Integrated approach for dietary exposure and biomarker measurements in aetiological models

  • Topic: Cancer-related outcomes
  • Institution: International Agency for Research on Cancer (IARC)
  • Country: France
  • Status: Ongoing

Scientific abstract

(View plain language abstract)

Background

In nutritional epidemiology, one important limitation is the challenge of assessing accurately individual's diet, particularly in observational contexts. With very few exceptions, the vast majority of dietary biomarkers do not reflect exposure to absolute intake of corresponding foods or nutrients. It is well acknowledged that all measurements of intake contain components of random and systematic errors. Although correction methods have been proposed, the techniques generally make heavy use of statistical assumptions on the different observed quantities, ie. dietary questionnaires (DQ), 24-hour dietary recalls (24-HDR), biomarker concentrations. The association between dietary exposures and cancer risks is most often carried out independently, either using dietary variables based on self-reported estimates or biomarker measurements of dietary exposure. Results from the two approaches are distinctly presented and discussed. A common setting that integrates information on dietary exposures from different assessments into a unique model is actually lacking in nutrition epidemiology.

Objectives

The aim of this proposal is to develop an integrated statistical model to accurately evaluate the association between dietary exposure(s) and risk of disease, and simultaneously explore the measurement error structure of observed dietary exposures, ie. biomarker levels, DQ and 24-HDR measurements.

Setting & Methods

An integrated latent factor hierarchical model will be developed in a Bayesian framework. Bayesian estimation will be carried out by Markov Chain Monte Carlo (MCMC), which generates samples from the joint posterior distribution of all parameters. Detailed dietary and biomarker data from the European Prospective Investigation into Cancer and Nutrition (EPIC) study will be utilized. EPIC has a vast resource of already existing nested case-control studies with available DQ, 24-HDR and biomarkers measurements for different cancer sites. For this proposal, we will focus on the challenging relationship between dietary fat and risk of breast cancer, which to date is still inconclusive. The association has been extensively evaluated independently using fat intake estimated from dietary questionnaires, or using biomarkers of fatty acids measured in blood. An integrated model will be built for poly-unsaturated (total omega-3 PUFA) and mono-unsaturated fatty acids, for which both dietary variables and plasma phospholipid biomarker values are available. In EPIC, plasma phospholipid values are currently being measured among 4,852 breast cancer case-control sets. Similar models will be used to combine dietary and serum levels of folate and vitamin B6, in order to explore their association with the risk of lung cancer in a nested case-control study with 899 lung cancer cases and 1,770 matched controls.

Impact

Our proposal to use information from different dietary assessments to investigate the association between dietary factors and risk of cancer into the same statistical framework will make use of all available information on dietary exposure(s). This will lead to more accurate risk estimates. New insights on the complex measurement structure of observed quantities will be provided. This will increase our knowledge into the system of dietary factors and their inter-relationships, which constitutes a key step towards better understanding of aetiological relations in nutritional epidemiology. Also, this work will set the basis for further developments of complex modelling with dietary data.

Plain language abstract

Background

One of the major limitations faced by any researcher engaged in nutritional epidemiology is the difficulty to measure the diet of a group of individuals, one of the reasons being that is cognitively very challenging. Therefore, estimates of dietary consumption are usually not very accurate. An advocated alternative is to use measures that do not rely on study subjects' capacity to remember their dietary consumption, i.e. biochemical markers, often measured in blood or in urine samples. Unfortunately, with very few exceptions, the vast majority of dietary biomarkers do not provide information on the absolute level of dietary exposure. There is general consensus that all types of dietary assessment contain components of measurement errors. Although correction methods have been proposed, these techniques generally make use of statistical assumptions that in practice are very difficult to hold. Thus, the correction techniques employed to overcome measurement errors are not always effective. The evaluation of the relationships between the role of dietary factors on the probability of developing a specific cancer is customarily carried out independently using either measures of self-reported intake (prone to mis-classification) or biomarker levels (prone to lack of specificity). This often results in rather inconsistent scientific evidence.

Aims & Objective

The aim of this proposal is to combine all available estimates of dietary consumption, i.e. self-reported and biomarker levels, into the same integrated statistical model. The rationale behind is that each quantity will contribute substantial information to the model, whereas the specificity of each dietary measurement will be accounted for. The model will, thus, produce more accurate estimates of association between dietary exposure(s) and risk of cancer.

How it will be done

The integrated statistical model will use the most recent advancements in the field of statistics, i.e. a Bayesian model, very often applied to complex situations. The model will be hierarchical in the sense that a very complex networks of relationships among different variables will be broken down into simpler relationships, so that each variable in the model will be related to only a few other variables. The model will be developed using dietary and biomarker data from the European Prospective Investigation into Cancer and Nutrition (EPIC) study. EPIC has a vast resource of already existing nested case-control studies with available self-reported and biomarker measurements. For this proposal, we will focus on the challenging relationships between dietary fat and the risk of breast cancer, and between dietary and serum levels of folate and vitamin B6 and the risk of lung cancer.

Potential Impact

Our proposal to use information from different dietary assessments to investigate the association between dietary factors and risk of cancer into the same statistical framework will make use of all available information on dietary exposure(s). This will lead to more accurate risk estimates. New insights on the complex measurement structure of observed quantities will be provided. This will increase our knowledge into the system of dietary factors and their inter-relationships, which constitutes a key step towards better understanding of aetiological relations in nutritional epidemiology.