Choosing a dietary assessment method
It is not possible to decide which dietary method to use until the purpose of the study has been clearly defined, since this will determine the kind of information and the length of time for which it needs to be collected from each individual. Often, the purpose of the study also determines the level of precision that is required to meet the study objectives and therefore the sample size. These two considerations are the most important ones in determining the method to be used, because both the method and the size of the sample have implications for the human and financial resources needed for the study.
When dietary data are collected to describe the diet of a group for comparison with that of another group or groups, it is possible to use either a short-term method such as a 24 hour recall or record, or a longer-term method such as food records obtained over several days, a diet history, or a questionnaire about habitual intake. The final choice will depend on factors such as the importance of a representative study sample, the resources available, and the level of precision required. Usually, the most efficient approach is to measure the diet of as many individu-als as possible for 1 day.
However, if the purpose of the dietary study is to determine the proportion of individuals in the group who are at risk of dietary inadequacy or excess, rela-tive to some standard of reference, then a single day of information on each individual is no longer ade-quate because it is necessary to have a reliable esti-mate of the distribution of habitual intake in the group. As Figure 10.2 shows, a single day of intake is generally not a reliable measure of an individual’s habitual intake.
Figure 10.2 Energy intake of one indi-vidual from weighed records obtained for 1 day every sixth day over 1 year. —, overall mean;
To determine the distribution of habitual food intake in a group, at least 2 days (preferably not consecutive) of information from each individual or a representative subsample of individuals from the group of interest are needed. If several days of intake are available they can be used to derive a mean intake for each individual and from this the distribution of average intakes for the group. Alternatively, statistical techniques can be used to adjust 1 day intake data, for the day-to-day variation that occurs in individuals, to provide a better estimate of the underlying distribu-tion of habitual intake for the group than is given by the 1 day data (Dodd et al., 2006). While the use of appropriate statistical techniques can improve esti-mates of the proportion of individuals at risk of defi ciency by adjusting for within-person variation, they do not enable at-risk individuals to be identifi ed.
When the purpose of the study is to assess the diet of specific individuals it is necessary to obtain dietary information over at least a week and preferably longer. This is best done by obtaining either multiple 24 hour recalls or 24 hour food records over an extended period. The minimum number of days needed to obtain an estimate of nutrient intake with a specified level of confidence differs for different nutrients. Information on energy intake, which tends to show less day-to-day variation than other nutrients, can be obtained over a shorter period (days) than informa-tion on a nutrient for which day-to-day intake is much more variable, such as vitamin A (weeks).
In studies of groups, precision is primarily a function of sample size, while in studies of individuals it is a function of the number of days of information avail-able. Precision increases with sample size and with the number of days for which information is collected, but so does the cost of the study. Precision therefore needs to be defined in relation to the purpose of the study.
Usually, what is required of the nutritionist is to be able to provide the statistician with an estimate of the level of difference that it is important to be able to detect (in nutritional, not statistical terms) and an estimate of the variance or standard deviation for the measurement(s) in question. For example, when looking for differences in energy intake between two groups, would a difference of 500 kJ or 1500 kJ be regarded as biologically significant?
Since the variance of a dietary measurement depends not only on the real variation within or between respondents but also on the error of the mea-surement, the precision of a dietary estimate can be improved not only by increasing sample size but also by reducing measurement error.
It is inevitable that the resources available, both finan-cial and human, also influence the choice of method. They should not, however, be the primary consider-ation. The method used should be determined by the question to be answered. If the method or methods needed to answer the question are beyond the resources available it is better either to abandon the study or to redefine the question than to collect inad-equate data.