Dietary assessment methods in epidemiological research: current state of the art and future prospects

Introduction

Adequate exposure assessment is a prerequisite in all epidemiologic investigations and presents particular challenges in studies of nutritional epidemiology. Diet represents an unusually complex exposure with strongly inter-correlated components. Early efforts to understand diet–disease associations focused on the role of specific nutrients, but later on it became evident that in several instances dietary exposures may act synergistically¹. Moreover, our eating habits may not only affect the way our genetic disposition is expressed but also probably participate in interplay with other lifestyle factors, such as physical activity and smoking². Trying to record what people eat is not an easy task, and, even when the best possible method or combination of methods are selected, some measurement error is introduced and needs to be accounted for in the analysis and interpretation of results³. Adequate dietary intake assessment is important not only in the study of associations between diet and health-related outcomes but also for nutritional surveillance and the evaluation of the nutritional status of patients in clinical settings. In the following paragraphs, we will refer to the most commonly used methods of self-reported dietary intake and we will review biomarkers of diet. We will also discuss sources of errors in the assessment of dietary intake as well as approaches to address them.

Methods of self-reported dietary intake

Methods of self-reported dietary intake can be grouped into two broad categories: methods of real-time recording and methods of recall. Real-time recording methods consist of food diaries (with or without weighing of foods consumed) and the duplicate portion method. When using food diaries, individuals are asked to record every food or beverage they consume in real time; in food diaries with weighing, individuals are also required to record the actual quantities consumed. In the duplicate portion method, pairs of daily portions are used: one is consumed by the individual, and the second is chemically analyzed for content. Because of their cost and complexity, the demanding duplicate portion method and the dietary diaries are less often used in large-scale epidemiologic investigations of diet–disease associations⁴.

Methods of recall include dietary histories, food frequency questionnaires (FFQs), and single or multiple daily recalls (24-hour dietary recalls [24-HDRs]). In the vast majority of the nutritional epidemiology literature, two dietary assessment methods prevail: the FFQs, mainly in studies assessing diet and disease associations, and the 24-HDRs, mostly in nutrition surveillance studies, which monitor the populations usual intake and can help identify subpopulations in need of dietary guidance.

The FFQ relies on the principles of the diet history method. It records the frequency by which an individual consumes foods and beverages—listed either collectively (e.g. green leafy vegetables) or individually (e.g. lettuce)—over a long period of time, most frequently a year. The FFQ may or may not include questions on the usual quantity consumed (semi-quantitative, quantitative or non-quantitative FFQs, respectively); information on quantities is often collected with the use of photographs of various portions and household or standard units. The questionnaire can be administered by an interviewer or filled in directly by the participant, who is usually asked to indicate the frequency of consumption through pre-determined options. A short FFQ may underestimate the true variation in dietary intake, but a very long and detailed one can be time and resource consuming and the burden on the responder may jeopardize data quality. Researchers balance their FFQ choices between these extremes so as to cover foods of their particular interest adequately but also to include enough information that would allow them to estimate the individual’s total daily energy intake⁵.

In a 24-HDR, participants are asked to recall and describe in detail and in an open-ended manner the foods and beverages consumed over one day, preferably the day before. Data collection can be either interviewer- or self-administered and it often follows the psychometric principles of structured, multi-faceted interviews that facilitate participants in their descriptions⁶. The use of technology further allows the integration of databases assisting the recording of the type and quantity of foods consumed. These databases include, for instance, specific information on recipe ingredients, packaging material, products’ characteristics (e.g. low-fat or vitamin D fortified) or even brand names, as well as multiple portion size measurement aids. However, since one or a selection of individual days may not be representative of a person’s usual diet, studies employing this assessment method should be carefully designed to include multiple administrations and cover both seasonal and weekly variations in intake⁵.

Sources of error are inherent to both FFQs and 24-HDRs. Both are prone to recall bias, particularly FFQs, since individuals are asked to report their intake retrospectively and usually refer to prolonged periods of time. Additionally, both can be hampered by the individual’s intentional misreporting of their consumption of certain foods, which can be affected by their personal characteristics (e.g. age, gender, overweight, or obesity) and could result in a differential misclassification with unpredicted consequences in the estimated associations⁷. Other errors in the assessment of diet through FFQs and 24-HDRs can be introduced via the use of food composition databases to calculate energy, nutrient, and alcohol intake; natural variations or limited information regarding the composition of processed and packaged products as well as foods prepared out of home are the main culprits⁸.

Biomarkers of dietary intake

A biomarker is a biological specimen that serves as an indicator of intake or metabolism of dietary constituents or an indicator of nutritional status^9,10.

A classification distinguishes biomarkers into recovery, concentration, replacement, and predictive biomarkers, although some biomarkers can fall into more than one of these categories¹¹.

Biomarkers can also be categorized into short-term (reflecting intake over hours/days and usually measured in urine, plasma, or serum), medium-term (reflecting intake over weeks/months and usually measured in red blood cells or adipose tissue), and long-term biomarkers (reflecting intake over months/years, and usually measured in hair, nails, or teeth)¹⁰.

In addition to the “traditional” dietary biomarkers, the field of omics is opening up new perspectives for identifying novel biomarkers. Hence, in the context of genomics, polymorphisms in the lactase gene, for instance, can be used to reflect milk consumption in Mendelian randomization analyses (these analyses rely on the assumption that genotype distribution is unrelated to confounders and use variation in genes of known function to examine the effect of a modifiable exposure on disease). Similarly, in epigenomics, we can examine if the epigenetic regulation of specific genes can be affected by food intake; in transcriptomics, we can compare differences in the gene expression profile among individuals following different dietary patterns; and in lipidomics, proteomics, and metabolomics, we can compare differences or changes in the respective profiles cross-sectionally or following particular dietary interventions¹⁰.

The use of dietary biomarkers is often recommended to overcome the errors of self-reported dietary intake and the bias introduced by the use of food composition tables¹¹. Nevertheless, using biomarkers to assess dietary intake also has its limitations¹³. Several inter-individual factors can operate and generate variation in biomarker levels, which does not reflect solely differences in dietary intake. Hence, in addition to the individual’s gender and age, tobacco smoking, alcohol consumption, medication, and physical activity can also affect the measurement¹⁴. Furthermore, interactions between dietary components, the type and handling of the biological samples (e.g. conditions related to blood drawing or urine collections and sample transport and storage), and the characteristics of the laboratory methods used (precision, detection limits, and inter-laboratory variations) may also contribute to the observed variations^10,13. For these reasons, the use of a biomarker in nutritional epidemiology should be preceded by an assessment of its validity, reproducibility, ability to detect changes, and suitability for the population under study. According to recent reviews^10,14, however, the majority of studies concentrate on the validity of biomarkers, whereas much less attention has been paid to the evaluation of their reproducibility and sensitivity in detecting changes in intakes and over time. In 2013, the US National Institute of Child Health and Human Development, supported by a consortium of NIH Offices and Divisions as well as the private sector, launched the Biomarkers of Nutrition for Development (BOND) program¹⁵. This program aims to harmonize aspects related to dietary biomarkers and provide advice to researchers, clinicians, and policy makers on the process of making decisions about their best use in individual situations. In 2014, the Food Biomarkers Alliance (FoodBAll) project was launched in Europe with the support of the EU Joint Programming Initiative “A Healthy Diet for a Healthy Life”. The program aims to develop strategies for food intake biomarker discovery and validation, and to identify and validate biomarkers for a range of foods consumed across Europe (http://foodmetabolome.org/).

Dealing with errors in dietary intake assessment

Well-designed studies on diet–disease associations often provide inconsistent findings. Over the years, the discussion has evolved on whether this could also, at least partially, reflect limitations of the dietary assessment methods that generate measurement error of different magnitude. Kipnis and colleagues¹⁶ described two potential components of the dietary measurement error. The first one reflects the correlation between error and true intake (“intake-related” bias), while the second is independent of true intake and represents errors related to the participant’s personal characteristics (“person-specific” bias). Measurement error can be systematic or random. Systematic errors reflect methodological weaknesses, appear at the group level, and generate differential misclassification. Random errors occur at the individual level, generate non-differential misclassification, and generally attenuate relative risk estimates and reduce statistical power to detect them³. That said, even random measurement error can lead to biases towards or away from the null depending on multiple factors^17,18.

Conclusion

In nutritional epidemiology, several methods for the assessment of dietary intake are available, but they all have their own limitations. To overcome possible sources of error, current practice recommends the concurrent administration of different dietary questionnaires and the collection of biological specimens to estimate levels of dietary biomarkers. Recent advances in the omics field that could enrich the list of reliable nutrition biomarkers as well as novel technologies applied to reduce respondent burden and reporting bias will open up new prospects in the field. In the analyses, food and nutrient intakes always need to be adjusted for total daily energy intake to account for errors related to reporting. Moreover, dietary pattern analyses are increasingly applied; they have shown consistent findings across cohorts and mitigate the issues related to food composition tables. Notwithstanding the limitations in exposure assessment, nutritional epidemiology has generated evidence which has contributed to the formulation of public health policies and related measures^30,41.

Abbreviations

24-HDR, 24-hour dietary recall; DLW, doubly labeled water; FFQ, food frequency questionnaire; NHANES, National Health and Nutrition Examination Survey; NIH, National Institute of Health; OPEN, Observing Protein and Energy Nutrition.