Use of a food frequency questionnaire in American Indian and Caucasian pregnant women: a validation study

The importance of nutrition during pregnancy has long been appreciated. Overall nutritional adequacy, i.e., a sufficient supply of calories and protein, is a major determinant of weight gain during pregnancy [1, 2]. Maternal intake of carbohydrates and protein [3], fatty acids [4], and micronutrients such as zinc [5‐7], iron [2], magnesium [8‐10], calcium [11], riboflavin [4], and vitamin C [12]may also have important effects on fetal growth and perinatal outcomes.

Food frequency questionnaires (FFQs) have been shown to be valuable tools for evaluating long-term dietary intake in the context of epidemiologic research [13], and some investigators have examined their reliability and validity among pregnant women [14‐22]. While early studies focused on the validity of FFQs for assessing intakes of calories, protein, and a small number of micronutrients during pregnancy [21, 22], several recent studies have examined a larger group of nutrients [15‐17]. The results generally indicate that FFQs can be used to classify pregnant women according to their nutritional intake with a reasonable degree of accuracy, although this may vary according to the population and the number of food items on the instrument.

Low-income pregnant women may have poor nutritional status [23], and those with fewer years of formal education may have more difficulty completing a FFQ [24]. Food frequency questionnaires should be developed and tested among low-income pregnant women, as they could be useful for examining associations of nutrient intakes with pregnancy outcomes and for identifying groups of women who would benefit most from nutritional interventions. Block reported on the validity of two different FFQs (Block and Harvard) for assessing the diets of low-income pregnant, lactating, and non-pregnant, non-lactating women enrolled in the Women, Infants, and Children (WIC) Program in New York, California, Texas, and Ohio [19]. The Harvard Service FFQ (HSFFQ) has also been validated among low-income pregnant women attending prenatal health clinics in Massachusetts [16, 22]. These studies, however, only included white, African American, and Hispanic women. The validity of FFQs for assessing dietary intake among pregnant women of other racial/ethnic backgrounds has yet to be determined.

The purpose of this study was to further expand the validity of the HSFFQ for use in the assessment of relative dietary intake among low-income American Indian and Caucasian pregnant women.

Study participants ranged in age from 16 to 40 years, with a mean age of 24. Of the 279 women who completed Phase 1, 104 (37%) were American Indian and 175 (63%) were Caucasian. Sixty-three percent of women who completed Phase 1 were at 100% or less of the poverty level, and 43% had no previous livebirths. Population characteristics were similar for women who completed Phase 1, those who completed Phase 2, and those who completed the entire study (Table 1).

Pearson correlation coefficients for energy-adjusted nutrient intakes estimated from the FFQ and the 24-hour diet recalls for all participants ranged from 0.03 to 0.52 at 12 weeks and from 0.21 to 0.48 at 28 weeks; the average correlations for both Phase 1 and Phase 2 were 0.35 (Tables 2 and 3). After correction for random within-person variation, the deattenuated correlations for all participants ranged from 0.09 for polyunsaturated fat (95% CI: -0.13, 0.31) to 0.67 for calcium (95% CI: 0.53, 0.77) for Phase 1 and from 0.27 for sucrose (95% CI: 0.09, 0.43) to 0.63 for total fat (95% CI: 0.47, 0.75) for Phase 2. Average deattenuated correlations for the two phases were 0.48 and 0.47, respectively. Sixteen nutrients had deattenuated correlations of 0.50 or greater at 12 weeks and 11 nutrients reached correlations of 0.50 or greater at 28 weeks (Tables 2 and 3). The greatest percentage agreement for being ranked in the lowest quartile by both methods was 59% at 12 weeks (vitamin B2) and 57% at 28 weeks (zinc). The greatest percentage agreement for being ranked in the highest quartile by both methods was 49% at 12 weeks (calcium and phosphorus) and 50% at 28 weeks (vitamin B6).

Deattenuated correlation coefficients for most nutrients were fairly similar across subgroups when computed separately by ethnicity and poverty level (Tables 2 and 3). However, average deattenuated correlations comparing nutrient intakes estimated from the FFQ and the diet recalls were slightly lower for American Indians than for Caucasians (0.46 vs. 0.51 for Phase 1, 0.37 vs. 0.50 for Phase 2) and slightly higher for women at 100% or less of the poverty level than for those at 101% or greater (0.51 vs. 0.48 for Phase 1, 0.54 vs. 0.40 for Phase 2). Deattenuated correlation coefficients were also higher for women with one previous livebirth than for those with none and those with two or more previous livebirths. The average deattenuated correlations comparing nutrient intakes estimated from the FFQ and the diet recalls for Phase 1 were 0.47 for those with no previous livebirths, 0.58 for those with one previous livebirth, and 0.41 for those with two or more previous livebirths, and these were similar for Phase 2 (data not shown). Deattenuated correlation coefficients for omega 3 fatty acids and polyunsaturated fat could not be computed in some subgroups due to small numbers of participants and high within-person variation; hence, these nutrients were not included in average deattenuated correlations for any of the subgroups.

When nutrient intakes estimated from the week-12 set of recalls and the week-28 set of recalls were compared to one another to examine the stability of diet over the course of pregnancy, correlations between energy-adjusted intakes at 12 weeks and 28 weeks ranged from -0.02 for omega 3 fatty acids to 0.51 for calcium, with an average correlation of 0.33 (Table 4). The correlations between intakes at 12 weeks and 28 weeks were also similar for most nutrients within subgroups defined by ethnicity and poverty level. Nutrients with the greatest percentage agreement for the extreme quartiles at 12 weeks and 28 weeks were vitamin B2 and folate, with 54% agreement for the lowest quartile, and dietary fiber, with 53% agreement for the highest quartile (Table 5).

The results of this study indicate that the Harvard Service FFQ can provide a reasonable assessment of relative nutritional intake among low-income American Indian and Caucasian women during pregnancy. In validation studies conducted in other adult populations, average correlations between nutrient intakes from FFQs and diet recalls have typically been in the range of 0.6 to 0.7 [13], which are slightly higher than those observed in this study. However, it is not surprising that correlations would be somewhat lower among pregnant women, given that diet during pregnancy may be less stable than during other time periods. The correlations in this study are comparable with those reported among other groups of pregnant women and suggest that the HSFFQ may be a useful tool for classifying these women according to nutritional intake.

Food frequency questionnaires are generally considered to be the most appropriate method for assessing diet in the context of epidemiologic studies. Besides offering practical advantages over more expensive and time-consuming methods, such as 24-hour diet recalls and food records, FFQs can provide better assessments of usual intake over longer periods of time, such as weeks or months, rather than a single day [13]. Because pregnant women's diets may change within each trimester, the assessment of diet during the past month is likely to be more accurate than during the past year. Previous studies have examined the validity of FFQs among pregnant women. A community-based study of 569 women in the United Kingdom reported correlations ranging from 0.27 to 0.55 for 20 energy-adjusted nutrients assessed by a 100-item FFQ and four-day food diaries at 15 weeks of pregnancy [18]. Brown et al. reported an average deattenuated correlation of 0.45 for changes in energy-adjusted intakes of 15 nutrients during pregnancy among 56 well-educated white women, assessed by a FFQ and by four-day food records [17]. In a study of 113 Finnish women in their third trimester of pregnancy, Erkkola et al. obtained an average deattenuated correlation coefficient of 0.53 for 45 nutrients assessed by a 181-item FFQ and two five-day food records [15].

Low-income pregnant women may have poorer nutritional status than other pregnant women. A recent study reported that diet quality among 2063 pregnant women in North Carolina who were between 24 and 29 weeks of gestation was significantly lower among poorer women and those with less education [23]. In addition, Kristal et al. assessed the validity of a FFQ among 1015 postmenopausal women according to race/ethnicity and level of education, and their results indicated that the validity of the FFQ increased with years of education and was higher among whites compared to blacks [24].

Only a few studies have evaluated the validity of FFQs among low-income pregnant women. Suitor et al. examined the validity of the HSFFQ for the assessment of total calories and seven nutrients (protein, calcium, iron, zinc, and vitamins A, B6, and C) among 95 low-income pregnant women in Massachusetts, comparing intakes estimated from the FFQ to those estimated from three 24-hour diet recalls [22]. With the exception of vitamin A, all of the deattenuated correlation coefficients for nutrient intakes were greater than 0.50. Wei et al. extended these results in the same group of women by examining the validity of the HSFFQ for the assessment of 17 additional nutrients; they reported a mean deattenuated correlation of 0.47, with correlations ranging from 0.03 for vitamin B12 to 0.90 for zinc [16]. In a study among African American, Hispanic, and white pregnant women in the WIC Program in four states, deattenuated correlation coefficients between energy-adjusted intakes assessed by the HSFFQ or the Block FFQ compared to 24-hour diet recalls were reported for five nutrients [19]. The highest correlation for the HSFFQ was for calcium (0.47) and the lowest correlation was for vitamin C (0.08); the correlations for the Block FFQ were very similar.

These previous studies did not examine whether the performance of the HSFFQ varies according to stage of pregnancy. The results of the present study indicate that the HSFFQ has similar validity during the first and second trimesters, as shown by the average deattenuated correlation coefficients of 0.48 and 0.47 for the week-12 and week-28 FFQs compared to the 24-hour diet recalls. These correlations are comparable to those observed among other groups of pregnant women [15‐17, 22]. This study is also unique because it examined the validity of a FFQ among American Indian pregnant women. Although the observed correlations were slightly lower for American Indians than for Caucasians, the results still suggest that the HSFFQ can be used in a variety of populations to assess nutritional intake during pregnancy.

Sixteen of the nutrients examined had deattenuated correlations of 0.50 or greater between energy-adjusted intakes estimated from the FFQ and the diet recalls at 12 weeks, and 11 nutrients had correlations of this magnitude at 28 weeks. In contrast, the correlation for polyunsaturated fat was very low in Phase 1 (deattenuated r = 0.09), and the 95% confidence intervals for polyunsaturated fat and omega 3 fatty acids included negative values either in Phase 1 or Phase 2. This was a particular problem in the stratified analyses and may have been a consequence of small sample size, high day-to-day variation in intake of some nutrients among individuals in this population, and reduced access to certain foods that are sources of these nutrients, such as fish [34]. The unstable correlations observed for polyunsaturated fat and omega 3 fatty acids may also be explained by the fact that the HSFFQ was designed specifically for the WIC population to assess intakes of seven nutrients (protein, calcium, iron, zinc, and vitamins A, B6, and C) but not fat, whereas the Minnesota Nutrient Data System used for the diet recalls was designed to assess fat content. Another limitation of the HSFFQ is its inability to estimate absolute nutrient intake. Because the HSFFQ has a limited number of items and minimal information about portion size, it is not intended to provide accurate estimates of absolute intake; therefore, our findings apply only to its use as a research tool for classifying these women according to their relative nutritional intake [13]. It is also possible that the correlations between nutrient intakes from the HSFFQ and the diet recalls could be overestimated, because the two methods have correlated sources of error (i.e., both rely on memory) and the process of completing the recalls prior to the HSFFQ could have improved awareness of food intake [13].

The present study complements past and recent research on the reproducibility and validity of the Harvard Service FFQ. Based on our results, the HSFFQ may be a useful research tool for assessing relative nutritional intake among low-income American Indian and Caucasian women during different stages of pregnancy. It could provide important information for epidemiologic studies of pregnancy diet in relation to fetal outcomes and also serve as a starting point for identifying groups of women who may benefit most from nutritional interventions. Further research is needed to evaluate the validity of FFQs for the assessment of nutrients with high day-to-day variation and for particular subgroups of pregnant women.