OUP user menu

Characterizing whole diets of young children from developed countries and the association between diet and health: a systematic review

Lisa G Smithers, Rebecca K Golley, Laima Brazionis, John W Lynch
DOI: http://dx.doi.org/10.1111/j.1753-4887.2011.00407.x 449-467 First published online: 1 August 2011

Abstract

Early childhood is an important nutritional period that involves the transition from a milk-based diet to ordinary foods. A systematic review was conducted of studies that applied whole-of-diet analysis of children aged 1−5 years to examine associations between diet and nutrition, health, and development. Literature searches identified 40 articles using dietary indices, principal component analysis, or cluster analysis. Reports that applied indices (n = 23, 18 indices) were cross-sectional, and most measured diet quality or variety. Articles reporting principal component or cluster analyses (n = 17) described between two and six dietary patterns, and most identified healthy, unhealthy, and traditional patterns. In cross-sectional analyses, mixed associations were found between index or pattern scores and nutrient intake (n = 10), nutritional biomarkers (n = 1), and anthropometry (n = 10). Five reports from two birth cohorts showed healthier dietary patterns were associated with better lean mass, cognition, and behavior, but not with bone mass or body mass index at later ages. Few studies have characterized the diets of children under 5 years of age and linked diet with health. Given the limited evidence, research establishing the predictive validity of whole-of-diet methods in childhood is needed.

  • children
  • cluster analysis
  • diet index
  • predictive validity
  • principal component analysis
  • toddler

INTRODUCTION

Over the last two decades, it has become increasingly acknowledged that events occurring in early life may have a lasting effect on an individual's health.13 Diet in infancy is considered an important and modifiable exposure that has short- and long-term implications for health and development. For example, breastfeeding has been associated with a slower growth trajectory and enhanced cognitive development,4,5 and sodium intake in the first few months of life has been positively associated with blood pressure in childhood and adolescence.6,7 The types of diets fed to infants change from milk as the sole food source to the introduction of solids and foods that reflect the family diet. It is plausible that dietary exposures during this transitional period contribute to future food and taste preferences.8,9 Tracking of dietary patterns from age 3 to 9 years10 and from age 6 years into adulthood11 suggests the preferences established prior to primary-school age may influence longer-term food choices. Therefore, understanding whether early dietary exposures shape later health and development is an important area of research. A prerequisite for this research, however, is the ability to reliably and validly characterize diets in children under 5 years of age.

While a great deal of nutrition research is centered on the role of specific nutrients or foods, summarizing dietary data into an overall measure offers the alternative of studying the relationship between whole diets and health. A number of data-reduction methods have been used, and, as described by Kant,12 whole-of-diet analysis has taken a priori or a posteriori approaches. A priori methods involve the assessment of food intake data against a dietary index, which is determined prior to analysis and usually reflects dietary guidelines or current nutrition knowledge. A posteriori or data-driven techniques examine foods that are often consumed together to arrive at a dietary pattern score or to group together people who consume similar types of foods. In the adult literature, both indices and dietary patterns have been linked with chronic disease outcomes, thus demonstrating their predictive validity.1315 For example, healthy dietary patterns, but not Western dietary patterns, are associated with lower odds of breast cancer.14

Although the association between whole-of-diet and health has been reviewed for adults,1218 there has not been a systematic review of studies involving children under 5 years of age. Given the potential importance of identifying modifiable early-life factors that contribute to later health, the primary objective of this review was to evaluate whether whole-of-diet patterns of children between 1 and 5 years of age are associated with later health and development. Secondary aims were to describe the whole-of-diet patterns in children 1 to 5 years of age via a priori or a posteriori methods and evaluate whether whole-of-diet patterns from 1 to 5 years of age are associated with children's nutrient intake and sociodemographic characteristics.

METHODS

Information sources and search strategy

The Medline, EMBASE, and ISI Web of Knowledge databases were searched for relevant articles. Search terms were pilot tested for accuracy and tailored to each database. The final search strategy used for each database is available from the authors upon request. All searches included a combination of key words that refined the number of hits to studies involving diet, foods, statistical methods involving data-reduction methods, and the population group of interest. For example, the search string included diet-related terms (diet* with score, index, pattern, quality, or variety; feeding pattern*; nutrient adequacy; nutritional adequacy; food variety), data analysis methods (factor analysis, cluster analysis, principal component* analysis, or multivariate analysis) and was limited by studies conducted in humans and by infant or preschool age (asterisks in search terms indicate a wild-card suffix). The search was not restricted by language or date. Reference lists of included studies were searched for additional articles. The last search was performed in December 2009.

Eligibility and exclusion criteria

Randomized trials and cross-sectional and prospective observational studies were eligible for inclusion in the review, provided that data-reduction methods for assessing whole diets were measured. Retrospective studies were excluded due to the increased risk of error in dietary recall, which places a greater reliance on memory.19 Articles reported as abstracts were excluded. Only studies that involved children born full term and that measured diet at least once in the preschool period from 1 to 5 years of age were eligible. No restriction was placed on the method of assessing diet (e.g., diet diary, 24-h recall, food frequency questionnaire), provided that total diet was measured (e.g., assessments limited to individual foods or nutrients, meal patterns, or behaviors were excluded). Studies from countries that do not have dietary recommendations for infants and toddlers were included; this meant that indices developed for reasons other than adherence to dietary guidelines could be included. Studies from low-income developing countries according to the World Bank Criteria (based on per capita gross national income)20 or that involved children with known disease states were excluded, as the review was conducted with a view of generalizability to developed countries and to healthy, well-nourished children.

Study selection

Titles and abstracts identified by the search were independently screened by two authors (LGS and RKG), and any articles that did not meet the eligibility criteria were excluded at this stage. The full text of the article was retrieved if it was unclear whether the study met the inclusion criteria. All full-text articles included in the review were screened by at least two authors. Any discrepancies were resolved by consensus.

Data extraction and synthesis

Data from included studies was extracted using a standard form by one author (LGS) and verified by other authors (RKG, LB). Data extraction was informed by the STrengthening the Reporting of Observational Studies in Epidemiology (STROBE) checklist, which is designed to enhance reporting of cohort, case-control, and cross-sectional studies.21 Based on the STROBE checklist, items extracted included study design; cohort details (country and setting, enrollment year, participant characteristics, eligibility); method of assessing diet; statistical information applied to evaluate dietary pattern scores, including any adjustment for confounders; index components and structure; and study findings. Studies were unable to be combined in a meta-analysis due to the large disparities in methodological approaches and outcomes.

RESULTS

After combining all searches, a total of 834 titles were screened, 98 full articles were retrieved for further evaluation, and a total of 40 articles were included in the review (Fig. 1).2261 Of the excluded studies, 21 were from developing countries, 19 involved schoolchildren over 5 years of age, 10 were methodology or review articles, 7 involved nutrients or foods rather than whole diets, and 1 was a retrospective design.

Figure 1

Summary of the systematic searches of databases involving studies utilizing data-reduction methods for quantifying diet in preschool-aged children. Titles and abstracts were screened by two authors for inclusion. “Other” reasons for exclusion (n = 8) included papers of dietary analysis methodology or review articles.

Studies using dietary indices

Table 1 describes the 23 studies that applied an index in order to characterize the whole diet of infants, toddlers, or preschool-aged children.2244 All dietary data were cross-sectional, with 13 studies using food consumption data from national samples. A further three studies utilized data from large prospective cohorts but only applied an index at one point in time. Index studies were predominately from North America (14 studies, 9 from national nutrition surveys spanning 1977–2004), with South Africa, Europe, and Latin America also represented. Sample sizes ranged from 82 to 15,423. Six studies included over 1,000 participants,2536 although other large studies did not report the sample size for children in the eligible age range.2640 Indices were most commonly applied to dietary data collected by 24-h recall, followed by diet diaries or weighed food records. A food frequency questionnaire was used in only one study.26 While the number of days of intake assessed varied from one to four, a single 24-h recall is appropriate to reflect usual intake of samples.62

View this table:
Table 1

Studies examining indices of diet quality among preschool-aged children, by year of publication. All studies were cross-sectional. Refer toTable 2 for details of the content and structure of the indices.

Reference; country/setting of study, study name, & year diet assessed (if available)Sample size and ageDietary intake measurement; dietary data processingIndex scoreAssociation with diet or nutrient intakeAssociation with child, family, or sociodemographic factorsAssociation with health outcomesAdjustment of analysis
Kranz et al. (2009)22; USA, community health centers, 2004–05n = 104, 2–5 years (48 plausible reporters)3 × 24-h recall; nil processing details providedServings/day. 10% of children (4% of plausible reporters) met all recommendations12% of children <85th BMI percentile met all recommendations versus 6–7% >85th BMI percentile. Of plausible reporters, only 20% of children >95th BMI percentile (obese) met recommendationsInfluence of misreporting of dietary intake explored in sensitivity analysis
Manios et al. (2009)23; Greece, Growth, Exercise and Nutrition Epidemiological Study In preSchoolers (GENESIS), 2003–04n = 2,287, 2–5 yearsWFR + 24-h recall or food diary; processing based on US Food Guide PyramidHEI. Mean/maximum score ± SD 58.7/100 ± 8.2. Component scores ranged from 1.2/10 for vegetables & saturated fat to 9.9/10 for sodium. Subcomponent scores increased across quartiles of HEI (sodium not reported)Intake of energy and nutrients increased across quartiles of HEI scores for total energy (Q1 = 1,192, Q4 =1,597 kcal/d), CHO (% E), protein (% E), folate, fiber, iron, vitamin C, Mg, P, Zn, Ca. HEI score was correlated with nutrients (e.g., protein −0.05, Ca 0.09, fiber 0.6)HEI scores were 1–2 points higher for 4–5-year-olds, boys, rural locality, maternal education over 12 years, & maternal employmentNo difference in HEI scores by BMI-for-age <85th (normal), 85–94th, (overweight), and ≥95th (obese) percentilesStepwise regression adjustments unclear. Bonferroni adjustment for multiple comparisons.
Fungwe et al. (2009)24; USA, National Health and Nutrition Examination Survey (NHANES), 2003–2004n = 763, 2–5 years24-h recall; nil processing details providedHEI−2005. Mean/maximum score 59.6/100. Component scores ranged from 0.6/5 (legumes, dark green & orange vegetables) to 5/5 (total grains, fruits) and 10/10 (dairy)Adjusted for population level energy intake.
Libuda et al. (2009)25; Germany, Dortmund Nutritional and Anthropometric Longitudinally Designed (DONALD) Study, 1985–2009n = 851, 2–4 years3 days WFR, mean intakes calculated for energy from added sugars & macronutrients; 11 vitamins & 6 mineralsNQI. Mean scores reported by gender onlyNQI scores were lower with increasing sugar-sweetened beverage intake (2–4-year-olds not reported separately). NQI more strongly associated with nutrient density than nutrient intakeMean score 83.2 ± 10.0 for boys and 81.6 ± 10.0 for girlsChild factors and period enrolled in study.
Kleiser et al. (2009)26; Germany, German Health Interview & Examination (KiGGS) Study, 2003–2006n = NR for 3–6 years54-item semi-Q FFQ; nil processing details providedHuSKY. Mean/maximum score 59.3/100 (95% CI 58.8−59.8)Adjusted HuSKY scores correlated with serum folate (0.068, P < 0.001) and homocysteine (−0.068, P < 0.001) but not Ca (−0.017), Fe (−0.001), Mg (0.008), vitamin D (−0.011), or B12 (0.017) (unadjusted correlation P >0.05)Higher scores (1–8 points) positively associated with younger child age, girls, higher SES, parental education, family income, non-immigrants, urban households, residence in East GermanyNo significant association between HuSKY score and BMISampling design, child, dietary and SES factors
Crombie et al. (2009)27; Scotland, deprived communities, 2005–06n = 300, 2 yearsStructured questionnaire; portions of food groups consumed per day (portion sizes not reported)DS. 15% classified as having a “good diet” (i.e., met all 5 score criteria). Adherence ranged from only 34% of children meeting high fat/sugar snacks criteria to all children meeting dairy products criteria“Poor diet” was associated with mothers who do not limit sweets, have difficulty providing fruit, worry about child's intake, and do not provide breakfastRegression adjustments unclear
Kranz et al. (2008)28; USA, National Health and Nutrition Examination Survey (NHANES), 1999–2002n = 1,521, 2–5 years24-h recall; dietary data converted to US Food Guide Pyramid servings (cups/day)RC-DQI. Mean/maximum score 59/90 (range 21–86). The proportion of children awarded maximum points on subcomponents ranged from 8% for whole grains and 17% energy balance to 71% iron and 93% fat quality (DHA + EPA)Higher RC-DQI scores were associated with younger child age, higher family income, and Mexican-American compared with non-white Hispanic ethnicityThe proportion of children with BMI between 85th and 94th percentiles (overweight) or >95th percentile (obese) decreased from lowest to highest RC-DQI quartiles (i.e., higher diet quality)Sampling design, regression adjustments unclear
Kranz et al. (2006)29; USA, CSFII (1994–96, 1998)n = 5,437, 2–5 years2 × 24-h recall; nutrient and food servings converted to US Food Guide Pyramid serving size/100 gRC-DQI. Mean/maximum score 64/95 (range 28–93). The proportion of children awarded maximum points on subcomponents ranged from 5% whole grains and energy balance to 76% iron and 90% fat quality (DHA + EPA)Subcomponent scores increased across quartiles of RC-DQI for energy intake (Q1 1,416 versus Q4 1,612), CHO, saturated fat, protein, fiber, and Ca. Increasing quartiles of RC-DQI associated with lower proportion of children with vitamin and mineral intakes below the EARSampling design and multiple children from one family
Glanville & McIntyre (2006)30; Canada, low-income single mothers, 1999–2000n = 82, 1–3 years4 × 24-h recall; nil processing details providedHEI-C. Mean/maximum ± SD score 69.9/100 ± 12.1. Component scores ranged from 5.3 ± 3 saturated fat, 11.3 ± 5.2 fruit & vegetables, to 8.9 ± 1.3 for varietyMothers' HEI-C scores (54.0 ± 10.7) were significantly lower than children's HEI scores. No association between HEI-C scores and maternal education or food securityAny regression adjustments unclear
Hoerr et al. (2006)31; USA, rural, low-income familiesn = 100, 11–25 months2 × 24-h recall; nutrient intake and servings from US Child & Adult Care Food ProgramMAR. Mean/maximum ± SD score 94.7/100 ± 8.1Higher MAR was associated with lower intake of fruit (not including juice) and higher intake of vegetables and dairy food (all P < 0.01)
Steyn et al. (2006)32; South Africa, National Food Consumption Survey, 1999n = 795, 1–3 years24-h recall; nil processing details providedFVS & DDS. Mean/maximum ± SD scores FVS = 5.4/8 ± 2.2, DDS = 3.5/8 ± 1.2FVS and DDS were correlated with MAR (0.65 and 0.62 respectively, P < 0.0001)FVS and DDS were correlated with HAZ (r = 0.21, 0.15) and WAZ (r = 0.14, 0.10) but not WHZ
Dewey et al. (2006)33; Brazil, Ghana, India, Norway, Oman, & USA, World Health Organization Multicentre Growth Reference Study (MGRS), 1997–2003n = 903, 1–2 yearsFFQ + 24-h recall; foods grouped into 8 categoriesDDS. Mean ± SD (maximum for all = 8) at: 1 year = 4.6 ± 1.4; 1.5 years = 5.2 ± 1.3; 2 years = 5.3 ± 1.3At all times, DDS were lowest in Brazil and highest in Ghana
Kranz et al. (2004)34; USA, Nationwide Food Consumption Survey (NFCS) 1977–79 and the CSFII 1989–91, 1994–96, & 1998n = 8,628, 2–5 yearsNFCS & CFSII 1989: 1 × 24-h recall + 2 day food record; CSFII 1994: 2 × 24-h recall. Food data converted to US Food Guide Pyramid servings (portion size is ⅔ the adult portion)C-DQI. Mean ± SE score (maximum for all = 70). 1977–79, 2–3 years, 45.7 ± 0.2; 4–5 years, 41.9 ± 0.2. 1989–1991, 2–3 years, 46.6 ± 0.5; 4–5 years, 43.4 ± 0.5. 1994–98, 2–3 years, 47.7 ± 0.2; 4–5 years, 43.8 ± 0.2At 1977–78 and 1994–98, the highest tertile of C-DQI score had lower intake of added sugar, total fat, and juice and higher intake of grains, fruit, vegetables, iron, and energySampling design and Bonferroni adjustment for multiple comparisons
Knol et al. (2004)35; USA, CSFII 1994–96, 1998n = 1,242, 2–3 years2 × 24-h recall; based on 5 foods groups in US Food Guide PyramidHEI. Mean/maximum ± SE score 8.0/10 ± 0.1HEI variety score was not associated with food sufficiency. Higher HEI score was associated with WIC program involvementMaternal and SES factors
Ruel et al. (2002)36; Latin American countries, demographic & health survey, 1994–99n = 15,423, 1–3 yearsNRCFI. Mean/maximum scores of the 7 countries ranged from 6.5–8.9/12Higher CFI scores were associated with child age (in 4/7 datasets) , higher maternal education (6/7 datasets), higher SES (6/7 datasets), fewer children <5 years (all datasets), and rural residence (6/7 datasets)HAZ increased across tertiles of CFI scoreMaternal, child, and SES factors
Basiotis et al. (2002)37; USA, National Health and Nutrition Examination Survey (NHANES) (1999–2000)n = NR for 2–3 years24-h recall; nil processing details providedHEI. Mean/maximum score 75.7/100. Mean subcomponent scores ranged from 5.9 (saturated fat) to 8.9 (cholesterol and grains)Sampling design
Lee et al. (2001)38; USAn = 192, 5 years, girls only3 × 24-h recall; number of servings/d of food groups according to US Food Guide PyramidHEI. Mean score for total sample not reported. Mean/maximum ± SE score was 69.6/100 ± 0.8 for diets >30% E as fat and78.7/100 ± 0.8 for diets <30% E as fat (P < 0.05)
Carlson et al. (2001)39; USA, CSFII Supplemental Children's Survey, 1998n = NR for 2–3 years24-h recall; nil processing details providedHEI. Mean/maximum score 74.4/100, subcomponent scores ranged from 5.4 (saturated fat) to 8.7 (sodium)
Lino et al. (1999)40; USA, CSFII, 1994–96n = NR for 2–3 yearsNRHEI. Mean/maximum score 73.8/100, subcomponent scores ranged from 5.4 (saturated fat) to 9.0 (cholesterol)
Cox et al. (1997)41; USAn = 124, 2–3 years2-day diet diary + 24-h recall; US Food Guide Pyramid serving size adapted to toddlersVIT. Mean/maximum score 0.8/1.0VIT correlated with total E (range by age group 0.54 to 0.66) but not % E from fat. VIT correlated with MAR (r = 0.74, P < 0.01)
Campbell & Sanjur (1992)42; Canadan = 160, 2–4 years3-day diet diary; no processing details providedDQS1 & DD. Mean/maximum ± SD scores: DQS1 13.7/18 ± 2.0, DD 15 ± 2.8 (no maximum score)DD positively associated with DQS1Higher DD score was associated with licensed childcare, higher income, less job/family strain, and not using food for non-nutritional purposes (i.e., to reward or pacify child)Maternal, child, and SES factors
Krebs-Smith et al. (1989)43; USA, Nationwide Food Consumption Survey, 1977–78n = 151, 1–3 years24-h recall plus 2-day diet diary; nil processing details providedNAS. Mean NAS scores not reportedNAS correlated with MAR (r = 0.68, P < 0.01). NAS had 80% sensitivity to detect children with MAR<0.8
Caliendo et al. (1977)44; USA, New York well-baby clinicsn = 113, 1–4 years24-h recall; 134 food items condensed to 20 major items (e.g., milk, other dairy, peanut butter, eggs, potato, etc.)Diversity score & DQS2. Mean/maximum diversity score 9/20 (range 4–13). Mean DQS2 scores not reported; 18% of children achieved highest DQS2 level (6 points)Diversity score positively associated with DQS2 scores.Higher DSQ2 score was associated with diet diversity, girls, ordinal position in family, higher maternal education, & homemaker attitudeMaternal, child, and SES factors
  • *  Variables included in adjustments were grouped into the following categories: maternal (e.g., maternal age, education, race, ethnicity, parity), child (age, sex), dietary (e.g., energy intake), and socioeconomic status (urban/rural, family income, residence, employment).

  • Abbreviations: BMI, body mass index; C-DQI, Children's Diet Quality Index; CFI, Child Feeding Index; CHO, carbohydrate; CSFII, Continuing Survey of Food Intakes by Individuals; DD, dietary diversity; DDS, dietary diversity score; DHA, docosahexaenoic acid; DQS, diet quality score; DS, diet score; E, energy; EAR, Estimated Average Requirement; EPA, eicosapentaenoic acid; FFQ, food frequency questionnaire; FVS, food variety score; HAZ, height for age z score; HEI, Healthy Eating Index; HEI-C, Healthy Eating Index Canada; HuSKY, Healthy Nutrition Score for Kids & Youth; MAR, mean adequacy ratio: NAS, nutrient adequacy score; NQI, nutrient quality index; RC-DQI, Revised Children's Diet Quality Index; SD, standard deviation; SE, standard error; Semi-Q FFQ, semi-quantitative food frequency questionnaire; SES, socioeconomic status; VIT, variety index for toddlers; WAZ, weight-for-age z score; WHZ, weight-for-height z score; WIC, Women, Infants and Children; WFR, weighed food record.

Eighteen unique indices were identified that assessed varying aspects of compliance with current nutrition knowledge or recommendations regarding the quality, variety, and combination of foods that may support health (Table 2). The Healthy Eating Index (HEI) was the most commonly used index, with the three versions of the HEI applied in eight studies.2340 The Children's Diet Quality Index (C-DQI), Revised Children's Diet Quality Index (RC-DQI),2834 and indices of dietary diversity were also applied in multiple studies.3244 The remaining indices were only applied in a single study. Although there is considerable interest in diet quality very early in life,3 only seven studies ranging in sample size from n = 82 to n = 2,506 have applied an index to the diets of children less than 2 years of age.3044 Eight studies specifically considered the nutritional needs of children, with other studies extrapolating adult-based guidelines to young children, probably because no quantitative guidelines exist for children less than 2 years of age.2543 Studies cited the lack of applicable dietary guidelines as the reason for excluding very young children.

View this table:
Table 2

The content and structure of the indices used to characterize the diet of 0–5-year-old children.

Index, countryAspect characterizedComponentsScoring systemWeighting or adjustment for energy intakeIndex validation
Child Feeding Index (CFI), internationalQuality of child feeding behaviors7 components: breastfeeding, does not use bottle, dietary diversity, food frequency (egg/fish/poultry, meat, grains/tubers), meal frequencyCategorical scoring (yes/no). 1 point awarded for presence of child-feeding variables, with bonus points for important behaviors (breastfeeding, high dietary diversity, more than 2 meals per day). Maximum 12 points for each age group (6–9 months, 9–12 months, 12–36 months)Weighted to score higher for important behaviors. Nil energy adjustment
Dietary diversity score (DDS), internationalNutrient adequacy8 or 9 food-group components: cereals, roots, & tubers; vitamin-A rich fruit & vegetables; other fruits and vegetables; legumes & nuts; meat & alternatives; fats & oils; dairy; eggs (fruit and vegetables kept separate for 9-food-group DDS)Continuous scoring (no maximum). One point for each food consumed over the last 24 h from the 9 componentsNilConcurrent validity against nutrient adequacy ratio
Diet Quality Index for Children (C-DQI), USAOverall diet quality8-component index: % of total energy as added sugars; total fat; saturated fat; number of servings of grains; fruit & vegetables; dairy; excessive juice; iron (mg/d)Proportional scoring (maximum score 70). 0 to 5 or 10 points allocated per component for intake meeting recommendations (details not provided)Authors designed components to have equal weighting. Nil energy adjustmentNil
Diet Quality Score 1 (DQS1), CanadaDiet quality based on Canadian Food Guide6 components: milk; meat; fruit and vegetables; bread and cereals; additional vegetables; vitamin-A-rich vegetablesProportional scoring (maximum 18 points). Up to 4 points each for recommended 2 servings of milk and meat, 4 points for 4 servings of fruit and vegetables, 4 points for 3 servings of bread & cereals. Additional 1 point for 2 servings of vegetables and 1 serving of vitamin-A-rich vegetablesAuthors designed scoring for equal weighting of each food group. Nil energy adjustmentNil
Dietary Quality Score 2 (DQS2), USANutritional status using dietary quality6 components: vegetables; fruit; bread & cereals; meat & milk; citrus fruit; dark green & yellow vegetablesCategorical scoring (total 6 points). 1 point = <1 serving from each of the food groups. 2 points = 1 serving from each group. 3 points = limit of 2 servings from any group. 4 points = 2 servings from meat & milk group, 4 servings each from bread group and vegetable/fruit groups but nil vitamin C/A source. 5 points = level 4, with one fruit being citrus. 6 points = level 5, with one vegetable being dark green/yellowAuthors designed scoring for equal weighting of components. Nil energy adjustmentNil
Diet score (DS), UKComparison of diet against UK food group portion recommendations5 components: bread, cereals, or potatoes; fruit or vegetables; dairy products; meat, fish, or alternatives; high-fat or high-sugar snacksCategorical scoring (poor diet yes/no). Classified poor if did not have ≥2 portions bread, cereals, or potatoes and ≥2 fruit or vegetables and ≥1 dairy products and ≥1 meat, fish, or alternatives and <2 high-fat or high-sugar snacksNilNil
Diversity score, USANutritional status using food diversity1 component: dietary diversity using items consumed by 20% or more of the study sampleContinuous scoring (maximum score = 20). One point for every food item consumed from a list of 20 food itemsNilNil
Food variety score (FVS), South AfricaNutrient adequacy1 component: diet diversityContinuous scoring (no maximum). One point for every food consumed from a 45-item listNilConcurrent validity against nutrient adequacy ratio
Healthy Eating Index (HEI), USAOverall diet quality10 components: grains; vegetables; fruits; milk; meat; total fat (% total calories); saturated fat (% total calories); total cholesterol; sodium; varietyProportional scoring (maximum score 100). 0–10 points allocated per component. Food component scoring based on number of servings in US Food Guide Pyramid. Fat, cholesterol, and sodium based on recommendation cut points. Variety based on number of different foods: less than 3 different foods = 0, 3–8 different foods scored proportionally, >8 different foods given 10 pointsNilNil
Healthy Eating Index-Canada (HEI-C), CanadaOverall diet quality9 components: grains; fruits & vegetables; milk; meat; other foods (high in fat, sodium, and sugars); total fat; saturated fat; cholesterol; varietyProportional scoring (maximum 100). 0–10 points per component using HEI cutoff points modified for Canadian recommendations. Serving size = half adult portion. Food component scoring based on servings. Fat and cholesterol based on recommended cutoff points. Variety score based on servings from food groups (rather than discrete foods) and servings from "other foods" substituted for sodium in original HEINil. All components adjusted for three levels of energy intakeNil
HEI-2005, USAOverall diet quality (including moderation criteria)12 components: whole fruit (not juice); total vegetables; dark green & orange vegetables and legumes; total grains; whole grains; milk & milk products; meat & alternatives and beans; food oils; saturated fat; sodium; extra calories from solid fats (including fat in milk); added sugarsProportional scoring (maximum score 100). 0 to 5 or 10 points allocated per component. Food component scoring based on number of servings in US Food Guide Pyramid. Fat, cholesterol, and sodium based on recommendation cutoff points. Variety based on number of different foods (0.5 serving minimum). Less than 3 different foods = 0, 3–8 different foods scored proportionally, >8 different foods given 10 pointsNil weighting. All components adjusted for energyNil
Healthy Nutrition Score for Kids and Youth (HuSKY), GermanyNutritional adequacy11 components: beverages; vegetables; fruit; fish; bread & cereals; other starchy foods; dairy products; eggs; meat & sausage; fats; sweets & fatty snacks & soft drinksScoring standardized (0–100). Food group intake converted to ratio ([intake/age- & sex-specific recommendation] × 100). Beverage, fruit, vegetables, fish above recommendations capped (100%). Component scores proportionally reduced for intake of bread & cereals or other starchy foods that exceeded recommendations. Double the recommended intake of meat; eggs; dairy; fats; sweets/fatty snacks/sugar-rich soft drinks awarded score 0NilCorrelation with biomarkers
Mean adequacy ratio (MAR), USANutritional adequacyNutrients included in ratio score varies according to research interestsMean of nutrient adequacy ratio (NAR) for all included nutrients (NAR = nutrient intake/Recommended Dietary Allowance, capped at 100% to reduce influence of high nutrient intakes to overall score). Maximum score = 100, scores >85 considered adequateNilCorrelation with food intake
Nutrient adequacy score (NAS), USANutrient adequacy based on US 1989 Food Guide (excluding fats, sweets, and alcohol)12 components: milk & milk products; whole grains; enriched grains; total grains; citrus fruit; other fruit; total fruit; green & yellow vegetables; starchy vegetables; other vegetables; total vegetables; meat & alternativesContinuous scoring. One point for intake of a food within the 12 components. An upper limit of points set for 8 of the components. No limit for intake of whole grains, both fruit components, green & yellow vegetables, starchy vegetables. Points summedNilConcurrent validity (mean adequacy ratio). Sensitivity and specificity
Nutrient Quality Index (NQI), GermanyDiet quality17 nutrients: vitamins A, E, K, B6, B12, C, thiamine, riboflavin, niacin, pantothenic acid, folate; minerals calcium, magnesium, iron, phosphorus, potassium, zincIntake Quality Scores (IQS) calculated using German Nutrition Society age- and sex-specific dietary reference values. IQS = mean intake/reference intake × 100 (truncated at 100%). Index equal to the harmonic mean of the IQS for each of the 17 nutrientsNil. Ratio truncated so higher energy intake not advantagedNil
Revised Children's Diet Quality Index (RC-DQI), USAOverall diet quality13 components: added sugar; total fat; fat quality (linoleic, eicosapentaenoic, and docosahexaenoic acids); total grains; whole grains; vegetables; fruits; 100% fruit juice; dairy; iron intake; energy balance (television and energy intake interaction)Proportional scoring (maximum score 95). 0 to 5 or 10 points allocated per component for intake meeting cutoff points based on 2001–02 Dietary Reference Intakes, US My Pyramid Food Guide, and American Pediatric Society toddler guidelinesAuthors designed scoring for equal weighting of components. Nil adjustment for energyAbility of quintiles of RC-DQI scores to discriminate different levels of food and nutrient intake
Servings/day, USAAssessment of diet against US My Pyramid5 components: fruit, vegetables, grains, milk/dairy, meat/alternativesCategorical scoring (yes/no). Number (%) meeting recommended servings/day of each food group according to US My Pyramid Food GuideNilNil
Variety index for toddlers (VIT), USADietary adequacy based on US Food Guide Pyramid (minus fats, sweets, and oils)5 components: bread group; vegetable group; fruit group; dairy group; meat groupMean of intake of toddler-sized servings for each food group (maximum score 1). Calculated as ratio against recommended (servings) for bread (6), vegetable (3), fruit (2), dairy (4), meat (2). Component scores: 0 (did not have any servings) and capped at 1 (met all recommended servings)NilNil

The indices varied in regard to which aspects of diet were assessed, how diet quality was defined, and the intended purpose of the index (Table 2). Most indices assessed diet quality based on intake of particular foods or food groups, with two using nutrient intakes25,31 and others involving both food and nutrient intake.2340 Food-based indices used a five-food-group concept (i.e., dairy foods, grains, fruit, vegetables, meat and alternatives) to assign foods to groups. Cutoff values for intake of foods or nutrients were guided by country-specific food selection guides (e.g., US “My Pyramid”) that were relevant to the year in which the study was conducted. Reflecting the evolution of dietary recommendations, more recent indices included food-group subcategories that are considered to be associated with good diet quality, such as dark green/deep yellow vegetables and whole-grain foods. Furthermore, five indices included items associated with poor diet quality, such as intake of excess energy, high fat, or sugary foods.2444 Portion sizes from food selection guides were used as cutoff values to quantify intake against index components. Where these were not available for very young children, a variety of arbitrary portion sizes were applied, ranging from one-half to two-thirds of the recommended adult portion. The portion size has implications for scores that younger children can achieve on an index and the ability to compare scores across studies.

The descriptive findings of studies applying an index show some consistency (Table 1). Mean scores for indices based on dietary guidelines (e.g., HEI, Health Nutrition Score for Kids & Youth [HuSKY], C-DQI, RC-DQI) show modest adherence to recommendations, indicating that diet quality in young children could be improved. In general, lower scores were from index components measuring vegetables, meat, whole grains, saturated fat, and sodium. In comparison, children achieved higher scores on fruit, total grains, and dairy foods components. Scores on indices assessing compliance with nutrient recommendations were higher than food-based scores. For example, the mean score on the Nutrient Quality Index among 2−4-year-old boys was 83.2 ± 10 out of 100,25 whereas the mean score on the predominantly food-based RC-DQI was 59 out of 90 for similar-aged children.28 Dietary diversity and variety were also relatively high; as reported by Knol et al.,35 mean HEI variety scores were 8 out of 10 for a nationally representative group of US children aged 2–3 years. In terms of diet quality in very young children, Child Feeding Index scores ranged between 5.9 and 8.9 out of 12 and were reasonably stable in children between 1 and 3 year of age and across Latin American countries.36

The most commonly used index, the HEI, was developed to assess compliance with US dietary guidelines. HEI scores tend to be fairly consistent in the US population (Table 1), although the revised version, the HEI–2005, resulted in the lowest mean score of 59.6 out of 100 in a nationally representative sample of 2−5-year-old children24; this is likely due to changes in index structure and scoring (Table 2). The HEI has also been used in a sample of Greek preschool children, who had a mean score of 58.7 ± 8.2,23 and it has been adapted to reflect dietary guidelines in Canada, where a small sample (n = 153) of 1−3-year-old children from low-income families had a mean score of 69.9 ± 12.1.30 It is difficult to determine whether the differences in HEI scores from different countries reflect true differences in diet quality, background characteristics of the samples, or the population-specific nature of dietary guidelines.

Scores on different indices within the same sample may reflect differences in index components, method of analysis, or measurement of different aspects of diet quality. For example, application of the RC-DQI compared with the HEI tended to result in lower estimates of diet quality (Table 1).2840 A key difference between these two indices is that the C-DQI index was developed to capture the nutritional needs of toddlers, whereas the HEI was developed using adult dietary guidelines. Finally, the study by Kranz et al.34 provides useful insight into changes in diet quality of preschoolers over the period 1977−1998. Overall diet quality scores on the C-DQI for 2−3-year-old children increased from 45.7 ± 0.2 in 1977–1979 to 47.7 ± 0.2 in 1994–1996 plus 1998. Over this period, component scores worsened for excess juice, added sugar, dairy food, and iron components, while scores for fat, saturated fat, fruits, and vegetables improved.

Dietary indices: relationship with sociodemographic characteristics

The relationship between diet quality and sociodemographic characteristics such as ethnicity, poverty, food security, and family circumstance (i.e., mother's education or work status) were examined in eight studies.2344 Nationally representative data from the United States showed that HEI variety scores did not differ according to maternal perceived access to sufficient food. However, HEI variety scores were higher in a subgroup of disadvantaged families participating in government-supported nutrition programs.35 In Canada, Glanville and McIntyre30 found no relation between socioeconomic circumstance and diet quality using a version of the HEI adapted to the Canadian population, although the small sample (n = 82) of low-income single mothers was relatively homogenous in socioeconomic characteristics, which may have attenuated the potential to detect such differences. By comparison, Caliendo et al.44 found that higher maternal education was positively associated with the Diet Quality Score, but maternal employment detrimentally affected diet quality in childhood. Likewise, Campbell and Sanjur42 found that children's diet diversity improved with licensed childcare, higher family income, less job-family strain, and better child-feeding practices, such as not using food for non-nutritional purposes (e.g., using food as a reward) and greater child control of eating. These findings, which suggest a relationship between sociodemographic characteristics and diet quality, are supported by results in larger studies involving nationally representative cohorts from Germany26 and Greece.23

Dietary indices: relationship with nutrition and health outcomes

Whether an index is associated with children's nutrient intake or nutritional biomarkers indicates whether the index score is a meaningful measure of diet quality or can differentiate variation in underlying nutrient exposure. Five studies assessed whether index scores were associated with differences in food or nutrient intake,2343 with a further three studies assessing the relationship between an index score and a measure of nutrient adequacy.3243 In all of these studies, higher index scores were positively associated with higher nutrient and food intake. For example, comparisons between the lowest and highest tertiles of C-DQI scores were associated with higher intakes of whole grains, fruit and vegetables, dairy foods, and iron.34 Similar associations between HEI scores and nutrient intake have been observed as part of the development of the HEI; however, data for US children were not reported separately.63 Manios et al. showed positive associations between HEI scores and nutrient intake in Greek children.23 Energy and nutrient intakes were weakly (r = 0.09 calcium, 0.25 iron) to moderately (0.56 energy, 0.60 fiber) correlated with HEI total score.23 Additionally, the highest C-DQI score quartile was associated with a greater percent of energy from total, but not saturated, fat, showing the RC-DQI distinguishes the contribution that fat makes to energy intake, but not the type of dietary fat associated with serum cholesterol.29 In relation to biomarkers of nutrient status, HuSKY scores were statistically significantly correlated with iron, magnesium, vitamin B12, folate, and homocysteine but not calcium and vitamin D. Although weak, associations with calcium (partial Pearson's correlation coefficient, r = 0.02 P = 0.068), folate (r = 0.07, P < 0.001), and homocysteine (r = 0.07, P < 0.001) remained after adjustment for age, sex, and energy intake.26

The association between indices of whole-of-diet and health outcomes is limited to cross-sectional relationships with anthropometry in six studies (Table 1).2236 For example, tertiles of the Child Feeding Index were positively associated with height-for-age z-scores after adjustment for confounders.36 After the C-DQI was revised to include measures of energy intake, higher scores on the RC-DQI were associated with a lowered risk of obesity.28 This is in contrast to the absence of a relationship between body mass index (BMI) and scores on the HuSKY index, although this is not surprising because the HuSKY index focuses on nutrient adequacy rather than energy balance.26

Studies using dietary pattern analysis

There were 17 publications that reported using data-driven methods to analyze whole dietary patterns, 14 of which used principal component analysis (PCA)4558 and three of which used cluster analysis5961 (Table 3). The 14 publications utilizing PCA were from six cohorts, two of which contributed to 10 publications.4857 These two cohorts are large, prospective birth cohorts based in the United Kingdom (the Avon Longitudinal Study of Parents and Children [ALSPAC] and the Southampton Women's Study). Remaining cohorts were from European countries (n = 3),4558 the United States (n = 2),60,61 and South Korea (n = 1).46 Although samples ranged in size from n = 109 to n = 27,763, 11 of the 17 publications involved samples exceeding n = 1,000.4560 Of the larger cohort studies, only the Norwegian Mother and Child Cohort Study (MoBa) was sampled from the entire population.45 Irrespective of the sampling frame, the large cohort studies achieved a study sample considered broadly representative of the wider population.4561 The studies included in this review involved children ranging in age from 1 to 5 years.46,48

View this table:
Table 3

Studies assessing dietary patterns among 0–5-year-old children using principal component analysis and cluster analysis.

First author (year) reference; study design, country/setting of study, cohort name (year diet assessed [if available])Sample size (participation rate), age diet assessedDietary intake measurementDietary data processing prior to pattern analysisDescription of PCA patterns or clustersAssociations with diet or nutrient intakeAssociations with child, family, or sociodemographic factorsAssociations with health or developmentAdjustment of analysis*
Studies involving principal component analysis
Ystrom et al. (2009)45; survey, Norway, Norwegian Mother and Child Cohort Study (MoBa) (1999–2008)n = 27,763/28,242 (98%), 1½ yearsFFQ36 items2 patterns: 1. Unhealthy, chocolate, sweets, soda, desserts, ice cream, juice, fruit drinks, cakes, cookies, waffles, bread and jam or honey, pancakes; 2. Wholesome: raw & boiled vegetables, fish, fruit, plain yogurt, rice, peas, beans, bread with fish, cheese or meat products, soured milk, pasta, and meatMaternal negative affectivity associated with Unhealthy but not Wholesome pattern. Wholesome pattern negatively associated with maternal BMI, smoking, several children, use of childcare, or sex (boy) & positively associated with maternal age & education. Unhealthy pattern associated with lower maternal age, education, & income and positively with smoking, BMI & several childrenSES, child, and maternal factors
Shin et al. (2007)46; survey, South Korea, Practical Approach for Better Maternal and Child Nutrition & Health Study (2001–2005 )n = 1,441/1,724 (84%), mean age 5.2 yearsSingle semi-quantitative FFQ100-item FFQ sorted into 33 food groups based on nutrient profile3 patterns: 1. Korean Healthy: vegetable, seaweed, beans, fruits, milk, dairy; 2. Animal Foods: beef, pork, poultry, fish, fast foods; 3. Sweets: ice cream, soda, chocolate, cookiesHigher scores on all patterns associated with higher total energy intake. Korean Healthy pattern scores associated with higher energy adjusted intakes of protein, fiber, iron, calcium, vitamin A, carotene, and vitamin C. Sweets pattern scores inversely associated with nutrient intakes. Mixed associations found with Animal Foods pattern scoresIncreasing quartiles of Animal Foods pattern scores were associated with boys. Greater household income and expenditure on food was associated with higher scores on all dietary patternsAnimal Foods pattern associated with overweight (OR 1.77 95% CI [1.06–2.94])SES, child, and dietary factors
Sepp et al. (2002)47; survey, Sweden (1998)n = 109/131 (83%), 3–6 years7-day diet diary35 food items sorted into 16 food groups4 patterns: 1. Milk & cheese; 2. Breads & breakfast cereals; 3. Meat, potato, cooked cereals; 4. Confectionery, buns, soft drinksPattern 1 negatively correlated with patterns 2 & 3. Pattern 3 correlated with pattern 4. PCA scores negatively correlated across weekdays and weekends
Robinson et al. (2007)48; prospective cohort study, UK, Southampton Women's Survey (1998–2002)n = 1,434/1,981 (72%), 12 monthsFFQ56 items in PCA (78 FFQ reduced to 46 food groups, plus 10 milk & open text groups)5 patterns at 12 months but only 2 meaningful: 1. Infant Guidelines–12 months: home-prepared foods, cooked & salad vegetables, beans, meat, fish, egg, cheese, fresh fruit, 2. Adult Foods–12 months: cow's milk, white bread, french fries, potato chips, processed meat, tinned vegetables, biscuits, sweetsPattern 1 scores associated with higher maternal education, prudent diet, & lower birth order. Pattern 2 scores negatively associated with maternal age, education, & prudent diet, and positively associated with maternal smoking, higher birth order, and earlier introduction of solidsSES, child, maternal, and dietary factors
Baird et al. (2008)49; prospective cohort study, UK, Southampton Women's Survey (1998–2002)n = 1,342/1,841 (73%), 12 monthsFFQAs per Robinson et al. (2007)48 (above)Infant Guidelines and Adult Foods patterns at 12 months reported by Robinson et al. (2007)48 (above)No effect of either pattern score at 12 months on weight, length, or skinfoldsSES, child, maternal, and dietary factors
Harvey et al. (2009)50; prospective cohort study, UK, Southampton Women's Survey (1998–2002)n = 599/1,841 (33%), 12 monthsFFQ12 months data from Robinson et al. (2007)48 (above)Infant Guidelines pattern reported by Robinson et al. (2007)48 (above)No association between Infant Guidelines pattern score and bone mass at 4 yearsSES, child, maternal, and dietary factors
Robinson et al. (2009)51; prospective cohort study, UK, Southampton Women's Survey (1998–2002)n = 536/782 (69%), 12 monthsFFQ12 months data from Robinson et al. (2007)48 (above)Infant Guidelines pattern reported by Robinson et al. (2007)48 (above)Higher Infant Guidelines pattern score at 12 months associated with increased (better) lean mass, but not fat mass or BMI at 4 yearsSES, child, maternal, and dietary factors
Gale et al. (2009)52; prospective cohort study, UK, Southampton Women's Survey (1998–2002)n = 241/1,981 (12%), 12 monthsFFQ12 months data from Robinson et al. (2007)48 (above)Infant Guidelines and Adult Foods patterns reported by Robinson et al. (2007)48 (above)Higher Infant Guidelines pattern score at 12 months associated with full-scale IQ at 4 years. Adult Foods pattern scores not associated with IQSES, child, and maternal factors
North & Emmett (2000)53; prospective cohort study, UK, Avon Longitudinal Study of Parents and Children (ALSPAC) (1991–1992)n = 10,139/14,541 (70%), 38 monthsFFQ43-item FFQ4 patterns: 1. Junk: soft drinks, chocolate, sweets, french fries, burgers, pies, potato chips, white bread, pizza, flavored milk drinks; 2. Healthy: pulses, vegetarian foods, rice, pasta, salad, fruit juice, water, eggs, cheese, & fish; 3. Traditional British: meat, poultry, green & root vegetables, peas, sweet corn; 4. Snacks: puddings, biscuits, cakes/buns, squash, fruitJunk pattern associated with younger, less educated mothers & older siblings. Healthy pattern associated with ethnicity, maternal education, & vegetarianism. Traditional British pattern associated with girls & no siblings. Snacks pattern associated with higher maternal education & older siblings, white ethnicity, and fewer financial difficultiesSES, child, maternal, and dietary factors
Northstone et al. (2005)54; prospective cohort study, UK, Avon Longitudinal Study of Parents and Children (ALSPAC) (1991–1992)n = 9,550/14,541 (66%). 54 months, n = 8,626/14,541 (59%), 81 monthsFFQ90-item FFQ reduced to 57 groups3 patterns (similar at both ages): 1. Junk: sausages, burgers, potato chips, sweets, chocolate, ice lollies & ice creams; 2. Traditional: meat, vegetables; 3. Health conscious: salad, rice, pasta, fruit, cheese, fish, meat substitutes, pulses, nutsJunk pattern associated with fussy eating, number of siblings, white ethnicity, and negatively associated with maternal education, age, & vegetarianism. Traditional pattern negatively associated with boys, vegetarianism, fussy eating, maternal education, single mother. Health conscious pattern negatively associated with fussy eating, younger mothers, lower education, maternal smoking, and positively associated with nonwhite ethnicitySES, child, maternal, and dietary factors
Wiles et al. (2009)55; prospective cohort study, UK, Avon Longitudinal Study of Parents and Children (ALSPAC) (1991–1992)n ∼ 4,000/12,783 (31%), 54 monthsFFQAs per Northstone et al. (2005)54 (above)Junk pattern identified at 54 months by Northstone et al. (2005)54 (above)Junk pattern associated with hyperactivity subscale scores but not other subscales or total score on the Strengths & Difficulties Questionnaire at 7 years. Association attenuated after adjustment but remained significantSES, child, and maternal factors
Northstone & Emmett (2008)56; prospective cohort study, UK, Avon Longitudinal Study of Parents and Children (ALSPAC) (1991–1992)n = 6,177/14,541 (43%) with data at 38, 54, 81, 103 monthsFFQAs per North & Emmett (2000)53 and Northstone et al. (2005)543 patterns described in North & Emmett (2000)53 & Northstone et al. (2005)54 (above). Junk pattern now referred to as Processed.Mean scores for Processed pattern increased over time, but other patterns remained similar. Across ages, factor score correlations ranged 0.46–0.65 for Processed, 0.35–0.61 for Traditional, 0.41–0.69 for Health conscious (all P < 0.0001)SES, child, and maternal factors
Feinstein et al. (2008)57; prospective cohort study, UK, Avon Longitudinal Study of Parents and Children (ALSPAC) (1991–1992)n = 5,741/14,541 (39%) at 38, 54, 81 monthsFFQAs per North & Emmett (2000)53 and Northstone et al. (2005)543 patterns described in North & Emmett (2000)53 & Northstone et al. (2005)54 (above)Junk pattern scores at 38 months associated with poorer school attainment at 6–7 years and 10–11 years, even after adjustment. Health conscious pattern at 38 months associated with attainment at 10–11 years. No relationship between dietary pattern scores at other ages with attainmentSES, child, maternal, and dietary factors
Friedman et al. (2009)58; prospective cohort study, Ukraine, European Longitudinal Study of Parents & Children (ELSPAC) (1993–1996)n = 883/4,510 (20%), 3 yearsFFQ104-item FFQ, 22 food items included in PCA analysis6 patterns: Food loadings not reported. Patterns named: 1. Meat; 2. Staples; 3. Noodles & pasta; 4. Fruit & vegetables; 5. Breakfast foods; 6. SnacksEnvironmental and social factors associated with BMI more strongly than dietMeat, but not other patterns, associated with increased odds of BMI >85th percentile (OR 1.37, 95% CI 1.04−1.81)SES, child, and maternal factors
Studies involving cluster analysis
Pryer and Rogers (2009)59; survey, UK, Survey of British Children (1992–1993)n = 1,675/1,859 (90%), 1½–4½ yWFR19 food and beverage groups3 clusters: Traditional: cakes, puddings, meat dishes, confectionery, soft drinks (n = 115); Healthy: whole-grain cereals, low-fat diary, egg dishes, vegetables, fruit, nuts, fruit juices (n = 931); Convenience: refined cereals, cakes, puddings, fat spreads, bacon/ham, candy, french fries/potatoes, and soft drinks (n = 629)Traditional cluster had higher energy intake from carbohydrate and fat and lowest nutrient density for most vitamins and minerals, compared with Healthy cluster. Nutrient density of Convenience diet cluster was intermediateFor boys, clusters differed on the proportion of parents in manual/nonmanual employment and type of housing. For boys and girls, clusters differed on whether parents received benefits and on wellness of the child at assessmentSES, child, maternal, and dietary factors
Knol et al. (2005)60; survey, USA, Continuing Survey Of Food Intakes in Individuals (CSFII), low-income families (1994–1998)n = 1,242, 2–3 yearsTwo 24-h recalls24 food variables based on US Food Guide Pyramid6 clusters: Big eaters (n = 625); Light eaters (n = 439); Bean eaters (n = 95); Substituters (n = 68); Low-cost eaters (n = 9); Semi-vegetarians (n = 6)No healthy eating cluster identified. HEI scores differed across the 4 key clusters. Two most populous clusters (n = 1,064/1,242 [86%]) adhered poorly to US dietary guidelines, and energy intakes exceeded estimated requirements for active childrenSES, child, and maternal factors
LaRowe et al. (2007)61; survey, USA, National Health & Nutrition Examination Survey (NHANES) (2001–2002)n = 541, 2–5 years24-h recall8 beverage groups according to intake (g): high-fat milk; reduced-fat milk; fruit juice; soda; diet soda; sweet beverages; coffee & tea; water4 clusters: Mix/light drinker (n = 249); High-fat milk drinker (n = 91); Water drinker (n = 128); Fruit juice drinker (n = 73)HEI scores for fruit juice higher than other groups. Total energy from protein and intakes of minerals (iron, zinc, calcium) and vitamins (riboflavin, folate, A and C) differed by clusterPercentage with BMI>85th percentile differed according to cluster (Mix/light 15%, Fruit juice 20%, Water 26%, High-fat milk 27%), but not after adjustmentSES, child, and maternal factors
  • *  Variables included in adjustments were grouped into the following categories; maternal (e.g., maternal age, education, race, ethnicity, parity), child (age, sex), dietary (e.g., energy intake), and SES (urban/rural, family income, residence, employment).

  • Abbreviations: BMI, body mass index; CI, confidence interval; FFQ, food frequency questionnaire; HEI, Healthy Eating Index; OR, odds ratio; PCA, principal component analysis; SES, socioeconomic status; WFR, weighed food record.

Two of the prospective cohort studies included in the review had collected dietary data at multiple ages during early childhood.4864 Thirteen studies used food frequency questionnaires (FFQs) for the collection of dietary data, with fewer using 24-h recall, weighed food records, or diet diaries. FFQs are considered appropriate for population-based evaluations of dietary patterns in childhood48 and adulthood65,66 and are favored in large-scale studies because they are less burdensome to participants and reduce post-collection processing of dietary data. However, both the size and the items included in the FFQs varied widely between studies. Furthermore, there was no consistency in the amount of processing of FFQ data prior to PCA analysis. In some cases, dietary pattern analysis was conducted on raw data from a 36-item FFQ,45 while in others, a 100-item FFQ was condensed into 32 food groups prior to analysis.46

For most PCA studies, important detail on the statistical methodology was often incomplete or missing. For example, the use of rotation in PCA analysis is thought to improve the interpretation of dietary patterns, but whether the dietary data were rotated, the type of rotation, or the effect of rotation on interpretability was not always described. It is also unclear how the optimal number of components was determined because, although most studies report consulting Scree plots or Eigenvalues, the specific criteria applied to determine the PCA model (termed “solution”) was generally not reported. Despite some uncertainties in statistical methodology, the included studies reported that the overall PCA solution explained between 13% and 58% of the total variance in dietary intake,48,58 although this information was not reported in one study.45 Cluster analysis was performed on between 8 and 24 food or beverage items using Wards or k-means methods for estimating distances between diet items and cluster membership. Additional multivariate methods were applied in two cluster studies to confirm the number and stability of clusters by comparing with PCA60 or by testing cluster membership by using discriminant analysis.59 Interestingly, all studies applying a posteriori analyses pretreated dietary intake data by grouping food variables, largely without justification or examination of the consequences of such data manipulation.67

Between two and six dietary patterns were identified by PCA and cluster analysis, with the majority of studies deriving a three-component solution. Some common themes emerged in the dietary patterns characterized in each study (Table 3). Among the cohorts that evaluated the diets of children aged 3–5 years, most identified a type of “healthy” and “unhealthy” pattern.4559 The names applied to pattern scores were determined by researchers and based on the foods that loaded most heavily on the pattern. For example, the “junk” food pattern identified by Northstone and Emmett56 at 3 and 4 years of age and the “unhealthy” diet pattern identified by Ystrom et al.45 at 18 months of age were both characterized by chocolate, sweets, and soft drinks, whereas foods associated with the “healthy” pattern or the “wholesome” pattern included rice, pasta, fish, fruit, and cheese. Other dietary patterns characterized by traditional foods in the UK cohort and animal foods in the Korean cohort were also identified (Table 3).4656 Although the naming of food patterns is arbitrary, there appears to be some consistency across studies in the identification of healthful and unhealthful dietary patterns. Ystrom et al.,45 North and Emmett,53 Shin et al.,46 and Robinson et al.48 reported the foods that were entered into the PCA and the loadings of each food on each component, which allows for a careful comparison of foods and patterns across studies.Complete presentation of data in this way revealed that fish is associated with healthy dietary patterns in studies from the United Kingdom and Europe, but in South Korea, fish loaded strongly on two patterns, the “Korean healthy” and “animal foods” components.

Dietary patterns over time

Data from the ALSPAC study demonstrated that three of the four patterns identified at 3 years of age are correlated with similar dietary patterns at 4−9 years of age.56

Dietary patterns: relationships with sociodemographic characteristics

The relationship between dietary patterns and various sociodemographic characteristics was reported in nine articles.4561 Higher maternal age and educational status were positively associated with healthier dietary patterns during the toddler and preschool years.4553 A finding common to three of the included studies was that the number of siblings a child has appears to influence dietary patterns, as higher scores on unhealthy patterns and lower scores on healthy patterns are associated with a greater number of siblings.4553 The association between higher household income and healthier dietary patterns was mixed. Although North and Emmett53 showed financial difficulties were positively related to “junk” food pattern scores, other studies did not report similar relationships.59,61 In fact, Shin et al.46 found that expenditure on food was associated with higher scores on all dietary patterns, suggesting that having a higher level of disposable income for food does not always discriminate between healthy and unhealthy dietary patterns.

Dietary patterns: relationship with nutrition, health, and developmental outcomes

A comparison of dietary pattern scores against nutrient intakes or biomarkers provides an important verification that pattern analysis can detect meaningful differences in nutrient intake. Only Shin et al.46 reported relationships between PCA and nutrient intakes and found that total energy intake was associated with higher scores on all patterns. Furthermore, energy-adjusted intake of protein, fiber, iron, calcium, vitamin A, carotene, and vitamin C was associated with higher scores on the “Korean healthy” dietary pattern. In contrast, scores on the “sweets” pattern were inversely associated with nutrient intakes. Pryer and Rogers59 reported similar nutrient relationships across the “healthy” and “convenience” clusters, with the “traditional” cluster having intermediate nutrient intakes. None of the studies compared dietary pattern scores against biomarkers.

Of the included studies, the outcomes associated with dietary patterns are broadly grouped into studies evaluating body composition or weight (n = 6)4661 and studies on psychometric outcomes (n = 3).5257 At 1 year of age, closer adherence to a dietary guidelines pattern of food intake was not associated with current weight, length, or skinfold thickness.49 Follow-up after 4 years revealed that higher scores on the dietary guidelines pattern was associated with better lean mass, but no patterns were associated with fat or bone mass.50,51 The odds of being overweight at preschool age were associated with higher scores on meat-dominated dietary patterns in both European and Asian settings.46,58

In a small subset of children from the Southampton Women's Study cohort (n = 248), scores on the “infant guidelines” pattern at 1 year were positively associated with scores on full-scale intelligence quotient measured using the Wechsler Preschool and Primary Scale of Intelligence at 4 years of age.52 Adding to this evidence are findings from the ALSPAC study suggesting that diet at 3−4 years of age is a stronger predictor of school success than diet at 7 years of age57 and that higher scores on the “junk” food pattern at preschool age are associated with increased risk of hyperactivity measured using the Strengths and Difficulties Questionnaire at age 7 years.55 For all longitudinal studies, the sample followed up was only a subset of the main cohort, ranging between n = 248 (22%)52 to approximately 5,700 (40%).57

DISCUSSION

The primary aim of this systematic review was to determine what is currently known about the measures of whole diet in early life and their effect on subsequent child health and development. Forty articles were identified in which the diets of children aged between 1 and 5 years were characterized; 23 of these articles utilized dietary indices and 17 utilized an a posteriori dietary pattern or cluster analysis. Of the dietary index studies, eight related index scores to estimates of food or nutrient intake2343 and six related index scores to concurrent anthropometry.2236 No studies examined whether index scores predicted later health or development. By comparison, three studies utilizing dietary pattern or cluster analysis examined relationships with nutrient intakes4661 and four studies examined associations with concurrent anthropometry.4661 Five articles reported on whether dietary patterns predicted later health, with all of these longitudinal associations derived from only two UK birth cohorts.

Early life diet and health or development

The predictive validity of dietary patterns revealed that some, but not all, dietary patterns were associated with health outcomes. For example, dietary patterns reflecting adherence to feeding guidelines at 1 year of age were positively associated with lean mass at 4 years of age but not with bone or fat mass, whereas patterns associated with adult foods showed no association.4951 Dietary patterns associated with feeding guidelines are consistent with the knowledge that breastfeeding influences child growth4 but extends our understanding beyond breastfeeding to the weaning diet. Breastfeeding is also associated with better cognitive development in childhood,5 and the studies included in this review support the idea that factors common to weaning and early childhood diet are associated with later cognitive development and child behavior.52,55 Although reliable psychometric assessments were used and analyses were adjusted for many confounders known to influence child development, the sample examined in longitudinal studies tended to differ from the wider cohort on dietary pattern scores at 1 year of age and on other factors known to influence child development (maternal education, social class, and breastfeeding).52 Given that predictive validity of whole-of-diet analysis in early life has only been explored using dietary pattern analysis among UK cohorts, studies involving other measures of whole diets (especially dietary indices), different populations, and a broader range of outcomes are needed to establish the effect of early-life diet on later health.

Studies describing cross-sectional associations between early-life diet and indicators of adiposity were inconsistent. Three of the five index studies showed no association between index score and BMI, weight >95th percentile, or weight-for-height,2332 whereas other studies reported both positive22 and negative associations.28 Likewise, a posteriori studies also showed mixed findings. Dietary patterns were not associated with weight or skinfolds at 12 months.49 In preschool-aged children, dietary patterns characterized by animal foods or meat intake (but not other patterns) were positively associated with concurrent BMI or overweight.46,58 The association between meat-dominated food patterns and adiposity warrants further investigation in longitudinal studies. Overall, the mixed findings between whole diets and measures of adiposity of children are similar to a systematic review of the adult literature showing inconsistent associations between index scores, dietary patterns or clusters, and concurrent measures of BMI or overweight.68 Interestingly, the only index study to report that better diet quality was associated with lower BMI included estimates of energy balance as a subcomponent of the index.28 Thus, indices that incorporate measures of energy intake or physical activity may be necessary to detect associations between diet quality and adiposity.

The scant literature linking whole-of-diet in children under 5 with later health is in contrast to the adult literature, where many studies have related dietary index or pattern scores to a broad range of outcomes that include mortality, cancer, obesity, and cardiovascular disease or risk factors (e.g., cholesterol, blood pressure).1218 For example, Kant12 systematically reviewed the adult literature and identified 25 prospective studies (and many other cross-sectional, case control, and intervention trials) in which dietary indices and patterns were related to morbidity and mortality. While there is increasing awareness of the role of early-life diet on later health and development,13 the lack of prospective studies that measure diet, health, or development outcomes limits the literature in children.

Relationship between whole-of-diet indices or patterns and nutrient intake

It is difficult to summarize the association between whole diets and nutrient intakes because there is no agreed method for quantifying such relationships and little consistency in the manner in which this has been evaluated. Indices and patterns were related to nutrient intakes using methods as disparate as the mean of a ratio of estimated intake to recommended intake for a variable numbers of nutrients (i.e., mean adequacy ratio), correlations with individual nutrients, or examination of nutrient intakes across categories (e.g., quartiles) of index or pattern score. Despite the heterogeneous nature of the included studies, nutrient intakes generally increased in the expected direction. For instance, higher RC-DQI scores were associated with a lower risk of deficiency of numerous vitamins (as indicated by the proportion of the sample meeting the Estimated Average Requirement), and nutrient intakes increased across quartiles of RC-DQI scores (including carbohydrate, protein, fiber, long-chain fatty acids, iron, and calcium).29 However, total energy intake also increased across quartiles,2341 and the controversy around whether data-reduction techniques measure true differences in nutrient density or reflect greater food consumption remains unresolved.1671 Clusters and tertiles of dietary patterns were associated with differences in nutrient intakes, as patterns labeled as healthy were associated with better nutrient intakes compared with processed, convenience, or traditional patterns.4661 Nevertheless, unexpected relationships were found, as iron intake did not differ across quintiles of “animal foods” pattern scores.46 Further work to demonstrate that dietary indices or patterns detect meaningful differences in nutrient intakes is an important complement to our understanding of whole-of-diet analysis.

Measures of whole diet in early life

Overall, the studies included in this review that characterized the diets of children were very heterogeneous in design, sample size, and sociodemographics characteristics of the participants. While the validity of some indices have been explored extensively in the target population,2972 the association between index scores and underlying variation in food or nutrient intake is untested for many indices. Despite these limitations, studies that applied the HEI and RC-DQI included reasonably large samples that were broadly considered to be representative of the wider population. These studies have been useful for describing the diet quality of 2−5-year-old children at a population level, over time, and across different age groups. In general, moderate scores were found for preschool-aged children, indicating there is room to improve diet quality in this age group. Interestingly, the social gradient associated with diet quality emerges from early childhood,2344 and family factors such as larger family size and maternal employment may also influence diets of young children.42,44 Secular trends indicate that some aspects of preschool-aged children's diets have improved over a 20-year period, including the lowering of saturated fat intake.34 Consistent with other work,73 children's intake of added sugars appears to have increased over the same period.34

Unfortunately, the quality of the a posteriori analyses was not assessable for many of the included studies because of inadequate reporting of the methodological details. This included information on use (or method) of rotation, how the number of dietary patterns was determined, or the impact of collapsing foods into groups prior to analysis. It is difficult to judge how these methodological differences might have influenced the findings, but transparent reporting of these decisions is needed to advance the field.18 Nevertheless, the studies utilizing data-driven analyses of whole diets show consistency in identifying particular types of dietary patterns or clustering of children, indicating that these methods may be relatively robust. The spectrum of dietary behaviors present in adulthood appears to emerge from early childhood, as most of the PCA studies identified at least one “healthy” and one “unhealthy” pattern. Healthy dietary patterns were typically characterized by fruit, vegetables, and whole grains, and unhealthy patterns included soft drinks, potato chips, and sweets.4559 Despite these similarities, dietary pattern analyses are likely to be age and culture dependent. For example, the type of dietary pattern reported at 1 year of age48 was quite different from that reported at 3 years of age,53 which was similar to the patterns seen in older children. Furthermore, the considerable differences in the foods which load on dietary patterns from the Korean and Western studies warrant caution when interpreting findings across cultural settings.

This review highlights opportunities for new research around early-life nutrition. In particular, studies examining the dietary patterns of children below 2 years of age are needed. Dietary exposures in the first years of life are difficult to measure because of the rapid transition from a milk-based diet to table foods. Some researchers have acknowledged that the absence of quantitative recommendations for children below 2 years of age is a limitation for designing indices for infants and toddlers.72 Another potential area for future research is to draw upon the dietary data from longitudinal studies and generate dietary trajectories rather than using data in a cross-sectional manner. A better understanding of the development of dietary patterns known for their health benefits, such as the Mediterranean diet, would also make a useful contribution to the field. Finally, the utility of whole-of-diet analysis would be improved by evaluating associations with a broader range of later health outcomes than has previously been studied. This will improve our understanding of which aspects of health and development are most vulnerable to early-life diet and provide evidence needed to improve dietary advice given to parents of young children.

CONCLUSION

A diet that meets minimal nutrient requirements is not necessarily synonymous with a high-quality diet, and a challenge for future research is how to measure the health impact of different diets among children whose nutritional needs are met. The relationship between whole diet and health is likely to be more informative than studies of individual nutrients and later health, because associations with whole diets may be stronger and may capture more information about overall diet quality. It is reassuring that the “healthy” patterns identified in several studies appear consistent with diets that would score high on an index, indicating there may be some crossover in how dietary exposures can be quantified. However, despite the widespread use of various dietary indices, the predictive validity of these indices has not been established, and associations with concurrent adiposity of young children from developed countries are inconsistent. As only two cohorts have examined longitudinal associations between dietary patterns and child development, further confirmatory evidence is needed. Thus, the evidence gathered within this review suggests the utility of indices and dietary pattern analysis has not been fully explored.

Acknowledgments

The authors thank Dr. Murthy Mittinty for helpful comments on an earlier draft of this manuscript.

Funding

Conduct of the systematic review and preparation of the manuscript was not supported by grant funds. Authors JWL and RKG are supported with fellowships from the National Health and Medical Research Association of Australia.

Declaration of interest

The authors have no relevant interests to declare.

REFERENCES

View Abstract