点击显示 收起
1 From the Division of Cardiovascular Epidemiology, Institute of Environmental Medicine, Karolinska Institutet, Stockholm (MSR and UHdF); the Center for Clinical Cardiovascular Research, Department of Cardiology, Karolinska Hospital, Stockholm (M-LBH and UHdF); and the Department of Food and Nutrition, University of Umeå, Umeå, Sweden (GKJ).
2 Supported by grants from the Swedish Council for Working Life and Social Research, the Swedish Heart-Lung Foundation, the Swedish Society for Medicine, the Swedish Medical Research Council, and Unilever Bestfoods. 3 Address reprint requests to MS Rosell, Institute of Environmental Medicine, Karolinska Institutet, Box 210, S-171 77 Stockholm, Sweden. E-mail: magdalena.rosell{at}imm.ki.se.
ABSTRACT
Background: Underreporting is a common problem in dietary surveys. Few studies have shown the implication of this when investigating diet-disease relations.
Objective: We investigated how underreporting affects the associations between dietary factors and the metabolic syndrome.
Design: Dietary intake measured with a 7-d food record, fasting insulin concentrations, and other variables of the metabolic syndrome were assessed in a cross-sectional study of 301 healthy men aged 63 y. Biological markers for intakes of protein, sodium, and potassium were measured in 24-h urine samples. Underreporters (URs, n = 88) were identified by Goldberg’s equation, which compares energy intake with energy expenditure, both expressed as multiples of the basal metabolic rate. Physical activity level was estimated, and individual cutoffs were calculated.
Results: The URs had higher nutrient and food densities in their diet than did the non-URs, which suggested that they followed a healthier diet. The URs had a higher prevalence of the metabolic syndrome than did the non-URs (18% and 9%, respectively; P = 0.029). The biological markers confirmed a low validity of the dietary data in the URs. The correlations between fasting insulin concentrations, a central component of the metabolic syndrome, and the intakes of polyunsaturated fats, n-6 fats, and fat from milk products were stronger in the URs than in the non-URs, which indicates that inaccurate data can introduce spurious associations.
Conclusion: The association between diet and fasting insulin differed between URs and non-URs in this study of 301 healthy men aged 63 y. If URs are not identified and excluded or treated separately in studies in nutritional epidemiology, spurious diet-disease relations may be reported.
Key Words: Diet • underreporting • metabolic syndrome
INTRODUCTION
There is a tacit assumption that we will know what people eat once we have performed a dietary survey. However, numerous studies have shown that it is difficult to obtain a representative picture of what people usually eat, mainly because of large day-to-day variations in food intake, misreporting, and changes in dietary habits during a given study. Underestimation of energy intake by 20% is common (1). Among obese subjects, this figure may rise to 50% (2). Although underreporting is more common in persons with a high body mass index (BMI; in kg/m2), other factors—eg, sex, age, smoking, educational level, and dieting behavior—and psychological factors—eg, self-image of body shape—have also been found to be related to underreporting (3–8). Unfortunately, a potential solution to underreporting is hindered by what appears to be a selective reporting of various nutrients and foods. For instance, protein is usually not underreported to the same degree as are carbohydrates and fats (9, 10), and between-meal snacks and foods considered to be unhealthy seem to be underreported to a greater extent than are those considered healthy (3, 4, 8, 11–13).
The fact that underreporting is not random and is selective for different foods raises serious concerns in investigations of the relation between diet and health. However, few studies have highlighted the implications of this problem. The aim of this investigation was to explore how underreporting affects the association between dietary factors and the metabolic syndrome in a sample of healthy 63-y-old men (n = 301). The approach was, first, to characterize underreporters (URs) and nonunderreporters (non-URs) with regard to the density of nutrients and foods in their reported diet; the validity of their reported energy, protein, sodium, and potassium intakes; and the presence of components of the metabolic syndrome, and, second, to investigate the relations between fasting insulin, which is a central part of the metabolic syndrome, and the intakes of fats, protein, and carbohydrates in the URs, the non-URs, and the group as a whole.
SUBJECTS AND METHODS
Study design and subject recruitment
The analyses are based on a cross-sectional study of 301 healthy men aged 63 y that was conducted between March 2000 and October 2001. The subjects were recruited from a cohort of men and women who had attended a baseline investigation regarding risk factors for cardiovascular disease in 1997–1999. Every third 60-y-old person in Stockholm County was invited to participate in that investigation, and 78% (n = 4232) agreed to do so. The examination included anthropometric measurements, the drawing of fasting blood samples, and a comprehensive questionnaire. Forty-nine percent of the men (n = 995) met the following criteria for participating in the present study: they were born in Sweden; they had no diagnosis of cardiovascular disease; they had had no pharmacologic treatment of diabetes, hypertension, hypercholesterolemia, or cancer; they had a BMI of 20–35; and they had no other serious disease. These men were divided into 3 groups by the tertiles of their fasting insulin concentrations. Requests to participate in a study concerning diet and the metabolic syndrome were continually sent out until the number of positive responders reached 100 in each group. This classification was used only to recruit subjects with a wide range of insulin concentrations and not for the analyses in this report. The participation rate was 71%. Two men did not complete the 7-d food record and were therefore excluded, which left 301 subjects for the present study. The ethics committee at the Karolinska Institutet approved the study.
Clinical procedures
The participants visited the Karolinska Hospital in a fasting state in the morning. Body weight was measured to the nearest 0.1 kg. Length and waist circumference were measured to the nearest 0.5 cm. BMI was computed. Systolic and diastolic blood pressures were measured to the nearest 2 mm Hg with a mercury manometer after 5 min of rest; the mean value of 2 measurements was calculated. Blood samples were drawn. Information about medications, smoking, and physical activity was recorded during a structured interview. Written and oral instructions on filling in a 7-d food record and collecting a 24-h urine sample were given individually. After 1 wk, the participants returned to the hospital with the completed food record and a urine sample. The food record was examined, and any ambiguities were resolved.
Dietary assessment
The food record that the subjects completed during 7 consecutive days was an optically readable version of a questionnaire used by the Swedish National Food Administration and Statistics Sweden in a national dietary survey performed in 1989 (14). We made small adjustments to match the printed food list to a later version of the food-composition data. We added 2 pages for recording between-meal eating, with both blank lines and printed alternatives for coffee, tea, sugar, buns, and small cakes. The predefined portion sizes were altered according to a validation study (15). The record contained printed alternatives for foods and dishes commonly eaten at main meals. Space was included for recording foods and snacks other than those in the printed list. The subjects estimated the amount of food with the use of household measures (eg, servings, glassfuls, cupfuls, and spoonfuls). Eight photos were used to estimate sizes of cooked food and the amount of fat spread on bread. The food quantities listed in the space for free text were converted to grams with the use of a weight guide (16). We calculated the intake of food and nutrients by using the food-composition database of the Swedish National Food Administration (PC-DIET, version 1/99; Swedish National Food Administration, Stockholm) (17), and SAS software (version 8.1; SAS Institute Inc, Cary, NC).
Urine collection
All 301 subjects collected one 24-h urine sample. The collections were made either before the week of food recording or during the first days of the week of food recording, except for 18 persons who made their collection at day 5 or later in the food record week. These 18 collections were excluded because the urinary nitrogen, sodium, and potassium concentrations were supposed to reflect the habitual diet, and the food habits might have changed during the week of food recording. We used the para-aminobenzoic acid (PABA)–check method to verify that the urine collection was carried out properly (18). Subjects were given 3 tablets containing 80 mg PABA to be taken evenly distributed during the day. We rejected collections containing < 50% PABA (n = 2). In the collections with a PABA recovery of 50–85% (n = 46), the contents of nitrogen, sodium, and potassium were adjusted according to Johansson et al (19). We excluded 11 because 4 of them had collections with a PABA recovery of > 100%, 6 missed taking 1 or 2 tablets, and 1 took paracetamol on the day the urine was collected. A total of 270 collections were accepted for the study.
Biological markers of food intake
Nitrogen content in 24-h urine was used as a biological marker for dietary intake of protein (20). The nitrogen content in urine was converted to dietary protein by multiplying by 7.72 (20). The sodium content in urine was used without alteration as a biological marker of dietary intake of sodium (21). The potassium content in urine was divided by 0.77 and used as a biological marker for dietary potassium intake (21).
Physical activity level
Physical activity during the previous year was recorded during the interview. The subjects were categorized in 4 levels of physical activity at work—very light (eg, sitting at the computer most of the day or sitting at a desk), light (eg, light industrial work, sales, or office work that comprises light activities), moderate (eg, cleaning, staffing a kitchen, or delivering mail on foot or by bicycle), and heavy (eg, heavy industrial work, construction work, or farming)—and 5 levels of physical activity during leisure time—very light (almost no activity at all), light (walking, nonstrenuous cycling, or gardening approximately once a week), moderate [regular activity at least once a week (eg, walking, bicycling, or gardening) or walking to work 10–30 min/d), active (regular activities more than once a week, eg, intensive walking or bicycling or sports), and very active (strenuous activities several times a week). The physical activity level (PAL) was systematically estimated for each subject according to a new method developed by Johansson and shown in Table 1.
View this table:
TABLE 1 . A scheme for estimating physical activity levels
Classification of underreporters
We classified URs and non-URs according to the Goldberg cutoff (22, 23). In this method, the reported energy intake is compared with the energy expenditure, and both are expressed as multiples of the basal metabolic rate (BMR). A cutoff for the reported energy intake is calculated, which indicates when the energy intake is too low to represent the habitual intake after taking variations in energy expenditure and energy intake into account. Goldberg’s equation is based on the principle that the energy intake should equal the energy expenditure, assuming weight stability. From this follows the principle that the food intake level (FIL) (24), which is calculated as the reported energy intake divided by the predicted BMR, should equal PAL, which is the ratio of energy expenditure divided by BMR (25). We estimated BMR from equations based on age, sex, and body weight (26). The lower 95% CI for FIL was calculated as follows:
RESULTS
Characteristics
The characteristics of the 301 men are shown in Table 2. There were no significant differences in smoking habits or educational level between the URs and the non-URs (data not shown). The prevalence of overweight, defined as BMI > 25, was 68% in the URs and 52% in the non-URs (P = 0.007). History of dieting or trying to lose weight was reported in 32% of the URs and 19% of the non-URs (P = 0.02). The mean body weight did not change in either the URs or the non-URs during the period of food registration (data not shown).
View this table:
TABLE 2 . Characteristics of the subjects1
Nutrient and food density
Compared with the non-URs, the URs reported a lower energy intake; lower intakes of energy-adjusted total, saturated, and monounsaturated fats, saccharides, and disaccharides; and higher intakes of protein and alcohol (Table 3). The URs had a higher food density of several water-soluble vitamins and minerals than did the non-URs (Table 4). The intakes of butter and margarine, buns and pastry, and chips and snacks were lower, and the intakes of bread, potatoes, meat and poultry, and fish and shellfish were higher in the URs than in the non-URs (Table 5).
View this table:
TABLE 3 . Daily energy intake and energy-adjusted nutrient intake in underreporters (URs) and nonunderreporters (non-URs)1
View this table:
TABLE 4 . Energy-adjusted intake of vitamins and minerals in underreporters (URs) and nonunderreporters (non-URs)1
View this table:
TABLE 5 . Energy-adjusted daily intake of food groups in underreporters (URs) and nonunderreporters (non-URs)1
Validity of energy, protein, sodium, and potassium intakes
In both groups, FIL was less than the Goldberg cutoff at a group level, which implies that the energy intake was underreported in both the URs and the non-URs (Table 6). The ratios of FIL to PAL and of diet to biological markers for protein, sodium, and potassium were significantly (P<0.001) < 1.00 in the URs and the non-URs (which indicated underreporting), except for sodium in the non-UR group, for which the significance was P > 0.05.
View this table:
TABLE 6 . FIL, PAL, Goldberg cutoff on individual and group level, and validation of energy, protein, sodium, and potassium intakes in underreporters (URs) and nonunderreporters (non-URs)1
The metabolic syndrome
The URs had a significantly greater waist circumference and higher systolic and diastolic blood pressures than did the non-URs. The prevalence of the metabolic syndrome was twice as high in the URs as in the non-URs (Table 7).
View this table:
TABLE 7 . Components in the metabolic syndrome and the prevalence of the metabolic syndrome in underreporters (URs) and nonunderreporters (non-URs)1
Associations between diet and fasting insulin
The correlations between fasting insulin and the intake of polyunsaturated fats, n-6 fat, and fat from milk products were significantly different in the URs and the non-URs. The correlations between fasting insulin and saturated fat, n-3 fat, and dietary fiber were strengthened when the URs were excluded (Table 8).
View this table:
TABLE 8 . Correlations between fasting plasma insulin and energy-adjusted nutrient intakes (as a percentage of energy) in underreporters (URs), nonunderreporters (non-URs), and the entire cohort1
DISCUSSION
The objective of this study was to investigate how underreporting affects the associations between diet and outcome variables with focus on diet and the metabolic syndrome. We identified the URs by using the Goldberg cutoff. This way of classifying the URs entails risks of misclassification, mainly because of uncertainties in choosing an appropriate PAL to include in the equation. PAL is usually not known in larger studies, and it is relatively common to use in the Goldberg equation a PAL of 1.55 (sedentary lifestyle) for the entire group or to use other limits, such as a PAL of 1.2 (sitting or lying) or 1.35 (the former Goldberg cutoff 1). However, these definitions will lead to an imprecise classification of URs at an individual level. According to Black (23), subjects should be assigned PALs that are at least somewhat different, eg, low, medium, and high. Black showed that the sensitivity for identifying URs was increased from 0.46, when a PAL of 1.55 was used for all individuals, to 0.70 when 3 different levels of PAL were used in a sample of 165 men, according to a golden standard based on a doubly labeled water technique (28). In each case, the specificity for detecting non-URs was 0.99. These results suggest that there is much to be gained if participants are categorized differently for PAL rather than the same PAL being used for all. The method for estimating PAL in the present study was developed by Johansson on the basis of available literature on how PAL corresponds to different degrees of physical activity (29). A limitation of this study is that this method remains to be properly validated against the doubly labeled water measurements. Nevertheless, the sensitivity for identifying URs should be better compared with if only one PAL was used for all individuals. In addition, the biological markers confirmed that the groups of URs and non-URs classified in this study differed in the validity of the dietary data. In a comprehensive analysis including all doubly labeled water measurements available until 1994, the mean PAL for men aged 40–64 and 65–74 y was 1.64 and 1.61, respectively (30). These figures are lower than the PAL of 1.72 found in our study, which indicates that we could have overrated the physical activity of our subjects. Although our method for estimating PAL may need further refinement, to our knowledge, this is the first time Black’s recommendation to obtain individual PAL values has been followed in a study of this size (23).
A history of dieting or trying to lose weight was more common in the URs than in the non-URs, a difference that was also reported in other studies (5, 6, 11). An association between BMI and underreporting has been noted in many studies (3–6, 8, 13, 31, 32) and was also seen in this study. Therefore, the characteristics of the URs in this study correspond with other observations, which indicates that URs are unevenly distributed in the population. This may have implications if corrections for underreporting are to be performed.
When the association between diet and health is studied, the nutrient intake is commonly adjusted for energy intake. Therefore, our study describes the dietary density of nutrients and foods. Lower intakes of fats (particularly of saturated fat), higher intakes of protein, and lower intakes of saccharides were seen in the URs than in the non-URs (Table 2). The intake of most water-soluble vitamins and minerals was higher in the URs than in the non-URs (Table 3). Among the food groups, there was a lower intake of butter and margarine, buns and pastry, and chips and snacks and a higher intake of bread, potatoes, meat and poultry, and fish and shellfish in the URs than in the non-URs (Table 4). Altogether, the nutrients and foods give an impression of a healthier diet for the URs than for the non-URs. This supports the hypothesis that the underreporting of foods is selective and that this selective underreporting affects the energy-adjusted nutrient intake in a biased way (3, 6). This in turn affects the obtained diet-disease relations (4, 8, 9, 12, 13).
Energy intake, energy expenditure, and the biological markers verified that the validity of the dietary data were low in the URs (Table 5). Although the mean energy intake in both the URs and the non-URs was too low to represent the habitual intake, the mean values of FIL and the biological markers were much lower in the URs than in the non-URs. The apparent underreporting of energy (0.60 and 0.86 in URs and non-URs, respectively) might not have resulted only from a too low FIL, but also from an overestimation of PAL. According to doubly labeled water studies, a typical PAL for this population is 1.63 (30). If this PAL is compared with the average FIL in the URs and non-URs, the ratio of FIL to PAL will be 0.64 and 0.90, respectively. These ratios are still lower than the corresponding ratio for protein in the URs and the non-URs (0.77 and 0.95, respectively), which suggests selective underreporting. This bias has also been shown in other studies (9, 10, 12).
When underreporting was related to the metabolic syndrome, a greater waist circumference and higher systolic and diastolic blood pressures were seen in the URs than in the non-URs. When the components were combined in the metabolic syndrome, a higher prevalence was seen in the URs than in the non-URs. Because hyperinsulinemia is thought to play a central role in the metabolic syndrome (33), we chose to correlate variables of dietary intake to fasting insulin. We found significantly different correlations in the intakes of polyunsaturated fat, n-6 fatty acids, and fat from milk products in the URs and the non-URs. The stronger correlations in the URs than in the non-URs indicate that inaccurate dietary data can introduce spurious associations. The data also suggest that underreporting can mask associations, because the correlations between fasting insulin and saturated fat, n-3 fatty acids, and dietary fiber became somewhat stronger when the URs were excluded. These results illustrate the importance of the validity of reported data to the obtained associations between diet and disease outcome variables.
Although underreporting is a well-known phenomenon in nutritional epidemiology, there are few studies that show the clinical implications of this problem. Conclusions can be altered when "low energy reporters" are excluded (34–36). Therefore, validation of dietary data is essential to make it possible to exclude or separately treat those who report data of poor validity. However, by excluding URs, who may represent people with poor health, the possibility of finding associations between diet and disease may be reduced. This problem becomes even more important in the light of the study carried out by Black and Cole (37), which showed that subjects classified as URs at one occasion also tended to be classified as URs when the measurement was repeated even if different assessment methods were used. It appears that it is more difficult to retrieve valid dietary data from some people than from others. Evidently, there is a need for biological markers as objective measures of dietary intake so that more valid conclusions can be reached in epidemiologic studies (38).
In conclusion, our study suggests that the association between dietary factors and the metabolic syndrome is greatly affected by underreporting. If URs are not identified and excluded or if they are treated separately in studies in nutritional epidemiology, spurious diet-disease relations may be reported.
ACKNOWLEDGMENTS
We thank Merja Heinonen for clinical assistance, Inga-Britt Gustavsson for developing the optical version of the food record, Bengt Vessby for allowing the use of the food record and the optical reader, and Lars Berglund and Rawya Mohsen for technical support with the optical reading of the food records. We thank Anders Hamsten, Karin Danell-Toverud, and Birgitta Söderholm for the ultracentrifugation of lipoproteins in serum.
MSR contributed to the data collection, analyzed and interpreted the data, and wrote the manuscript. M-LBH participated in the concept and design of the study, data collection, and edited the manuscript. UHdF was responsible for the concept and design of the study and edited the manuscript. GKJ contributed to the study protocol, formulated the hypothesis, interpreted the results, and contributed to the writing and editing of the manuscript. None of the authors had any conflict of interest with regard to the companies and organizations sponsoring the research.
REFERENCES