Improving the estimation of flavonoid intake for study of health outcomes

Abstract

Imprecision in estimating intakes of non-nutrient bioactive compounds such as flavonoids is a challenge in epidemiologic studies of health outcomes. The sources of this imprecision, using flavonoids as an example, include the variability of bioactive compounds in foods due to differences in growing conditions and processing, the challenges in laboratory quantification of flavonoids in foods, the incompleteness of flavonoid food composition tables, and the lack of adequate dietary assessment instruments. Steps to improve databases of bioactive compounds and to increase the accuracy and precision of the estimation of bioactive compound intakes in studies of health benefits and outcomes are suggested.

INTRODUCTION

Flavonoids are bioactive, polyphenolic, non-nutrients in plants^1,2 that are ubiquitous in diets.^3–5 Figure 1 shows the major classes of flavonoids commonly found in plant foods. Table 1 lists some common food sources of the various flavonoid classes and references figures illustrating some of these (Figure 2 and Figures S1–S3, which are available in the Supporting Information for this article online).^6–11 Although there are thousands of flavonoids found in nature, only several dozen are common in foods. Emerging science from some observational and metabolic studies suggests that flavonoid-rich diets may lower the risk of some diet-related chronic degenerative diseases, such as cardiovascular disease, type 2 diabetes, and certain cancers,^12–16 but a few clinical and laboratory reports indicate that very high doses of certain flavonoids may have adverse effects, such as interference with common medications and, for some green tea extracts, liver toxicity.^17–24 Therefore, it is important to accurately assess flavonoid intakes from the perspectives of both disease prevention and safety.^25–29 Adverse effects and safety issues, however, are beyond the scope of this article.

Figure 1.

Flavonoid classes commonly found in plant foods, with representative examples of each. Except for flavan-3-ols and proanthocyanidins, most compounds in the other classes have sugars attached (flavonoid glycosides). Here, they are all presented as aglycones (without sugars).

Table 1.

Flavonoid classes, common compounds, and their plant sources

Flavonoid class (compounds)	Common plant sources (botanical families)
Anthocyanidins (cyanidin, delphinidin, malvidin, pelargonidin, peonidin, petunidin)	Blueberries (Ericaceae), red wine (Vitaceae), strawberries (Rosaceae). Usually present in any pink to purple fruit or vegetablea
Flavonols (isorhamnetin, kaempferol, myricetin, quercetin)	Blueberries (Ericaceae), garlic and onions (Amaryllidaceae), kale and broccoli (Brassicaceae), spinach (Chenopodiaceae), red wine (Vitaceae), tea (Theaceae), tomatoes (Solanaceae). Ubiquitous in plant families
Flavones (apigenin, luteolin)	Celery (Apiaceae), garlic (Amaryllidaceae), green peppers (Solanaceae), peppermint (Lamiaceae). Common in any leafy plant, particularly the parsley family (Apiaceae)
Isoflavones (daidzein, genistein, glycitein, biochanin A, coumestrol,b formononetin)	Soy products, peanuts, beans (Fabaceae). Present in members of the legume family (Fabaceae), especially the genus Glycine (soy), but also in small amounts in other plants7–11
Flavanones (eriodictyol, hesperetin, naringenin)	Citrus fruits and juices (Rutaceae), peppermint (Lamiaceae)
Flavan-3-ols (epicatechin, epicatechin 3-gallate, epigallocatechin, epigallocatechin gallate, catechin, gallocatechin)	Apples, apricots, peaches, pears, strawberries (Rosaceae), black and green tea (Theaceae), blueberries and cranberries (Ericaceae), chocolate (Malvaceae), grapes and red wine (Vitaceae)
Proanthocyanidins (monomers,c dimers, trimers, 4–6 mers, 7–10 mers, polymers)	Apples, apricots, peaches, pears, strawberries (Rosaceae), blueberries and cranberries (Ericaceae), chocolate (Malvaceae), grapes and red wine (Vitaceae), peanuts (Fabaceae), pecans and walnuts (Juglandaceae)
Theaflavins and thearubigins. Theaflavins are flavan-3-ol dimers, and thearubigins are flavan-3-ol polymers. Both are derived tannins6	Formed during the processing of tea (Theaceae). Theaflavins (Figure 2) are red, and thearubigins (Figure S3) are reddish brown

^aOther pink to red to violet pigments are lycopene (a tetraterpene or carotenoid) in tomatoes and red peppers (Solanaceae, potato family, where eggplant has anthocyanins), watermelon (Cucurbitaceae, squash family), and grapefruit (Rutaceae, citrus family, where blood oranges have anthocyanins), and the betalains (indole alkaloids) in the Chenopodiaceae family (beets, quinoa, spinach, Swiss chard, etc.) (Figure S1 in the Supporting Information online)

^bCoumestrol is a coumestan; coumestans are structurally similar to isoflavones (Figure S2 in the Supporting Information online).

^cMonomers are flavan-3-ols that are free flavan-3-ols or cleaved from oligomers and polymers during analysis (extraction and/or isolation)

Figure 2.

Tannins: condensed (proanthocyanidins), derived (theaflavin), and hydrolyzable (tannic acid). Until or unless databases that include the condensed, derived, and hydrolyzable tannins are developed,⁶ it is probably wise to include theaflavins with proanthocyanidin dimers and thearubigins with proanthocyanidin polymers. There are a few flavone dimers, particularly in medicinal plants, as well as a few flavan-3-ol-anthocyanin and proanthocyanidin–anthocyanin pairs in wine.

This article summarizes flavonoid intakes from recent observational epidemiologic studies in Europe and the United States. It explores some possible sources of flavonoid intake misestimation in these studies, including the variability of flavonoids in foods due to differences in growing conditions and processing, the challenges in laboratory quantification of flavonoids in foods, the incompleteness of flavonoid food composition tables, and the lack of adequate dietary assessment tools. It concludes with steps scientists can take to improve databases, reduce methodological misestimations, and increase the accuracy and precision of studies assessing the associations between flavonoids and health outcomes. This article does not address reporting requirements for studies of bioactive compounds such as flavonoids or the many pitfalls that must be avoided in experimental and clinical studies. These issues are covered in a recent article³⁰ documenting reporting requirements by the International Life Sciences Institute Bioactive Committee³¹ using the CONSORT framework developed^32–33 and endorsed by leading researchers here and abroad.

FLAVONOID INTAKES IN EUROPEAN AND US COHORTS

Table 2 ^3,4,34–58 reports flavonoid intakes based on means and medians from recent US, European, and Australian cohort and population-based studies of adults. Only studies that measured intakes of proanthocyanidins^6,59 (condensed tannins, Figure 2) and at least 4 monomeric flavonoid classes were included (indicating that more complete flavonoid data were used). As seen in Table 2, intake estimates varied greatly across cohorts by country. Since flavonoid intake distributions are usually skewed to the right, mean intakes tended to be higher than median intakes. When proanthocyanidins, particularly the thearubigins⁶⁰ (derived tannins; see Figure S3 in the Supporting Information for this article available online), were not measured, total flavonoid intakes were usually lower by at least a third. Note that, in Table 2, values in boldface indicate thearubigins were included; in the remaining studies, these compounds were excluded.

Table 2.

Mean and median flavonoid intakes reported in recent European, US, and Australian cross-sectional and cohort studies

Country	Year	Cohort	Sex	No.	Diet method	Databases useda	Total flavonoidsb (mg per day)	Monomeric flavonoidsb (mg per day)	Reference
Means
Australia	2013	CIFOARES	F	1063	FFQ	f2007 i2008, PE	775	560	Ivey et al. (2013)37
Denmark	2012	EPICc	M + F	3917	24-h recall	f2007 i2008, PE	599	200	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
Finland	2008	Findiet 2002	M + F	2007	24-h recall	f2007, local	209	92	Ovaskainen et al. (2008)41
France	2011	SU.VI.MAX	M + F	4942	24-h recall	PE	423	187	Perez-Jimenez et al. (2011)42
France	2012	EPICc	F	4735	24-h recall	f2007 i2008, PE	529	189	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
France	2012	SU.VI.MAX	M + F	2574	24-h recall	PE	528	274	Kesse-Guyot et al. (2012)38
Germany	2012	EPICc	M + F	4415	24-h recall	f2007 i2008, PE	482	203	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
Greece	2012	EPICc	M + F	2687	24-h recall	f2007 i2008, PE	236	108	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
Italy	2012	EPICc	M + F	3953	24-h recall	f2007 i2008, PE	429	163	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
Norway	2012	EPICc	F	1797	24-h recall	f2007 i2008, PE	483	151	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
Scotland	2011	LBC	M + F	882	FFQ	Kyle	302	250	Butchart et al. (2011)35
Scotland	2011	BMD	F	3226	FFQ	f2004, Kyle	307	259	Hardcastle et al. (2011)36
Spain	2010	EPICc	M + F	40 683	History	f2007 i2007	313	122	Zamora-Ros et al. (2010)44
Spain	2012	EPICc	M + F	3220	24-h recall	f2007 i2008, PE	410	150	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
Spain	2013	EPICc	M + F	40 622	History	f2007 i2008, PE	387	133	Zamora-Ros et al. (2013)45
Sweden	2012	EPICc	M + F	6050	24-h recall	f2007 i2008, PE	378	166	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
The Netherlands	2012	EPICc	M + F	3980	24-h recall	f2007 i2008, PE	576	202	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
United Kingdom	2012	EPICc	M + F	974	24-h recall	f2007 i2008, PE	1017	311	Knaze et al. (2012)39; Zamora-Ros et al. (2011, 2011, 2012, 2013)46–49
United States	2007	NHANES	M + F	8809	24-h recall, FFQ	f2003 i2002	285	190	Chun et al. (2007),4 Wang et al. (2011)43
United States	2011	NHS I	F	46 672	FFQ	f2007	357	124	Cassidy et al. (2011)3
United States	2011	HPS	M	23 043	FFQ	f2007	377	139	Cassidy et al. (2011)3
United States	2011	NHS II	F	87 242	FFQ	f2007	412	129	Cassidy et al. (2011)3
United States	2012	ACS CPSII	M + F	98 469	FFQ	f2007 i2002	268	83	McCullough et al. (2012)40
United States	2014	NHANES III	M + F	17 900	24-h recall	f2011 i2008	353	245	Bai et al. (2014)34
Mean							435	193
Range							209–1,017	83–560
Medians
Greece	2010	EPICc	M + F	28 572	FFQ	f2007, i2007	161d	86d	Dilis et al. (2010)51
United States	2007	Iowa	F	34 489	FFQ	f2003, i1999	239	70d	Mink et al. (2007)53
United States	2012	NHS	F	69 622	FFQ	f2007	232	100d	Cassidy et al. (2012)50
United States	2013	Framingham	M + F	2915	FFQ	f2007	225	87d	Jacques et al. (2013)52
Mean							214	86
Range							161–239	70–100

Abbreviations: ACS CPSII, American Cancer Society Cancer Prevention Study II Nutrition; BMD, bone mineral density population study; CIFOARES, Calcium Intake Fracture Outcome Age-Related Extension Study; EPIC, European Prospective Investigation into Cancer and Nutrition; Findiet 2002, National Findiet 2002; Framingham, Framingham Offspring Study; FFQ, food frequency questionnaire; HPS, Health Professionals Follow-up Study; Iowa, Iowa Women's Health Study; LBC, Lothian Birth Cohort 1936; NHANES, National Health and Nutrition Examination Survey 1999–2002; NHANES III, National Health and Nutrition Examination Survey 1988–1994; NHS, Nurse’s Health Study; NHS I, Nurses’ Health Study I; NHS II, Nurses’ Health Study II; SU.VI.MAX, Supplémentation en Vitamines et Minéraux Antioxydants Cohort.

^aDatabases: f2003, f2007, f2011 – versions of the USDA flavonoid database⁵⁵; i1999, i2002, i2003, i2007, i2008 – versions of the USDA isoflavone database⁵⁴; Kyle – Kyle & Duthie (2006)⁵⁷; local – local flavonoid data⁴¹; PE – Phenol Explorer;⁵⁶ proanthocyanidin data are from the USDA 2004 database.⁵⁸

^bBoldface indicates thearubigins are included (see Figure S3).

^cEPIC 10 countries: Denmark, France, Germany, Greece, Italy, The Netherlands, Norway, Spain, Sweden, and the United Kingdom.

^dMedian values summed for illustrative purposes.

Thus, intakes may not be directly comparable because the studies used various flavonoid databases and dietary assessment tools (food frequency questionnaires of varying length, histories, recalls, and records). Despite these variable approaches to measuring diet and assessing flavonoid intakes, some generalizations are possible. First, the range of mean total flavonoid intakes between and within these Australian, European, and US adult populations was wide, from 209 to 1017 mg/d (mean 435 mg/d). The highest mean intake (1017 mg/d) was reported in studies by Knaze et al.³⁹ and Zamora-Ros et al.^46–49 for a British cohort, and the lowest mean intake (209 mg/d) reported was for the National Findiet cohort.⁴¹ Both of these studies used a single 24-hour recall, which may be too short in duration to reflect usual diets and may have captured unrepresentative (outlier) days that do not reflect usual diets.⁶¹ Four studies^35,40,44,45 provided means and medians for many of the flavonoid classes, and 3 of these provided mean and median total flavonoid intakes. The median total intakes in these 3 studies were 44 mg/d, 57 mg/d, and 66 mg/d, lower than the means reported, illustrating that estimates tend to be skewed to the right (Table S1).

Second, the flavonoids consumed in the greatest amounts in these studies were the polymeric proanthocyanidins (ranging from 48 to 706 mg/d, with a mean of 242 mg/d). Proanthocyanidins were rarely included in studies before 2004,⁶² when a provisional database on the proanthocyanidin content of foods first became available.^{58,59,63–65} In these recent studies, the mean total intake of the monomeric flavonoid classes was 193 mg/d (range, 83–560 mg/d). Intakes of the monomeric flavonoids in these studies were as follows: flavan-3-ols, 102 mg/d (range, 12–431 mg/d); flavanones, 34 mg/d (range, 13–53 mg/d); anthocyanidins, 27 mg/d (range, 3–59 mg/d); flavonols, 27 mg/d (range, 5–52 mg/d); flavones, 5 mg/d (range, 0.2–34 mg/d); and isoflavones 1 mg/day (range, 0.1–5 mg/d) (Table S1).

Third, it should be noted that mean isoflavone intakes of individuals in these European and US cohorts (<2 mg/d) differed strikingly from intakes reported in studies of Asian populations (≈26 mg/d).^66–69 Mean intakes of other classes of flavonoids appear to be more similar among European,^{35,36,38,39,41,42,44,46–49} US,^3,4,34,40,43 and possibly Asian populations.^70–74 However, very few Asian studies^70–74 have measured intakes of the other flavonoid classes, and none appear to have included proanthocyanidins. It is not known whether intakes from other classes of flavonoids differ between Asian and Western populations.^13,75

Fourth, even the highest flavonoid intake levels for individual compounds or classes in these cohorts were typically lower than those in intervention studies that have achieved clinical effects.⁷⁶ For example the dose used in most intervention studies exceeded the usual intakes from diets by ≈3- to 15-fold.^77–83 In the recent FLAVURS intervention study, doses ranging from 20 to 70 mg flavonoids over baseline flavonoid intake levels (that were stated to be low but not reported) showed some effects on microvascular reactivity.⁸⁴

POSSIBLE REASONS FOR ACROSS-STUDY VARIATIONS IN ESTIMATED FLAVONOID INTAKE

Flavonoid intakes vary greatly within and between populations, owing not only to real differences in intakes⁸⁵ but also to other causes of variability. These variations may lead to inconsistent associations with health outcomes.^12,86–95 Sources of variability include the variability of plants themselves, the methods of processing plant foods for consumption, the chemical analyses used to measure the flavonoids present, and the completeness of flavonoid databases. In addition, inadequate dietary assessment tools can cause variability between studies. Each source of misestimation is discussed in greater detail below.

Inherent variability of flavonoid content in plants

A plant’s capacity to produce flavonoids varies intrinsically by its genus, species, family, and even within subspecies and cultivars or varieties.^96–101 Most plants contain more than one class of flavonoids. Table 3 shows some particularly rich sources (e.g., total flavonoids ≥50 mg/serving) common in US diets,¹⁰² including apples, blueberries, dark chocolate, grapefruit juice, grapes, oranges, pears, red wine, strawberries, and tea (black, black decaffeinated, and green). Variability in flavonoid content, even within the same family, genus, and species, occurs because flavonoids are secondary metabolites produced in varying amounts in response to stressors such as invading microorganisms, insects, plant diseases, season, climate, geography, or other environmental influences.^103–110

Table 3.

Flavonoid content of individual foods commonly listed in food frequency questionnaires^a

Food	Serving	Total flavonoids (mg per serving)	Total flavonoids (mg per 100 g)	Anthocyanins (mg per 100 g)	Flavonols (mg per 100 g)	Flavones (mg per 100 g)	Isoflavones (mg per 100 g)	Flavanones (mg per 100 g)	Flavan-3-ols (mg per 100 g)	Proanthocyanidins (mg per 100 g)	Genus	Species	Botanical family
Beverages
Tea, blackb	1 c (8 fl oz)	315	133		4	0	0		28	101	Camellia	sinensis	Theaceae
Tea, decaffeinated blackb	1 c (8 fl oz)	137	58		5	0			3	50	Camellia	sinensis	Theaceae
Tea, greenb	1 c (8 fl oz)	339	143		5	0	0		132	7	Camellia	sinensis	Theaceae
Wine, red	5 fl oz serving	141	96	19	2	0.2	0.01	2	11	62	Vitis	vinifera	Vitaceae
Fruit
Apples	1 medium	218	120	2	4	0.1	0	0	9	105	Malus	domestica	Rosaceae
Blueberries	1 c fresh	529	357	163	11	0.2	0	0	7	176	Vaccinium	augustifolium	Ericaceae
Grapefruit juice	1 c	54	22		0.4	0	0	21			Citrus	X paradisi	Rutaceae
Grapefruit	White, 0.5 medium	28	22		0	0	0.17	22		0	Citrus	X paradisi	Rutaceae
Grapes	Red, 1 c fresh	144	114	48	1	1	0		2	62	Vitis	vinifera	Vitaceae
Orange juice	1 c	37	15	0	0.3	0	0.11	14			Citrus	sinensis	Rutaceae
Oranges	1 medium	57	43		1	0.2	0	43	0	0	Citrus	sinensis	Rutaceae
Pears	1 medium	66	40	2	1	0	0	0	5	32	Pyrus	communis	Rosaceae
Raisins	1 sm box, 1.5 oz	0.4	1	0.05	0.3	0.01	0.1	0	1	0	Vitis	vinifera	Vitaceae
Strawberries	1 c fresh halves	258	175	27	2	0	0	0.26	5	142	Fragaria	× ananassa	Rosaceae
Candy, nuts, soy
Dark chocolate	0.5 bar (1.75 oz)	144	287						53	234	Theobroma	cacao	Malvaceae
Milk chocolate	0.5 bar (1.75 oz)	84	167						15	152	Theobroma	cacao	Malvaceae
Peanuts	1 c halves & whole	12	16	0	0	0	0	0	1	16	Arachis	hypogaea	Fabaceae
Walnuts	0.5 c pieces or chips	42	70	3	0	0	0.03	0	0	67	Juglans	regia	Juglandaceae
Soy milk	1 c	24	10				10				Glycine	max	Fabaceae
Soy tofu	0.5 c tofu firm	40	32		1	0	31				Glycine	max	Fabaceae
Vegetables
Broccoli	0.5 c chopped, cooked	9	12	0	11	1	0	0	0	0	Brassica	oleracea var. italica	Brassicaceae
Celery	1 c chopped, raw	5	5	0	1	4	0	0	0	0	Apium	graveolens var. dulce	Apiaceae
Garlic clove	1 clove	0.1	4		4		0.1				Allium	sativum	Amaryllidaceae
Garlic powder	1 tsp powder	0.3	8		8		0.2				Allium	sativum	Amaryllidaceae
Onion as garnish	1 medium slice	4	26		26	0.03	0		0	0	Allium	cepa var cepa	Amaryllidaceae
Onion as vegetable	0.5 c raw	21	26		26	0.03	0		0	0	Allium	cepa var cepa	Amaryllidaceae
Pepper, green	1 medium	8	7	0	2	5	0		0	0	Capsicum	annuum	Solanaceae
Spinach, raw	1 c raw	3	11		11	1	0			0	Spinacia	oleracea	Chenopodiaceae
Squash	0.5 c mashed, cooked	1	1		1		0		0		Cucurbita	maxima	Cucurbitaceae
Tomato sauce	0.5 c	1	1		1						Lycopersicon	esculentum	Solanaceae
Tomato	1 medium fresh	2	1	0	1	0	0	1	0	0	Lycopersicon	esculentum	Solanaceae

Abbreviations: c, cup; fl, fluid; oz, ounces; sm, small; tsp, teaspoon.

^aSources: USDA databases^54,55,58; foods listed are from a modified version of the Willett Food Frequency Questionnaire,¹⁰² except green tea. Serving sizes are predefined on the food frequency questionnaire (available at www.cancer.org) and are generally similar to those in the USDA Nutrient Database for Standard Reference. Blanks mean no chemical analysis was available for that particular class of flavonoids in that food item.

^bFor black and green teas, the proanthocyanidins include theaflavins and thearubigins.

Animals do not synthesize flavonoids. The few flavonoids present in animal foods are usually due to the animal’s diet (e.g., isoflavones in milk due to alfalfa, clover, other Fabaceae plants, and soy meal eaten by cows)¹¹¹ or to processing.¹¹

Flavonoids in foods are often conjugated with gallates, glycosides, or methoxy groups (Figures S4–S6 available in the Supporting Information online), which may affect bioavailability.^112–116 Although plant genera and species differ in the specific aglycones, sugars, and the amount of compounds that are present (which are affected in turn by the degree of ripeness and other environmental factors),^9,117–122 each genus and species is fairly consistent in the type and amount of one or more “signature” flavonoid classes it contains. For example, blueberries¹⁰¹ are rich in anthocyanins and proanthocyanidins, dark chocolate^123,124 is rich in flavan-3-ols and proanthocyanidins, and citrus^125,126 is rich in flavanones. This considerable variability in food flavonoid content, even within specific species and varieties, cannot be entirely eliminated. For example, in the US Department of Agriculture (USDA) databases, the mean (weighted by number of data points), number of data points, standard error of the mean, minimum value from data points, maximum value from data points, and confidence estimate for the quality of the data, along with sources of data, are all provided for each item and compound. As databases grow to include more and better flavonoid analyses, the variability of estimated values should decrease, improving precision in measuring intakes.

Processing of foods

Flavonoid content is also affected by food processing. Ignoring processing losses leads to overestimates of food flavonoid content. Processing losses depend on the individual flavonoid and its chemical properties, the food and its matrix, and the processing method used.^127–144 Milling (particularly for grains) often removes flavonoids.^145–147 Boiling may leach flavonoids into water that is discarded, decreasing the flavonoid content.^127,148 Microwaving, freezing, and frying appear to have less effect on flavonoid content.^{127,131,149–152} Juicing may increase the citrus flavonoid content, if the processing techniques that are used extract flavonoids from the peel and deposit them into the juice.^144,153,154

Databases sometimes provide data only on the raw food item because that is what was analyzed. Now, the USDA’s Nutrient Data Laboratory and the Food Composition and Methods Development Laboratory¹⁵⁵ and scientists compiling Phenol-Explorer¹⁵⁶ are providing flavonoid “retention factors” that account for common processing techniques for foods and flavonoids so that food tables in the future will contain improved food flavonoid composition estimates for both raw and cooked items by type of cooking method.

Although a detailed discussion of the influence of food preparation and processing methods on flavonoid values is outside the scope of this article, it is important to mention that the lack of quantitative data on non-nutrient bioactive compounds such as flavonoids in processed foods potentially lessens the ability to detect associations between flavonoid intakes and health outcomes.⁸ Dekker and Verkirk¹⁵⁷ investigated the data on bioactive glucosinolates from the broccoli family (Brassicaceae). With the data and computer modeling, they estimated that, if information about home preparation of foods was quantified, the statistical power of a study could at least be doubled.¹⁵⁷ If all variation of the food production chain could be accounted for, a study’s statistical power could be increased by a factor of 5.

For the flavonoids, values are needed for processed foods such as cereals, breads, and crackers. This might improve the estimated relative risks observed in studies of flavonoids and health.

Varying precision of chemical analyses used to determine flavonoid content

Until recently, the gaps that existed in food composition data were major impediments to accurate estimation of flavonoid intake. Today, better chemical analytic methods are available and databases are more complete, with duplicate analyses of more compounds and a wider variety of foods, including cooked and processed items.

The laboratory methods used to determine flavonoid content affect the precision and accuracy of the values in and the quality of databases.¹⁵⁸ The analytical methods for measuring flavonoids involve extraction, separation, characterization (i.e., detection and identification), and quantification. Newer extraction methods reduce the amount of solvent used as well as flavonoid losses due to degradation. Although solvent extraction systems are effective, no single extraction or separation system is suitable for all the flavonoid compounds in any one class or group of classes because of differences in polarity among individual compounds.^1,159,160

Acid hydrolysis is used for reliable identification of aglycones because mass spectroscopy fragmentation may not identify some similar flavonoids. It also separates the O-glycosides (Figure S5 in the Supporting Information online), which are easily hydrolyzed from C-glycosides (Figure S6 in the Supporting Information online), which are not hydrolyzed.¹⁶¹ Alkaline (basic) hydrolysis is used to determine where an acyl group (acetyl, malonyl, p-coumaryl, etc.) is attached to a sugar (Figure S6 in the Supporting Information online).^{1,160,162–166} High-performance liquid chromatography is the most robust method for separation and quantification, and better separation is now available using ultra high-performance liquid chromatography. Ultraviolet detection, fluorimetry (for flavonoids with natural fluorescence), and mass spectroscopy are used for characterization and quantification.

Mass spectroscopy can determine the following: 1) molecular mass (and, if high-resolution mass spectroscopy is used,¹⁶² even more accurate identification of a compound’s molecular formula and structure is possible); 2) structure of the aglycone (pattern of hydroxylation, attachment of ring B to ring C [see Figure 1]); 3) information about acylation of sugar hydroxyl groups and possible methylation or sulfation of aglycone hydroxyl(s); and 4) the number and the configuration of sugars and, sometimes, the glycosidic bond positions. However, mass spectroscopy does not provide information about the stereochemistry of the glycosidic linkage or distinguish between diastereomeric sugar units, even though such characteristics may be important for biological activity.¹⁶⁷ Mass spectroscopy can be used to quantify monomeric flavonoids in foods but not polymeric proanthocyanidins because the number and similarity of all their isomers, both structurally and in mass spectroscopy fragmentation, is so great.^{160,168–171}

Table 4 ^{1,125,146,159–162,165–220} summarizes some of the preferred analytical methods used today. Particular analytical challenges remain for each flavonoid class,^{216,217,221–223} including the lack of standards for many individual flavonoid compounds in foods and the cost of developing and using standards. The quantification difficulties involving flavonoids are being resolved with the use of one standard for several similar compounds,^173,217,219 mass spectroscopy extractable chromatogram methods,²¹⁶ and ultraviolet molar relative response factors.^160,169 When evaluating the chemical data on foods and plants in the literature for inclusion in databases, database developers must determine if the analytical methods used were appropriate.

Table 4.

Current methods used for chemical analyses of flavonoid classes in foods and dietary supplements

Flavonoid class	Extraction	Hydrolysis	Separation	Identification	Quantification	Comments
Anthocyanidins146,162,165,170,171,173,177,179,202,217–219	Acidic alcoholic such as methanol/water/acetic acid 85:15:0.5 v/v	Sugars present, but hydrolysis not usually done due to rearrangement and degradation	RP-HPLC: Binary elution usually either 1) aqueous 5% formic acid and methanol or 2) water-formic acid, 9:1 (v/v), and acetonitrile formic acid 9:1 (v/v)	UV maxima 265–275 and 465–560 nm. MS – peaks for compound and aglycone; very poorly ionized in negative ion mode, quantified in positive ion mode	An individual anthocyanin standard is often used to quantify several anthocyanins. MS can be employed for qualitative and quantitative analysis of anthocyanins in foods	Acid medium stabilizes anthocyanins. Anthocyanins more stable than anthocyanidins (aglycones) because of sugars
Flavonols1,160,162,170,175,178,187,189,190,195,200,202,210,213	Flavonoid glycosides more soluble in water and flavonoid aglycones more soluble in methanol. 50% aqueous ethanol most efficient; 60% or 70% methanol often used	Sugars usually present; hydrolysis often done using 1.2 MHCl, 50% methanol/water (v/v). C-glycosides are not appreciably acid hydrolyzed, and no glycosidase is yet known to hydrolyze carbon-carbon linked sugars	RP-HPLC: Binary eluant usually water with 1% formic acid, 0.01% trifluoracetic acid, or 10% acetic acid and acetonitrile with the same acid and percentage	UV maxima 240–280 and 300–380 nm. Fluorimetry (more sensitive) used for some flavonols. MS – compound ion usually most intense peak; O- and C-glycosides fragment by losing their sugars; better ionized in negative ion mode	Individual standards for aglycones are common. Usually quantified using UV at 270 nm and 360–370 nm or using MS. Quantitative analyses of flavonol monoglycosides can be done using calibration curves of quercetin 3-O-glucoside	C-glycosides are sometimes present but not often separated, identified, and quantified
Flavones125,160,161,167,181,183,189,194,195,200,202	Flavone glycosides usually extracted with flavonol glycosides. Methanol–water (60:40, v/v) has high extraction efficiency for glycosylated flavonoids, greatest simplicity, and least cost	Sugars usually present; often done with HCl in 50% aqueous methanol and, at appropriate temperatures, aglycones stable in acid medium. The major drawback of acid hydrolysis is isomerization of unsymmetrically substituted flavones (e.g., vitexin and isovitexin, Figure S6). C-glycosides resistant to hydrolysis	RP-HPLC: Binary eluant usually aqueous acetic or formic acid and acetonitrile (or methanol) with the same acid and percentage (e.g., 5% aqueous formic acid and methanol with 5% formic acid)	UV maxima 240–280 and 300–380 nm. 334–338 nm for apigenin, 342–350 nm for luteolin, 213–216 nm for polymethoxyflavones (Figure S4). Fluorimetry used for polymethoxyflavones. MS – O- and C-glycosides fragment by losing their sugars. Compound ion most intense peak; better ionized in negative ion mode	Individual standards for apigenin and luteolin available	C-glycosides very common but not often separated, identified, and quantified; usually only apigenin and luteolin determined out of at least 20 flavone aglycones found in food
Isoflavones166,179,180,197,199,203–205,207,208,211,212,214	80% methanol or aqueous acetonitrile. 58% acetonitrile extracts significantly higher amounts of malonyl glucosides than 80% methanol or 83% acetonitrile	Sugars usually present; commonly done because aglycones are stable during heat and acid treatment. Coumestrol (Figure S2) and genistein most sensitive to acid treatment. Most aglycones are liberated using enzyme hydrolysis	RP-HPLC: Binary eluant usually acetonitrile and water, both with acid (0.1–1% acetic, formic, or trifluoroacetic acid)	UV maxima 250–275 and 300–340 or 465–560 nm. Fluorimetry only for daidzein, formononetin, and coumestrol. MS usually removes sugars from the isoflavone glycoside. In other words the isoflavone aglycones tend to remain intact when subjected to MS.	Individual standards for aglycones common. Standard curves for most isoflavones show high linearity when standard concentration is plotted as function of peak area from HPLC analysis. Internal standards commonly used	Legume-containing foods (such as lentil soup and baked beans) are rarely analyzed for their isoflavone content or for other flavonoids
Flavanones1,125,160,165,169,178,179,181,194,198,201,202	Methanol, ethanol, acetone, water, ethyl acetate (to a lesser extent, propanol, dimethylformamide) solvent combinations frequently used for extraction. Using methanol–water (e.g., 60:40, v/v), extraction efficiencies for glycosylated flavanones in orange peel were <80%	Sugars present, but hydrolysis not usually done. Acid hydrolysis destroys flavanone aglycones nearly as fast as the flavanone glycosides are hydrolyzed. Alkaline hydrolysis can cause the formation of chalcones (Figure S7)	RP-HPLC: Binary eluant usually acetonitrile and water, both with acid (0.1–1% acetic, formic, or trifluoroacetic acid)	UV maxima 270–295 and a weak peak or shoulder at 300–360 nm. MS – flavanone glycosides provide parent and aglycone ions, and aglycone fragments in both positive and negative ion mode. Rutinosides (Figure S5) tend to fragment easier than neohesperidosides	Individual standards for most glycosides readily available	Presence of sugars needed for stability during analysis. Analyses are usually limited to certain glycosides, so aglycones may be underestimated. Flavanones have diastereoisomers that require chiral columns to separate
Flavan-3-ols169,171,174,176,182,183,192,200,201,206,208	Commonly, methanol–water or acetone/water/acetic acid (e.g., 70:28:2 v/v/v). 90% aqueous methanol most efficient compared with 90% ethanol, 90% acetone, or water-saturated ethyl acetate	Not done; sugars not present but can be gallated. If flavan-3-ols are present in an extract exposed to acidic hydrolysis, destruction and/or rearrangement can occur	RP-HPLC: Binary eluant usually acetonitrile (and/or methanol) and water, both with acid (0.05–2% acetic, formic, or trifluoroacetic acid)	UV detection at 210, 240–280 nm. Fluorimetry also used. MS – catechin gallate esters cleave at the gallate (Figure S4). In negative-ion mode, mass spectra dominated by the molecular ion	Individual standards available. Quantified by comparing peak height of sample with those of standards	Each compound is separated and quantified
Proanthocyanidins159,168,172,184–186,188,191,193,196,206,215,216,220	Aqueous acetone (e.g., 70% v/v) or methanol–water (e.g., 60:40 v/v). Use of acidified aqueous acetone increases extraction of procyanidins by weakening hydrogen bonds between procyanidins and polar fibrous matrixes	Not done intentionally. Acid degradation can occur due to organic acids present in separation and extraction	NP-HPLC: Ternary eluant most common, usually dichloromethane, methanol, and acetic acid/water (e.g., 1:1 v/v); separates proanthocyanidins according to degree of polymerization up to decamers.	UV maxima at 200–220 nm, 240–280 nm, and 300–380 nm. UV detection commonly done at 280 nm and is not specific in presence of other polyphenols. Fluorimetry also used. MS – fragments are formed by loss of monomeric residues, and only low-molecular-mass procyanidins are quantified. The positive mode is successful in detecting procyanidin oligomers through pentamers. MS does not distinguish different stereoisomers	For NP-HPLC, quantification is based on area under the curves using cocoa standards and does not use individual compound standards. Area summation of peak grouping is used to include contributions from all isomers within an oligomeric class. Peak breadth of certain oligomers is due to number of isomers. For RP-HPLC, each compound is separated and quantified	Individual compounds are not separated and quantified in NP-HPLC; monomers may be free flavan-3-ols or cleaved from oligomers and polymers during analysis (extraction and/or isolation). Neither NP-HPLC nor MS separates isomers
			RP-HPLC: For lower-molecular-weight proanthocyanidins, order of elution not related to degree of polymerization. Binary eluant is acetonitrile (or methanol) and water, both with acid (acetic, formic, or trifluoroacetic acid)

Abbreviations: HPLC, high-performance liquid chromatography; MS, mass spectrometry; NP-HPLC, normal-phase high-performance liquid chromatography; RP-HPLC, reverse-phase high-performance liquid chromatography; UV, ultraviolet.

Completeness of flavonoid databases

The primary, or core, flavonoid databases provide extensive and precise data on the compounds in many foods, with a great deal of detail on variability, preparation, and processing. Table 5 compares the three primary flavonoid databases currently available: the USDA databases^203,224,225 for flavonoids,⁵⁵ isoflavones,⁵⁴ and proanthocyanidins⁵⁸ and the Phenol-Explorer^{56,114,156,226–228} and EuroFIR-eBASIS^229–234 databases. The values in these USDA and European databases are generally similar because they are based largely on a common group of analytical data for food flavonoids in the literature. The most recently updated databases contain the greatest number of foods and the greatest amount of flavonoid analytical data points for each food.

Table 5.

Characteristics of the different primary flavonoid databases as of 2014

	USDA database54,55,58,224	Phenol Explorer56,114,156,226–228	EuroFIR ePlantLIBRA and eBasisa 229–234
Coverage	Flavonoids, isoflavones, proanthocyanidins	All phenols	All major bioactive compounds in food plants and supplements
Specific classes included	Anthocyanidins, flavan-3-ols, flavanones, flavones, flavonols, proanthocyanidins; isoflavones, and 1 coumestan	Anthocyanins; chalcones; dihydrochalcones; flavan-3-ols; flavanones; flavones; flavonols; isoflavonoids; phenolic acids, lignans, stilbenes, other polyphenols (curcuminoids, furanocoumarins, tyrosols, etc.)	30 classes, including: flavonoids, proanthocyanidins, biogenic amines, carotenoids, cyanogenic glycosides, glucosinolates, oligosaccharides, oxalates, phytosterols, and other polyphenols
Total no. of compounds included	26 individual flavonoids; 3 major and 2 minor isoflavones and the coumestan coumestrol; proanthocyanidins grouped by degree of polymerization monomers, dimers, trimers, 4–6 mers, 7–10 mers, and >10 mers; >24 132 original data pointsb	518 polyphenols (282 flavonoids) in 6 classes and 31 subclasses, >37 636 original data points; >20 347 data points for flavonoids	>25 000 data points covering 240 plants, and >500 distinct compounds in 30 classes
Date released	Flavonoids: March 2003; latest version (3.1) May 2014, with glycoside data in MS Access	2009	EuroFIR eBASIS started in 2005, ePlantLIBRA started in 2010
	Isoflavones: April 1999; latest version (2.0) September 2008
	Proanthocyanidins: August 2004
Total no. of foods included	Flavonoids: 506 food items	>455 foods	Database includes 370 plants and plant food supplement ingredients commonly consumed in Europe
	Isoflavones: 557 foods
	Proanthocyanidins 205 foods
Available as	Free pdf or MS Access database	Free, by Internet in MySQL or MS Access database, reports exported in Excel spreadsheets	By subscription, reports exported in Excel spreadsheets
Database data sources	CALS, FSTA	Article citations, FSTA, Web of Knowledge	CAS SciFinder, Embase, PubMed, WoS
Sources	Flavonoids: 307 articles, 1 industry source	>636 articles in peer-reviewed journals	Composition data extracted from primary research in peer-reviewed journals
	Isoflavones: 102 articles, 2 unpublished academic data
	Proanthocyanidins: 30 articles
Acceptable chemical methods	Flavonoids: HPLC, CC, capillary zone electrophoresis, micellar electrokinetic capillary chromatography	HPLC, GC, capillary electrophoresis; Folin assay, pH differential method for anthocyanins. For chromatographic methods, content values are reported separately as glycosides and esters (without hydrolysis) or as aglycones (after hydrolysis)	Chromatography, colorimetry, fluorimetry, GC, GC-MS, high-performance thin-layer chromatography, HPLC, LC-MS, near-infrared spectroscopy, NMR
	Isoflavones: Murphy et al.203 reference method
	Proanthocyanidins: NP-HPLC for all oligomers, RP-HPLC only to trimers
Data	mg/100 g fresh weight of edible portion of food; beverages as mg/100 g	mg/100 g fresh weight for solid foods and oils, and mg/100 ml for beverages, sauces, and seasonings.	mg/kg dry or fresh weight
Trace valuesc	Limit of quantification, if available, is multiplied by 0.71225		Trace amounts listed
Lack of valuesd	If zero value is provided, it is a true zero; if data are unavailable, compound is not listed	True zeros available if selected	If zero value is provided, it is a true zero; if data are unavailable, compound is not listed
Variability	By cultivar, location, agricultural practices, processing and storage conditions, preparation methods, environmental stress	By plant varieties, environmental factors, agricultural practices, food processing, food storage, and cooking	Subspecies/cultivar, maturity, country of origin, region, season, growing condition, processing
Data evaluation	5 criteria: sampling plan, sample handling, no. of samples, analytical method, analytical quality control to generate a quality index and confidence code	3 criteria (inclusion/exclusion criteria): samples, analytical methods, expression of results	6 criteria: food description, component description, sample plan, sample handling, analytical methodology, and analytical performance
Botanical names	Botanical name or Nutrient Databank no. provided	Botanical name and family provided	Botanical name, family, plant part used as food, common name (15 languages), and a color image provided
Presentation of flavonoid food data	Description, mean (weighted) mg/100 g, total no. of data points, SEM, minimum value from data points, maximum value from data points, confidence code, sources, standardized food item no. (NDB no. or other no.)	Weighted mean value, standard deviation, minimum and maximum, no. of original data points for mean value, total no. of individual samples analyzed, no. of papers. Data sorted according to analytical methods	User chooses fields: plant information (cultivar, part, origin), processing, analytical method, compound class, extraction and preparation, identification, mean, standard deviation, standard error, maximum level, minimum level
Ease of use	The USDA databases are the easiest databases to use in developing food flavonoid databases for epidemiologic studies because all amounts are in mg/100 g and all data calculations are done appropriately (e.g., glycoside to aglycone, dry weight to fresh weight, etc.). Information for individual data such as glycosides, dry weight, fresh weight, etc., is available in a separate file	Phenol Explorer has the most complete flavonoid glycoside data, which can be converted to aglycone values. Provides flavonoid data in mg/100 ml (volume) for liquid items. Includes metabolic, retention, structural, and other chemical data on flavonoids and other polyphenols. Data sources are easily and directly accessed	The EuroFIR-eBASIS database provides raw data from the literature in mg/kg dry or fresh weight. The investigator must parse, aggregate, and compile the data
Comments	Flavonoid database release 3.1 and isoflavone database release 2.1 are now expanded to include ≈2900 foods in the FSRG's FNDDS 4.1, released September 2014. An updated proanthocyanidin database is in process, and retention factors for flavonoids are in process	Phenol Explorer includes structures and chemical data. All compounds are linked to other chemical databases (CAS, ChEBI, PubMed). All mean values are linked to original publications. Original data used to calculate mean values are available. Of the 3 databases, it is the richest	This database contains information on beneficial and adverse bioeffects. It includes data on plant food supplements as well as plant foods. It also lists the individual compounds found in each plant
URL	https://www.ars.usda.gov/Services/docs.htm?docid=24953	http://phenol-explorer.eu/	http://eplantlibra.eurofir.org/
			http://ebasis.eurofir.org/

Abbreviations: CALS, Current Awareness Literature Service; CAS, Chemical Abstracts Service; CC, column chromatography; ChEBI, Chemical Entities of Biological Interest; Embase, Excerpta Medica dataBASE; FNDDS, Food and Nutrient Database for Dietary Studies; FSRG, Food Surveys Research Group; FSTA, Food Science and Technology Abstracts; GC, gas chromatography; HPLC, high-performance liquid chromatography; LC, liquid chromatography; MS, mass spectrometry; NMR, nuclear magnetic resonance; NP, normal-phase; PubMed, public access to MEDLINE (Medical Literature Analysis and Retrieval System Online, or MEDLARS Online); RP, reverse-phase; SEM, standard error of the mean; USDA, US Department of Agriculture; WoS, Web of Science.

^aeBasis (BioActive Substances in Food Information System) database, which covers bioactive compounds in food plants, has been enlarged and extended to ePlantLIBRA (PLANT food supplements: Levels of Intake, Benefit and Risk Assessment), which includes data on plant food supplements. All eBASIS composition data are included within ePlantLIBRA.

^bData point: one value for one compound in one food item.

^cTrace values or trace amounts: the item is analyzed, but the flavonoid values are below the limit of quantification of the method

^dLack of values: value is unknown or is missing because no analyses have been reported or published for that item and flavonoid class. True zeros: a zero where the absence of a flavonoid class or compound is based on chemical analyses.

For example, the first USDA release in 2003 had 225 food items, the release in 2007, 385 food items, and the release in 2011, 500 food items for 26 predominant flavonoids. The 2013 update includes 506 foods. Using blueberries as an example, in 2003 the USDA release had 2 items, in 2007 it had 4 items, in 2011 it had 5 items, and in 2013 it had 5 items but more data points. As a result, the total anthocyanidin value of raw blueberries (USDA National Nutrient Database for Standard Reference #09050) has varied: 113 mg/100 g in 2003, 164 mg/100 g in 2007, 131 mg/100 g in 2011, and 163 mg/100 g in 2013. Table 3 shows some common foods, which vary both in their flavonoid content and in the frequency with which they are consumed. Flavonoid intake estimates using updated databases tend to be higher than older estimates. Omissions of certain foods very rich in flavonoids, such as tea, cocoa, and berries, and also foods that are lower in flavonoid content but are eaten frequently in large amounts, such as tomatoes and potatoes, may lead to underestimates of flavonoid intakes.^227,228

Flavonoid-rich dietary supplements are often not included in dietary assessment instruments or, if they are included, only a generic question on supplement use is asked (e.g., use of vitamin-mineral or botanical supplements). Detailed information on form and dose is needed when individuals report the use of flavonoid-containing dietary supplements.¹⁰

Flavonoid-rich supplements include black cohosh, blueberry extracts, chaste tree, citrus extracts, dong quai, evening primrose, Gingko biloba, green tea extracts, kava, lemon balm, licorice root, red clover, St. John’s wort, saw palmetto, soy, and valerian. Although single-ingredient flavonoid products (such as chrysin, daidzein, diosmin, genistein, hesperidin, quercetin, and rutin) are sold over the counter, most supplements, including soy, green tea, citrus, or bioflavonoids, are blends, extracts, or mixtures containing several flavonoids (often in more than one class). Regulations do not require the precise composition of blends to be validated by chemical analysis and stated on labels. Supplement products are reformulated often, making it difficult to know what is actually present. A dietary supplement label database that provides information on labeled ingredients for many products sold in the United States is now available.²³⁵ At present, survey data are insufficient to estimate the prevalence of the use of flavonoid-rich dietary supplements, but if it is high in certain groups^236,237 such as postmenopausal women,^26,238–240 this would be an incentive to include specific questions in surveys of such groups.

Inadequacy of dietary assessment tools

Dietary assessment of flavonoid intakes is unique because there are so many flavonoids in foods and the food composition tables upon which assessment depends are still quite incomplete, particularly for flavones and proanthocyanidins. Inadequate dietary assessment tools also contribute to misestimation in exposure assessment.^{61,95,241–249}

In most epidemiologic studies, very large numbers of participants (often tens of thousands) report their usual diets using semiquantitative food frequency questionnaires that maximize information while reducing both participant burden and the costs of coding and analysis. Even so, food frequency questionnaires are particularly challenging for the assessment of flavonoid intakes. Most food lists for food frequency questionnaires used in US cohort studies were constructed on the basis of contributions to energy intakes or other key micro- or macronutrients of foods and beverages consumed by the study population or a sample of the entire US population (e.g., National Health and Nutrition Examination Survey [NHANES]) without consideration of bioactive compounds. Since analytical data on flavonoids were not available during the construction of the original lists used in most food frequency questionnaires, older questionnaires still in use today may not be optimal to reflect the variability in dietary flavonoid intake in a population. This is primarily due to the paucity of information that would have allowed earlier researchers to identify key contributors of different flavonoids in the diet and place them in food frequency questionnaire food lists. Moreover, early food frequency questionnaires were designed to provide a rank order of usual intakes and not to provide quantitative estimates of absolute amounts of flavonoid intakes. Data from food frequency questionnaires developed in this way were often later used for studies of the association between flavonoids and health outcomes.^250–252

Thus, the content of food frequency questionnaires is important. Some major flavonoid sources have been added to updated food frequency questionnaires as scientific interest in them has grown (e.g., tofu, tea, soy milk, blueberries). Some investigators have developed questions that focus on foods or dietary supplements that are known to be high in the bioactive compounds of particular interest, such as isoflavones in soy products (tofu and soy milk), or that contribute high amounts of flavonoids because they are consumed very frequently (e.g., potatoes, which are low in flavonoid content but are eaten very frequently in large amounts).^252–254 Some questionnaires now include questions on seasonally available fruits (apricots, blueberries, cherries, nectarines, strawberries). The appropriateness of questions on seasonality will depend upon the food availability in the populations studied.

A common problem in food frequency questionnaires is the combination of multiple foods with varying flavonoid content in the same line item. On some generic food frequency questionnaires designed to assess the intakes of many nutrients and not specifically the intakes of flavonoids, the criteria for combining certain foods in one line item were based on similar content for other nutrients (carotenoids, vitamin C, etc.) rather than for flavonoids. Such food groupings for composite items may not be appropriate for the precise assessment of flavonoid intake. For example, one early food frequency questionnaire grouped melons with berries, which would lead to underestimates of anthocyanidins, flavonols, and proanthocyanidins.²⁵⁵ Similarly, composites such as “citrus fruits” and “wines” may vary greatly in their flavonoid values. Another issue with composite items is that the weights given to the different items may vary from questionnaire to questionnaire (based on underlying intake in the study), so researchers should be careful to apply appropriate weights when deriving flavonoid values.

Table 6 illustrates how descriptions of some composite items (weighted equally for illustrative purposes) differ on existing food frequency questionnaires. For example, depending on which fruits are included, values for total flavonoids may vary from 113 mg/100 g (for peaches, apricots, and plums) to 60 mg/100 g (for peaches and nectarines) to 48 mg/100 g (for peaches, apricots, and nectarines). The differences are also notable for composites of the mustard family (Brassicaceae), ranging from 61 mg/100 g for kale, chard (actually in the beet or Chenopodiaceae family), and mustard greens, to 45 mg/100 g for kale, collards, mustard, and turnip greens, to 29 mg/100 g for collards, mustard, and turnip greens. For squashes and other vegetables, the values for the composites ranged from 29 mg/100 g (summer squash, eggplant, zucchini) to 10 mg/100 g (summer squash, okra, green pepper) down to 1 mg/100 g (summer squash, yellow squash, zucchini). Protocols for weighting the foods in a composite item vary and may increase the differences between values on food frequency questionnaires for individual foods and composite items. Standardization would facilitate comparisons across studies.²⁴⁶

Table 6.

Examples of composite items listed on three food frequency questionnaires commonly used in epidemiologic studies

Composite items and questionnaire source						All items in one botanical family?	Comments
Harvard (Willett) food frequency questionnairea	Total flavonoids, mg/100 g	Block food frequency questionnaire, 1998b	Total flavonoids, mg/100 g	National Cancer Institutec	Total flavonoids, mg/100 g
Fresh apples or pears	80d	Apples or pears	80	Raw apples or pears	80	Yes	Same botanical family (Rosaceae)
Orange juice	15	Real 100% orange juice or grapefruit juice, including fresh, frozen, or bottled	18	Fresh or frozen orange or grapefruit juice	18	Yes	Same botanical family (Rutaceae) but very different flavanone content, both in aglycones and in sugars
Peaches, apricots, or plumse	113	Raw peaches, apricots, or nectarines, while in season	48	Fresh peaches or nectarines, in season	60	Yes	Same botanical family (Rosaceae)
Kale, mustard, or chard greens	61	Mustard greens, turnip greens, collards	29	Collards, mustard or turnip greens, or kale	45	No	All belong to Brassicaceae family except chard (spinach family, Chenopodiaceae)
Eggplant, zucchini, or other summer squash	29	Any other vegetable, like okra, squash, or cooked green peppers	10	Zucchini, yellow, or summer squash in season	1	No	All belong to Cucurbitaceae family except eggplant and green peppers (Solanaceae) and okra (Malvaceae)

^aThree sources of the Willett food frequency questionnaire. 1) Diet Assessment. Copyright 1988, Brigham and Women’s Hospital. All rights reserved worldwide; 2) Harvard Medical School Nurses’ Health Study. Copyright 1998, Brigham and Women’s Hospital. All rights reserved worldwide; and 3) Harvard Medical School 2002 Nurses’ Health Study Questionnaire.

^bFood questionnaire. Block 98.2. Copyright 1998, Block Dietary Data System. www.nutritionquest.com.

^cEsophagus questionnaire. National Cancer Institute (oldest questionnaire; appears to have been administered by an interviewer).

^dTotal flavonoids (mg/100 g) for raw food items weighted equally for illustrative purposes using USDA databases.^54,55,58

^eA 2007 Harvard food frequency questionnaire has peaches or plums (with apricots listed separately). Total flavonoids for “peaches or plums” average 159 mg/100 g.

Several investigators have developed food frequency questionnaires to assess isoflavone intakes more specifically.^256–263 In the United States, the Block phytoestrogen questionnaire^260,264,265 has been used in the Study of Women’s Health Across the Nation,²⁵² the Women’s Health Initiative (Fred Hutchinson Cancer Center),²⁵⁷ the Cancer Research Center of Hawaii,²⁶⁶ and the Adventist Health Study²⁶¹ to measure isoflavonoid intakes. Some specialized questionnaires with a few flavonoid classes and compounds have been developed and used in some studies,^{254,267–271} and other more complete flavonoid questionnaires^272–274 are being developed. Recently, the USDA released flavonoid data from the NHANES surveys for foods reported in 24-hour recalls, starting with the 2007–2008 survey years,^275,276 simplifying the task of identifying foods that are important contributors of flavonoids in the US diet.^4,43,277 Similar data are, or soon will be, available for populations in European countries, Australia, and elsewhere.^85,278,279 This information is critical for identifying important sources of flavonoids to modify food frequency questionnaires and to develop food lists for targeted interviews. More work is needed in food frequency questionnaire development to capture important sources of additional flavonoids in US and other Western diets.

The alternative to food frequency questionnaires is the use of multiple food recalls and records to gather reports of all foods and beverages consumed. Multiple dietary recalls or records have the advantage that all foods and dietary supplements eaten are reported and quantified in 1-day time periods in replicate. Such data provide an estimate of intraindividual variation of intakes. Depending on the size of the study, however, this type of dietary assessment may not always be feasible, as daily recalls and records can be expensive and time intensive. Newer online technologies may make the use of multiple dietary recalls more acceptable.^280,281 A few studies, such as the European Prospective Investigation Into Cancer and Nutrition–Norfolk^61,269 and the Women's Health Initiative,²⁸² have collected both multiple food recalls and food frequency questionnaires. This use of dual assessments is helpful for capturing infrequently consumed flavonoid-rich foods and for developing even more complete food frequency questionnaires in the future.

OTHER ISSUES UNRELATED TO DATABASES: BIOAVAILABILITY AND BIOMARKERS OF INTAKE

Other sources of variation in exposure include the unknown bioavailability of flavonoid compounds.²⁸³ The bioavailability of flavonoids is thought to differ depending on the flavonoid, the food matrix, and the influence of intestinal bacteria in biotransforming the compounds in the gut, all of which contribute to intra- and interindividual variability of biological flavonoid exposure.^284–294 The gallate, methoxy, or sugar groups (Figures S4–S6) as well as the structure of the aglycone are important in determining flavonoid bioavailability.^{112–115,295} The Phenol-Explorer database provides pharmacokinetic data from animal and human studies of various compounds.¹¹⁴

The lack of biomarkers of flavonoid intake is another problem. Although work is continuing and consensus is developing on the best candidates, at present there are no agreed upon biomarkers of intake for all the various flavonoid classes, owing in part to the interindividual variation in the metabolism of flavonoid compounds.^{95,249,296,297}

RECOMMENDATIONS FOR IMPROVING ESTIMATES OF FLAVONOID EXPOSURE

It is clear that the estimated flavonoid intakes in US and European cohorts vary greatly, owing in part to databases that vary in their completeness and to the use of different dietary assessment tools that were not designed to estimate flavonoid intakes precisely. In addition, there are difficult-to-quantify influences of cultivar, cooking method, degree of fruit or plant maturity at consumption, etc. These methodological limitations may obscure true differences in intakes, limit the ability to observe associations between flavonoid intakes and chronic degenerative diseases, and contribute to inconclusive findings in published studies.

Many of the shortcomings of flavonoid food and dietary supplement composition tables are remediable. For example, the development of improved chemical analyses and more complete flavonoid food and supplement databases is advised. Dietary assessment can be improved by adding flavonoid foods and supplements to food frequency questionnaires and using checklists to probe dietary recall and record responses to make sure all flavonoid-rich sources were mentioned and sources of variability ascertained (e.g., cooking method). The recommendations in Table 7 provide cross-disciplinary suggestions to advance the knowledge of food flavonoids,^298–301 starting with suggestions to improve methods of analyzing the flavonoid content of foods and supplements. In addition, standardization across laboratory methods is recommended. More complete data on the flavonoid content of foods and dietary supplements should be included in the available databases, and researchers should use these up-to-date flavonoid databases. Major food and supplement sources of flavonoids included in dietary assessment instruments should be based on population-based studies (such as NHANES) and reflect the consumption patterns of the study population.

Table 7.

Recommendations to improve the assessment of flavonoid exposure

Investigators	Recommended action	Comment
Analytical chemists	Use up-to-date and appropriate chemical analyses of flavonoids	For flavonoids, HPLC is currently the most robust method for separation and quantification. Analyses that include the C-glycosides and methoxylated flavonoids as well as the O-glycosides are needed. More data on flavonoid glycosides are now becoming available
	Utilize standard reference materials (when available) for quality control	Reference materials for flavonoids are usually developed in-house. Some external standard reference materials are now being provided through the National Institute of Standards and Technology and other agencies
Food composition database developers	Fill existing gaps on commonly eaten foods that contribute to total flavonoid intakes	Include the most popular varieties of each food item (e.g., Delicious, Granny Smith, McIntosh varieties of apples) due to variability by cultivar and variety
	Ensure that all appropriate foods are included in flavonoid databases	All food items and recipe constituents must be identified botanically. Some foods with the same name not only are different plants but are often in different botanical families, depending on the community. For example, “spinach” is usually Spinacea oleracea (Chenopodiaceae) in the United States, Rumex patientia (Polygonaceae) in Europe, and Ipomea aquatica (Convolvulaceae) in Asia
		Use an up-to-date primary database that is relatively complete and based on current analytical methods. Teas and other foods vary in their flavonoid content, depending on their preparation; values should be included for commonly used varieties and variants
Nutritional epidemiologists	Ensure that dietary assessment tools are appropriate for obtaining estimates of flavonoid intakes	When serial 24-h recalls or diet records are available and representative groups queried, use the data to construct purpose-built FFQ food lists if those foods are consumed at least weekly in the population. 24-h recalls and food records are most useful if they are obtained randomly on numerous occasions so that factors such as seasonality are taken into account. However, this may not always be feasible
		Check the scientific literature for flavonoid content of foods not listed in primary databases. A food’s absence from a primary database does not mean it does not contain flavonoids. In some cases, flavonoid data for a specific food can be imputed from another botanically related food
		Focus on greater precision of estimated intakes. If possible, standardize FFQs across studies to optimize pooled analyses and replication. Develop new and better purpose-built FFQs that focus on flavonoid-rich or commonly eaten foods. Commonly consumed flavonoid-rich foods or supplements that capture variation across the population are particularly important and should be added to FFQs whenever possible. See Tables 1 and 3 for sources of flavonoid-rich foods
		Certain details such as cultivar, herbs, and spices may be less feasible to assess using a FFQ except with a flavonoid-focused survey. In this case, significant sources of flavonoid intake (foods, supplements, and medicines), including recipes and composite items, may be queried (see Tables 1 and 3 for some key food sources)
	Validate estimates of flavonoid intakes with biomarkers when possible	Whenever available, biomarkers of flavonoid intakes should be used to validate intake estimates. Continue to identify new and improved biomarkers of flavonoid intake
	Use additional statistical techniques in assessing the relationship between flavonoid exposures and health outcomes	Avoid multiple statistical testing of dozens or even hundreds of individual flavonoid compounds, as this increases the likelihood of false-positive findings, particularly if adjustments such as the Bonferroni correction are not done
		Because associations of risk for flavonoids and other nutrients or non-nutrients may not be linear, consider statistical techniques (e.g., restricted cubic splines) to evaluate the shape of the association and to evaluate the linearity assumption
	Explore the gains to be made by controlling misestimations in flavonoid exposure estimates	Estimate the effects of improving flavonoid exposure estimates on relative risks and health outcomes in epidemiologic studies. It would be useful to carry out simulations to determine how much effect sizes and risk estimates improve by doing so
	Conduct both observational and intervention studies of flavonoids and disease risk	Observational studies are useful for examining associations between relative flavonoid consumption and chronic disease outcomes that require years to develop. Randomized double-blinded clinical trials of foods or components rich in flavonoids will be needed to prove causality but may not be feasible for studying chronic disease outcomes. Intervention studies are more feasible for studying shorter-term, modifiable biomarkers such as blood lipids, intermediate markers of disease such as blood pressure, and markers of glucose homeostasis and inflammation. Intervention research requires careful consideration of exposure form, dose, and length of intervention
	Devise flavonoid-rich dietary patterns associated with decreased health risks	Recently, health recommendations from the American Institute of Cancer Research,298 the USDA,299 the American Cancer Society,300 and other organizations have focused on food patterns or groupings rather than specific nutrients in order to simplify communications and to deal with the issue of intercorrelated nutrients.301 With computer programs that permit categorization of food intakes by family, genus, and species, it may be possible to better identify flavonoid-rich dietary intake patterns that are most beneficial for health

Abbreviations: FFQ, food frequency questionnaire; HPLC, high-performance liquid chromatography; USDA, US Department of Agriculture.

Ideally, future work would provide estimates of the impact of food form, processing, and preparation on flavonoid values (as is provided by Phenol-Explorer). Randomized intervention studies should also examine the doses and forms of compounds consumed in a flavonoid-rich diet, along with their impact on blood biomarkers. Food frequency questionnaire line items with more than one food should consider flavonoid content whenever possible and be appropriately weighted for the study population. Clearly, decisions in the development of dietary assessment instruments will be difficult because researchers must balance the need to assess multiple dietary factors with keeping the instrument relatively brief so as not to tire or overwhelm participants.

The harmonization of dietary assessment instruments across studies will facilitate future pooling of data, improve statistical power, and enable examination of rarer outcomes and disease phenotypes. Prespecification of hypotheses, the judicious use of statistical tests, and corrections for multiple testing are also recommended because of the multiple compounds involved and the many exposures of interest in flavonoid research. Finally, researchers should include more complete descriptions of the methodology used in assessing and analyzing flavonoid data, including the procedures used in estimating and categorizing flavonoid and proanthocyanidin classes and individual compounds.

CONCLUSION

Better food composition tables can strengthen studies of the links between intakes of bioactive compounds, such as flavonoids, and health outcomes. However, to determine whether flavonoid intakes (or intakes of other bioactive compounds) lower the risk of chronic degenerative diseases, the aggregate evidence from many types of data must be examined. These data include in vitro, animal, clinical, and observational data and, when feasible, the results of randomized intervention studies. Complete food and supplement composition databases for flavonoids and appropriate dietary assessment methods are vital to any food-based research on flavonoids. The recently available flavonoid data from NHANES 24-hour recalls and other such studies highlight flavonoid-rich food and supplement sources for epidemiologic research. If these methodological issues are addressed, more precise estimates of flavonoid consumption can be evaluated in relation to health outcomes, and recommendations on optimal intakes for health can proceed more rapidly from speculative to evidence-based advice.