Open AccessDose‐response relationships in health risk assessment of nutritional and toxicological factors in foods: development and application of novel biostatistical methods
Author(s) -
Vinceti Marco,
Filippini Tommaso,
Malavolti Marcella,
Naska Androniki,
Kasdagli MariaIosifina,
Torres Duarte,
Lopes Carla,
Carvalho Catarina,
Moreira Pedro,
Orsini Nicola
Publication year - 2020
Publication title -
efsa supporting publications
Language(s) - English
Resource type - Journals
ISSN - 2397-8325
DOI - 10.2903/sp.efsa.2020.en-1899
Subject(s) - medicine , random effects model , meta analysis , clinical study design , breast cancer , cohort study , risk assessment , clinical trial , environmental health , statistics , computer science , mathematics , cancer , computer security
Dose‐response meta‐analyses are of key relevance to identify causal relations between exposure and health‐related endpoints based on epidemiologic evidence, thus providing a powerful tool for both investigators and risk assessors. So far, their use has been limited in food safety risk assessment and in epidemiologic studies with either experimental or non‐experimental design, also since they could not be applied in studies with two levels of exposure only. However, the growing number of epidemiologic studies, the need to identify and shape relations between exposure and endpoints that are not linear, such as those L‐, U‐ and J‐shaped, and to locate possible thresholds of exposure which may characterize beneficial and adverse effects of dietary constituents, have increased the need of biostatistical tools for dose‐response modelling in meta‐analyses.We addressed these issues in two case‐studies, the relation between cadmium exposure and breast cancer risk in non‐experimental cohort studies, and between potassium exposure and blood pressure in randomized controlled trials, using a recently developed methodology for ‘one‐stage’ dose‐response modelling. This statistical methodologyis based on restricted cubic spline models fit with a generalized least‐squares regression, combining study‐specific estimates with a restricted maximum likelihood method within a multivariable random‐effects meta‐analysis. Such method allows to use studies based on less than three categories of exposure, such as trials based on two arms only.The implementation of such modelling in our two case studies hasshown that cadmium exposureis not generally related with breast cancer risk in cohort studies, and that potassium intake has a U‐shaped relation with both systolic and diastolic blood pressure, also depending on hypertensive status.Overall, the availability and implementation of the one‐stage dose‐response meta‐analytic approach yields a flexible and powerful tool to comprehensively summarize and model the relation between dietary constituents and health endpoints based on epidemiologic evidence, greatly favouring the implementation of the risk assessment process.