European Journal of Clinical Nutrition (2014) 68, 316–323 & 2014 Macmillan Publishers Limited All rights reserved 0954-3007/14 www.nature.com/ejcn

ORIGINAL ARTICLE

The biomarker-based validity of a food frequency questionnaire to assess the intake status of folate, pyridoxine and cobalamin among Iranian primary breast cancer patients S Pirouzpanah1,2, F-A Taleban2, P Mehdipour3, M Atri4, A Hooshyareh-rad5 and S Sabour6 BACKGROUND/OBJECTIVES: Folate, pyridoxine and cobalamin are coenzymatically essential in one-carbon methyl metabolism, and their deficiencies could explain some alterations during breast carcinogenesis. We aimed to evaluate the validity of folate, pyridoxine and cobalamin estimates from a food frequency questionnaire (FFQ) on the basis of their corresponding fasting plasma biomarkers, in breast cancer (BC) patients. SUBJECTS/METHODS: In a prospective, consecutive case series, 149 women with primary BC aged between 30 and 69 years as a representative sample of Iranian women with BC were recruited. The 136-item FFQ was used for the validity assay. Fasting plasma folate and cobalamin were tested by automated electrochemiluminescence. The high-pressure liquid chromatography with fluorescence detection was used to determine the plasma levels of pyridoxal-5’-phosphate (PLP) and total homocysteine (tHcy). RESULTS: Area under the curve (AUC) for assessing the diagnostic accuracy of folate-related data through an FFQ was 0.74 (Po0.01) in the reference model (folate plasma levelo5.9 ng/ml), with sensitivity and specificity of 68% and 63%, respectively. The positive and negative predictive values (PPV and NPV) were 96.9% and 96.8%, respectively. The AUC for cobalamin intake in the reference model (plasma cobalamino260 pmol/l) was 0.64 (Po0.01), with 60% sensitivity and 61% specificity. Although tHcyX10.0 mmol/l was used as reference indicator, the folate intake (AUC ¼ 0.71, Po0.01) and cobalamin intake status (AUC ¼ 0.67, Po0.05) were also determined appropriately by FFQ. CONCLUSIONS: Dietary folate and cobalamin estimates from FFQ were significantly correlated with their fasting plasma concentrations. Our data supported the validity of new FFQ to rank individuals by dietary intake status of folate and cobalamin. European Journal of Clinical Nutrition (2014) 68, 316–323; doi:10.1038/ejcn.2013.209; published online 30 October 2013 Keywords: food frequency questionnaire; folate; cobalamin; validation study; total homocysteine; breast cancer

INTRODUCTION Several lines of epidemiologic evidence and preclinical investigations have shown that diminished folate status in longitude could increase predominantly the risk of breast cancer (BC) carcinogenesis.1–4 In addition, some findings have suggested that inadequate intake of cobalamin chronically might contribute to BC risk, meanwhile this condition was considered often in keeping with folate deficiency status.5–7 However, among B vitamins, less information exists regarding the nutritional status of pyridoxine among BC participants in previous epidemiologic studies.8 There is abundant evidence concerning the intake status of these nutrients in association with cancer risk.2,9 However, in terms of nutrient estimation using questionnaires, the accuracy of data remains an issue of a widespread epidemiologic debate in many studies.10,11 Despite the fact that the validity of the nutrients is so often investigated in non-malignant participants in various studies,2,4,8,10 to our knowledge, little attention has been paid to the accuracy of dietary measurements on the micronutrients related to one-carbon metabolism among BC patients.

It is widely evident that the accuracy of data obtained by means of a dietary questionnaire, that is, food frequency questionnaire (FFQ), in a target population, is supposed to be verified before launching a survey of the dietary trends and associations in epidemiologic studies.12 For this concept, the validity of certain nutrients from FFQ is usually attained through a comparing test method with a more accurate reference method (that is, true gold standard).12,13 Even though the sources of errors derived from using the questioning-based instruments as reference are similar and could result in some extent of overestimation, using the corresponding biomarkers attributed to the relative nutritional intake status is increasingly regarded as accurate reference criteria in dietary validation approaches.4,14–17 The plasma folate level was suggested as a suitable indicator for reflecting recent dietary intake status.4,18 Although the red blood cells’ (RBC) folate level has also been shown as an appropriate biomarker for evaluating folate intake status in long term, various recent studies have suggested that the RBC folate concentration could rarely provide privileged information for identifying

1 Department of Community Nutrition, Faculty of Health and Nutrition, Tabriz University of Medical Sciences, Tabriz, Iran; 2Department of Clinical Nutrition & Dietetics, Faculty of Nutrition Sciences and Food Technology/National Nutrition and Food Technology Research Institute, Shahid Beheshti University of Medical Sciences, Tehran, Iran; 3Department of Medical Genetics, School of Medicine, Tehran University of Medical Sciences, Tehran, Iran; 4Cancer Institute, Tehran University of Medical Sciences/Day General Hospital, Tehran, Iran; 5Nutrition and Food Technology Research Institute, Shahid Beheshti University of Medical Sciences, Tehran, Iran and 6Department of Clinical Epidemiology/Safety Promotion and Injury Prevention Research Centre, Faculty of Dentistry, Shahid Beheshti University of Medical Sciences, Tehran, Iran. Correspondence: Dr S Pirouzpanah, Department of Community Nutrition, Faculty of Nutrition, Tabriz University of Medical Sciences, Attar-Neyshapouri Ave., Golgasht Ave, Tabriz, 5166614711, Iran. E-mail: [email protected] Received 13 December 2012; revised 8 September 2012; accepted 12 September 2013; published online 30 October 2013

The validity of FFQ in breast cancer S Pirouzpanah et al

317 suboptimal folate status as compared with the serum folate level.18–20 Recently, fasting plasma total homocysteine (tHcy) has also been shown to serve as a sensitive inverse indicator of serum or plasma folate status.19 In addition, the interrelationships of cobalamin and folate in homocysteine metabolism might also be synergized by plasma status of pyridoxal-50 -phosphate (PLP).21–23 These suggest that tHcy, as a reference determinant, could additionally be used to evaluate the diagnostic accuracy of folate, and even possibly for cobalamin, and pyridoxine derived from FFQ. By taking into account the particular Iranian dietary habits, which is rather comparable to the Mediterranean diet,24 and the exceptional diversity of foods providing these nutrients, it would be noteworthy to probe the certain nutrients among BC patients by means of validated FFQ.17,25 Therefore, the aim of present study was to evaluate the validity of FFQ to assess the intake levels of folate, pyridoxine and cobalamin on the basis of plasma biomarkers and also fasting plasma tHcy in primary BC patients. MATERIALS AND METHODS Study population In this prospective, consecutive case series study, 149 women within the range of 30 to 69 years of age diagnosed with histopathologically confirmed malignant breast tumor were recruited from 2008 to 2010 at the Day General Hospital, Tehran. The eligibility criteria were defined previously.7,26 Particularly, patients having used certain vitamin supplements (folate, biotin, cobalamin and pyridoxine) during a month before sampling were excluded. A well-trained questioner was conducting face to face interviews with each subject. Informed consent was obtained from each subject in the hospital before questioning. An institutional review board approval was granted for this study. The existence of underreporting was evaluated by calculating the physical activity level defined by the ratio of reported energy intake and a mean basal metabolic rate.27,28 The basal metabolic rate was estimated according to the Schofield equations.29

Nutritional assessments Food frequency questionnaire. The present semi-quantitative FFQ (FFQ as test method) is a 10-page instrument consisting of 136 food items, 25 questions about food preparation and 25 fields for open-ended questions. The content and face validities were confirmed previously.30 The focus of the present questionnaire was on 10 specified food categories for enhancing the recall accuracy. These food clusters included bread and cereals, dairy products, meat, legumes, fruits, vegetables, oils, nuts, beverages and finally ended with seasonings. Some edible green leafy fresh vegetables and herbs are regularly consumed in side dish (such as Tarreh as leek-like leaves, mint, parsley, water cress-like leaves, radish, onion, dill, tarragon, scallion, local basil, coriander, spinach, garlic leaves, Parpin, respectively, arranged on the basis of the percentage of usage among the participant), stewed vegetables (spinach, Tarreh, parsley, Kanagr, Bamieh, fenugreek, carrot, green bean, celery, eggplant, gourd, quince and so on) and salads are major dietary sources of folate in Iranian dietary habits.31,32 The food items within the Iranian habitual diet without any equivalent English word have been written in Farsi, while the first character is shown in uppercase. Respondents of the FFQ were asked on their intake over the previous one year. Set of color photographs and customary household utensils were also used to depict different portion sizes. Seasonal variations of fruits and vegetables were weighted for the season during analyses. The consumption frequency of each food item was responded in daily, weekly, monthly or yearly period of time. The amount of dietary consumption was asked on the basis of the standard serving size provided for each food with a privilege to average portion size. For assessing nutrient intakes, the time-dependent frequency of each food item was multiplied by the amount of intake in servings and then converted to the daily magnitude in grams.31 The Nutritionist IV Software (version 3.5.2; 1994, N-Squared Computing, San Bruno, CA, USA) was implemented to analyze the intake level of folate, coblamin, pyridoxine and total calorie for each participant from the included list of foods. The US drug administration (USDA) national nutrient database for standard reference is the leading data bank in this software.33 The supported & 2014 Macmillan Publishers Limited

database in the software for some particular foods was also modified and adjusted on the basis of nutritive contents of some Iranian foods.33 Residual dietary intake was also calculated to adjust the effect of total calorie intake for certain dependent nutrients, which was described by Willett.14,34

Laboratory and biochemical analysis (biomarker reference method) The plasma aliquots were provided from overnight fasting venous blood and frozen at  70 1C. Folate and cobalamin plasma concentrations were measured using an automated electrochemiluminescence immunoassay as previously described.26 The plasma level of tHcy was measured using high-pressure liquid chromatography with a fluorescence detector. More details about the estimated coefficient variations were reported previously.26 The PLP was measured using high-pressure liquid chromatography with the ClinRep Complete Kit for PLP (ClinRep complete kit, Munich, Germany) and florescence detector by following the instructions on the kit. The within and between coefficient variations for PLP measurements were determined as 3.7% and 3.2%, respectively. All assays were carried out in Laboratory of Day General Hospital, Tehran. In BC patients, fasting plasma concentration 410.0 mmol/l was defined as high-plasma tHcy status.18 The plasma concentrations of folate p5.9 ng/ml,35,36 PLP p10.0 mg/l37 and cobalamin p260 pmol/l38 were specified as the reference criteria for deficiency status of the nutrients (the reference models).

Statistical analysis Data were analyzed by using SPSS statistical package for Windows (version 11.5; SPSS Inc., Chicago, IL, USA). An independent sample t-test was performed to compare the means of two groups. A dual-dimensional scatter plot was used to show bivariate correlation between the plasma and daily intake levels of each nutrient. The agreement between the test method (FFQ) and the reference variable (plasma level) for each nutrient was determined using Pearson’s or Spearman’s correlation coefficient in Figure 1. The least squares fit of a line to data and its linear equation were calculated using Microsoft Office Excel 2003. A 2  2 contingency table was created for categorical analysis and tested with a two-tailed w2 test (Table 3). In the reference model, the binary classification of plasma variables was carried out according to the most addressed criteria for the plasma level of each nutrient.35–38 The median level of each nutrient of our population was also considered for categorization in the median model. The receiver operating characteristic (ROC) analysis is a graph used in testing the diagnostic accuracy.39,40 In an ROC curve, the sensitivity of the test in y-axis was plotted against the false-positive error rate (1-specificity) in x-axis for a binary classification model.39 In this regard, the overall diagnostic accuracy was measured by the area under the ROC curve (AUC) for a test in order to show the ability of the test (each nutritional factor in FFQ) to correctly classify and discriminate those with and without the deficiency, based on biochemical characteristics (dichotomous outcome). The suitable range for AUC was defined in the range of 0.5 to 1.0.39,40 The sensitivity (used to test the ability of the model to determine true-positive cases) and specificity (used to test the ability of model to identify true-negative cases) were also evaluated. The probability of true insufficiency according to the FFQ (individual test) versus being at risk of biochemical deficiency based on plasma level (golden standard) was evaluated by likelihood ratios as well. The tests with LR þ (sensitivity/false-positive error rate; similar to the ROC curve) larger than one are helpful for screening patients with better sensitivity and dependent on choosing a high cutoff point, whereas the LR- (falsenegative error rate/specificity) less than one could reflect a high degree of specificity and is suitable for confirming the diagnosis of deficiency.39 Consequently, LRs are independent of the prevalence of disease. LR values near one indicated a result that did not substantially change deficiency likelihood. The ratio of LR þ to LR  was calculated to estimate the diagnostic odds ratio as another global measure for diagnostic accuracy. Indeed, a test with high specificity and sensitivity with low rates of false positives and false negatives could result in a high odds ratio. For example, a diagnostic odds of 10 means that the true results of a test is 10 times higher than its false results. The positive predictive value and negative predictive value were also calculated. All statistical tests were two tailed and P-values less than 0.05 were considered statistically significant. European Journal of Clinical Nutrition (2014) 316 – 323

The validity of FFQ in breast cancer S Pirouzpanah et al

318 (A-I) Plasma folate level, ng/ml

20

15 10 5

200

600 400 200

3 6 9 12 Daily intake of cobalamin, g

Pyridoxal-5-phosphate plasma level, g/I

(C-I)

20

10

Y=14.52-0.47*X, r=-0.004; P=0.964

1.0 1.5 2.0 2.5 3.0 Daily intake of pyridoxin, mg

3.5

(B-II)

800 600 400 200 Y=312.3+11.78X, r=0.240; P=0.007

0 5 10 15 20 Daily intake of cobalamin (residual), g

15

30

0.5

Y=6.53+0.008*X, r=0.324; P=0.001

0

Y=289.9+16.90*X, r=0.266; P=0.001

0

5

1000

800

0

10

200 400 600 800 Daily intake of folate (residual), g

(B-I)

0

15

400 600 800 Daily intake of folate, g

Plasma cobalamin level, pmol/l

Plasma cobalamine level, pmol/L

1000

(A-II)

0

Y=6.48+0.008*X, r=0.322 ; P=0.001

0

Plasma level of pyridoxal 5phosphate, g/l

Plasma folate level, ng/ml

20

(C-II) 30

20

10

0

Y=13.0+0.10*X, r=0.033; P=0.695

1.0 2.0 3.0 4.0 5.0 0.0 Daily intake of pyridoxin (residual), mg

Figure 1. The correlation scatter plots and Pearson correlation coefficient values of daily dietary and residual intake levels of nutrients from FFQ versus the plasma level of nutrients in 142 BC patients.

RESULTS The mean age at diagnosis of BC patients was 49±8 years (mean±s.d.). The general characteristics of 148 eligible participants with BC are described in Table 1. The invasive ductal carcinoma (IDC: 82.3%) was the predominant histopathological subtype. A higher frequency of pathological grades III (49.0%) and II (39.2%) was determined among BC patients (Po0.01). Smoking was seen in 8.3% (12/144) of participants (Table 1). In Table 2, despite the inadequate intake levels of folate in the present sample population (55% consumed lower than 400 mg/d of folate intake, n ¼ 149), adequate dietary intakes of cobalamin and pyridoxine were notable compared with the criteria specified in DRI (dietary reference intake) (Table 2). The scatter plots and correlation coefficient values (r) between dependent plasma concentrations and dietary intake levels of folate, pyridoxine and cobalamin are shown in Figure 1. Daily dietary intake of folate was positively associated with its plasma level at r ¼ 0.32 (Po0.01; Panel AI). Similarly, energy-adjusted folate intake (residual intake) was correlated significantly with the plasma folate level at r ¼ 0.32 (Po0.01; Panel AII). The daily dietary intake of cobalamin was associated significantly with the plasma cobalamin level at r ¼ 0.27 (Po0.01; Panel BI). After considering the residual cobalamin intake, this relationship remained significant with r ¼ 0.24 (Po0.01; Panel BII). Stratified analysis was performed to show the arrangement of dichotomous classification of dietary folate, pyridoxine and cobalamin intakes versus stratified plasma levels of the corresponding biomarkers (Table 3). In the reference model, European Journal of Clinical Nutrition (2014) 316 – 323

the significant number of individuals who were screened in deficient category of intakes for folate (89%, Po0.01) and cobalamin (60%, Po0.05) were noticeably classified in plasma deficiency status. In the median model, the similar categorical assignments of patients for deficiency status of folate and cobalamin were also obtained significantly (Po0.01). Although the predominant part of the BC patients was at reproductive ages, the carried-out stratification analysis within either menopausal status was not capable to attenuate the residual variance caused by confounding effect of menopause. In Figure 2a, typical ROC curve illustrates the diagnostic performance based upon the plasma concentration of folate p5.9 ng/ml. It was demonstrating qualitatively a suitable specificity and discriminative power of the test (large area of 0.5 indicated with being farther than the diagonal straight line). The AUC for assessing the diagnostic accuracy of intake status evaluated by FFQ versus biochemical determinants and also the sensitivity, specificity, LR  and LR þ of models are presented in Table 4. The evaluated AUC for folate from FFQ was estimated to be significant against plasma status of folate in either the reference model at 0.74 (95%CI: 0.63–0.85; Po0.01) or in the median model at 0.61 (95%CI: 0.50–0.70; Po0.01). The reference model for predicting the folate insufficiency (p400 mg/d) showed the highest sensitivity and specificity at 68% and 63%, respectively. The estimated positive predictive value and negative predictive value of folate intake in the reference model were determined to be 96.9% and 96.8%, respectively. The calculated LR þ was not high enough, but the LR  was preferably small to improve the test of determining the folate status with possibly a & 2014 Macmillan Publishers Limited

The validity of FFQ in breast cancer S Pirouzpanah et al

319 less false-negative error rate in the reference model (Table 4). The estimated diagnostic odds ratio showed that the true discrimination of patients in dietary folate deficiency category in this model was 9.08 times higher than placing patients in adequate status (false results). The AUC for cobalamin intake derived from FFQ in the reference model upon plasma cobalamin (p260 pmol/l) was 0.64 (95%CI: 0.54–0.74; Po0.05). The sensitivity and specificity of cobalamin intake in this model were 60% and 61%, respectively. The negative predictive value equal to 82% was high enough to predict the lower false-negative results derived from cobalamin intake obtained from FFQ in the reference model. The cobalamin intake in the median model on the basis of plasma cobalamin

Table 1. General characteristics of studied sample population with BC (n ¼ 148) Total patients (n)

The relative frequency (%)

P-value

Age at diagnosis (years) o48 X48

74 69

51.7 48.3

0.676

Histopathology Ductal Lobular Others

79 9 8

82.3 9.4 5.3

0.001

110 38

74.3 25.7

0.001

Grade I II III

12 40 50

11.8 39.2 49.0

0.001

BMI (kg/m2) o20 20–24.9 25o

64 50 30

44.4 34.7 20.8

0.002

12 132

8.3 91.7

0.001

Variables

Menopausal status Pre-menopause Post-menopause

Smoking status (%) Ever smokers Never smokers

Some missing data existed in histopathological variables.

Table 2.

(p331 pmol/l) received sufficient AUC equal to 0.67 (95% CI: 0.58– 0.76; Po0.01). Either sensitivity or specificity of median model for cobalamin intake was also similar to the former model. The accuracy of pyridoxine data from FFQ was not significantly predicted by using plasma PLP status (Table 4). The AUC (0.55) was relatively non-discriminative for the reference model with sensitivity and specificity of 61% and 58%, respectively. Table 4 also summarizes measures of diagnostic accuracy for dietary folate, pyridoxine and cobalamin intakes (FFQ) using the plasma levels of tHcy as a metabolic indicator of the nutrients. The folate intake model showed an AUC equal to 0.71 (95% CI: 0.55–0.86; Po0.05) with an estimated positive predictive value of about 78.5% (Table 4). The cobalamin intake was observed to have significant AUC ¼ 0.67 (95%CI: 0.51–0.83; Po0.05) and LR þ about 3.42 indicating relatively good diagnostic value of the cobalamin intake model with respect to tHcy status (Table 4).

DISCUSSION Our findings revealed that the intake assessments of folate and cobalamin by FFQ could receive reasonable credibility for representing the true nutrient intake status among BC patients. In consistent with previous findings, the folate intake level was significantly correlated with the plasma folate concentration. Similarly, the residual folate intake was also in an appropriate concordance with plasma folate. Similarly, Owens et al.15 supported a positive correlation coefficient between folate intake and folate RBC (r ¼ 0.35). A wealth of evidence exists and suggested identical trend of associations (r) within the range of 0.1–0.5.4,22,35,41,42 However, sometimes the correlations between dietary folate intakes derived from FFQ and serum folate could be elusive and fail to reach statistical significant outcomes.43 These conditions might be influenced by several reasons such as information bias, various sample size among studies and also using different laboratory techniques for the estimation of folate levels.4,15,35,44 Eventually, our findings supported the fact that the new designed FFQ could provide valid measurements to rank BC subjects regarding folate intake in the present sample population. A significant positive correlation between daily dietary intake of cobalamin and its plasma level exists confirming the acceptable agreement between these two data sets. Accordingly, the diagnostic accuracy for determining the dietary intake status of cobalamin intake was also significantly attained by appropriate validity parameters in either reference or even median model. In detail, the performance of the FFQ was rather sensitive with a

The average amount of plasma and dietary intake levels of folate, cobalamin and vitamin B6 among BC patients (n ¼ 149)

Characteristics (unit)

Total patients

percentile

(n)

Mean

2.5

90th

97.5

Plasma concentration tHcy (mmol/l) Folate (ng/ml) PLP (mg/l) Cobalamin (pmol/l)

141 140 143 141

7.38±2.78a 9.95±3.89 13.74±5.84 371±179

3.22 2.82 5.47 131

11.2 15.9 21.9 642

14.70 16.98 28.41 864

Nutrient-dietary intake/d Total calorie (kcal) Folate (mg) Pyridoxine (mg) Cobalamin (mg)

148 149 148 149

2419±869b 385±145 1.68±0.68 4.91±2.98

1306 137 0.56 1.37

3431 576 2.78 9.1

4815 691 3.20 13.94

Abbreviations: tHcy, Total homocysteine; PLP: Pirydoxyl-5-phospahte. aMean±s.d. bThe age and sex dependent dietary reference intakes (DRIs) were used to detect the intake insufficiency based on FFQ data. In case of folate p400 mg/d, cobalamin p2.4 mg/d and pyridoxine age dependently less than 1.3–1.5 mg/d were defined as the nutrient inadequacies.

& 2014 Macmillan Publishers Limited

European Journal of Clinical Nutrition (2014) 316 – 323

The validity of FFQ in breast cancer S Pirouzpanah et al

320 Table 3. Categorical analysis for dichotomously ranked dietary intakes of folate, pyridoxine and cobalamin (FFQ, test outcome) versus the plasma levels of folate, pyridoxine and cobalamin as biochemical indicators (n ¼ 141) Plasma levels of nutrients Reference modela Dietary intake status of nutrientc

Low

High

Total

All BC subjects Folate (mg/d) o400 X400

17d 2

58 62

75 64

Vitamin B6 (mg/d) o1.5 X1.5

36 26

31 48

Vitamin B12 (mg/d) o2.4 X2.4

21 14

Premenopausal status Folate (mg/d) o400 X400

Median modelb P-value

Low

High

Total

P-value

0.001e

46 25

29 39

75 64

0.009

67 74

0.026

40 35

32 34

72 69

0.373

41 64

62 78

0.031

43 28

26 43

69 71

0.007

15 1

41 45

56 46

o0.001

35 19

21 27

56 46

0.033

Vitamin B6 (mg/d) o1.5 X1.5

31 20

25 28

56 48

0.164

33 27

23 21

56 48

0.783

Vitamin B12 (mg/d) o2.4 X2.4

14 9

33 47

47 56

0.096

32 20

21 30

53 50

0.039

Postmenopausal status Folate (mg/d) o400 X400

11 6

8 12

19 18

0.134

2 1

17 17

19 18

0.529

Vitamin B6 (mg/d) o1.5 X1.5

5 6

11 15

16 21

0.860

7 8

9 13

16 21

0.783

Vitamin B12 (mg/d) o2.4 X2.4

7 5

8 17

15 22

0.127

10 8

51 6

16 21

0.141

Abbreviations: FFQ, food frequency questionnaire; BC, breast cancer; PLP, pirydoxyl-5-phospahte; DRI, dietary reference intake. aThe term of reference model was referred to the classification of plasma variables on the basis of normal reference laboratory values of folate p5.9 ng/ml,35,36 PLP p10.0 mg/l,37 and cobalamin p260 pmol/l.38 They were defined as inadequacy status in reference model. bIn median model, the low plasma status of variables were stratified on the basis of the median of our population, that is, folate p10.0 ng/ml, PLP p12.0 mg/l and cobalamin p331 pmol/l. cRanking was undergone on dietary intake level of the nutrients based on the consensus criteria from DRI. dNumber of individuals. eFisher’s exact test was performed.

convincing specificity in detecting the cobalamin status from FFQ. However, pyridoxine intake level from FFQ was incapable of capturing the plasma PLP status in either model. Despite the compelling supportive evidence for folate validity, less comparable information regarding the validity of cobalamin and pyridoxine was available in previous investigations on nonmalignant subjects. Yoshino et al.,22 in Japanese population, suggested lesser extent of correlation coefficients between pyridoxine and cobalamin intakes estimated from a brief FFQ and their corresponding serum concentrations. In the case of cobalamin, Johansson et al.,45 demonstrated significant but weak agreements between dietary cobalamin from FFQs versus plasma cobalamin (r ¼ 0.15 and 0.18 for longer and shorter versions of their FFQ, respectively). More specifically, they just demonstrated this correlation significantly in women (r ¼ 0.32). The lower correlations for cobalamin in different studies may represent inter-individual variations in absorption, due to differences in European Journal of Clinical Nutrition (2014) 316 – 323

gastric acidity or intrinsic factor secretion.45 However, achlorhydria and atrophic gastritis had been considered as the exclusion criteria in the present study. Moreover, in contrast to our findings, Johansson et al.45 found a positive correlation between pyridoxine and its plasma status. However, the validation of pyridoxine was met just in a very few studies with significant outcomes in adults.45 Some discrepancies in several findings could be likely due to the limited number of food items in FFQ, weakly defined or fixed portion size, questioning-related recall bias, seasonal variations, population sample size and so forth.17,42,46 In addition, low correlations are expected as the biomarker measurement is often subject to laboratory errors such as interindividual metabolic variations and laboratory bias.17,42,47 Given the laboratory-related bias, patient misclassification and selection of inappropriate criteria in the models, any interpretation in biomarker-based validity should be made cautiously. Regardless of some previous equivocal proves for using FFQ to estimate & 2014 Macmillan Publishers Limited

The validity of FFQ in breast cancer S Pirouzpanah et al

321 cobalamin intake status in non-malignant subjects, the present findings supported the notion that the FFQ could be a valid tool to measure cobalamin intake level in women with BC. However, the accuracy of data obtained from FFQ for pyridoxine intake was not supported according to our results. Indeed, the plasma PLP is a predominant metabolite of pyridoxine in the circulation and probably the best single measure of pyridoxine, because it reflects tissue stores.48 However, some inter-individual variations such as recent dietary intake, probable prolonged fasting, raised alkaline phosphatase and inflammation have been addressed to have confounding effects in association with intake levels.49,50 Despite taking chronic inflammatory defects and Ramadan fasting as an ineligibility criteria for participation, they might probably in part affect our results in pyridoxine validity. There is very confined evidence concerning the use of the measures of diagnostic accuracy in addition to correlation 1.0

Sensitivity

0.8

0.6

0.4

0.2

0.0 0.0

0.2

0.4 0.6 1 - Specificity

0.8

1.0

Figure 2. A typical ROC curve to show a pair of diagnostic sensitivity and specificity values of individuals regarding the dietary folate state as a test variable versus plasma folate status.

coefficient in questionnaire-validity assays. In this respect, the present findings suggested that the diagnostic accuracy of folate intake from FFQ in the reference model conveyed the rather higher capability of the test to discriminate folate status by which AUC was significantly estimated to be 0.74. The reference model also provided acceptable sensitivity to rank individuals on the basis of the folate status from the FFQ. The specificity of the model was also well comparable to prior studies.15,19 Baric et al.19 documented consistent findings in sensitivity (61%) and specificity (56%) of FFQ to determine folate intake status of individuals (o200 mg/d) by serum folate level. Similarly, Owens et al.15 supported a positive correlation coefficient between folate intake and folate RBC (r ¼ 0.35), with appropriate specificity (91–96%) to validate their FFQs (based on a cutoff value of 200 mg/day; the estimated average requirements of folate). However, they indicated very weak sensitivity to figure out true folate inadequacy status.15 In this respect, they also suggested a relatively similar diagnostic ability for their FFQ (AUC ¼ 0.68) on the basis of the HTLC-MS/MS method for quantification of RBC folate as the reference. One of the major strengths of the present study is that we calculated sensitivity, specificity, likelihood ratios and predictive values as validity statistical tests to evaluate and ascertain the usefulness of FFQ to measure intake status of the nutrients on the basis of the plasma status, as the reference. Accordingly, AUC analysis was also undertaken to determine the diagnostic accuracy of a model to test the true insufficiency of a nutrient. Indeed, studies on diet–disease relationships frequently stratified nutrient intakes into categories. In the meantime, Bruner et al. speculated that in nutritional epidemiologic studies, the primary need for performing the analysis is considering individual in its correct rank order, rather than to make accurate estimates of absolute intake.17,46 In the majority of earlier studies, the correlation coefficient was mostly the subject of analyses to seek the agreement of two sets of quantitative data in studies on FFQ validation.12,15,19,45,51 Using certain measures of diagnostic accuracy preferentially provided a more useful statistical approach to assess the discriminative potential of nutritional assessing tools rather than even performing categorical analysis. Indeed, simple categorical analysis just limited to describe positive and negative

Table 4. The area under curve (AUC) for dietary folate, pyridoxin and cobalamin intake (FFQ) obtained by using the plasma levels of folate, PLP and cobalamin as biochemical indicators (n ¼ 141) AUC

SE

P-value

95%CI

Sensitivity (%)

Specificity (%)

PPV (%)

NPV (%)

LR þ

LR 

OR

Folate (ng/ml) 5.9 (reference) 10.0 (median)

0.74 0.61

0.05

0.001 0.007

0.63–0.85 0.50–0.70

68 54

63 64

96.9 61.3

96.8 60.9

1.85 1.52

0.20 0.61

9.08 2.50

PLP (mg/l) 10.0 (reference) 12.0 (median)

0.55 0.51

0.05 0.05

0.306 0.903

0.45–0.65 0.41–0.60

61 51

58 55

53.7 55.5

64.9 49.3

0.69 1.09

1.48 0.85

2.15 1.28

Cobalamin (pmol/l) 260 (reference) 331 (median)

0.64 0.67

0.05 0.05

0.013 0.001

0.54–0.74 0.58–0.76

60 60

61 62

34.0 62.3

82.0 60.5

1.54 1.60

0.65 0.63

2.37 2.54

tHcys (mmol/l)b Folate intake Pyridoxin intake Cobalamin intake

0.71 0.55 0.67

0.05 0.09 0.08

0.011 0.515 0.034

0.55–0.86 0.38–0.73 0.51–0.83

67 59 64

64 54 57

78.5 55.5 46.1

51.8 57.1 86.0

1.63 1.29 3.42

0.41 0.77 0.65

3.94 1.67 5.24

Plasma concentrationa

Abbreviations: n, number of participants; OR, Odd ratio; tHcy, total homocysteine; PLP, pyridoxal-5-phospahte; ROC, receiver operator characteristic. Plasma concentration of folate p5.9 ng/ml,35,36 PLP p10.0 mg/l,37 and cobalamin p260 pmol/l,38 were defined as deficiencies reference values. The risk of high tHcy was defined as plasma level X10.0 mmol/l. aThe validity assessment of each nutrient from FFQ data set was analyzed on the basis of plasma status of the same nutrient as standard reference. bThe tHcy as a sensitive indicator of certain nutrients intake status was used as standard reference for evaluating the validity of FFQ to assess certain nutrient intakes.

& 2014 Macmillan Publishers Limited

European Journal of Clinical Nutrition (2014) 316 – 323

The validity of FFQ in breast cancer S Pirouzpanah et al

322 diagnoses. Therefore, it is widely expected that testing the accuracy of obtained data in categorized form contributes additionally to what has been previously published regarding the validity of FFQ in estimating the nutrients. Our findings suggested that the plasma tHcy status could predict the folate or cobalamin intake status with appropriate accuracy. Although the plasma tHcy status was substantiated as an inverse determinant of folate status in earlier observations,19,22,52 the diagnostic ability of tHcy measurement in detecting the intake status of either folate or even cobalamin was denoted and proven at present study among BC participants. The major limitation of our study was that a second FFQ was not undertaken for reproducibility analysis due to the impact of diagnosis bias. This source of errors referred to conditions related to malignancy diagnosis and neoadjuvant therapies. The studied sample size was also another potent limiting factor for providing enough statistical power of analysis in this study. Despite the restricted number of FFQ items in other investigations, the variety and number of our FFQ items seem to be supportive for the agreement between the two methods. Other advantage of this study was the estimation of folate and cobalamin intakes in the absence of routine mandatory fortification strategies in Iran; in the meantime, almost the results of various studies could be attenuated by confounding effects of certain dietary fortifications.25,41 As far as we knew, alcoholic beverages were not consumed by the Iranian women because of cultural and religious prohibitions. In addition, the frequency of smoking in this study population was low (8.3%). Therefore, the confounding effects of alcohol consumption and smoking in the metabolism of B vitamins, particularly folate, were apparently attenuated due to the nature of the population studied.9,53 In addition, in this questionnaire-based measurements of nutrients, under-reporters, who could be attributed to the response bias, were excluded from our analyses.27 Among different questioning-based tools, FFQ received further credential among studies to be used in nutritional assessments in large-scale studies on diet and disease. The privileges of this FFQ to estimate micronutrient related to one-carbon metabolism include relatively cost-effective and easily being administered for larger population-based studies over an extended period of time. Validation reports suggested that the performance of a FFQ at least partly depends on the ethnical differences and diversity in dietary habits consumed in different population.17,46 In this regard, the Iranian people consume various indigenous green leafy vegetables in their daily dietary usage (rich in folate). Therefore, these could give rise to the consequential necessity of conducting validity assessment for probing the status of concerning nutrients in upcoming studies. CONCLUSION These data could support the reasonable validity of FFQ as a useful nutritional assessing tool to be administered in estimating the intake status of folate and cobalamin among BC patients. Preliminarily, linear agreements were obtained between dietary intake variables and biochemical concentrations of folate or cobalamin, which lie within those range of correlation coefficients obtained in non-malignant populations. Ranking individuals in the finest nutritional category is a concerning clue in association with the risk assessments in numerous diet-cancer studies. However, override usages of correlation coefficients as a major characteristic of validity in various nutritional epidemiologic studies seems to be an insufficient method of choice and has assisted little to the resolving the debates in the field. Implicating further diagnostic concepts from AUC analysis, LRs and even diagnostic odds ratio could be informative to enhance the aptitudes of testing the performance of a questionnaire, for example, FFQ. In this respect, our findings ascertained good performance for the likelihood of European Journal of Clinical Nutrition (2014) 316 – 323

defining dietary folate inadequacy with clearer advantage for the reference model and suitable range of validity for discriminating cobalamin deficiency among individuals with BC based on the corresponding measurements in plasma specimens. In addition, the accuracy of dietary intake level of folate or cobalamin was also determined appropriately with plasma tHcy status as the reference biomarker. However, laboratory-based errors, which often account for epidemiologic misclassifications, might cause our interpretation to be made cautiously, particularly in discrimination of cobalamin. Interestingly, the particular lifestyle and dietary habits among Iranian population emphasize the importance of further validation studies in the future investigations. Moreover, information regarding the validity on the FFQ might convey more accurate data for translation purpose into dietary recommendations in diet-cancer associations. CONFLICT OF INTEREST The authors declare no conflict of interest.

ACKNOWLEDGEMENTS We thank all participating patients and our valued colleagues in the Day General Hospital. We are also grateful to research affair of National Nutrition & Food Technology Research Institute, Shaheed Beheshti University of Medical Sciences, Iran, for providing financial support to conduct this study. The financial support for conception, design, data analysis and manuscript drafting comes from Faculty of Nutrition, Tabriz University of Medical Sciences, Tabriz, Iran. The author’s responsibilities were as follows: SP and FAT, MA and AH: study design; SP, MA and PM: data ollection; SP, AH and SS: laboratory and data analysis; and SP, and SS: writing of manuscript. The authors are grateful to Dr. Mohammad-Amin Tabatabaifar for his valuable reading of this manuscript.

REFERENCES 1 Chou YC, Wu MH, Yu JC, Lee MS, Yang T, Shih HL et al. Genetic polymorphisms of the methylenetetrahydrofolate reductase gene, plasma folate levels and breast cancer susceptibility: a case–control study in Taiwan. Carcinogenesis 2006; 27: 2295–2300. 2 Larsson SC, Giovannucci E, Wolk A. Folate and risk of breast cancer: a meta-analysis. J Natl Cancer Inst 2007; 99: 64–76. 3 Johnson IT, Belshaw NJ. Environment, diet and CpG island methylation: epigenetic signals in gastrointestinal neoplasia. Food Chem Toxicol 2008; 46: 1346–1359. 4 Verkleij-Hagoort AC, de Vries JH, Stegers MP, Lindemans J, Ursem NT, SteegersTheunissen RP. Validation of the assessment of folate and vitamin B12 intake in women of reproductive age: the method of triads. Eur J Clin Nutr 2007; 61: 610–615. 5 Choi SW. Vitamin B12 deficiency: a new risk factor for breast cancer? Nutr Rev 1999; 57: 250–253. 6 Stempak JM, Sohn KJ, Chiang EP, Shane B, Kim YI. Cell and stage of transformation-specific effects of folate deficiency on methionine cycle intermediates and DNA methylation in an in vitro model. Carcinogenesis 2005; 26: 981–990. 7 Pirouzpanah S, Taleban FA, Abadi AR, Atri M, Mehdipour P. The association of plasma folate, vitamin B12 and homocysteine levels on hypermethylation status of RARb2 gene in primary breast carcinoma. Iran J Epidemiol 2009; 5: 17–19(Persian). 8 Lin J, Lee IM, Cook NR, Selhub J, Manson JAE, Buring JE et al. Plasma folate, vitamin B-6, vitamin B-12, and risk of breast cancer in women. Am J Clin Nutr 2008; 87: 734–743. 9 Kim YI. Does a high folate intake increase the risk of breast cancer? Nutr Rev 2006; 64: 468–475. 10 Ahn Y, Kwon E, Shim JE, Park MK, Joo Y, Kimm K et al. Validation and reproducibility of food frequency questionnaire for Korean genome epidemiologic study. Eur J Clin Nutr 2007; 61: 1435–1441. 11 Jackson M, Walker S, Younger N, Bennett F. Use of a food frequency questionnaire to assess diets of Jamaican adults: validation and correlation with biomarkers. Nutr J 2011; 10: 28. 12 Flood VM, Smith WT, Webb KL, Mitchell P. Issues in assessing the validity of nutrient data obtained from a food-frequency questionnaire: folate and vitamin B12 examples. Public Health Nutr 2004; 7: 751–756.

& 2014 Macmillan Publishers Limited

The validity of FFQ in breast cancer S Pirouzpanah et al

323 13 Carlsen MH, Karlsen A, Lillegaard ITL, Gran JM, Drevon CA, Blomhoff R et al. Relative validity of fruit and vegetable intake estimated from an FFQ, using carotenoid and flavonoid biomarkers and the method of triads. Brit J Nutr 2011; 105: 1530–1538. 14 Willett W. Issues in analysis and presentation of dietary data. In Nutritional Epidemiology. 2nd edn, Oxford University Press: New York, pp 321–346, 1998. 15 Owens JE, Holstege DM, Clifford AJ. Comparison of two dietary folate intake instruments and their validation by RBC folate. J Agric Food Chem 2007; 55: 3737–3740. 16 Signorello LB, Buchowski MS, Cai Q, Munro HM, Hargreaves MK, Blot WJ. Biochemical validation of food frequency questionnaire-estimated carotenoid, alpha-tocopherol and folate intakes among African Americans and non-Hispanic Whites in the Southern Community Cohort Study. Am J Epidemiol 2010; 171: 488–497. 17 Brunner E, Stallone D, Juneja M, Bingham S, Marmot M. Dietary assessment in Whitehall II: comparison of 7 d diet diary and food-frequency questionnaire and validity against biomarkers. Br J Nutr 2001; 86: 405–414. 18 Cheng CH, Tsai TP, Chen WS, Huang YC. Serum folate is a reliable indicator of hyperhomocysteinemia and borderline hyperhomocysteinemia in young adults. Nutr Res 2009; 29: 743–749. 19 Baric IC, Satalic Z, Keser I, Cecic I, Sucic M. Validation of the folate food frequency questionnaire with serum and erythrocyte folate and plasma homocysteine. Int J Food Sci Nutr 2009; 60: 10–18. 20 Jaffe JP, Schilling RF. Erythrocyte folate levels: a clinical study. Am J Hematol 1991; 36: 116–121. 21 Melnyk S, Pogribna M, Pogribny IP, Yi P, James SJ. Measurement of plasma and intracellular S-adenosylmethionine and S-adenosylhomocysteine utilizing coulometric electrochemical detection: alterations with plasma homocysteine and pyridoxal 5’-phosphate concentrations. Clin Chem 2000; 46: 265–272. 22 Yoshino K, Nishide M, Inagawa M, Yokota K, Moriyama Y, Ikeda A et al. Validity of brief food frequency questionnaire for estimation of dietary intakes of folate, vitamins B6 and B12, and their associations with plasma homocysteine concentrations. Int J Food Sci Nutr 2011; 61: 61–67. 23 Lee BJ, Huang MC, Chung LJ, Cheng CH, Lin KL, Su KH et al. Folic acid and vitamin B12 are more effective than vitamin B6 in lowering fasting plasma homocysteine concentration in patients with coronary artery disease. Eur J Clin Nutr 2004; 58: 481–487. 24 Da Silva R, Bach-Faig A, Raidio Quintana B, Buckland G, Vaz de Almeida MD, Serra-Majem L. Worldwide variation of adherence to the Mediterranean diet, in 1961–1965 and 2000–2003. Public Health Nutr 2009; 12: 1676–1684. 25 Fayet F, Flood V, Petocz P, Samman S. Relative and biomarker-based validity of a food frequency questionnaire that measures the intakes of vitamin B(12), folate, iron, and zinc in young women. Nutr Res 2011; 31: 14–20. 26 Pirouzpanah S, Taleban FA, Atri M, Abadi AR, Mehdipour P. The effect of modifiable potentials on hypermethylation status of retinoic acid receptor-beta2 and estrogen receptor-alpha genes in primary breast cancer. Cancer Cause Control 2010; 21: 2101–2111. 27 Goldberg GR, Black AE, Jebb SA, Cole TJ, Murgatroyd PR, Coward WA et al. Critical evaluation of energy intake data using fundamental principles of energy physiology: 1. Derivation of cut-off limits to identify under-recording. Eur J Clin Nutr 1991; 45: 569–581. 28 Black AE. Physical activity levels from a meta analysis of doubly labeled water studies for validating energy intake as measured by dietary assessment. Nutr Rev 1996; 54: 170–174. 29 Schofield WN. Predicting basal metabolic rate, new standards and review of previous work. Hum Nutr Clin Nutr 1985; 39: 5–41. 30 Pirouzpanah S, Taleban FA, Sabour S, Mehdipour P, Atri M, Farrin N et al. Validation of food frequency questionnaire to assess folate intake status in breast cancer patients (In Faris). Razi J Med Sci 2012; 18: 31–41. 31 Ghaffarpour M, Housyar-Rad A, Kianfar T. The Manual for Household Measures, Cooking Yields Factors and Edible Portion of Food. Keshavarzi Press: Tehran, 1999. 32 Gallagher ML. Intake: the nutrients and their metabolism. In Mahan LK, Escott-Stump S, Raymond JL (eds) Krause’s food and the nutrition care process. 13 edn, Elsevier: Missouri, 2012.

& 2014 Macmillan Publishers Limited

33 Gebhardt SE, Lemar LE, Pehrsson PR, Exler J, Haytowitz DB, Patterson KK et al. USDA National Nutrient Database for Standard Reference, vol. Release 22, 2009. 34 Willett WC. Invited commentary: a further look at dietary questionnaire validation. Am J Epidemiol 2001; 154: 1100–1102. 35 Ihara H, Watanabe T, Aoki Y, Nagamura Y, Totani M, Hashizume N. Dietary folate intake and serum folate status in Japanese women of childbearing age. J Anal Bio-Sci 2009; 32: 181–185. 36 Pellis L, Dommels Y, Venema D, Polanen A, Lips E, Baykus H et al. High folic acid increases cell turnover and lowers differentiation and iron content in human HT29 colon cancer cells. Brit J Nutr 2008; 99: 703–708. 37 Lui A, Lumeng L, Aronoff GR, Ting-Kai LI. Relationship between body store of vitamin B6 and plasma pyridoxal-P clearance: metabolic balance studies in humans. J Lab Clin Med 1985; 106: 491–497. 38 Van Asselt DZ, de Groot LC, van Staveren WA, Blom HJ, Wevers RA, Biemond I et al. Role of cobalamin intake and atrophic gastritis in mild cobalamin deficiency in older Dutch subjects. Am J Clin Nutr 1998; 68: 328–334. 39 Jekel JF, Katz DL, Elmore JG. Epidemiology, Biostatistics and Preventive Medicine. 2nd edn., W.B. Saunders company: Philadelphia, 2001. 40 Swets JA. Measuring the accuracy of diagnostic systems. Science 1998; 240: 1285–1293. 41 Iso H, Moriyama Y, Yoshino K, Sasaki S, Ishihara J, Tsugane S. Validity of the self-administered food frequency questionnaire used in the 5-year follow-up survey for the JPHC Study to assess folate, vitamin be and B12 intake: Comparison with dietary records and blood level. J Expo Anal Environ Epidemiol 2003; 13: S98–S101. 42 Willett WC, Lenart E. Reproducibility and validity of food frequency questionnaire. In: Willett WC (ed) Nutritional Epidemiology. 2nd edn, Oxford University Press: New York, pp 101–147, 1998. 43 Yen J, Zoumas-Morse C, Pakiz B, Rock CL. Folate intake assessment: validation of a new approach. J Am Diet Assoc 2003; 103: 991–1000. 44 Kipnis V, Subar AF, Midthune D, Freedman LS, Ballard-Barbash R, Troiano RP et al. Structure of dietary measurement error: results of the OPEN biomarker study. Am J Epidemiol 2003; 158: 14–21. 45 Johansson I, Van Guelpen B, Hultdin J, Johansson M, Hallmans G, Stattin P. Validity of food frequency questionnaire estimated intakes of folate and other B vitamins in a region without folic acid fortification. Eur J Clin Nutr 2010; 64: 905–913. 46 Mirmiran P, Hosseini-Esfahani F, Mehrabi Y, Hedayati M, Azizi F. Reliability and relative validity of an FFQ for nutrients in the Tehran Lipid and Glucose Study. Public Health Nutr 2009; 13: 654–662. 47 Nelson M. The Validity of Dietary Questionnaires. In: Margetts B, Nelson M (eds) Design Concepts in Nutritional Epidemiology. Oxford University Press: New York, USA, 1991. 48 Liu A, Lumeng L, Aronoff G, Li TK. Relationship between body store of vitamin B6 and plasma pyridoxal-P clearance: metabolic balance studies in humans. J Lab Clin Med 1985; 106: 491–497. 49 Bates CJ, Thurnham DI, Bingham SA, Margetts BM, Nelson M. Biochemical markers of nutrient intake. In: Margetts BM, Nelson M (eds) Design concepts in nutritional epidemiology. Oxford University Press: New York, US, 1995. 50 Morris MS, Sakakeeny L, Jacques PF, Picciano MF, Selhub J. Vitamin B-6 intake is inversely related to, and the requirement is affected by, inflammation status. J Nutr 2010; 140: 103–110. 51 Daures JP, Gerber M, Scali J, Astre C, Bonifacj C, Kaaks R. Validation of a food-frequency questionnaire using multiple-day records and biochemical markers: application of the triads method. J Epidemiol Biostat 2000; 5: 109–115. 52 De Bree A, Verschuren WMM, Bjrke-Monsen AL, van der Put NMJ, Heil SG, Trijbels FJM et al. Effect of the methylenetetrahydrofolate reductase 677C-T mutation on the relations among folate intake and plasma folate and homocysteine concentrations in a general population sample. Am J Clin Nutr 2003; 77: 687–693. 53 Van Engeland M, Weijenberg MP, Roemen GMJM, Brink M, de Bruı¨ne AP, Goldbohm RA et al. Effects of dietary folate and alcohol intake on promoter methylation in sporadic colorectal cancer: the Netherlands Cohort Study on diet and cancer. Cancer Res 2003; 63: 3133–3137.

European Journal of Clinical Nutrition (2014) 316 – 323