Baseline" (high red and processed meat, potatoes), "NutRich Baseline" (high nutrient-dense animal products, nuts, vegetables), and "LowClim Baseline" (high low-GHGE foods, vegetables, pulses). Baseline diets showed deficiencies in carbohydrates, fiber, iron, and vitamin D (except for LowClim), and excesses in saturated fatty acids and sodium. In the first set of models (meeting DRVs and FBDGs only), GHGE reductions ranged from 8–24%, with minor cost increases and low average relative deviations (ARDs), except for the Classic diet (ARD of 20%). Adding the CO₂eq constraint (second set of models) resulted in GHGE reductions of 43–53%, slight cost decreases (8–13%), and only marginally higher ARDs (5.8–22.8%). All CO₂eq-constrained optimized diets showed lower shares of animal-based foods and higher amounts of vegetables, potatoes, and fruits. The Classic+ diet demonstrated the most significant reduction in animal products and a substantial increase in pulses. The optimized cluster diets differed significantly from the diet optimized for the total population, suggesting that a cluster-based approach may lead to more realistic and acceptable dietary changes. Optimized diets did not closely align with the EAT-Lancet diet, potentially due to differences in constraint implementation, food categorization, and the scope of environmental factors considered.

The findings address the research question by demonstrating that a cluster-based optimization approach is superior to optimizing for the population average in developing sustainable dietary recommendations. The results highlight the feasibility of achieving significant reductions in dietary GHGE while maintaining nutritional adequacy, cultural acceptability, and even reducing cost in a high-income country context. The discrepancy between the optimized diets and the EAT-Lancet diet underscores the importance of considering cultural context and specific national dietary guidelines. The study's methodology offers a valuable framework for tailoring sustainable dietary recommendations to diverse populations.

This study introduces a novel cluster-based optimization approach for generating nutritionally adequate, health-promoting, and climate-friendly diets while considering cultural acceptability. It shows that simply meeting existing nutritional recommendations and FBDGs is insufficient to meet climate targets. The optimized diets successfully stayed within planetary boundaries for climate change, involved minimal changes to existing diets, and resulted in lower costs. Future research could explore the inclusion of novel climate-friendly food alternatives and a broader range of environmental impact factors.

The study's limitations include focusing solely on GHGE as an environmental indicator, neglecting other factors like land use and water footprint. The model did not include novel climate-friendly food replacements that might offer better solutions. The assumption that similarity to existing diets equates to cultural acceptability needs further investigation. The study's use of data from 2010–2011 might not fully reflect current consumption patterns.