Dietary Assessment

Food intake consumption can be measured at the population, group or individual level and different methods are suited to these different needs. At the population level, food disappearance data are often used. For example food balance sheets are constructed by FAO (Food Agriculture Organisation) from national accounts of the supply and use of foods. This type of data is not based on individual measures of intake. Rather, aggregate data are based on surveys of groups of people. Some surveys rely on household purchases as a proxy for food consumption, but this may not take into account food eaten outside the home or food waste or within-household variation.

There are several methods available for assessing dietary intake at the individual level. These include dietary records, questionnaire methods and interviews. Methods commonly used in nutritional epidemiology are described below:

Dietary records/food diaries

Food diaries or dietary records involve the recording of types and quantities of food consumed over a specific period of time, typically 3-7 days. Portion sizes can be estimated, using for example household measures, or in some cases foods are weighed. A comparison of weighed versus estimated food diaries is currently being undertaken for the National Diet and Nutrition Survey in the UK.

Overall, the process is burdensome for the subjects taking part in the study and consequently highly motivated individuals tend to take part in these studies. Subjects need to be relatively numerate and literate, and there is a risk that usual diets might be modified to make the recording process easier (for example, avoiding eating out). However, in the past this method has been regarded as the most accurate method of dietary assessment. This is because there is no dependency on the memory of the subject, direct measurement of portion sizes can be done and variation in day-to-day intake can be assessed easily compared to other methods of dietary assessment, particularly if both week days and weekend days are included in the recording period.

Food frequency questionnaires (FFQs)

FFQs are usually self-administered and are therefore designed to be easy to complete by the study subjects. FFQs usually comprise a list of foods or food groups, and a corresponding frequency response section (e.g. never, once per week, twice per week etc). FFQs often rely on assumptions regarding portion size, and are limited by the amount of detail that it is feasible to include in the questionnaire. FFQs are often designed to gain information about specific aspects of diet, such as dietary fats or particular vitamins or minerals, and other aspects may be less well characterised. It is possible for the questionnaires to be semi-quantitative where subjects are asked to estimate usual portion sizes. In epidemiology, FFQs are often filled in with reference to the previous year in order to ascertain usual food consumption patterns for that period.

The structured design of the questionnaire means that data can be easily processed and computerised. Furthermore, the relative ease of administration and affordability makes FFQs appropriate for use in large scale studies and they are frequently used in cohort studies. An important advantage of this method of dietary assessment is the low burden on the study subjects, compared to recording methods. However, accuracy of measurement of absolute intakes is lower than for other methods (Gibson, 1998). FFQs are mainly used in epidemiology for ranking individuals into broad categories, e.g. high, moderate and low consumers of vegetables.

24 hour recalls

In a 24 hour recall interview subjects are asked to report their food intake from the previous day. Trained interviewers will ask the subject, in a structured manner, for a list of foods consumed, probing as appropriate, and then proceed to ascertain further details such as brands and cooking methods used. Portion sizes will be estimated, using, for example, household measures or food photographs. If habitual dietary intake is required for the study then multiple 24 hour recalls can be undertaken. In such cases it is important to include both week and weekend days in the recall periods. The repeat visits make this method more costly, although it is possible to conduct the recalls by telephone where appropriate.

A key advantage to using this method is that subjects require no training and need only spend minimal effort in providing the information requested by the interviewer. Furthermore, processing data from a 24 hour recall is cheaper and less time consuming than processing data from 3-7 day dietary records. 24 hour recalls are useful for assessing average usual intakes of a large population, and are therefore often used for large dietary surveys (Gibson, 1998). However, a single 24 hour recall cannot take account of day-to-day variation in an individuals’ food consumption and, due its retrospective nature, the 24 hour recall is less suitable for use with children and the elderly. Diet histories A diet history is an interview method whereby a trained interviewer conducts a 24 hour recall with a subject, followed by a discussion on usual eating behaviour and possibly a food frequency questionnaire to further clarify this data. This method can be used to obtain retrospective information from the previous month or year. The effectiveness of dietary assessment by diet history is very dependent on the skill of the interviewer. It is labour-intensive and time-consuming, and relies on the memory of the subject. The unstructured nature of the enquiry renders it more useful in clinical settings than research.

The choice of dietary assessment method in research depends on the study objectives together with the time-scale and resources available for the project. None of the available methods is without limitations, and it is imperative to take into account all considerations at the design stage. Bearing in mind the objectives of the study and the population under investigation, it is important to clarify the detailed format of the chosen dietary assessment method. For example the considerations for developing a FFQ include:

• Self-administered or administered (by an interviewer) questionnaire
• Computerised or paper form.
• Location of questionnaire completion. This might affect the amount of time available to complete the questionnaire.
• Open or closed format of questions. For example, ‘how often do you eat bread?’ with a free text response, or ‘How many slices of bread do you eat per day?’ with frequency options to select.
• Sequence and wording of questions. Questions should be unambiguous and it is important to start a questionnaire with a simple question
• Data coding for analysis, i.e. how answers will be converted into numerical values
• Range and compatibility checks on data. Rules need to be developed so that unlikely and contradictory data can be identified.

Fehily & Johns (2004)

In all cases, it is crucial to conduct a pilot study before a large scale study is undertaken.

Measurement error

All dietary assessment methods involve measurement error. Random measurement error refers to the precision of methods, whereby increasing the number of measurements will reduce random error and improve precision. Systematic measurement error cannot be minimised by extending the number of measurements. This type of error is important because it can introduce bias e.g. where respondents to a questionnaire may overestimate consumption of ‘good’ foods, such as fruit and vegetables.

Errors arise from assessment of the frequency of consumption of foods, portion size, daily variation, failure to report usual diet (either due to changing habits whilst taking part in a study or misreporting foods consumed), and the use of food composition data (Nelson and Bingham, 1997).

FFQs are frequently associated with errors due to overestimation of consumption, possibly resulting from the use of lists. Participants using the FFQ may have difficulty in choosing a frequency category so that over-estimation tends to occur.

The estimation of portion size rather than direct weighing is associated with imprecision at the individual level and to differing degrees for different foods. This is a greater issue for measuring intake of foods rather than nutrients.

Daily variation in intake is one of the main factors introducing imprecision to diet records and 24 hour recall methods of dietary assessment. A 7 day record is probably sufficient to rank the distribution of energy and macronutrients in a population, but longer periods are necessary for investigating alcohol, vitamins and minerals.

Problems of under reporting have been demonstrated by studies that have compared dietary assessment methods of food intake with biological measures (biomarkers); under reporting in obese individuals is now well recognised (Mendez et al. 2005; Rennie et al. 2004).

Many study objectives relate to the effects of food components, e.g. vitamins or phytochemicals, on disease outcomes. Food composition data are required to convert the information from dietary assessment methods to intake data on nutrients and other food components. Errors can be introduced at this stage because of an assumption that all subjects eat foods with the same standard composition and portion size. In reality, food composition varies widely depending on soil (in the case of plants), harvesting conditions, storage, processing and preparation of foods. Unless portion sizes are directly assessed, non-standard portion sizes may also lead to bias. Furthermore, food composition tables can be incomplete, for instance, in respect of phytochemicals or fatty acids. Incomplete food lists are also problematic, especially for some countries where indigenous foods are not analysed and their composition recorded, or where analyses on new food products or recently modified foods (e.g. lower in fat) are not available.

The EuroFIR project

In assessment of complex diets, measurement error can change the direction of estimated relationships, or produce spurious results. Measurement error may also cause observed associations to appear smaller than the true effect. As discussed above, food composition data can be an important source of error in the investigation of diet and disease relationships. The EuroFIR (European Food Information Resources Network) project has been set up to develop and integrate a comprehensive, coherent and validated databank to provide a single authoritative source of food composition data for Europe. A European food composition database can increase the quality of epidemiological data and the comparability of nutrition data across European countries.

References

Fehily AM & Johns AP (2004) Designing questionnaires for nutrition research. Nutrition Bulletin 29(1): 50-56.

Gibson G (1998) Dietary assessment. In: Essentials of human nutrition (eds Mann J & Truswell S), pp 409-426, Oxford University Press: New York

Mendez MA, Wynter S, Wilks R et al. (2004) Under- and over-reporting of energy is related to obesity, lifestyle factors and food group intakes in Jamaican adults. Public Health Nutrition 7(1): 9-19.

Nelson M & Bingham SA (1997) Assessment of food consumption and nutrient intake. In Design Concepts in Nutritional Epidemiology (eds Margetts BM ; Nelson M), 2nd edn, pp123-169, Oxford University Press: New York

Rennie KL, Jebb SA, Wright A et al. (2005) Secular trends in under-reporting in young people. British Journal of Nutrition 93(2): 241-247.

Further reading

Biro G, Hulshof KFAM, Oversen L et al.(2002) Selection of methodology to assess food intake. European Journal of Clinical Nutrition 56(2): S25-S32.

 Bingham SA, Gill C, Welch A et al. (1995) Comparison of dietary assessment methods in nutritional epidemiology: weighed records v 24 h recalls, food-frequency questionnaires and estimated-diet records. British Journal of Nutrition 72: 619-643.