Epidemiologic evidence supports an association between diet and several chronic diseases, including cardiovascular disease, type 2 diabetes, and several types of cancer (1-6). Intervention trials designed to promote dietary change are important to determine effective ways to implement change in populations with disparate risk behaviors and outcomes. The Healthy Directions studies within the Harvard Cancer Prevention Program Project (HCPPP) have implemented interventions through worksites and health care centers, focusing on a multiethnic, working-class (e.g., clerical, sales, skilled and unskilled labor) population (7-9). The aims of the interventions were to decrease red meat intake and increase fruit and vegetable intake, multivitamin use, and physical activity.

Differential overreporting is a potential problem when monitoring change in intervention studies. People who have been advised to change their diet may have an increased awareness of their intake and a desire to seem compliant (10,11). Self-reported change at the end of follow-up may reflect intervention-related bias rather than actual changes in diet. Therefore, validation studies are an important component of dietary intervention studies (12,13). An ideal validation study of a diet-related intervention would accomplish three goals: validate that the questionnaire is adequately measuring intake, validate the questionnaire-based dietary change from preintervention to postintervention, and finally, calibrate the primary data collection tool to eliminate potential intervention-related biases. These goals are often difficult to accomplish, particularly with budget restrictions common in many intervention studies.

To address these goals within the Healthy Directions studies, we conducted an ancillary study to examine plasma nutrient levels within a subset of participants at the end of follow-up. We present the results of these analyses, along with a discussion of the benefits and limitations of conducting such a study and suggestions for incorporating biomarker validation in future intervention studies.

Characteristics of the study population at blood collection by treatment group (within each study and within the whole ancillary study) are presented in Table 1. The proportion of women varied between the studies but was similar between groups when the studies were combined (46% in the usual care group and 51% in the intervention group). The usual care group in the HC study had a higher prevalence of black participants (44%) than the other groups, which had more similar racial distributions. Higher income was more prevalent in the HC study intervention group, and the HC study had a higher prevalence of college-educated participants than the SB study. The distribution of smoking was similar between intervention groups in each study, but current smoking was more prevalent in the SB study. The HC intervention group was older (54 years) than the usual care group (48 years), and the HC study participants were slightly older and had a higher BMI than the SB participants. At the end of follow-up, fruit and vegetable consumption was higher in the intervention group, not only in each study but also when studies were combined (3.8 servings per day for the intervention group and 3.1 servings per day for the usual care group). Similarly, when studies were combined, the prevalence of multivitamin use was higher in the intervention group (57%) than the usual care group (30%).

Concentrations of several nutrients differed between the usual care and intervention groups, combined across studies (Table 2). Significantly higher concentrations of retinol (P = .01) and α-carotene (P = .03) were detected in the intervention group than the usual care group, which in turn had higher levels of δ-tocopherol (P = .02) and γ-tocopherol (P = .002). 

When participants were stratified by reported multivitamin use (concordant pairs), those who used multivitamins 6 or more days per week had significantly higher concentrations of retinol (P < .001), cis-β-carotene (P = .01), trans-β-carotene (P = .02), total β-carotene (P = .02), and α-tocopherol (P < .001) (Table 3). γ-Tocopherol was significantly higher in those who were not regular multivitamin users (P = .002). When stratified by reported fruit and vegetable intake (concordant pairs), participants who consumed 4 or more servings of fruits and vegetables per day had higher concentrations of lutein and zeaxanthin (P = .05) and β-cryptoxanthin (P = .05) than those who ate fewer than 4 servings (Table 4).

Correlations between plasma nutrients and reported multivitamin intake varied by treatment group (Table 5). Retinol was correlated with multivitamin use overall (r = 0.36, P < .001) but was not significant in the usual care group (P = .08). β-Carotene was more strongly and significantly correlated with intake in the usual care group (r = 0.39, P = .01 for cis-β-carotene; r = 0.34, P = .02 for trans-β-carotene; and r = 0.33, P = .02 for total β-carotene). α-Tocopherol was significantly correlated with multivitamin use in both groups (usual care: r = 0.35, P = .01; intervention: r = 0.54, P < .001). γ-Tocopherol was significantly inversely associated with multivitamin use in both groups (usual care: r = –0.38, P = .01; intervention: r = –0.40, P = .002). 

Correlations between plasma nutrients and reported fruit and vegetable intake also differed by treatment group (Table 6). Fruit and vegetable intake was significantly directly correlated with α-carotene when both groups were combined, (r = 0.22, P = .02) and with cis-β-carotene (r = 0.34, P = .02), trans-β-carotene (r = 0.39, P = .006), and total β-carotene (r = 0.39, P = .006) in the usual care group but not in the intervention group (r = 0.14–0.17, P = .21–.28). In the usual care group, the correlations of lutein and zeaxanthin (r = 0.31, P = .03) and β-cryptoxanthin (r = 0.35, P = .01) with fruit and vegetable intake were higher than in the intervention group (both r = 0.23, P = .08, .09).

To confirm the questionnaire-based findings of dietary change, we examined plasma nutrient concentrations at the end of follow-up by treatment status and found higher levels of retinol and α-carotene in the treatment group. To further explore nutrient differences, we stratified the pairs by multivitamin use and fruit and vegetable consumption instead of treatment group. We found higher retinol, β-carotene, and α-Tocopherol in participants who reported regular multivitamin use than in those who used multivitamins infrequently or not at all and higher lutein and zeaxanthin and β-cryptoxanthin in participants who reported consuming 4 or more servings of fruits and vegetables per day than in those who ate fewer. To verify the data collected by questionnaire, we examined correlations between reported multivitamin use and fruit and vegetable consumption. Multivitamin use was significantly directly correlated with retinol, cis-β-carotene, trans-β-carotene, total β-carotene, and α-Tocopherol and significantly inversely associated with γ-Tocopherol. However, the β-carotene components were only significantly correlated with multivitamin use in the usual care group; retinol and γ-Tocopherol were more highly correlated in the intervention group. Correlations between reported fruit and vegetable intake and cis-β-carotene, trans-β-carotene, total β-carotene, lutein and zeaxanthin, and β-cryptoxanthin were only significant in the usual care group; α-carotene was significantly correlated with consumption when both groups were combined. We did not have adequate data to calibrate the survey responses to the measured plasma nutrient levels.

The nutrient differences and correlations by multivitamin use and fruit and vegetable intake were expected. Higher plasma α-carotene, β-carotene, and α-Tocopherol levels have been observed after multivitamin supplementation (17-20). Common multivitamin formulations often contain retinol, β-carotene, and α-Tocopherol but are less likely to contain zeaxanthin or β-cryptoxanthin (21). The significantly higher levels of δ- and γ-Tocopherol among nonusers is likely a result of supplementation with γ-Tocopherol (the most common tocopherol in multivitamins), because it reduces plasma concentrations of δ- and γ-Tocopherol (22). Plasma carotenoid levels reflect consumption of carotenoid-rich foods (23-25) and have been shown to be an effective measure of a dietary intervention (26,27). Participants who reported higher fruit and vegetable intakes had higher plasma concentrations of lutein and zeaxanthin (which are commonly found in dark green, leafy vegetables) and β-cryptoxanthin (which is found in citrus fruit and orange juice) (28).

The higher concentrations of some plasma nutrients at the end of follow-up in the intervention group support the findings of significant increases in reported fruit and vegetable consumption and multivitamin use in these groups in the HC and SB studies (29,30). However, given that some characteristics were not well-balanced between the treatment groups among the subset of participants in the ancillary blood study, the nutrient differences may have been influenced by factors unrelated to the intervention. For instance, the higher prevalence of more highly educated, higher income participants in the intervention group may have contributed to the differing nutrition status; for example, they may have had a higher intake of fruits and vegetables at baseline and follow-up but not as a result of the intervention. In addition, the intervention group also had a higher mean age and BMI and slightly higher level of physical activity. Older age has been associated with higher concentrations of retinol and carotenoids (31,32). Although micronutrient concentrations are less likely than macronutrients to be affected by caloric intake, it is still possible that people who consume more calories (e.g., those with higher BMIs, higher physical activity levels, or both) also consume more micronutrients and thus may have higher plasma concentrations of micronutrients (12). Given that blood samples were pooled only accounting for treatment group and analyzed in pairs, we were unable to adjust our results for these factors.

Pooling the blood samples also decreased our sample size when stratifying by multivitamin use and fruit and vegetable intake, because we restricted the analyses to pairs that were concordant with respect to the dichotomized stratification factors. Despite the decreased sample size, we were still able to detect differences between the concordant pairs. However, although differences in nutrient concentrations were found by reported fruit, vegetable, and multivitamin intake, it is difficult to determine whether these differences are the result of the intervention, because only postintervention samples were collected. If preintervention and postintervention samples were available, within-person changes in plasma nutrient concentrations could have been compared with changes in reported intake.

Overall correlations of reported intake of multivitamins and fruits and vegetables with plasma nutrients suggest that the 5-A-Day survey adequately discriminated between individuals. However, the difference in correlations between the intervention and usual care groups is a concern. The nonsignificant correlations in the intervention group suggest that the questionnaire is not adequately measuring intake in this group. Although it is possible that biomarkers are representative of longer-term intake and may not reflect dietary changes within 1 year, dietary supplement studies suggest that increases in plasma concentrations of retinol, carotenoids, and tocopherols are apparent within 4 to 16 weeks of supplementation (33,34). Perhaps a more plausible source of this error is that participants who did not change their eating behaviors as advised overreported their fruit and vegetable consumption on the follow-up survey. Thus, it is possible that the questionnaires incorporate a systematic bias in the intervention group. Similar results have been found in other intervention studies. For instance, in a dietary intervention study, the ratio of reported fiber intake to fecal fiber content increased from baseline to 12 months, suggesting that fiber intake was overreported at the end of follow-up (35). Similarly, systematic underreporting of caloric intake in intervention groups was reported in a study of childhood obesity (11). Thus, when participants strive to comply with difficult interventions such as dietary changes, systematic bias may affect the results.

The results of this ancillary blood study suggest that the behavioral change intervention altered intake of multivitamins, fruits, and vegetables. However, several limitations to this validation study prompt a few recommendations for future studies. First, given the potential for intervention-related bias, data could be collected to calibrate or correct survey responses to minimize the impact of the bias. For example, a third instrument not prone to systematic, intervention-related bias could be used to regress intake on the nutrient biomarkers to calculate a calibration factor that could be applied to the survey data (12).

Second, cost-reduction measures should be planned to maintain a breadth of analysis options in the validation study. Blood samples were pooled to reduce assay costs so that we could conduct an ancillary study within the budget. Pooling samples is an effective way to perform a preliminary screen by comparing means among groups or to develop more specific hypotheses (15). However, pairing prevented us from adjusting for individual characteristics and reduced our effective sample size for some analyses. If cost reduction is necessary, pooling could be accomplished with carefully matched pairs so that the individuals in a set would be more similar with respect to potential confounders or with respect to nutrient intake (e.g., by deciles of reported fruit and vegetable intake). Thus, samples should be collected and stored individually, allowing time to match individuals on factors collected over the course of the study, and pooled just before being sent to the laboratory for analysis. Alternatively, other measures might be taken to reduce costs in future studies, such as minimizing the number of assays. For instance, a few biomarkers could be chosen to reflect hallmark components of multivitamins, fruits, and vegetables, such as red blood cell folate and plasma α-Tocopherol, β-carotene, and lutein and zeaxanthin (12,17-20,23-25), instead of a more extensive panel.

Finally, given the difficulty of validating change in reported intake with only postintervention biomarkers, preintervention and postintervention blood samples could be used to compare biomarker changes within individuals, thus removing the variability caused by individual metabolic differences (12). For example, change in serum cholesterol concentration is a relatively good marker of change in saturated fat intake, whereas a cross-sectional measure of cholesterol is likely more closely related to individual metabolic differences than to saturated fat intake, given the strict homeostatic regulation within individuals (12,36,37). Thus, improvements could be made to produce more valuable data in future validation studies.

Overall, the ancillary blood study was an important component of the Healthy Directions behavior intervention studies. We achieved the first goal of a validation study in that we found correlations between plasma concentrations of several nutrients and reported intake of fruits, vegetables, and multivitamins. However, the correlation analysis also revealed potential intervention-related bias. We found group differences in the postintervention samples, but other factors may have contributed to the differences in nutrient concentrations. Without preintervention blood samples, we could not compare the effect of intervention measured by reported change with change in plasma nutrients. To incorporate a more thorough validation component, future intervention studies might include preintervention and postintervention blood samples, with fewer assays or pooling that accounts for more matching factors to save costs.