Dietary patterns of 6-24-month-old children are associated with nutrient content and quality of the diet.

Abstract We determined the associations of dietary patterns with energy/nutrient intakes and diet quality. Previously collected single 24‐hr dietary recalls for children aged 6–11 months (n = 1,585), 12–17 months (n = 1,131), and 18–24 months (n = 620) from four independent studies in low socio‐economic populations in South Africa were pooled. A maximum‐likelihood factor model, with the principal‐factor method, was used to derive dietary (food) patterns. Associations between dietary pattern scores and nutrient intakes were determined using Kendall's Rank Correlations, with Bonferroni‐adjusted significance levels. For both 6–11 months and 12–17 months, the formula milk/reverse breast milk pattern was positively associated with energy and protein intake and mean adequacy ratio (MAR). The family foods pattern (6–11 months) and rice and legume pattern (12–17 months) were positively associated with plant protein, fibre, and PU fat; both for total intake and nutrient density of the complementary diet. These two patterns were also associated with the dietary diversity score (DDS; r = 0.2636 and r = 0.2024, respectively). The rice pattern (18–24 months) showed inverse associations for nutrient intakes and nutrient densities, probably because of its inverse association with fortified maize meal. The more westernized pattern (18–24 months) was positively associated with unfavourable nutrients, for example, saturated fat and cholesterol. These results highlight that underlying dietary patterns varied in terms of energy/nutrient composition, nutrient adequacy, nutrient densities of the complementary diet, and dietary diversity.

distinct Western-like dietary pattern and health conscious dietary pattern are already present at this young age (Kiefte- de Jong et al., 2013).
Dietary patterns based on predefined dietary indices or derived from factor or cluster analyses examine the whole diet rather than individual foods and/or nutrients (Hu, 2002). In factor analyses, various arbitrary decisions are taken, including grouping of foods into food groups and the naming of the dietary pattern (Hu, 2002;Newby & Tucker, 2004).
Dietary patterns derived through factor analysis may therefore not necessarily be comparable between studies or even age groups, and associations between dietary patterns and nutrient intakes are complex and may be difficult to interpret. For example,  identified a home-made traditional pattern for young children at age 6-8 and 15 months, but the association of this dietary pattern with the nutrient profile was inconsistent between the two age groups.
Dietary patterns have been shown to be associated with infant growth outcomes such as length-for-age z-scores and BMI z-scores (Wen et al., 2014). Understanding the energy and nutrient content and nutritional quality of specific dietary patterns therefore will provide valuable insight that may guide the development of appropriate nutrition messages/policies in terms of infant and young child feeding, particularly against the background of the triple burden of malnutrition in South Africa (stunting, overweight/obesity, and micronutrient deficiencies).
In vulnerable populations in South Africa, dietary intake in 6-24month-old children can range from predominantly maize-based to predominantly based on commercial infant foods (Faber, 2005;Faber, Laubscher, & Berti, 2016;Swanepoel et al., 2018). Pooling diverse dietary intake data would potentially provide a dataset with sufficient variation to determine the nutrient profile of a variety of dietary patterns.
The aim of this study was to determine whether distinct dietary patterns are associated with energy/nutrient intakes and nutritional quality in 6-24-month-old South African children of low socio-economic status.

| Study design
This study consisted of pooled single 24-hr dietary recalls for 6-24 month-old children previously collected in four independent studies.
All study sites were of low socio-economic status. In Study 1 (Smuts et al., 2005) and Study 2 (Faber, 2005;Faber, Kvalsvig, Lombard, & Benadé, 2005), dietary intake data were collected for children who participated in two independent randomized controlled trials (RCT) that were done in rural sites in KwaZulu-Natal province. Study participants were recruited through an NGO-driven community-based health programme that operated through 12 health posts. Exclusion criteria were birth weight <2500 g and haemoglobin concentration < 80 g/L (both studies), premature birth (<37-week gestation), and weight-for-length z-score < -3 (Study 1 only). Data collection was done at baseline (at age 6-12 months) and follow-up (at age 12-18 months). In Study 1, additional data were collected 6 months after the completion of the RCT (at age 18-24 months). In Study 3 Swanepoel et al., 2018), dietary intake data were collected for children who participated in an RCT that was done in a peri-urban site in North West province. Study participants were recruited through primary health care facilities and house-to-house visits. Exclusion criteria included haemoglobin concentration <70 g/L, weight-for-length z-score < -3, severe congenital abnormalities, infant known to be HIV positive, and infants known to be allergic/intolerant to peanuts, soy, cow's milk protein, or fish. Data were collected at baseline (at age 6 months), follow-up (at age 12 months), and 6-month post RCT (at age 18 months). In all three studies, dietary intake data were missing for children whose caregiver could not provide reliable information because the child was not in her permanent care during the 24-hr recall period. Study 4 (Faber et al., 2016) was a cross-sectional dietary assessment study. Primary caregivers of randomly selected children, stratified per age category (6-11 months, 12-17 months, and 18-24 months), were recruited through house-to-house visits in two study sites, one rural and one urban, in KwaZulu-Natal province.
Previously collected 24-hr dietary recalls were recoded to ensure that coding and analysis were standardized across all dietary surveys and that all records were analysed with the same version of the food composition database. Estimated intake of breast milk was assumed according to age: 675 ml for partially breastfed infants at age 6-11 months, 615 ml at age 12-17 months and 550 ml at age 18-24 months (WHO, 1998). Exclusively breastfed or formula-fed infants were excluded. The complementary diet was defined as all foods and beverages consumed, excluding breast milk and formula milk feeds. Formula milk powder mixed into porridge/infant cereal may affect the nutrient density of the complementary diet and was therefore coded separately from formula milk feeds, using dummy food codes. This allowed for formula milk powder mixed into food to be included when calculating the nutrient density of the complementary diet. Food intake was converted to energy and nutrients using Stata software and the 2017 South African Food Composition Database (SAFOODS, 2017), which includes an updated section on infant foods.

Key Messages
• The association of formula milk/reverse breast milk pattern scores and MAR suggests that breastfeeding children are more likely to consume a diet of lower nutrient adequacy.
• Associations of formula milk/reverse breast milk pattern scores and nutrient densities of the complementary diet suggest that breastfeeding children consume a complementary diet of lower nutrient density.
• The more westernized dietary pattern was associated with unfavourable nutrients such as saturated fat, cholesterol, and sugar, as well as certain micronutrients.
• Although associations of dietary pattern scores with dietary quality indicators could be explained by the foods with high factor loadings in most cases, this was not always the case.
Nutrient adequacy ratios (NAR) were calculated using age appropriate estimated average requirements (EAR) or, where there is no EAR, the Adequate Intakes (AI) of the Dietary Reference Intakes (DRIs) (Otten, Hellwig, & Meyers, 2006 Individual food items were grouped into 36 foods (or groups) based on nutritional composition and similarity of foods. Energy contribution of the foods was calculated and expressed as a percentage of total energy intake. Daily energy intake values (expressed as percentage of total intake) for the 36 foods were used in a maximumlikelihood factor model, with the principal factor method to derive estimates of dietary patterns; a varimax (orthogonal) rotation of the factor-loading matrix was done to make interpretation easier. Derived components with an eigenvalue > 1.00 and also containing two or more original foods with loading factor ≥ 0.35 or ≤ -0.35 were retained in order to name the factors. Regression scoring was used for the set of retained factors. A higher factor score indicates higher adherence to the corresponding dietary pattern. These factor scores (continuous variables) were then used to determine associations of the dietary patterns with energy and nutrient intakes, MAR, nutrient densities of the complementary diet, and the DDS, using Kendall's rank correlations, with Bonferroni-adjusted significance levels.
Data were further explored by stratifying the children according to dietary pattern tertiles (Ts) and then calculating the percentage consumers for the 36 foods within each tertile. Differences across the tertiles were determined using the Fisher exact test.

Ethical considerations
Ethical approval was not required as we used pooled data from previous studies.

| Dietary patterns
In each of the three age categories, three dietary patterns were identified, which explained 38.6% (6-11 months), 37.8% (12-17 months), and 32.7% (18-24 months) of the variance ( Table 1). The percentage of children who consumed foods during the recall period according to dietary pattern tertiles is given inTables 2-4. For significant associations of these patterns with energy and nutrient intakes, MAR, DDS (Table 5), and nutrient densities of the complementary diet (Table 6), a correlation coefficient (r) of between -0.3 and 0.3 is considered weak, and associations with r ≤ -0.3 and r ≥ 0.3 will mostly be highlighted hereafter.

| Age 6-12 months:
Factor 1, named the 'formula milk/reverse breast milk' pattern, had a very high positive loading for formula milk and a very high negative loading for breast milk (Table 1), indicating an inverse association between formula milk and breast milk. In terms of pattern score tertiles (Table   2), 7.6% of children consumed formula milk in T1 versus 86.2% in T3.
The opposite was observed for breast milk, with all children in T1 receiving breast milk, versus 11.9% in T3. The 'formula milk/reverse breast milk' pattern was positively associated with energy, protein and most micronutrients, and ultimately MAR (Table 5), as well as with the nutrient density of the complementary diet for various nutrients (Table   6), although these associations were weak (r > -0.3 and r < 0.3).
Factor 2, named the 'family foods' pattern, had high positive loadings for maize meal, rice, and legumes and a high negative loading for infant cereal. The 'family foods' pattern was inversely associated with all commercial infant products and positively associated with several family foods (Table 2). In terms of nutrients, the 'family food' pattern was positively associated with plant protein, fibre, and PU fat; both for total intake (Table 5) and the nutrient density of the complementary diet (Table 6). This pattern was positively associated (r ≥ 0.3) with magnesium and vitamin B6, both for total intake and nutrient density of the complementary diet, and inversely associated with the nutrient density of the complementary diet for vitamin C and, to a lesser extent, calcium (r = -0.2742) and iron (r = -0.2476).
Factor 3, named the 'maize meal and sugar' pattern, had a high loading for maize meal. The 'maize meal and sugar' pattern was inversely associated with all commercial infant products (Table 2). This dietary pattern showed statistically significant correlations with a few nutrient intakes (Table 5) and nutrient densities for various micronutrients (Table 6), but most of these correlations were weak (r > -0.3 and r < 0.3), except for the nutrient densities for carbohydrates, magnesium, and folate (r ≥ 0.3).

| Age 12-17 months:
Factor 1, named the 'tea and sugar' pattern, had high loadings for sugar and rooibos tea. This pattern was not associated with energy and nutrient intakes, or MAR (Table 5), but it was associated with the nutrient density of the complementary diet for several micronutrients (Table 6).
Factor 2, named the 'rice and legumes' pattern, had high loading for rice, legumes, and tea (Table 1). This pattern was positively associated with plant protein, fibre, and PU fat, both for total intake (Table   5) and the nutrient density of the complementary diet (Table 6).
Factor 3, named the 'formula milk/reverse breast milk' pattern, had a high positive loading for formula milk and a high negative loading for breast milk (Table 1). In terms of pattern score tertiles (Table 3), 4.2% of children consumed formula milk in T1 versus 58.1% in T3. The pattern was positively associated with energy, protein and most micronutrients, and ultimately MAR (Table 5). This pattern was also positively associated with the nutrient density of the complementary diet for various nutrients (Table 6), although these associations were weak (r > -0.3 and r < 0.3).

| Age 18-24 months:
Factor 1, named the 'tea and sugar' pattern had a high loading for tea, rooibos tea, and sugar and a high negative loading for breast milk (Table 1). This dietary pattern showed several statistically significant inverse correlations with nutrient intakes, but these correlations were weak (r > -0.3 and r < 0.3). Factor 2, named the 'more westernized' pattern had high loadings for breakfast cereal and milk and high negative loadings for rice and legumes (Table 1), therefore indicating a less traditional but more westernized diet. This pattern was associated with a higher percentage consumers of unhealthy food items such as sweets, cake and cookies, cold drinks, and salty snacks (Table 4). In terms of nutrients (Table 6), this pattern was associated with saturated fat, cholesterol, and riboflavin intakes.
Factor 3, named the 'rice' pattern, had a high loading for rice and a high negative loading for maize meal (Table 1). This pattern showed several statistically significant inverse correlations but r ≥ 0.3 for only magnesium, thiamine, and folate.  T1  T2  T3  P-value  T1  T2  T3  P-value  T1  T2  T3  P-

| DISCUSSION
In this paper, we describe dietary patterns for 6-24-month-old children, using a large dataset of pooled single 24-hr recalls previously collected in four independent studies done in areas of low socioeconomic status. Distinct dietary patterns were identified, and the   and r < 0.3), suggest that breastfeeding children consume a complementary diet of lower nutrient density. We can only speculate on why this is the case. A study in South Africa reported that mothers who were breastfeeding were more likely to be unemployed compared with mothers who formula fed (Nieuwoudt, Manderson, & Norris, 2018) suggesting that income may be a factor. Nonetheless, these results suggest that a stronger focus is needed on the nutritional quality of the complementary foods for breastfeeding babies.
The 'family foods' pattern (age 6-11 months) was positively associated with plant protein and fibre for total intake as well as the nutrient density of the complementary diet, indicating a mostly plant-based diet. The association with PU fat can most probably be ascribed to oil used when preparing legumes. This pattern was positively associated with maize meal and inversely associated with infant cereals, both of to the foods with high loadings in these patterns.
The positive association of both the 'formula milk/reverse breast milk' pattern and the 'tea and sugar' pattern with the nutrient density of the complementary diet may suggest a perception that as long as children are being breastfed, the quality of the complementary diet is not of that high importance. Although this is pure speculation, it warrants further investigation.
Dietary patterns identified in our study are based on a single 24-hr recall, which has several inherent limitations (Murphy, Guenther, & Kretsch, 2006). Studies reporting dietary patterns in children of similar age group used either single 24-hr recall (Gatica, Barros, Madruga, Matijasevich, & Santos, 2012;Melaku et al., 2018) or a food frequency questionnaire (Betoko et al., 2013;Wen et al., 2014). Robinson et al. (2007) reported that princi- to different versions of the database being used to convert food intake data to nutrient intake data was therefore avoided.
In conclusion, dietary patterns varied in terms of energy and nutrient composition, MAR, nutrients densities of the complementary diet, and DDS. Interpretation of the associations between pattern scores and indicators of dietary quality is complex, for various reasons.
Firstly, although in most cases the associations could be explained by the foods with high loadings, this was not always the case. Secondly, some dietary patterns had both positive and negative associations with key micronutrients, particularly in the younger age group, probably because both infant cereals and maize meal are fortified.
Lastly, the associations of the 'formula milk/reverse breast milk' pattern score with various indicators of dietary quality need further attention, as these associations imply poorer dietary quality for breastfeeding babies.

ACKNOWLEDEGEMENTS
We acknowledge the role of the dietary coders and data capturers.

CONFLICT OF INTEREST
The authors declare that they have no conflicts of interest.

CONTRIBUTIONS
The authors' responsibilities were as follows: M.F. conceptualized the study, wrote the first draft, and was responsible for collecting dietary intake data for all the original studies; M.R. contributed to dietary coding and was involved in collecting dietary data in one of the original studies and writing of the manuscript; R.L. did the data analyses; and C.M.S. was the principle investigator for two of the original studies.
All authors read and approved the final manuscript.

FUNDING INFORMATION
The study was funded by the South African Sugar Association (Project 247).