Dietary patterns associated with fat and bone mass in young children

Abstract

Background: Obesity and osteoporosis have origins in childhood, and both are affected by dietary intake and physical activity. However, there is little information on what constitutes a diet that simultaneously promotes low fat mass and high bone mass accrual early in life.

Objective: Our objective was to identify dietary patterns related to fat and bone mass in children during the age period of 3.8–7.8 y.

Design: A total of 325 children contributed data from 13 visits over 4 separate study years (age ranges: 3.8–4.8, >4.8–5.8, >5.8–6.8, and >6.8–7.8 y). We performed reduced-rank regression to identify dietary patterns related to fat mass and bone mass measured by dual-energy X-ray absorptiometry for each study year. Covariables included race, sex, height, weight, energy intake, calcium intake, physical activity measured by accelerometry, and time spent viewing television and playing outdoors.

Results: A dietary pattern characterized by a high intake of dark-green and deep-yellow vegetables was related to low fat mass and high bone mass; high processed-meat intake was related to high bone mass; and high fried-food intake was related to high fat mass. Dietary pattern scores remained related to fat mass and bone mass after all covariables were controlled for (P < 0.001–0.03).

Conclusion: Beginning at preschool age, diets rich in dark-green and deep-yellow vegetables and low in fried foods may lead to healthy fat and bone mass accrual in young children.

INTRODUCTION

Obesity and osteoporosis are related to significant morbidity and mortality. Both diseases may originate in childhood. Childhood dietary intake influences both adiposity and bone mass, and these intake behaviors may track into later years (1, 2). Children who were obese between the ages of 3 and 5 y are >4 times more likely to be obese as adults than children who were not obese (3). Moreover, 3-y-old children with high body fat were shown to have less bone accrual over the next 4 y than those with lower body fat (4), even though others have shown that body fat is positively related to bone mass in children of various ages in cross-sectional studies (5, 6). Preventive measures early in life that promote low adiposity and high bone mass are needed to reduce the public health burden of obesity and osteoporosis. However, there is a notable scarcity of studies in young children that evaluated dietary intakes that simultaneously optimize these 2 outcomes, namely fat and bone mass accrual. Such studies are needed to discern the effects of dietary intake and to inform practices to support appropriate growth and fat and bone mass accrual early in life.

Overall health and growth are affected by dietary intake through single nutrients and interactions among nutrients. Detecting single-nutrient effects is difficult as the effects may be small, and the nutrient intakes themselves are often highly correlated. Furthermore, people eat foods, not single nutrients; therefore, nutritional recommendations are often given in terms of foods. Dietary patterning methods can be used to characterize total dietary intake in terms of foods and capture small effects on health from many single nutrients and nutrient-nutrient interactions. A small number of dietary patterns (typically 2 or 3) are identified from intakes of a large number (usually >20) of food groups. An individual's dietary pattern score reflects his or her intake of multiple foods most representative of the pattern relative to other individuals in a study sample.

Dietary patterning with principal components analysis has been used for ≈3 decades to describe the variation in intake of a large number of food items in terms of a small number of underlying patterns. However, such patterns simply explain the variation in intake, and there is no guarantee that the identified patterns will be related to specific health outcomes. The need for methods that identify dietary patterns that are related to health outcomes is recognized. The reduced-rank regression (RRR) method is such a tool (7). RRR can be used to identify dietary patterns that are related to multiple health outcomes simultaneously, provided the outcome measures are continuous variables (8).

Although the importance of childhood dietary patterns for lifelong health is increasingly recognized, there is little information about what constitutes a diet that promotes low adiposity and high bone mass early in life. It makes sense to study adiposity and bone health together because obesity and osteoporosis likely have common antecedents. An optimal diet for growing children would include foods that promote bone accrual but suppress excess fat mass accrual. A diet that is energy-dense, but not nutrient-dense, combined with low physical activity levels, may lead to excess fat gain and reduced bone accrual in children. One recent study that used RRR showed that an energy-dense, low-fiber, high-fat dietary pattern at age 7 y was related to more body fatness at age 9 y (9), but there are no such patterning studies of younger children to our knowledge. Our primary objective was to identify dietary patterns related to the 2 outcomes of whole-body fat mass and bone mass in children during the age period of 3.8–7.8-y by using RRR.

SUBJECTS AND METHODS

Study population

Data for the current analyses were from a longitudinal study of children in which serial measures of fat, lean, and bone mass and height, weight, physical activity, and dietary intake were taken every 4 mo from ages 3.8–7.8 y (13 visits) as previously detailed (4). Recruitment began in 2000. Of the 372 children enrolled, 308 completed the parent study. The numbers of children who contributed data for the current analyses by year were as follows: 295 children in year 1 [3 children refused the dual-energy X-ray absorptiometry (DXA) scan, 12 children had nonusable DXA scans, and 31 children were out of the age range]; 325 children in year 2 (2 children had nonusable DXA scans, and 4 children were out of the age range); 315 children in year 3 (1 child had a nonusable DXA scan, and 1 child was out of the age range); and 292 children in year 4 (2 children refused the DXA scan, and 14 children were out of the age range). The study was approved by an Institutional Review Board, and parents provided written informed consent for their children to participate in the study.

Anthropometric, body-composition, and bone measurements

Height and weight were measured with a wall-mounted stadiometer and digital scale. Body mass index z scores were determined by using the 2000 Centers for Disease Control and Prevention growth reference (10). Whole-body fat mass, lean mass, and bone mass were measured by whole-body DXA scans. We reported a measurement error of <1.9% for fat, lean, and bone mass by DXA for children (11). The skull was excluded from the bone mass measures to avoid the complication of the decline in head size relative to body size over time in young children. All scans were analyzed with low-density bone edge detection algorithms (software v12.4, DXA QDR4500; Hologic Inc, Waltham, MA). Trained personnel visually inspected all printed scans, and those scans with major limb or trunk movement were excluded (n = 13).

Dietary intake

Parents were asked to provide a 3-d diet record (2 weekdays and 1 weekend day; all consecutive) for each visit). Cards were given to parents to give to the child care provider with instructions to record the amount and type of foods eaten when the child was not with the parent. The Nutrient Data System for Research program (v2005; Nutrition Coordinating Center, University of Minnesota, Minneapolis, MN) was used to determine the average daily intake of energy (kcal), carbohydrate (g), protein (g), fat and 34 food guide pyramid–based food-group servings (seesupplemental material under “Supplemental data” in the online issue) for each 3-d record. We combined expertise on dietary recommendations for children (KSW) and clinical pediatrics in terms of what foods children commonly consume (KAC) to determine the food groupings that largely stayed with food guide pyramid–based food groupings but with special consideration of foods that were likely to be commonly eaten and clinically relevant such as nonwholegrains, fried meats, fried potatoes, whole milk, processed meats (franks, sausages, and lunch meats), protein source (eg, red meat, white meat, fish, and nonanimal protein), and sweetened beverages. Nutrient-supplement data were included in the estimates of calcium intake.

Only 10 of the 4258 records collected had <3-d records. We used the mean of the 3-d averages for food-group servings and nutrients over 4 visits for each of the 4 study years. Information for diet records in year 1 were collected at visits 1–4, year 2 at visits 4–7, year 3 at visits 7–10, and year 4 at visits 10–13. The double use of visits 4, 7, and 10 was done to capture a child's intake over the entire previous year, rather than over three-quarters of the previous year, and to keep the number of study visits consistent across all years. We realize our chosen method may have induced a correlation between years. However, this should not have seriously affected our findings because each year was treated separately in the RRR analyses. More than 93% of subjects had four 3-d records/y.

Physical activity and television viewing

For the same 3 d as the diet record at each visit, parents were asked to have the child wear a triaxial accelerometer (RT3; Stay Healthy Inc, Monrovia, CA) on the right hip except when sleeping, swimming, or bathing. The triaxial data were downloaded as vector magnitude units (ie, counts on the basis of all 3 axes). For each day, accelerometer counts per minute were calculated following an algorithm previously described in detail (4). A total of 9591 d from all 372 children qualified for analyses. We used the mean of the average daily counts per minute over 4 visits for each of the 4 study years. All usable days were included (≥85% of children had ≥6 usable days for each of the 4 study years).

For the same 3 d as the diet record at each visit, parents reported on the television viewing of their children as previously described (4). We calculated the mean hours per day that children spent viewing television for each of the 4 study years by using 4 visits per study year as we did with the diet and accelerometer data. Over 98% of the children had 3 or more 3-d television records for each of the 4 study years. Outdoor playtime was measured on the same questionnaire with the same method as used for recording television viewing (12) by giving the average hours per day spent in outdoor play for each of the 4 study years.

Statistics

Descriptive analyses

Analysis of variance (ANOVA) was used to describe mean differences by study year in children's anthropometrics, accelerometer counts per minute, television viewing, and outdoor playtime. Food-group intakes were reported as medians (ranges) because several food groups had a large number of children with a zero intake. Differences in median intakes by study year were assessed by the Kruskal-Wallis test. SAS v9.1 software (SAS Institute, Cary, NC) was used to conduct all analyses.

RRR analyses

The RRR model was constructed with the 2 outcomes of interest, fat mass and bone mass, as simultaneous response variables and with the intakes of the 34 food groups as predictor variables. Unlike factor analysis or principal components analysis, which focus on explaining the variation in food intake and, thus, do not define food patterns that are necessarily related to health outcomes, RRR by definition identifies diets that must be related to the outcome variables of interest. If patterns identified by RRR are not significantly related to the outcomes of interest, then none of the food groups would have a high coefficient. On the basis of current knowledge about relations between diet and health, we assumed that RRR would identify patterns related to our 2 outcomes of fat mass and bone mass. We used raw data (unadjusted for energy intake) for food-group intakes when running the RRR analyses. The age range for each year was very narrow and made it unlikely that the intakes of children varied enough for a few individuals to have a major effect on the identified patterns (13). RRR extracts as many patterns (factors) as response variables. Therefore, 2 dietary patterns were identified for each of the 4 study years for a total of 8 patterns. The RRR analyses were conducted with the PLS procedure in the SAS software (SAS Institute). A specific SAS code (SAS; SAS Institute) is available from the authors. For each year, dietary patterns were identified to explain the maximum possible variation in the 2 simultaneous outcome variables of fat mass and bone mass. The 4 study years were treated separately because RRR does not handle repeated measures. For example, a child's dietary intake for year 1 was calculated as the average of the 4 diet records that comprised year 1, and the DXA scan at the end of year 1 was used to measure fat mass and bone mass. After RRR analyses, we identified those food groups with a high loading (≥0.20) in the identified pattern, and all of these high-loading foods were reported as representative of the dietary pattern. A food group could have either a high-positive or high-negative loading on a pattern, which means that children with higher scores for the overall pattern consumed high (positive loading) or low (negative loading) amounts of that food group relative to the other children. We report the percentage variation in each response variable accounted for by each identified pattern. With the RRR method, the first identified pattern always explains more variation in the response variables than the second identified pattern.

A dietary pattern score for each child was calculated for each pattern identified by RRR. This was done by using the coefficients from the RRR analysis for all 34 food groups multiplied by the child's intake (servings) of the food groups. For example, a high-positive coefficient for fried foods in a dietary pattern would contribute to giving a child a higher score for the dietary pattern if the child consumed a large amount of fried foods. By definition, pattern scores are uncorrelated, which allows for their simultaneous entry in subsequent multivariable analyses.

Multivariable analyses

The dietary pattern scores were the primary independent variables. We used the residuals of the relation between energy intake and dietary pattern scores [dietary pattern 1 (DP1) and dietary pattern 2 (DP2)] so that we could evaluate the influence of the patterns separately from energy intake. We also ran models that used the dietary pattern scores unadjusted for energy intake. Multivariable analyses with fat mass and bone mass as separate dependent variables were conducted (ie, there were 4 multivariable models for fat mass, one multivariable model for each of the 4 study years, and 4 multivariable models for bone mass, one multivariable model for each of the 4 study years, for a total of 8 models). The covariables race, sex, height, and exact age were included in all models. The inclusion of annual household income as reported by the parent (<$20,000, $20,000–$49,999, $50,000–$74,999, or ≥$75,000) did not affect our findings and was not included in the final models. Calcium intake, which is related to total food intake, bone mass (14), and possibly fat mass (15) and the 3 activity-related variables of counts per minute, television viewing, and outdoor playtime were always included to minimize potential confounding.

RESULTS

Anthropometric measurements and food-group intakes

Anthropometric measurements and food-group intakes (servings) are given in Tables 1 and 2. Food-group servings were based on adult serving sizes from the Nutrient Data System for Research program. As expected, body size and the intake of most foods increased from years 1–4.

TABLE 1.

Descriptive information across the 4 study years¹

	Year 1	Year 2	Year 3	Year 4
Age range (y)	3.8–4.8	>4.8–5.8	>5.8–6.8	>6.8–7.8
n	295	325	315	292
Primary outcomes
Fat mass (kg)2	4.8 ± 1.33	5.4 ± 1.7	5.9 ± 2.3	6.9 ± 2.9
Bone mass (g)2	379 ± 50	462 ± 66	540 ± 75	626 ± 88
Covariables
Race [black/white (n)]	56/239	61/264	61/254	53/239
Sex [boys/girls (n)]	154/141	167/158	161/154	150/142
Height (cm)2	105 ± 5	112 ± 5	118 ± 5	125 ± 6
Weight (kg)2	18 ± 2	20 ± 3	23 ± 4	27 ± 5
BMI z score	0.4 ± 1	0.5 ± 1	0.5 ± 0.1	0.5 ± 0.1
Accelerometer counts per minute4	582 ± 131	606 ± 153	612 ± 157	586 ± 141
Television viewing (h/d)2	2.07 ± 0.73	2.27 ± 0.74	2.34 ± 0.72	2.31 ± 0.71
Outdoor play (h/d)2	1.75 ± 0.78	1.65 ± 0.76	1.65 ± 0.73	2.24 ± 0.90
Calcium intake (mg)	804 ± 271	851 ± 275	879 ± 257	892 ± 266

¹n, number of children who contributed data to the analyses for that year.

^2,4ANOVA (except for food groups) across years 1–4: ²P < 0.001, ⁴P < 0.05.

³Mean ± SD (all such values).

TABLE 2.

Intakes (servings/d) of food groups across the 4 study years¹

	Year 1	Year 2	Year 3	Year 4
n	295	325	315	292
Whole grains	0.24 (0–3.01)	0.29 (0–2.38)	0.26 (0–2.50)	0.33 (0–2.72)
Non–whole grains2	2.61 (0.49–7.47)	2.91 (0.40–7.48)	3.37 (0.61–7.71)	3.81 (0.99–7.81)
Dark-green vegetables	0.00 (0–1.11)	0.01 (0–2.16)	0.01 (0–1.98)	0.01 (0–0.70)
Deep-yellow vegetables	0.04 (0–0.85)	0.05 (0–0.76)	0.04 (0–0.85)	0.05 (0–0.84)
White potatoes	0.04 (0–0.73)	0.06 (0–1.04)	0.05 (0–1.42)	0.06 (0–0.74)
Other starchy vegetables	0.06 (0–0.82)	0.07 (0–1.13)	0.06 (0–0.83)	0.07 (0–1.00)
Tomatoes2	0.12 (0–0.68)	0.14 (0–0.79)	0.14 (0–0.86)	0.17 (0–0.83)
Other vegetables3	0.10 (0–1.19)	0.11 (0–0.94)	0.13 (0–1.40)	0.15 (0–1.35)
Citrus fruit	0.00 (0–0.68)	0.00 (0–0.76)	0.00 (0–0.77)	0.00 (0–0.51)
Noncitrus fruit	0.67 (0–3.81)	0.64 (0–3.63)	0.60 (0–2.80)	0.58 (0–3.57)
Non–whole milk3	0.89 (0–5.21)	1.03 (0–3.90)	1.06 (0–4.28)	1.10 (0–5.05)
Whole milk2	0.05 (0–3.99)	0.04 (0–3.21)	0.01 (0–2.96)	0.00 (0–1.95)
Yogurt	0.04 (0–0.67)	0.04 (0–0.74)	0.04 (0–0.66)	0.04 (0–0.68)
Cheese2	0.39 (0.01–1.24)	0.40 (0–1.86)	0.44 (0.02–1.90)	0.49 (0–1.75)
Meat2	0.42 (0–1.83)	0.49 (0–2.42)	0.55 (0–3.37)	0.67 (0–2.54)
Processed meats	0.61 (0–3.39)	0.64 (0–3.39)	0.65 (0–3.36)	0.69 (0–4.77)
Poultry (nonfried)2	0.22 (0–1.88)	0.26 (0–2.78)	0.29 (0–2.62)	0.40 (0–2.19)
Fish (nonfried)	0.00 (0–1.95)	0.00 (0–1.34)	0.00 (0–1.56)	0.00 (0–1.57)
Eggs3	0.08 (0–0.89)	0.08 (0–1.05)	0.08 (0–0.93)	0.10 (0–1.61)
Beans and peas	0.00 (0–0.48)	0.00 (0–0.85)	0.00 (0–0.67)	0.00 (0–0.60)
Meat and dairy alternates2	0.00 (0–0.62)	0.00 (0–0.99)	0.00 (0–0.82)	0.00 (0–1.88)
Nuts, seeds, and peanut butter4	0.21 (0–1.61)	0.22 (0–2.65)	0.28 (0–2.47)	0.27 (0–2.17)
Discretionary fats2,5	1.77 (0.22–6.03)	2.07 (0.20–5.86)	2.22 (0.16–6.10)	2.45 (0.21–12.12)
Added sugars4,6	0.45 (0.003–3.83)	0.48 (0–4.30)	0.55 (0.01–9.00)	0.54 (0.01–9.25)
Sweetened beverages4	0.73 (0–3.15)	0.87 (0–4.32)	0.87 (0–3.78)	0.89 (0–3.93)
Savory snacks2	0.32 (0–2.01)	0.39 (0–2.11)	0.45 (0–4.03)	0.51 (0–3.45)
Nondairy desserts2	1.02 (0.08–2.84)	1.07 (0.08–4.15)	1.11 (0.10–3.30)	1.25 (0–3.89)
Fried potatoes	0.14 (0–2.28)	0.15 (0–2.16)	0.16 (0–0.98)	0.17 (0–1.30)
Fried chicken and fish4	0.37 (0–2.08)	0.47 (0–2.05)	0.48 (0–2.19)	0.51 (0–2.58)
Miscellaneous	0.03 (0–0.84)	0.03 (0–1.05)	0.04 (0–1.95)	0.04 (0–1.85)
Frozen dairy desserts and pudding4	0.19 (0–0.98)	0.23 (0–1.54)	0.25 (0–1.35)	0.29 (0–2.00)
Fruit juices4	0.82 (0–7.14)	0.73 (0–5.42)	0.58 (0–5.70)	0.49 (0–4.77)
Crackers	0.18 (0–1.12)	0.17 (0–1.09)	0.14 (0–1.78)	0.14 (0–1.20)
Unsweetened/artificially sweetened beverages2	0.00 (0–4.08)	0.00 (0–4.50)	0.02 (0–3.58)	0.02 (0–2.50)

¹All values are medians; ranges in parentheses. n, number of children contributing data to the analyses for that year. Intakes are based on adult serving sizes.

^2–4Kruskal-Wallis rank sum test (food groups) across years 1–4: ²P < 0.001, ³P < 0.05, ⁴P < 0.01.

⁵Such as salad dressings and butter.

⁶Such as soda, candy, and cakes.

Dietary patterns

Food groups with high-factor loadings for the identified patterns, the direction of the relation of the patterns with fat mass and bone mass, and the percentage variation in fat mass and bone mass explained by the patterns are shown in Tables 3 (DP1) and 4 (DP2). DP1 was related to both high fat mass and high bone mass in all 4 y. DP1 explained 13.4–19.2% and 11.4–18.1% of the variation in fat mass and bone mass, respectively, for the 4 study years. Food groups for which high intakes were consistently (in ≥3 of the 4 study years) related to high fat mass and high bone mass for DP1 were nonwholegrains, cheese, processed meats, eggs, fried potatoes, discretionary fats, and artificially sweetened beverages.

TABLE 3.

Food groups with factor loadings ≥|0.20| for dietary pattern 1 for each of the 4 study years¹

	Year 1	Year 2	Year 3	Year 4
Age range (y)	3.8–4.8	>4.8–5.8	>5.8–6.8	>6.8–7.8
n	295	325	315	292
Relation of pattern scores with fat mass	+	+	+	+
Relation of pattern scores with bone mass	+	+	+	+
Percentage variation in fat mass explained2	16.6	13.4	19.2	16.0
Percentage variation in bone mass explained2	13.8	11.4	16.8	18.1
Non–whole grains3	0.30	0.41	0.31	0.41
White potatoes	—	—	0.23	—
Cheese3	0.25	0.20	—	0.32
Meat	—	—	0.22	0.23
Poultry (nonfried)	—	0.26	0.31	—
Processed meats3	0.28	0.22	0.24	0.35
Eggs3	0.32	0.31	0.33	0.34
Fried potatoes3	0.27	0.26	0.36	0.30
Fried chicken and fish	0.32	—	—	0.20
Nondairy desserts	0.29	0.32	—	—
Discretionary fats3	0.31	0.30	0.23	0.24
Artificially sweetened beverages3	0.21	0.20	—	0.25

¹n, number of children who contributed data to the analyses for that year; +, direction of the relation between dietary pattern 1 scores and fat mass or bone mass. We identified the food groups with a high loading (≥|0.20|) in the identified pattern, and all of these high-loading foods are reported as representative of the dietary pattern.

²For example, dietary pattern 1 for year 1 explains 16.6% and 13.8% of the variation in fat mass and bone mass, respectively (unadjusted for covariables).

³Items appeared in ≥3 of the 4 study years.

DP2 explained an additional (ie, after DP1) 3.3–5.2% and 3.9–5.8% of the variation in fat mass and bone mass, respectively, for the 4 y. We consider DP2 to be of high importance because it was related to the desired outcomes of lower fat mass and higher bone mass. Food groups for which high intakes consistently (in 3 of 4 study years) were related to lower fat mass and higher bone mass for DP2 were dark-green vegetables, deep-yellow vegetables, and processed meats. The fact that processed meats appeared as a positive coefficient in both DP1 and DP2 suggests that their effect was largely toward high bone mass. Food groups for which low intakes consistently were related to lower fat mass and higher bone mass were fried chicken and fish and fried potatoes. An example of a diet that would give a child a high score for DP2 in year 2 is as follows: a high intake (relative to the other children) of wholegrain breads and cereals; broccoli and spinach; carrots and sweet potatoes; tomatoes and tomato sauces; 100%-fruit juices; poultry (nonfried) ; beef, pork, and poultry processed meats; nuts, seeds, and peanut butter; and a zero or low intake of fried chicken (eg, chicken nuggets), fried fish (eg, fish sticks), and fried potatoes (Table 4). Diets that gave high scores for DP1 and DP2 for all other years can be similarly ascertained by looking at the food groups with high-loading coefficients ≥20 in Tables 3 and 4.

TABLE 4.

Food groups with factor loadings ≥|0.20| for dietary pattern 2 for each of the 4 study years¹

	Year 1	Year 2	Year 3	Year 4
Age range (y)	3.8–4.8	>4.8–5.8	>5.8–6.8	>6.8–7.8
n	295	325	315	292
Relation of pattern scores with fat mass	−	−	−	+
Relation of pattern scores with bone mass	+	+	+	−
Additional percentage variation in fat mass explained2	4.8	3.3	3.7	5.2
Additional percentage variation in bone mass explained2	5.8	3.9	4.3	4.6
Whole grains	0.30	0.28	—	—
Non–whole grains	0.30	—	0.29	—
Dark-green vegetables3	0.31	0.24	0.37	—
Deep-yellow vegetables3	—	0.25	0.20	−0.22
Tomatoes	—	0.24	—	—
Other vegetables	0.24	—	—	—
Noncitrus fruit	—	—	—	−0.36
Fruit juices	—	0.23	0.29	—
Poultry (nonfried)	—	0.21	—	—
Processed meats3	0.30	0.25	0.33	−0.37
Fish (nonfried)	—	—	0.27	—
Eggs	—	—	—	−0.23
Beans and peas	0.27	—	—	—
Nuts, seeds, and peanut butter	—	0.37	—	—
Meat and dairy alternates	0.26	0.26	—	—
Fried chicken and fish3	—	−0.32	−0.27	0.25
Fried potatoes3	−0.23	−0.22	—	0.37
Savory snacks	0.31	—	—	—
Sweetened beverages	0.26	—	—	—
Artificially sweetened beverages	—	—	−0.28	0.31

¹n, number of children who contributed data to the analyses for that year; + or −, direction of the relation between dietary pattern 1 scores and fat mass or bone mass. We identified the food groups with a high loading (≥0.20) in the identified pattern, and all of these high-loading foods are reported as representative of the dietary pattern. A food group can have either a high-positive or high-negative loading on a pattern, which means that children with higher scores for the overall pattern consumed high (positive loading) or low (negative loading) amounts of that food group relative to the other children. A negative-factor loading indicates that the pattern was characterized by low intakes of that food.

²Additional percentage variation explained after dietary pattern 1 (unadjusted for covariables).

³Items appeared in ≥3 of the 4 study years.

For DP2, we determined if and to what extent dietary pattern scores were different according to race, sex, and household income, with the intention of identifying target demographics, by using ANOVA or Tukey's honest significant difference as appropriate. DP2 scores were higher for blacks than for whites in year 1, for boys than for girls in all 4 y, and for children from families with a household income <$20,000 than for children from families with a higher household income (data not shown).

Because processed meats appeared as an important food group in all 4 study years for both fat mass and bone mass, we examined which factors were related to processed-meat intake by using ANOVA or Tukey's honest significant difference as appropriate. Processed-meat intake was higher in blacks than whites in all 4 y, in boys than in girls in year 1, and in children from families with a household income <$20,000 than in children from families with a higher household income (data not shown). Race and household income were strongly related in all 4 y, with blacks more likely than whites to be in the lower-income categories (data not shown).

Multivariable analyses

Mean fat mass and bone mass by quartile of DP1 and DP2 scores from adjusted models were used to illustrate findings (Figures 1 and 2). Across study years, DP1 scores were consistently and positively related to fat mass, whereas DP2 scores indicative of higher intakes of deep-yellow vegetables and processed meats and lower intakes of fried chicken and fish and fried potatoes were consistently negatively related to fat mass after adjusting for race, sex, height, exact age, energy intake, calcium intake, television viewing, accelerometer counts, and outdoor play (all P < 0.001). DP1 scores were significantly associated with higher bone mass after adjusting for all covariables (all P < 0.05). DP2 scores were related to higher bone mass in years 1, 3, and 4 (all P < 0.05) but not in year 2 (P = 0.2) in models adjusted for all covariables (Figure 2). In year 4, high intakes of deep-yellow vegetables and processed meats and low intakes of fried foods led to a lower DP2 score, which in turn was related to lower fat mass and higher bone mass (see the legends of Figures 1and 2 for explanations of the relation of DP2 scores with fat mass and bone mass in year 4). The RRR analyses selected the same food groups, with the same relation to fat mass and bone mass, across ≥3 of 4 study years.

FIGURE 1.

Fat mass of children with low (quartile 1), medium (quartiles 2 and 3), and high (quartile 4) scores for dietary pattern 1 (A, C, E, and G) and dietary pattern 2 (B, D, F, and H) at progressively higher ages. P values are for the effect of the continuous variables for dietary pattern score in the full model adjusted for the other dietary pattern score, height, exact age, race, sex, accelerometer counts per minute, television-viewing time, outdoor playtime, calcium intake, and energy intake; R² is for the full model. Plotted values are least-squares means from the full model with dietary pattern scores treated as categorical variables. The sample included 18–19% black and 51–52% male subjects at each of the 4 study years. In year 4, relevant coefficients for dietary pattern 2 are reversed compared with years 1–3 (Table 3): coefficients are negative for deep-yellow vegetables and processed meats and positive for fried chicken and fish and fried potatoes. This means that a child with high intakes of deep-yellow vegetables and processed meats and a low intake of fried foods has a low dietary pattern 2 score in year 4; thus, a low dietary pattern 2 score in year 4 is linked to lower fat mass. Therefore, the finding for year 4 is similar to the findings for years 1–3.

FIGURE 2.

Bone mass of children with low (quartile 1), medium (quartiles 2 and 3), and high (quartile 4) scores for dietary pattern 1 (A, C, E, and G) and dietary pattern 2 (B, D, F, and H) at progressively higher ages. P values are for the effects of the continuous variables for dietary pattern score in the full model adjusted for the other dietary pattern score, height, exact age, race, sex, accelerometer counts per minute, television-viewing time, outdoor playtime, calcium intake, and energy intake; R² is for the full model. Plotted values are least-squares means from the full model with dietary pattern scores treated as categorical variables. The sample included 18–19% black and 51–52% male subjects at each of the 4 study years. In year 4, relevant coefficients for dietary pattern 2 are reversed compared with years 1–3 (Table 3): coefficients are negative for deep-yellow vegetables and processed meats and positive for fried chicken and fish and fried potatoes. This means that a child with high intakes of deep-yellow vegetables and processed meats and a low intake of fried foods has a low dietary pattern 2 score in year 4; thus, a low dietary pattern 2 score in year 4 is linked to higher bone mass. Therefore, the finding for year 4 is similar to the findings for years 1–3.

We ran models that used the dietary pattern scores unadjusted for energy intake and compared the coefficients for the dietary pattern scores to those from models that used the dietary pattern scores adjusted for energy intake. We observed that the coefficients for dietary pattern scores unadjusted for energy intake were either similar to (≤5% difference) or lower than the coefficients for the dietary pattern scores adjusted for energy intake. This confirmed our decision to use dietary pattern scores adjusted for energy intake in the final models. Interestingly, in full models calcium intake was not related to fat mass in any year and was positively and significantly (P < 0.05) related to bone mass in years 1 and 2 but not in years 3 and 4. To confirm that the inclusion of calcium intake was not an overcontrol for the effects of a pattern, we compared the coefficients for the dietary pattern scores to the coefficients for the dietary pattern scores from models without calcium intake. We observed that the coefficients for dietary pattern scores unadjusted for calcium intake were either similar to (a ≤5% difference) or lower than the coefficients for the dietary pattern scores adjusted for calcium intake. Thus, calcium intake was retained in all final models.

DISCUSSION

We identified dietary patterns related to fat mass and bone mass in children aged 3.8–7.8 y. Our findings help to elucidate potential target foods that may promote low adiposity and high bone mass accrual in young children. For example, our results for DP2 suggest that increasing intakes of dark-green and deep-yellow vegetables and limiting fried-food intake may promote healthy fat and bone mass accrual. This suggestion is supported by other findings that older children who increased their fried-food intake had a greater rise in body mass index than those with low fried-food intake (16), and higher fruit and vegetable intake was related to higher whole-body bone accrual in boys (17). Whether or to what extent a low intake of fruits and vegetables and a high intake of fried foods by young children lead to lasting undesirable effects on bone is unknown.

DP1 and DP2 were related to fat mass and/or bone mass independent of energy intake in all 4 y. This suggests that the combination of foods in the dietary pattern, regardless of energy intake, may significantly affect young children's fat mass and bone mass. Whether these patterns hold true for children with energy intakes different from those observed in our cohort is unknown. Our findings generally align with a recent report in 5- to 9-y-olds that showed that a high-fat, low-fiber, energy-dense dietary pattern was associated with a higher fat mass 2–4 y later (9). To our knowledge, no studies have examined, as we have, the association of diet composition with both fat and bone mass simultaneously in younger children.

The biological process by which dark-green and deep-yellow vegetables (eg, spinach, romaine lettuce, broccoli, carrots, and sweet potatoes) affect bone mass remains unclear but may be related to their high content of alkalizing minerals such as potassium. Although single-nutrient intakes were not the focus of this study, subjects in the highest quartile of dietary pattern scores had higher intakes of calcium and potassium than did subjects in the other 3 quartiles. We decided a priori to include calcium intake in multivariate models, and it was shown to have no significant effect on the coefficients for dietary pattern scores. Higher dietary potassium intake is related to lower net endogenous acid production and to a higher bone mass in adults (18). Animal studies have shown that osteoclasts are stimulated to resorb bone at an acidic pH (19). Although dietary protein is acid promoting, a recent meta-analysis showed that protein intake had a beneficial effect on bone density in adults (20). Protein intakes within the adequate range, especially when combined with high fruit and vegetable intakes, were associated with lower age-related bone loss in the Framingham cohort (21). In children 6 to 18 y of age, the anabolic effect of protein on bone was shown to be partly negated when intakes of alkalizing minerals were low (22). To our knowledge, studies on net endogenous acid production and bone in young children do not exist. Our finding that a high intake of processed meats combined with a high intake of dark-green and deep-yellow vegetables promotes higher bone mass suggests that the net production of endogenous acid may influence bone mass accumulation, but further studies that involve measures of acid production are needed. Although we do not advocate processed meats as the primary protein source for children because of the high sodium and saturated-fat contents of processed meats, in this cohort, processed meats were a significant protein source and should be considered important when studying diet and health in young children because such foods are commonly eaten.

We observed that higher fried-food intake was associated with greater adiposity even when accounting for energy intake. A study of Spanish men and women showed that subjects with a higher percentage of energy intake from fried foods were most likely to have central obesity (23). Although it is unknown if high fried-food intake promotes central obesity in children, our data show that higher fried-food intake was related to higher overall fat mass. We speculate that high fat mass may lead to increased adipocytokine secretion that, combined with the known disruption to the growth hormone–insulin-like growth factor axis that occurs in obese children (24) may promote nutrient partitioning for fat rather than lean mass accrual. The role of hormonal shifts because of diet-induced central and/or overall adiposity on the long-term growth and health of young children needs further investigation.

In our study, the only case where the dietary pattern scores were not related to bone mass after adjusting for energy intake and other covariables was for DP2, which was characterized by a high intake of dark-green and deep-yellow vegetables and processed meats and a low intake of fried foods, at ages >4.8–5.8 y. It is possible that this dietary pattern has less influence on bone mass than on fat mass during this age period when adiposity rebound (the point in early childhood when body fatness reaches a nadir before increasing again) is likely to occur. The possibility that fat mass is preferentially influenced by diet during adiposity rebound was recently highlighted by Gunther et al (25) who focused on protein intake and showed that 5–6 y of age may be a critical period in which relatively high protein intake leads to more body fatness at 7 y of age in German children. We focused on food-group combinations rather than protein, and in DP2 we observed diet to be strongly related to fat mass but not bone mass during the adiposity-rebound age period. It is possible that the magnitude of the effect of food-group combinations on body composition varies with age in early childhood, although our data imply a consistent influence of dietary patterns on fat mass and bone mass in young children.

This study differs from others that have examined the relation between obesity and bone mass in children because we considered both fat mass and bone mass as dependent variables. Therefore, it is difficult to compare our findings with those that have shown either a positive (5, 6) or negative (4) relation between fat mass and bone mass. Instead, to our knowledge, we have provided new information about combinations of food groups associated with fat mass and bone mass simultaneously.

Strengths of our data include the narrow age spans for children and our frequent closely spaced measures of dietary intake. Each individual dietary intake estimate was based on the following four 3-d diet records: one diet record at the time of the DXA scan plus 3 diet records preceding it for the whole prior year's intake. We were able to account for physical activity, television viewing, and outdoor playtime by using measures obtained at the same time and with the same frequency as the diet records.

A limitation is that RRR, like principal component analysis, factor analysis, and cluster analysis, is data driven and nonlongitudinal. The patterns identified in this cohort are not exactly the same from year to year, and the patterns cannot be exactly reproduced in other cohorts. Subjectivity can enter the RRR process at various decision points such as how to group foods or the method to account for energy intake, which we decided a priori. Two authors consulted and came to agreement on how to group foods before conducting the analyses. Moreover, to avoid subjectivity in interpretation, we intentionally did not name the dietary patterns. Despite the limitations, the foods that consistently characterized the dietary patterns across all years can potentially inform further studies of target foods that may promote healthy fat and bone mass gain in young children.

Although our findings should be considered hypothesis generating, our data suggest that the diets of young children that are high in colorful vegetables, even with the low median intakes observed, and low in fried foods may lead to lower fat mass and greater bone accrual in young children. Further, we observed that white children, especially girls, had the lowest scores for DP2, which was related to both low fat mass and high bone mass, and may be important targets for intervention.