Retrospective dietary assessment methods, such as short-term recall, dietary history, and the food frequency questionnaire (FFQ), are designed to assess the subject's past diet. Short-term recall, however, is seldom suitable for epidemiological studies, because long-term dietary intake is often more relevant than the current diet when associations between exposures and diseases are examined (Willett 1998). Compared to an interview-based dietary history, a self-administered FFQ is cheaper, requires less time to complete and is easier to enter into a mainframe computer. Because FFQ is easy for participants, it leads to a high response rate, which is the most important reason why FFQ is becoming the main method in large population studies (Thompson and Byers 1994).
FFQs are based on the detailed dietary history interview developed by Burke (1947). That interview included a checklist on which subjects indicated frequencies and amounts of consumed foods and drinks during a specified period. Although the present form of food frequency questionnaire was developed during the 1960s (Wiehl and Reed 1960, Stefanik and Trulson 1962), it did not become common until the 1980s when large cohort studies between diet and chronic diseases were initiated (Willett 1998). The main purpose of a FFQ is not to estimate absolute intakes, but to rank subjects into exposure categories according to food consumption or nutrient intakes, often over the entire past year (Thompson and Byers 1994). Such ranking is adequate for most epidemiological studies to assess the overall relative risk of disease.
The important issues concerning FFQs are a well-designed food list and frequency categories. Foods included in the list should be consumed reasonably often in the basic population, have a high between-person variance and cover the relevant contributors of nutrients of interest (Willett 1998). Often 10 to 15 foods can explain over 80% of the intake of a special nutrient (Byers et al. 1985). Callmer et al. (1987) found that individuals in developed countries regularly consumed 80 to 120 food items. Therefore, about 100 food items or aggregate foods are usually enough to assess the intakes of most nutrients. An inquiry into the frequency of consumption has typically involved 5 to 10 frequency categories (Willett 1998). The increasing order of frequency categories, from never to frequently used food items or mixed dishes, is recommended because the decreasing order may increase the frequency responses for some food items (Kuskowska-Wolk et al. 1992). The overestimation was especially related to the consumption of bread, vegetables, and fish.
Three approaches on estimating portion sizes have been presented: a simple FFQ in which no information on portion sizes is collected, a semi-quantitative FFQ in which a certain portion size is specified for each food item or foods, and an open-ended FFQ in which subjects can select portion sizes freely (Willett 1998). Portion sizes are, however, difficult to evaluate (Guthrie 1984, Smith et al. 1991), mainly because of large within-person variation for each food (Hunter et al. 1988). Further, when FFQs with and without portion sizes were compared, no significant differences appeared (Humble et al. 1987, Tjønneland et al. 1992). Exceptions are studies by Block et al. (1990, 1992b) in which correlations between diet records and FFQs improved when gender and age-specific portion sizes were taken into account.
FFQs more often assess aggregate foods rather than individual food items. Questionnaires are also quite inflexible in measuring unusual diets (Thompson and Byers 1994, Lissner et al. 1998). FFQs are usually designed for specific populations, which implies that any new questionnaire should be evaluated against a more established method, such as diet records or biochemical indicators, in that population (Buzzard and Sievert 1994). The purpose of the evaluation is to yield guidance for the interpretation of the results in that particular study (Thompson and Byers 1994).
Reproducibility, i.e., reliability, refers to how consistently the same method repeats the diet of the same subject at different occasions without overlapping reference periods. In reproducibility studies of FFQ carried out in 1964-1997, correlation coefficients have ranged from 0.5 to 0.7 for nutrients and from 0.4 to 0.7 for food items (Willett 1998). Beverages have had the highest and foods seldom consumed the lowest correlation coefficients (Colditz et al. 1987). The reproducibility of dietary factors resembles the values of many biological measurements, such as serum cholesterol and blood pressure, widely used in epidemiological studies (Willett 1998).
Validation, on the other hand, means how well the method measures what it is supposed to be measured. Diet records are regarded as a "gold standard" against other dietary assessment methods, although they have some weaknesses of their own (Thompson and Byers 1994). Nevertheless, diet records describe more accurately a subject's diet than FFQs, and their inherent inaccuracies are independent of those of FFQs (Willett 1998). To estimate most nutritional factors adequately requires 7 to 14 diet-record days (Jeor et al. 1983, Willett et al. 1985, Nelson et al. 1989, Hartman et al. 1990), although the motivation of subjects may decrease if the records are kept for more than 7 record days (Gersovitz et al. 1978).
The relative risks of a disease are often measured according to quartiles or quintiles of the exposure. Therefore, it is important to rank subjects into categories correctly. Willett et al. (1985) and Pietinen et al. (1988) have reported that about 75% of the subjects were classified into the same or adjacent quintile according to the FFQ and diet records.
Use of biochemical indicators is a possible reference method to objectively assess the validity of dietary information furnished by the participants. In addition to validation, biochemical indicators can also be used as indicators of dietary intake or nutritional status, and for assessment of susceptibility to disease, and metabolic effects and risk of disease (Kohlmeier 1991).
Although human cells, tissues, and fluids offer a large number of biochemical parameters for monitoring of the nutritional status (Kohlmeier 1991, Kok and van't Veer 1991), only a few biochemical indicators are known to assess dietary intakes reasonably well. Good indicators exist for intakes of fatty acids, vitamins, and some minerals, whereas for intakes of total fat, total carbohydrate, and cholesterol, just to mention a few, no biochemical indicators are available (Kohlmeier 1991, Hunter 1998). One difficulty in finding good markers is that associations between nutrient intakes and biochemical indicators are not always linear because of variation among subjects in nutrient absorption and metabolism (Willett and Lenart 1998). Another problem may be that other nutrients such as fat and carbohydrate may change the association between biochemical indicators and nutrient intake.
Several biological sources are available for estimating selenium intake, for example, plasma, erythrocytes, urine, hair, and nails (Riboli et al. 1987). Of these, nails are best suited for epidemiological studies since they are easy to collect, transport, and store, and they are less exposed to environmental contamination than is hair. Nails also give a good estimate of long-term selenium intake (Longnecker et al. 1996, Hunter 1998).
Selenium intake has correlated well with toenail selenium concentration in some studies (Swanson et al. 1990), though not in all (Ovaskainen et al. 1993). A trial with three groups of four men consuming a high (4.9 µmol), a medium (2.6 µmol), or a control dose (0.41 µmol) of selenium per day showed that toenail selenium concentration provided an approximation of selenium intake over six to twelve months (Longnecker et al. 1993). It has also been reported that sex, obesity, current smoking, alcohol consumption, and intake of other antioxidants may affect the selenium status, and that these factors should thus be considered in the analyses (van't Veer and Alfthan 1991, Virtanen et al. 1996, Hunter 1998).
Food frequency questionnaires are widely used in large epidemiological studies since they are cheap, are easy for the participants, and they assess the past diet. Because the questionnaire is usually designed for a particular population, interpretation of the results requires information on the accuracy of the method (reproducibility and validity). Diet records are generally regarded as the best reference method for FFQs, because the errors in records differ from those related to FFQs. It has been shown that FFQ measures relatively well the overall levels of food consumption and nutrient intake, although it may reflect the diet which participants would like to consume rather than what they really eat (Kimmick et al. 1997).
Biochemical indicators are objective, and thus highly recommended for use together with FFQs in epidemiological studies. The reason for this recommendation is that some nutrients can be measured accurately by a questionnaire, whereas some others can be measured less accurately or sometimes only by use of biochemical indicators (Willett 1991). The problem is that biochemical indicators have not yet been found for all dietary factors, for example, for total fat, total carbohydrate, and cholesterol. Good indicators exist for fatty acids, and for some vitamins and minerals, such as selenium.