Risk-benefit assessment of food substitutions

Abstract

Risk-benefit assessment (RBA) of foods is a tool used for weighting the adverse and beneficial health effects associated with food consumption. RBA offers support to policy makers by providing evidence on the overall health impact of e.g. dietary interventions or food policies, and thus can provide guidance for public health promotion. So far, most existing RBAs have been conducted by only considering one food in isolation from the whole diet. However, the change in consumption of one food will likely change the consumption of other foods in the diet, which may influence human health. Only very few RBAs have accounted for substitution of foods; most of these compared exposures to nutrients and contaminants to established dietary reference values and health-based guidance values, respectively, without accounting for the overall health impact in terms of morbidity, mortality, severity or duration of the associated diseases. The overall aim of this thesis was to develop an approach for integrating food substitution in RBA that allows for the quantification of the associated long-term health impact. Case studies based on the Danish food-based dietary guidelines were conducted and described in three manuscripts to fulfill this aim. The health impact of substituting red and processed meat by fish to reach the recommended intake of fish (350 g/week) in the Danish adult diet was investigated in two manuscripts. Two different approaches to model the substitution were taken. Manuscript I based the substitution on a deterministic approach, assuming the same substitution behavior across the Danish population. The influence of the choice of fish species consumed on the overall health impact was investigated in four alternative scenarios, assuming also a change in the fish species consumed before substitution. The results showed that the overall health impact of substituting red and processed meat by fish varies depending on the fish species consumed. The largest health gain was observed when all the fish consumed was either a mix of fatty and lean fish (according to average preferences in Denmark) or only fatty fish. A smaller health gain was estimated when only consuming lean fish, while a marked health loss was estimated when all fish consumed was tuna. Women in the childbearing age were identified as a particularly sensitive subgroup based on the estimated health impact of maternal fish consumption on the neurodevelopment of unborn children. The substitution model proposed in this study did not account for variability in substitution behaviors in the population. The variability in the final health outcome was not assessed either. In order to get more detailed insight in which subgroups will experience a health loss and a health gain, respectively, the variability in the overall health impact between individuals was assessed in manuscript II investigating the same substitution as manuscript I. Variability between individual substitution behaviors was simulated using a probabilistic approach. In order to reflect only between-individual variability in the distribution of long-term intakes before and after the substitution, without the day-to-day variability for each individual, a model to adjust intake differences for within-individual variability was proposed. The approach taken for the substitution and modeling of long-term intake differences allowed for a quantification of the health impact at the individual level to reflect variability in the overall health impact of the theoretical intervention. The results showed that large variations exist in the overall health impact of substituting red and processed meat by fish in the Danish adult population. Elderly, in particular men (> 50 years of age), experienced the highest health gain, along with women in the child-bearing age (in particular those between 25 and 40 years of age), who were assigned the health impact experienced by their yet unborn children. However, a small fraction of these young women were assigned an overall health loss, primarily due to the adverse effects of methyl mercury on fetal neurodevelopment. This thesis proposed methods to integrate food substitution in RBA. Both the deterministic and the probabilistic substitution models suggest approaches to account for substitution of foods, and the choice of model will depend on which questions need be answered. The deterministic substitution model is valuable for defining extreme or worst-case scenarios, whereas the probabilistic substitution model is useful for investigating the variability in substitution behavior and the resulting health impact to identify sensitive subgroups in the population. We found that the overall health impact of the change in consumption of one food may depend on which other foods are changed as a consequence, stressing the need for accounting for substitution in RBA. Challenges related to the integration of substitution in RBA remain. The estimates obtained from RBA are often associated with uncertainties. Dose-response models were identified as a particular source of the quantified uncertainty in the RBA studies conducted in this thesis. Uncertainties associated with the choice of data for the dose-response modeling are difficult to quantify but contributes to the overall uncertainty of the estimated health impact. The implications of the choice of data to model the association between intake of a food or food component and a health effect were investigated in manuscript III. Large differences between dose-response functions based on data associating a health effect with either an intake of a food component, a non-specified substitution of a food, or a specified substitution between foods was illustrated. By estimating the health impact only based on the effects associated with the intake of a food component, the potential impact of the food matrix on the diet-disease association is neglected, while calculations based on association measures for non-specified substitutions ignore underlying food substitutions. When not specified in the statistical analysis of observational studies, it is unclear which substitution is reflected in the association measures and thus in the results of RBAs, building on these association measures. Furthermore, the health impact estimated in RBAs of food substitutions may in fact be a mixture of food substitutions, including other substitutions than that investigated. It was concluded that scenario analyses should be encouraged when different dose-response data with sufficient strength of evidence are available. Assumptions and limitations of the epidemiological study behind the dose-response data should be acknowledged and communicated along with the final results. Finally, the need for specification of food substitution in human observational studies was emphasized.