Biostatistical modelling for nutrition
Statistical methods are indispensable for high quality research. Sometimes standard methods are sufficient, but in other cases methods have to be specifically tailored to a particular research question, e.g. in handling measurement error in dietary assessment. Furthermore, in order to give advice on nutrition, the positive effects of a food need to be balanced against its disadvantages. This requires mathematical modelling and quantification of effects.
Research of the Division of Human Nutrition and Health in this field targets methods that present results of studies in a way that shows their relevance for Public Health. Research focusses first on the estimation of the effects of foods on health: such estimates should not be biased by effects of measurement error in dietary assessment. Development of statistical methods to correct for this bias is therefore a central issue. However, as devising better dietary assessment methods is always preferable over such corrections, there is also strong interest in statistical methods supporting the development of new dietary assessment methods, including prediction models. Secondly research targets estimation of outcomes such as population attributable risk or gains in healthy life expectancy that are informative for public health.
Statistical methods for handling measurement error in dietary assessment
Estimates of health effects of dietary exposures are biased by measurement error in dietary assessment. We therefore contribute to the development of methodology to obtain unbiased estimates.
Prediction of dietary intake from biomarkers
With increasing analytical possibilities - e.g. in metabolomics – possibilities arise to estimate dietary intake from a large number of related biomarkers, instead of using only a single biomarker. Statistical prediction modelling is an integral part of such an endeavour.
Calculation of outcomes relevant to Public Health (population attributable risks, life expectancy) from individual data
In epidemiological research that aims at finding etiological relationships, mostly relative outcome measures, such a relative risks are used. For the impact on public health, however, absolute measures relating to the entire population (and not just to risk groups) are more relevant. We contribute to methodology to calculate this type of measures both from individual studies and from population level data.
Statistical methods for combining data from different data sources
For decision making, risks and benefits of consumption of particular foods need to be balanced. This requires quantification of risks and benefits on the same scale. Healthy life years and disability weighted life years are measures that form such a scale on which risks and benefits can both be measured. Healthy life years and disability weighted life year of foods can be estimated using chronic disease modelling. Part of such modelling is the integration of data from many different sources. This challenge includes development of methods of statistical linkage of unrelated datasets and Bayesian modeling.