Food intake measurements: their validity and reproducibility

This thesis reports on the validity and the reproduciblity of methods assessing food consumption in groups of free-living Dutch adults consuming self-selected diets. The validity of a method, that is the demonstration that a method measures what it is intended to measure, can only be assessed by comparing it with an independent method of indisputable accuracy. There Is no such absolute method because of the nature of the data to be collected. Instead, it may be possible to establish either the relative validity or the concurrent validity. The relative validity evaluates the method in terms of another generally accepted method, designed to measure the same concept. The concurrent validity compares the results of a method with a biological marker. Validity studies refer to information bias and systematic response errors.A method is commonly called reliable or reproducible if it gives the same results when used repeatedly in the same situation. The problem in food consumption surveys is that the situation is never absolutely identical. Reproducibility refer to the biological within- person variation (a true day-to-day variation) as well as to random response errors, because these two sources of variation can hardly be separated.Many objectives can be served in conducting food consumption studies. It is essential to distinguish between food consumption surveys mainly used for national food planning and administration and surveys in which the emphasis is primarily on the relation between nutrition and health. Distinction is necessary because each objective demands a different type of information. In this thesis only methods assessing food consumption used in studies on the relationship between diet and health have been examined.Chapter 2 describes the state of the art of methods assessing food consumption. There are many sources of error in these types of studies as Table 4 in Chapter 2 clearly shows. For epidemiological studies, however, it is not always necessary to produce accurate results for each individual, but methods yielding valid data on groups of individuals would be of great value. To be able to obtain such valid data on a group level it is of the utmost importance to specify beforehand which data the study should supply. In Chapter 2, Table 6, the data required for three purposes have been formulated, viz.:A. Average energy and nutrient intake data of a group, to make group-to- group comparisons. In such a survey a representative sample of the population under study as well as a representative period of observation are required.B. The distribution of usual energy and nutrient intake data within a population, to detect groups of individuals at risk, if the risk is marked by a high or a low intake of a certain nutrient.C. The usual energy and nutrient intake of groups of individuals throughout time to make a correlation and regression analysis relating independent and dependent variables in order to make inferences on an aggregate level.For all three purposes data have to be collected on an individual level, but they are analysed and interpreted on an aggregated level. Chapters 3 to 8 describe the evaluation of methods used for the above-mentioned purposes in groups of Dutch adults. In all methods sizes of portions of the foods most frequently consumed have been checked by weighing by the interviewer. Other portions have been estimated in standard household measurements. The amounts of foods consumed have been converted into energy and nutrients using the national nutrient data base "UCV".In Chapter 3 the seven-day record method has been described assessing the energy and dietary fiber intake of 100 adults (44 men and 56 women). In the analysis the emphasis was on the within-person and the between-person variation in the intake of energy and dietary fiber. Information on these components of variance is desired to determine in study designs the number of persons and the number of records per person required to assess with a certain precision the group mean in the intakes of nutrients. The findings in this study show that the precision for the measurement of dietary fiber intake obtained by using one-day food records from twice the number of individuals is comparable to that obtained from a seven-day food record. For the measurement of energy intake, only 1.5 times the number of individuals is required for the one-day record method to obtain the same precision as with the seven-day record method. However, a one-day record method does not yield information on the usual food consumption of an individual. To be able to distinguish within the study population groups of individuals with a very high or a very low intake of a certain nutrient more food records per person are needed. The number of records required depends on the ratio of the within-person variation to the between-person variation.Chapter 4 describes the validity and the reproducibility of a dietary history method assessing the usual food consumption during one month. Forty-four young adults (aged 19-32 years) participated in this study. The concurrent validity of the method was assessed by means of the 24-hour urine nitrogen excretion. The mean difference between nitrogen intake and nitrogen excretion was 0.0 g with a 95 per cent confidence limit of ±1.1 g. These limits for the mean difference between excretion and intake indicate a valid assessment of the protein intake of this group.The reproducibility was evaluated in the same group through a test- retest design. The intra-class correlation coefficients were high over a weighted average of week-days and for an average workday as regards the intakes of energy and selected nutrients. As to the Saturday and Sunday intakes, the correlation coefficients were lower for the energy intakes and most of the nutrients (except alcohol) indicating a poorer reproducibility for week-end assessments.Chapter 5 describes a dietary history method assessing the food consumption In retrospect. In 1983 the relative validity of the retrospective dietary history method (DH'R) was assessed against a current dietary history taken seven years earlier, in 1976 (DH'76), and a second current dietary history taken in 1983 (DH'83). In total 44 men and 58 women, aged 38-62 years, participated in the study. For the intake of energy and most of the nutrients the relative difference between DH'R and DH'76 was below 15 per cent. The difference between energy and nutrient intake reported contemporaneously and retrospectively appears to be comparable with or slightly more than the reported change in food consumption between 1976 and 1983. Similar results were found for food groups. The results indicate that for group- to-group comparisons past dietary patterns can be ascertained retrospectively with some degree of success. Such data can be useful in case-control studies examining the relationship between diet and cancer.Seasonal variation in the intakes of energy and nutrients may affect the validity as well as the reproducibility of results in studies examining food consumption.In Chapter 6 the effects of the season on the energy balance of 114 young adult women are described. Energy intake and pattern of physical activity were assessed monthly fourteen times with the 24-hour recall method. The day following each interview before breakfast body weight without clothes was measured by the participants themselves. After these 14 months, in the second year the same estimates were made with intervals of 2-3 months to check if the observed seasonal variations were not a casual effect.The study did not demonstrate seasonal variation in the mean energy intake of the group under study. A statistically significant effect of the season was observed in the intake of fat, dietary fiber and mono- and disaccharides. For mono- and disaccharides this seasonal effect could not be confirmed the second year. Small seasonal fluctuations were observed in body weight and time spent in the various physical activity categories. On the one hand these fluctuations are too small to indicate physiological significance, on the other hand the fluctuations are wide enough to be taken into account in many longitudinal studies assessing changes in intakes of nutrients and body weight.Chapter 7 describes the same study as Chapter 6, but reports on the validity of the monthly, fourteen times repeated, 24-hour food recall method. Validity was examined by comparing the mean daily energy intake with fluctuations in body weight over a period of 14 months and by comparing the mean daily protein intake with the protein intake as assessed from the nitrogen excretion in at least 11 collections of 24- hour urine. A hundred and twenty-three women participated in this study. The data confirm earlier findings that the 24-hour recall gives a valid estimate of energy and protein intake on a group level.Categorizing the group in approximately quintiles of the energy intake revealed that subjects reporting a very low energy intake tend to underestimate their energy intake and that subjects reporting a very high energy intake tend to overreport their energy intake. In 20 subjects (16 per cent) the mean difference in daily protein intake as assessed with the fourteen 24-hour recalls and as assessed from the nitrogen excretion in 24-hour urine exceeded 20 grams. Sixteen of these subjects reported either a very high (n = 7) or a very low energy intake (n = 9). The results indicate that it is very difficult to determine within a group under study - solely on the basis of food consumption data - the percentage of subjects with a very high or a very low intake of certain nutrients. This implicates that it is also difficult to determine within such a group the percentage of subjects with a high or a low health risk, if the risk is marked by a high or a low intake of a certain nutrient.Chapter 8 reports on the validity of the fatty acid composition of subcutaneous fat tissue microbiopsies as an estimate of the long-term average fatty acid composition of the diet of separate individuals. Fifty-nine women taking part in the study on seasonal variation (Chapter 6) and whose food consumption data were considered valid (Chapter 7) participated in this study. Three months after the food consumption study had been finished subcutaneous adipose tissue samples were collected during a home visit. Highly significant correlations were found between the linoleic acid content of diet and fat tissue (r = 0.70) and also between the linoleic acid-to-saturated fatty acid ratio (linoleic/S) of fat tissue and diet (r = 0.62). This confirms the hypothesis that on an individual level the fatty acid composition of the adipose tissue is a valid index for the habitual dietary fatty acid composition of free-living individuals.When using one 24-hour recall instead of the average of 19 recalls, the correlation coefficient between the linoleic/S ratio of the diet and that of the adipose tissue was substantially lower. This demonstrates the weakening effect of the wide day-to-day variation in within-person intakes on the correlation between a short-term assessment of the nutrient intake of an individual and a biochemical indicator of a long-term nutritional status.In the final discussion of this thesis (Chapter 9) it is concluded that the methods assessing food consumption described appear to be relatively valid for estimating the mean intakes of energy and of the described nutrients In groups. If only data on food consumption are available, however, it is difficult to categorize unbiased within the study population groups with a low or a high intake of a certain nutrient. This is so even if, based on the precision required, the ratio of the witrhin-person to the between-person variation, enough food records or recalls are obtained per participants in a sample of sufficient size.If energy and nutrient intake data are to be related to health indicators, it is advocated to oversample and to check the intake data with an independent biochemical or physiological indicator of the intake of a nutrient. This makes it possible to discard data that are considered not to be trustworthy. The remaining data, considered valid, and collected within an adequately formulated research design, will be very useful in finding out the relation between nutrition, health and disease