Longitudinal Assessment of Childhood Dietary Patterns: Associations with Body Mass Index z-Score among Children in the Samoan Ola Tuputupua'e (Growing Up) Cohort

Abstract

Background: Intervention strategies to prevent childhood obesity in the Pacific Islands encourage eating a variety of local and traditional foods, but context-specific data to support this approach are limited. The objective was to assess the association between 2-year adherence to modern and neotraditional dietary patterns and body mass index z-scores (BMIz) among Samoan children.

Methods: A convenience sample of 214 Samoan biological mother-child dyads who participated in the Ola Tuputupua'e “Growing Up” cohort study in 2015 and 2017 was included. At each time point, modern and neotraditional dietary patterns were identified using food frequency data and factor analysis. Children were assigned to categories based on diet pattern adherence: consistently high, high to low, low to high, and consistently low. Associations between 2-year adherence to dietary patterns, BMIz (in 2017 and 2015–2017), and weight and height z-scores were examined using linear models adjusted for potential confounders.

Results: Consistently high adherence to the modern pattern was associated with a 0.36 standard deviation (SD)-adjusted change in BMIz between 2015 and 2017 (95% confidence interval [CI]: 0.04–0.69, p = 0.03). While the estimates for the individual components of BMI were imprecise, on average, children with consistently high adherence to the modern pattern had a 0.13 SD greater change in weight z-score (95% CI: −0.10 to 0.37) and 0.15 SD lower change in height z-score (95% CI: −0.43 to 0.13). The change in BMIz between 2015 and 2017 did not significantly differ by adherence to the neotraditional pattern.

Conclusion: A neotraditional dietary pattern, comprising local produce, should be encouraged as a possible childhood obesity prevention strategy in Samoa.

Introduction

Pacific Island countries experience a catastrophic prevalence of cardiometabolic noncommunicable diseases (NCD), with adult obesity prevalence as high as 80% and more than 70% of deaths attributed to NCDs at the regional level.^1,2 This phenomenon may be partly explained by changes in the diet over time, as countries in the region continue to undergo nutritional and epidemiological transitions. Extensive shifts in lifestyle and the globalization of the food supply have reduced food self-sufficiency and changed dietary patterns.^3,4

Since World War II, traditional foods in the Pacific Islands, including yams, fish, taro, and local fruits and vegetables, have been replaced and/or supplemented with canned foods, sugar-sweetened beverages, and micronutrient-poor processed foods.⁵ This shift in dietary patterns toward increased consumption of energy-dense, micronutrient-poor, highly processed, imported foods adversely affects health across the life course^6–8; children are particularly vulnerable since many imported foods are specifically marketed to them and they tend to be exposed to new foods entering the market earlier than adults.^9,10

The Ola Tuputupua'e study in Samoa is the first child cohort study in the Pacific Island countries and has generated longitudinal data to understand the influence of diet on children health outcomes. Samoa, like other low-middle income Pacific Island countries undergoing significant sociodemographic and nutritional transition, faces a dual burden of obesity and micronutrient deficiencies.⁵ In 2015, we documented an already low-moderate burden of overweight/obesity, as defined by body mass index-for-age z-score (BMIz) >2 based on World Health Organization (WHO) growth standards, 16.1% among the cohort at age 2–4 years.¹¹ While the majority of children were meeting or exceeding carbohydrate, fat, and protein requirements (based on US dietary recommendations), a significant proportion was not consuming adequate micronutrients like calcium, vitamin A, and vitamin E.¹²

Health protection and promotion strategies to address and/or prevent NCD in the Western Pacific region have historically focused on eating a variety of local fruits and vegetables. Recently, the Secretariat for the Pacific Community released the first Pacific Guidelines for a Healthy Diet and Lifestyle handbook and recommended eating a variety of nutritious local foods that are energy dense (starchy staples), protective (vegetables and fruits), and body building (protein-rich foods) in appropriate amounts every day.¹³ Consuming local produce and traditionally grown foods is emphasized for good health; however, Pacific Islander-specific data to support the validity of this approach are limited. Indeed, in a prior cross-sectional analysis, we observed a positive association between following a neotraditional pattern and overweight/obesity among Samoan children 2–4 years of age,¹⁴ contradicting results from other populations^15,16 and adult studies in Samoa.^17,18

Recognizing the limitations of our earlier cross-sectional approach and with the goal of providing the first data on prospective associations between diet and childhood growth in the Pacific region, we aimed to describe childhood dietary patterns over a 2-year period and evaluate their prospective association with BMIz.

Methods

Study Design, Subjects, and Data Collection

The initial recruitment of the Ola Tuputupua'e cohort, which took place between June and August 2015 on the Samoan island of Upolu, was previously described.¹¹ Briefly, we recruited a convenience sample of 319 biological mother-child pairs from 10 villages selected to equally represent three census regions and differential exposure to the nutrition transition based on urbanicity: the Rest of Upolu (n = 3 villages; rural with low exposure), Northwest Upolu (n = 3 villages; peri-urban with moderate exposure), and the Apia Urban Area (n = 4 villages; urban with high exposure).

Eligible mothers were not pregnant, had no severe physical or cognitive impairments, and were willing/able to complete the interview portion of the study. Eligible children were 2–4 years old (in 2015), had no severe physical or cognitive impairments, and were of Samoan ethnicity (four Samoan grandparents). To minimize relatedness, a single child from each household was randomly selected. The follow-up assessment was completed 2 years later (June and August 2017), across 11 villages in Upolu, with an additional Apia Urban Area village surveyed.

Yale and Brown University Institutional Review Boards and the Health Research Committee of the Samoan Ministry of Health approved all procedures involving human subjects (IRB no. 2000020519 and IAA no. 18–41 959). Written informed parental consent was obtained for all participants in 2015 and 2017.

Study Sample

Children without follow-up and those with missing data on the key variables of interest were excluded. Specifically, among the 319 children surveyed in 2015, we were able to re-contact 74.3% (n = 237) in 2017. We did not complete follow-up assessments on the other 82 children because mother or child was contacted, but were unavailable for assessment (n = 44), the family had moved out of the original study village (n = 37), or the child had died (n = 1).

Of the 237 children with follow-up data, six were missing BMIz data and 17 had no food frequency questionnaire (FFQ) data. There were 32 children who missed three or fewer food items in the FFQ and to preserve the analytic sample size, we imputed a frequency value of zero to the missing values. The proportion of children surveyed by sex and census region was similar in 2015 compared to 2017, and there were no significant demographic differences among those with and without missing data (data not shown). The final analytic sample included 214 children.

Assessment of Exposures

Two-year adherence to dietary patterns was the exposure of interest; for convenience, this is hereafter referred to as the “change in dietary patterns.” Individual food items came from an FFQ, with a 30-day reference period, which had been validated and adapted from previous adult studies in Samoa.^17,18 Mothers reported how often each food item was consumed by children using seven categories ranging from “never/less than once per month” to “more than six times per day.” Food items were placed into prespecified food groups based on culinary use and nutrient profile (Supplementary Table S1). There were 35 food groups in 2015 and 36 groups in 2017 due to the addition of coffee, which was widely consumed by children at this age in Samoa during our study visits in 2017.

We calculated the food group frequencies as the sum of individual food items and adjusted for total energy intake in the corresponding year using the residual method.¹⁹ Total energy intakes of children were calculated by multiplying the daily consumption frequency of each food by the nutrient content of a fixed, standard portion size from the FFQ.

We used factor analysis with the principal component method to identify empirical dietary patterns in 2015 and 2017 separately. This approach has been previously used to examine the change in dietary patterns when relatively similar population-level dietary patterns can be identified at two time points.²⁰ The number of factors retained for the final analysis was based on the scree plot,²¹ the eigenvalue >1 criterion, and the overall interpretability of the factors. We rotated the retained factors by orthogonal (varimax) transformation to achieve uncorrelated and more interpretable structures without mutual confounding. The factor score for each retained pattern was calculated based on the sum of the standardized food group frequencies weighted by their scoring coefficients, with higher factor scores indicating higher intake. Because the food group frequencies were adjusted for total energy intake, scores were independent of total energy intake.

We identified modern and neotraditional dietary patterns in both 2015 and 2017 (Table 1). The modern pattern explained 10.03% and 6.95% of the variation in food intake in 2015 and 2017, respectively. The neotraditional pattern explained 6.54% and 8.64% of the variation in food intake in 2015 and 2017, respectively. While the factor analyses were completed separately, the modern pattern included high intakes of french fries, unprocessed red meat, potatoes/sweet potatoes, cereals, noodles, and fruit juices, and low intakes of breadfruit and taro in both years. The neotraditional pattern included high intake of fruits, vegetables, tomatoes, banana-based dishes, and soups.

Table 1.

Factor Loadings for Child Dietary Patterns Derived from Factor Analysis^a

Food groups	Modern		Neotraditional
	2015	2017	2015	2017
French fries	0.40*	0.49*	−0.41*	−0.16
Unprocessed red meat	0.38*	0.61*	−0.53*	0.02
Potatoes/sweet potatoes	0.46*	0.46*	0.23*	0.04
Cereals	0.33*	0.44*	−0.19	−0.12
Noodles	0.36*	0.38*	−0.20	0.05
Breadfruit	−0.37*	−0.47*	0.05	0.28
Yam and taro	−0.47*	−0.31*	0.12	0.66*
Fruits	0.22	−0.25	0.50*	0.60*
Banana-based dishes	−0.16	−0.19	0.47*	0.54*
Tomato	0.02	0.31*	0.41*	0.31*
Soup	−0.10	−0.01	0.39*	0.45*
Fruit juices	0.23	0.46*	−0.37*	−0.03
Green, leafy vegetables	0.32*	−0.07	0.50*	0.29
Infant food items	0.42*	0.13	0.18	−0.03
Dairy products	0.41*	0.14	−0.03	−0.26
Butter	0.38*	0.16	−0.02	0.02
Margarine	0.38*	−0.04	−0.08	0.01
Condiments	0.30*	−0.07	−0.18	0.22
Sweets or desserts	0.35*	−0.11	−0.09	−0.55*
Snacks	0.32*	−0.03	−0.03	−0.37*
Nuts	0.24	0.26	−0.10	−0.19
Seafood	−0.26	0.37*	−0.04	0.14
Coconut and coconut products	−0.28	0.11	0.15	0.45*
Poultry	−0.41*	0.20	−0.13	0.35*
Fish	−0.50*	−0.24	0.16	0.03
Tea	0.13	−0.25	0.47*	−0.07
Other vegetables (not green-leafy)	−0.08	−0.12	0.20	0.42*
Refined grains	0.20	−0.13	0.15	0.01
Eggs	−0.03	0.38*	0.08	0.05
Sugary drinks	0.05	0.01	−0.04	−0.43*
Mixed dishes	−0.11	0.10	−0.18	0.15
Corn	0.08	0.22	−0.24	0.36*
Processed red meat	−0.01	0.43*	−0.27	−0.07
Pizza	0.08	0.26	−0.37*	0.05
Whole grains	0.21	0.14	−0.45*	0.03
Coffee	NAb	−0.01	NAb	−0.06

^aValues are factor loadings (i.e., correlation coefficients of each food group with the corresponding dietary pattern). Food groups with factor loadings ≥0.3 or ≤−0.3 are marked with an asterisk (^*). Bold indicates food groups present as high loadings for both 2015 and 2017. All food groups were adjusted for total energy intake in the corresponding year.

^bFood group items were not included in the questionnaire.

NA, not applicable.

For each retained pattern, factor scores were dichotomized as “low” or “high” according to the sample median in each year. It was expected that the underlying dietary patterns at the population level would not be identical for children at 2–4 years of age in 2015 and at 4–7 years of age in 2017, and by using the sample median, we were able to capture each child's relative standings in dietary patterns in each year and also overcome the potential lack of comparability in the continuous factor scores across time. This approach has been used in previous work to facilitate the evaluation of changes in dietary patterns between two time points that were 3 years apart.²⁰

We then categorized the children into four groups based on their change in dietary pattern factor scores from 2015 to 2017. Specifically, children who remained above or below the median in both years were categorized as “consistently high” or “consistently low,” respectively. Children who moved from above the median in 2015 to below the median in 2017 were categorized as “high to low” and vice versa as “low to high.”

Assessment of Outcomes

BMIz in 2017 and the change in BMIz from 2015 to 2017 were the two outcomes of interest; we also present change in weight and height z-scores. Standardized procedures were used to collect duplicate measures of standing height (Stadiometer Pfister Imports, New York, NY) and weight (HD 351 weight scale; Tanita Corporation of America, Arlington Heights, IL). We calculated z-scores using the WHO child growth standards or references depending on child age and sex.^22,23

Assessment of Covariates

Several potential confounders were identified a priori based on their known relationships with diet and/or child growth^11–12,14: age, sex, exposure to breastfeeding, physical activity, maternal overweight/obesity, years of maternal education, household census region, and material lifestyle score.

Child characteristics were reported by mothers. Child age was calculated by subtraction of the birth date from the date of the physical measurements. We asked mothers to report whether they ever breastfed or gave their pumped milk to determine whether a child was ever breastfed. Physical activity scores were calculated using the 35-item Netherlands physical activity questionnaire for young children,²⁴ with higher scores indicating higher levels of overall physical activity.

Maternal overweight/obesity was defined as BMI ≥26 kg/m² based on Polynesian BMI cutoffs, which are sensitive to the greater lean mass per kilogram in the adult Polynesian population compared with other ethnicities.²⁵ Duplicate measures of weight and height of the mother were measured using the same standardized procedures for the child. Mothers reported how many years of education they had received before the interview date.

Based on the village of residence, the household census region was categorized as rural Rest of Upolu, peri-urban Northwest Upolu, or urban Apia Urban Area. Mothers reported ownership of consumer durables, like a television or refrigerator, from which we calculated an 18-item material lifestyle score to estimate household socioeconomic status.^18,26 The material lifestyle score was categorized into quartiles based on score distribution.

Statistical Analyses

We first assessed differences in the child, maternal, and household characteristics by the factor analysis-derived, 2-year change in dietary patterns using generalized linear regression models for continuous variables and chi-square tests for categorical variables. The difference in BMIz of children measured in 2017 and the change in BMIz, weight, and height between 2015 and 2017 by the change in dietary pattern categories were assessed using linear regression models. Children who were consistently low for a dietary pattern between 2015 and 2017 were the reference group in all models. We adjusted for potential confounders using data from 2015 with the exception of total energy intake, where data from both 2015 and 2017 were included to reduce the extraneous variation in dietary intake data and to minimize dietary measurement error.¹⁹ All analyses were conducted at α = 0.05 using SAS 9.4 software (SAS Institute, Inc., Cary, NC).

Results

The overweight/obesity prevalence among the sample was 16.8% at age 2–4 years (in 2015) and increased to 24.8% at age 4–7 years (in 2017). Half (49.5%) of the children were female and 68.22% were ever breastfed (Table 2). The sample average weight increased from 15.3 ± 3.2 kg in 2015 to 20.0 ± 4.1 kg in 2017, while height increased from 94.3 ± 9.6 to 109.8 ± 8.6 cm. The sample average BMIz decreased from 1.14 ± 0.88 standard deviation (SD) in 2015 to 0.70 ± 0.98 SD in 2017, although many children remained above the median BMI for the WHO child standard and reference groups.

Table 2.

Baseline Characteristics of the Samoan Children in the Total Sample and by the 2-Year Change in Dietary Patterns^a

Baseline characteristicsb	Total (N = 214)		Modern patternc								pd	Neotraditional patternc								pd
			Consistently low (n = 62)		Low to high (n = 45)		High to low (n = 45)		Consistently high (n = 62)			Consistently low (n = 53)		Low to high (n = 54)		High to low (n = 54)		Consistently high (n = 53)
	% or mean	SD	% or mean	SD	% or mean	SD	% or mean	SD	% or mean	SD		% or mean	SD	% or mean	SD	% or mean	SD	% or mean	SD
Children
Age (years)	3.36	0.85	3.26	0.82	3.44	0.87	3.57	0.90	3.24	0.82	0.17	3.44	0.85	3.41	0.88	3.28	0.86	3.30	0.82	0.73
Female	49.5		50.0		57.8		48.9		43.6		0.55	47.2		50.0		53.7		47.2		0.89
Ever breastfed	68.2		58.06		66.7		77.8		72.6		0.14	62.3		64.8		64.8		81.1		0.14
Physical activity scoree											0.75									0.05
Tertile 1	32.2		33.9		28.9		40.0		27.4			28.3		25.9		40.7		34.0
Tertile 2	29.4		32.3		33.3		22.2		29.0			30.2		18.5		33.3		35.9
Tertile 3	38.3		33.9		37.8		37.8		43.6			41.5		55.6		25.9		30.2
Weight (kg)	15.3	3.2	14.5	3.0	15.5	3.5	16.1	3.6	15.5	2.8	0.06	15.5	2.9	16.2	3.9	15.1	2.9	14.6	3.0	0.05
In 2017	20.0	4.1	18.8	3.6	20.2	4.8	20.7	4.1	20.5	3.9	0.06	20.1	3.5	21.2	5.0	19.7	3.7	19.0	3.8	0.02
Height (cm)	94.3	9.8	91.6	9.4	95.3	10.9	96.8	10.3	94.4	8.4	0.04	95.0	8.9	96.1	11.1	93.7	9.2	92.3	9.6	0.08
In 2017	109.8	8.6	107.8	8.7	110.7	9.5	111.7	9.0	109.7	7.4	0.12	110.2	7.7	111.6	9.9	109.5	8.4	107.7	8.2	0.18
WHO BMIz (SD)	1.14	0.88	1.14	0.87	0.99	0.90	1.10	0.81	1.36	0.93	0.46	1.14	0.75	1.26	1.02	1.09	0.89	1.06	0.84	0.67
In 2017	0.70	0.98	0.49	0.93	0.58	1.05	0.70	0.76	1.00	1.05	0.022	0.70	0.76	0.90	1.28	0.63	0.95	0.58	0.82	0.35
Maternal
Overweight/obesity (BMI ≥26 kg/m2)	88.3		87.1		95.6		91.1		82.3		0.18	86.79		88.9		90.7		86.79		0.90
Years of education	12.2	2.17	11.1	2.63	12.5	1.75	12.4	1.97	12.98	1.61	<0.001	13.0	1.91	12.3	1.65	12.2	2.13	11.4	2.61	<0.001
Household
Census region											<0.001									0.34
Apia Urban Area	28.04		16.1		31.1		26.7		38.7			22.6		38.9		24.1		26.4
Northwest Upolu	34.5		27.4		24.4		46.7		40.3			32.1		25.9		42.6		37.7
Rest of Upolu	37.4		56.5		44.4		26.7		21.0			45.3		35.2		33.3		35.9
Material lifestyle scoref											<0.001									0.03
Quartile 1	22.9		40.3		20.0		22.2		8.06			7.55		22.2		33.3		28.3
Quartile 2	33.2		33.9		46.7		31.1		24.2			30.2		31.5		33.3		37.7
Quartile 3	20.09		17.7		11.1		24.4		25.8			22.6		20.4		16.7		20.8
Quartile 4	23.8		8.06		22.2		22.2		41.9			39.6		25.9		16.7		13.2

^aMeans and SDs are presented for continuous variables. Percentages for categorical variables may not sum to 100% due to rounding.

^bBased on baseline measures in 2015, unless specified otherwise.

^cDietary patterns were derived from factor analysis with the principal component method. Factor scores of each pattern in 2015 and 2017 were dichotomized as low or high according to the sample median value. The children were then categorized into four groups based on the change in dietary patterns from 2015 to 2017.

^dBased on generalized linear regression models for continuous covariates and chi-square tests for categorical covariates.

^eThe sum of items from the Netherlands Physical Activity Questionnaire for Young Children. The score is out of a maximum of 35, with higher scores indicating greater physical activity, and were classified into tertiles: 1 (mean: 23.52, SD: 4.29), 2 (mean: 29.08, SD: 0.75), and 3 (mean: 31.93, SD: 1.18).

^fThe sum of consumer durables owned (fridge, freezer, stereo, portable stereo, microwave oven, rice cooker, blender, sewing machine, television, VCR/DVD, couch, washing machine, landline telephone, computer/laptop, tablet, electric fan, air conditioner, and motor vehicle). The score is out of a maximum of 18, with higher scores indicating higher material lifestyle, and was classified into quartiles: 1 (mean: 1.22, SD: 0.80), 2 (mean: 4.07, SD: 0.83), 3 (mean: 6.40, SD: 0.49), and 4 (mean: 10.02, SD: 2.54).

BMIz, body mass index z-scores; SD, standard deviation; WHO, World Health Organization.

Children who had consistently high adherence to the modern pattern had a higher average BMIz than those who had consistently high adherence to the neotraditional pattern in 2015 (1.36 ± 0.93 SD vs. 1.06 ± 0.84 SD) and in 2017 (1.00 ± 1.05 SD vs. 0.58 ± 0.82 SD). Compared to children who had consistently low adherence to the neotraditional pattern, children who had consistently high adherence had a lower average weight in 2017 (19.0 ± 3.8 kg vs. 20.1 ± 3.5 kg). The majority of children had a mother who was overweight/obese (88.32%) and who had completed at least a high school (12.22 ± 2.17 years) education. The proportion of children in the total sample was similar across household census regions and material lifestyle score quartiles.

At age 4–7 years in 2017, children who consistently had a high adherence to the modern pattern during the prior 2 years had less of a decrease in BMIz from age 2–4 years in 2015 (−0.26 ± 0.96 SD) than those who had consistently low adherence to the modern pattern (−0.66 ± 0.68 SD) (Table 2). In comparison to children who consistently had low adherence to the neotraditional pattern, children with high adherence had a similar change in the magnitude of BMIz between 2015 and 2017. Consistently high adherence to the modern pattern was associated with higher maternal education and socioeconomic status, and living in the urban Apia Urban Area region compared to consistently low adherence, whereas, consistently high adherence to the neotraditional pattern was related to lower child physical activity score, maternal educational attainment, and socioeconomic status.

After adjusting for potential confounders, including BMIz in 2015 and total energy intake, the average BMIz in 2017 was 0.34 SD higher among children who had consistently high adherence to the modern pattern compared to those who had consistently low adherence (95% confidence interval [CI]: 0.03–0.65, p = 0.03) (Table 3). Consistently high adherence to the neotraditional pattern was associated with a 0.11 SD lower adjusted BMIz in 2017 compared to consistently low adherence (95% CI: −0.43 to 0.20, p = 0.48).

Table 3.

Associations between the 2-Year Change in Modern and Neotraditional Dietary Patterns and World Health Organization z-Scores for Body Mass Index, Weight, and Height among 214 Samoan Children^a

Two-year change	WHO BMI z-score				Weight z-score		Height z-score
	2017		Change between 2015 and 2017		Change between 2015 and 2017		Change between 2015 and 2017
	Crude	Adjustedb	Crude	Adjustedb	Crude	Adjustedb	Crude	Adjustedb
	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)	β (95% CI)
Modern pattern
Consistently low	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Low to high	0.10 (−0.28 to 0.47)	0.12 (−0.19 to 0.42)	0.25 (−0.06 to 0.55)	0.20 (−0.12 to 0.52)	−0.01 (−0.22 to 0.21)	0.05 (−0.18 to 0.28)	−0.25* (−0.52 to 0.02)	−0.10 (−0.37 to 0.18)
High to low	0.22 (−0.15 to 0.59)	0.10 (−0.21 to 0.41)	0.25* (−0.05 to 0.56)	0.13 (−0.19 to 0.46)	−0.05 (−0.27 to 0.16)	0.00 (−0.23 to 0.24)	−0.34** (−0.61- −0.07)	−0.12 (−0.40 to 0.16)
Consistently high	0.52*** (0.18 to 0.86)	0.34** (0.03 to 0.65)	0.39*** (0.12 to 0.67)	0.36** (0.04 to 0.69)	0.03 (−0.17 to 0.23)	0.13 (−0.10 to 0.37)	−0.32** (−0.57- −0.07)	−0.15 (−0.43 to 0.13)
Neotraditional pattern
Consistently low	Ref	Ref	Ref	Ref	Ref	Ref	Ref	Ref
Low to high	0.19 (−0.18 to 0.57)	0.02 (−0.28 to 0.31)	0.07 (−0.23 to 0.38)	−0.028 (−0.34 to 0.29)	0.06 (−0.15 to 0.27)	0.02 (−0.21 to 0.24)	0.04 (−0.24 to 0.31)	0.09 (−0.18 to 0.35)
High to low	−0.07 (−0.44 to 0.30)	−0.08 (−0.39 to 0.23)	−0.03 (−0.33 to 0.28)	−0.10 (−0.43 to 0.22)	0.02 (−0.19 to 0.24)	−0.04 (−0.27 to 0.19)	0.07 (−0.20 to 0.35)	0.06 (−0.21 to 0.34)
Consistently high	−0.13 (−0.50 to 0.25)	−0.11 (−0.43 to 0.20)	−0.05 (−0.36 to 0.25)	−0.12 (−0.45 to 0.21)	0.00 (−0.22 to 0.21)	−0.08 (−0.32 to 0.16)	0.05 (−0.23 to 0.33)	0.01 (−0.27 to 0.29)

^*p < 0.10, ^**p < 0.05, ^***p < 0.01, ^****p < 0.001 (compared to “consistently low”).

^aFactor scores of the dietary patterns in 2015 and 2017 were dichotomized as low or high according to the sample median value. The children were then categorized into four groups based on the change in dietary patterns from 2015 to 2017. Values are beta coefficients and 95% CI for the BMI-z scores based on the WHO child growth standard and reference groups (WHO BMI z-score) and for the BMI z-score internally consistent with the sample mean (Internal BMI z-score) in 2017 and the 2-year change between 2015 and 2017.

^bAdjusted for child age, sex, total energy intake in 2015 and 2017 based on food frequency questionnaire data, ever breastfed (yes or no), physical activity score tertiles, maternal overweight/obesity (yes or no), years of maternal education, household census region (Apia Urban Area, Northwest Upolu, and Rest of Upolu), and household material lifestyle score quartiles. The covariates measured at baseline in 2015 were used, except for total energy intake.

CI, confidence interval.

Consistently high adherence to the modern pattern was associated with less of a decrease in 2-year change in BMIz (least-square mean: −0.35 SD (95% CI: −0.59 to −0.11)) compared to consistently low adherence (least-square mean: −0.71 SD (95% CI: −0.96 to −0.46)), which resulted in a positive difference between the consistently high and low adherence groups (adjusted 2-year change in BMIz: β: 0.36 SD, 95% CI: 0.04–0.69, p = 0.03). Children who had consistently high adherence to the neotraditional pattern had a similar 2-year change in BMIz compared to low adherence (adjusted β: −0.12 SD, 95% CI: −0.45 to 0.21, p = 0.48). On average, the change in BMIz was −0.59 SD (95% CI: −0.85 to −0.32) and −0.47 SD (95% CI: −0.73 to −0.21) in consistently high and low adherence groups, respectively.

After adjusting for potential confounders, the average change in z-score for weight was 0.13 SD higher and the change in z-score for height was 0.15 SD lower for children who had consistently high adherence to the modern pattern compared to consistently low adherence; however, the estimates were imprecise (95% CI for weight z-score: −0.10 to 0.37 and 95% CI for height z-score: −0.43 to 0.13) The adjusted 2-year changes in z-scores for weight and height were also similar between children who had consistently high and low adherence to the neotraditional dietary pattern (p > 0.05).

Discussion

This study provides the first prospective evidence that consistent consumption of a “modern” diet, among Samoan children, promotes BMIz gain over a 2-year period. This suggests that the Pacific-wide focus on eating a variety of traditional food and local produce, rather than the high-fat, micronutrient-poor foods that characterize the modern dietary pattern, is justified to help promote optimal growth and weight maintenance. These notable associations exist independent of total energy intake, suggesting potential benefits of eating a variety of foods within the neotraditional pattern for childhood growth and consequences of adhering to a modern pattern beyond the amount of energy in food consumed.

Recommendations and strategies to promote a neotraditional diet rich in local produce have thus far not been based on empirical evidence gathered in this setting. To our knowledge, this is the first study to describe prospective associations of dietary patterns with BMIz change among children in Samoa and more widely, across Pacific Island countries. While not directly comparable, our data align with results from the Pacific Island Family study in New Zealand. At age 4 years, eating a diet high in fruits and vegetables was associated with a lower BMI and a lower 4-year weight gain.²⁷ Conversely, consumption of foods high in protein (eggs, meat, poultry, and fish) and dairy (milk, yogurt, and ice cream) were associated with higher BMI and 4-year weight gain.²⁷

In thinking about protecting well-being across the life course, the neotraditional pattern appears favorable because it includes foods that provide the diverse combination of micronutrients necessary for healthy growth and development. A typical modern pattern consists of foods high in refined sugar, white flour, saturated fats, salt, and numerous food additives, which may predispose children toward a greater risk of both obesity and micronutrient deficiencies.^28–30 On a population level, eating local produce helps to preserve culture, reduces food insecurity, boosts the agricultural economy, and benefits the environment and climate.^31,32

Interestingly, we observed some children who had a low-to-high and high-to-low adherence to both the modern and neotraditional patterns between 2015 and 2017. These dietary pattern changes may be partly explained by rapid economic development and the nutritional transitions toward increased access to and availability of the unprocessed meats, refined grains, and highly processed foods found in the modern pattern.^3,33 For many Pacific Island countries, agricultural and food production have slowed to the point at which subsistence farming cannot supply enough food to satisfy the needs of families at prices that are competitive with highly processed, imported foods.³⁴ With increasing dependence on food imports and food advertising proliferating in Samoa, more opportunities to adopt a modern pattern exist for children whose preferences are mostly based on taste, palatability, and the availability of foods. Acute exposure to food advertising has been shown to increase food intake and alter dietary behaviors in children,³⁵ but not adults.⁹

Alternatively, changes in adherence to dietary patterns may reflect changes in preferences of children as they age and transition from the home to school. Recent national policies and public health efforts likely shape the food environments in village communities and schools, and influence eating behaviors. In partnership with various organizations, the Samoa Ministry of Health leads many efforts to promote local-grown fresh foods, like the “Eat the Rainbow” initiative, which may encourage children to follow a neotraditional pattern.³⁶

The finding that the change in BMIz was similar among children who had consistently high and consistently low adherence to the neotraditional dietary pattern over 2 years contradicts our previous cross-sectional finding of a positive association with overweight/obesity in this cohort at age 2–4 years.¹⁴ Aside from acknowledging the limitations of our cross-sectional analysis in that publication, we noted the potential threat of reverse causality; that recognition of child overweight/obesity may have prompted mothers to make dietary changes or to give responses perceived to be more socially desirable to questions about child dietary intake.¹⁴ We believe that the analysis in this study addresses these key limitations.

While BMIz in 2017 and change between 2015 and 2017 were similar across neotraditional pattern adherence groups, we observed a lower BMIz compared to those with consistent adherence to the modern pattern and a potential trend toward lower BMIz in 2017 among those consistently adhering to the neotraditional pattern. The lower BMIz may reflect greater increases in height relative to weight and despite similar changes in z-scores for weight and height across the neotraditional pattern adherence groups in the sample, we observed a potential trend toward greater weight and lower height among children who had consistently high adherence to the modern pattern compared to low adherence. We believe that this still justifies continued promotion of the micronutrient-rich, local foods that comprise the neotraditional pattern.

We are currently completing a third follow-up of the children from the Ola Tuputupua'e cohort, which will allow us to continue observing differences emerging between those consuming modern and neotraditional diets. It is worthwhile to note here that the timing of this study spanned the years during which the second rise/rebound in adiposity may occur.^37,38 The timing of the adiposity rebound in the Samoan population is poorly understood and may have contributed to some of the outcomes we observed here in a way we could not account for. With the addition of the planned further time points in this cohort, we will be well equipped to determine the timing of the adiposity rebound and model growth curves to further investigate how changes in dietary patterns are related to changes in adiposity.

The current findings should be interpreted in the context of additional limitations. First, as reported previously, our convenience sample is not nationally representative, with a lower proportion of mothers who completed secondary education and lower socioeconomic status based on ownership of household consumer durables compared to the general population.^11,39 Second, factor analysis has limitations, most notably the subjectivity in determining the number of patterns to retain, and in naming and interpreting the retained patterns. Since we derived the dietary patterns in 2015 and 2017 separately, the dietary patterns at the two time points do not share identical food group compositions and the continuous factor scores were not directly comparable across time. To overcome the potential lack of comparability in the continuous factor scores, we used sample medians at each time point to capture the relative standings of the constructs of modern and neotraditional diets and then categorized the changes in dietary patterns.²⁰

Third, with two time points and our fixed sample size, a post hoc power calculation confirms that we are underpowered to investigate the relationship between changes in dietary patterns and childhood BMIz (data not shown). Even with this cohort data, there is no prior information available in Samoa to conduct a priori power calculations to determine what level of differences we might expect for this analysis. Many aspects of our work are exploratory and designed to generate new hypotheses. Finally, despite adjustment for various potential confounders in statistical models, the assessment of the change in BMIz outcome and the change in dietary pattern exposure occurred in the same time frame, and thus, we are unable to distinguish temporality between the changes in dietary patterns and BMIz and cannot infer causality.

There are, however, notable strengths. Compared to cross-sectional analyses, the methodology used in this study allows us to maximize the contemporary data as collected in the Ola Tuputupua'e cohort. The FFQ used captures the “usual” diet, acknowledging day-to-day variation in intake among children and was validated and adapted from previous adult studies in Samoa.^17,18

Using the dietary pattern approach rather than nutritional intake, we were able to provide a more comprehensive overview of the diet and better explain obesity risk. For the dietary pattern factor analysis, the food groups were used as continuous variables and this allowed us to include all the food groups identified in 2015 and 2017, and also capture the variation in consumption at each time point (especially if the vast majority of the children ate a food group at least once a day). We assumed that the underlying dietary patterns at the population would not be identical at each time point as children age between 2–4 years in 2015 and 4–7 years in 2017; thus, we captured the “modern” and neotraditional’ dietary patterns by using the median as the cutoff at each time point to identify the relative standings of each child over time.

The data also build upon the existing cross-sectional findings in 2015 and maintained the “modern” and “neotraditional” pattern categories at each time point to enhance the interpretability of the results to inform nutrition guidelines and public health strategies in Samoa.¹⁴ Moreover, the data provide empirical support for current public health messages, align with the current efforts of Pacific Island leaders to understand early determinants of childhood obesity, and may be useful in informing future interventions and policies.

In conclusion, we first report that consistent 2-year adherence to a modern dietary pattern in children was associated with greater BMIz gains between the ages of 2–7 years, while adherence to a neotraditional pattern was associated with no change BMIz over this period in Samoa. We hope that this evidence will accelerate the national adoption of the Secretariat of the Pacific Community's guidelines^13,40 and inform future recommendations and strategies to promote healthy diet and lifestyle for communities in Samoa and populations across the Western Pacific region.

Authors' Contributions

C.C.C., D.W., and N.L.H. conceived this study and drafted the initial article. C.C.C. and N.L.H. conducted the research with the support and assistance of T.N., C.S.-U., M.S.R., and R.L.D.. C.C.C. and D.W. completed study analyses under the supervision of N.L.H. and A.B., and with T.N. and C.S.-U., interpreted the data. All authors critically reviewed this article and gave their approval for submission and publication.

Funding Information

The Ola Tuputupua'e study received financial support from the following sources: Yale School of Public Health (Faculty Funding, David Dull Internship Fund, Jan A.J. Stolwijk Fellowship Fund, Yale Downs International Health Student Travel Fellowship, Thomas C. Barry Travel Fellowship), US National Institutes of Health (NIH) Minority and Health Disparities International Research Training Program (NIMHD T37MD008655), U.S. Fulbright Graduate Student Research Fellowship, Brown University School of Public Health (International Health Institute, Nora Kahn Award, and Framework in Global Health Program), and Brown University Population Studies and Training Center, which receives funding from the NIH for training (T32 HD007338) and general support (P2C HD041020). CCC was also supported by a training grant from the NIH National Heart, Lung and Blood Institute (F31HL147414-01) and the Fogarty Global Health Equity Scholars Program (FIC D43TW010540).

Author Disclosure Statement

No competing financial interests exist.

Supplementary Material

Supplementary Table S1