Our outcome was obesity defined as BMI ≥95^th percentile (age and sex specific) at age 3 years and was calculated from the child’s height and weight as self-reported by the mother on CUBS. All non-obese children, including those who were underweight (5.2% of non-Native children and 3.5% of Alaska Native children), were compared to obese children.

We analyzed obesity by maternal Alaska Native vs. non-Native race. The terms “Alaska Native children” and “non-Native children” are based on maternal race only, as reported on the birth certificate. Risk factors for childhood obesity were assessed using t-tests for continuous variables and chi-square tests for categorical variables. Pre-pregnancy, prenatal, postnatal (from 2–8 months postpartum), and childhood predictors at age 3 years were included for evaluation of risk for obesity. Variables that were significant at p<0.05 in bivariate analysis were included in the multivariate model. When variables were highly correlated with one another with r>0.5 (such as prenatal and postnatal maternal cigarette smoking, income status from the time of the birth of the child and at 3 years of age, and location of residence and urban/rural) we included in multivariate analysis either the variable more strongly associated with obesity in bivariate analysis or the variable most important to elucidate obesity trends in the state of Alaska (region of residence).

Variables that were known from previous studies to be associated with increased risk for obesity in early childhood were also included in the multivariate model process including maternal pre-pregnancy weight category (obese versus non-obese), breastfeeding duration (continuous), maternal age (continuous) and consumption of sugar-sweetened beverages including 100% fruit juice. Beverage frequency variables were analyzed both ordinally (how the questions were originally asked on the CUBS questionnaire), as well as dichotomously, with the cut point used corresponding to the approximate mean or median of the variable for the population. The cut point for beverages was either ≥1 cup/day or >1 cup/day. Food consumption questions were analyzed dichotomously with the cut point similarly corresponding to the approximate mean, with cut-points reflective of the number of times per day the item was consumed. All food and beverage questions were based on questions that asked about consumption yesterday with the exception of the milk consumption question, which asked about usual consumption.

Because of the high percentage of missing data for paternal education (10.5%), we did not include paternal education in the multivariate models even though they were significant in bivariate <0.05 for Alaska Native children. Due to a high percentage of respondents who were missing data on child BMI (16.5%), we compared the percentage of selective socio-demographic factors (maternal age, paternal education, maternal education and location of residence) in those who were missing for BMI compared with non-missing participants (for BMI) (as 16.5% was also missing for BMI).

Subsequently, an analysis was conducted to compare daily consumption of milk, 100% fruit juice, soda and SSB, as well as usual type of milk child drank, based on regional residence (Anchorage, Interior, Northern/Southwest and Gulf Coast/Southeast). This analysis was conducted so as to better understand the protective association found between fruit juice consumption and obesity in multivariate analysis for Alaska Native children only and not non-Native children. Residence was obtained from the mother’s address at the time she completed the CUBS survey, and regional groups were a modified version of those defined by the Alaska Department of Labor (http://labor.alaska.gov/research/index.htm)—Northern and Southwest regions, and Gulf Coast and Southeast regions, respectively, were combined due to small numbers and because these regions are geographically contiguous. We also evaluated associations between 100% fruit juice consumption and consumption of other beverages and food using chi-squared tests to better understand the protective relationship between 100% fruit juice consumption and obesity. Additionally, we examined differences in bivariate associations for non-Native and Alaska Native children using an interaction term between the predictor of interest and being either Native or Non-Native and child obesity at age 3 in logistic models. All data analysis used Stata 13.0 using “svy” statements to allow for the sampling weights provided by PRAMS. Sampling weights are necessary in survey analysis to get the survey weighting, clustering and stratification accounted for so as to get standard errors of estimates correct. Data are expressed as means ±standard errors (SE). Specific details about the PRAMS and CUBS variables used in the analysis are included as a supplement to this article (S1 Appendix: Additional Methodological Details on PRAMS and CUBS Variables).

This was primarily a hypothesis-generating study and we did not do any multiple comparisons to reduce the chance of Type II errors [16].

Similar to other studies that have focused on early life exposures and risk for obesity in the preschool years [19,20], we found increased breastfeeding duration was associated with lower odds of obesity. In Alaska Native children, however, the relationship between obesity and duration of breastfeeding was no longer statistically significant controlling for other confounders in the multivariate model. It is possible that we lacked statistical power in the multivariate model of Alaska Native children to demonstrate the association of obesity with early termination of breast-feeding found in studies of American Indian children [21]. An alternative hypothesis is that the higher prevalence of obesity in Alaska Native compared to non-Native children results from unmeasured differences in eating patterns and that these may overwhelm any potential benefit from breastfeeding. While we investigated some feeding practices (e.g. fried potato and beverage consumption patterns), future studies should conduct more detailed dietary analysis of early feeding practices in Alaskan families and could consider the contribution of traditional foods as well as non-traditional or modern foods.

We also evaluated the role of exposure to domestic violence and other forms of violence on odds for obesity in early childhood. Our study found an association for a child witnessing abuse or violence in person prior to age 3 years among Alaska Native children in bivariate and multivariate analysis. Other studies have found an association between exposure to chronic intimate partner violence and risk of later childhood obesity at age 5 years [22, 23], including a study, which found that boys exposed to domestic violence before age 5 years, were twice as likely to be obese as adolescents [24]. Exposure to violence may directly impact obesity by causing depression, hopelessness, and subsequently lack of exercise or interest in overall health [25, 26]. Alternatively, this may be a marker for the type of food choices available in that house, the possible presence of fast foods and the overall level of disorder in a child’s life. Future obesity interventions with Alaska Native children, in addition to focusing on diet and exercise alone, may take into consideration the psychosocial context of the family so as to potentially reduce obesity risk.

Among Alaska Native and non-Native children, we found that pre-pregnancy maternal obesity was associated with significantly increased odds for obesity at age 3 years in bivariate analysis although the relationship was attenuated in multivariate and non longer reached statistical significance in Alaska Natives but maintained statistical significance in non-Native children. Previous studies have found that maternal obesity is a significant risk factor for obesity in young children [27, 28]. Our study may have been underpowered to assess any difference based on maternal BMI pre-pregnancy category in Alaska Natives. Addressing maternal obesity prior to pregnancy and even during pregnancy can have inter-generational benefits in terms of reducing overall odds of obesity.

In bivariate analysis, increased consumption of 100% fruit juice was associated with obesity at age 3 in Alaska non-Natives but protective against obesity in Alaska Natives with similar results in multivariate analysis for Alaskan Native children. It is not clear why 100% fruit juice consumption was protective against obesity in Alaska Native children, although there may have been unmeasured confounding factors or interaction with region of residence that explain this relationship. We did find that increased 100% fruit juice consumption was associated with higher reported intake of fruits and vegetables in Alaska Native children but not non-Natives. It is possible that in this population the consumption of fruit juice may be a marker for reduced caloric intake of energy dense, low nutrient value foods. Increased consumption of fruits and vegetables is associated with reduced risk for obesity in adults, through potentially healthy overall lifestyles [29].

Our findings in Alaska non-Natives are consistent with previous studies that have found increased odds of childhood obesity with higher liquid calorie intake, including 100% fruit juice [30, 11]. On further investigation of the contribution of all energy-dense beverages (including 100% fruit juice), we found a higher frequency of consumption of all beverages in the Northern/Southwest region of the state where obesity prevalence was highest in Alaska Natives, which doesn’t correspond with our finding of a protective association between 100% fruit juice and obesity risk at age 3. Increased energy dense beverage consumption occurred among both obese and non-obese Alaska Native children from this region.

Previous studies suggest that all children [30], including those among indigenous populations [31] should avoid excessive consumption of energy dense, low nutrient value beverages and foods replace these with water and other healthier options. Consumption of water has been found to increase the resting energy expenditure in overweight and obese children [32]. Unfortunately, the Northern/Southwest region has a large number of villages that lack piped water to households [33, 34], and many residents rely on community wells or relatively expensive bottled water.

We found a significance increase in risk for obesity among non-Native Alaska children based on low-income status in the 12 months prior to birth. Previous studies in high risk populations have found an association between low household income and increased risk for childhood obesity [35, 36] and studies have warned that interventions that simply counsel individuals on which food types or physical activity regimens are appropriate will not be sustainable in communities that do not have access to these types of foods/beverages or opportunities for physical activity due to poverty [37]. The role of income and family livelihood likely permeates many risk factors for obesity including types of foods and beverages consumed. It is possible we did not find any independent association in Alaska Native children for income status after adjusting for other factors including WIC participation and location of residence, because the distribution of income is narrower in Alaska Native families with most families having a low income at birth (mean for Alaska Natives $22,993.90±814.35 versus $35,657±762.65 for non-Natives) and at 3 years of age (mean for Alaska Natives $30,207.72±1237.16 versus $49,72.34±1145.80 for non-Natives).

Limitations of our study include recall bias (questions were asked retrospectively, both in the postnatal period for PRAMS and for CUBS) as well as self-report and mode bias—some mothers may report certain behaviors more or less willingly than other mothers based on their experiences or they may be more or less willing to disclose certain behaviors during a mail survey [38, 39]. Our outcome of childhood obesity was based on self-reported measurements by the mother and previous studies have found some misclassification of child weight category based on parental report, with under estimates of child heights resulting in upward bias in obesity prevalence estimates [40, 41]. Some factors that we examined had missing values, including 16.5% of respondents missing information on the outcome of childhood BMI and 10.5% missing information on paternal education. When we compared percentages of socio-demographics between those who were missing BMI measurements and those that were not there were no differences in paternal education or child sex, but those who were missing BMI measurements were more likely to have less maternal education, live in rural areas and the Northern/Southwest part of the state and be Alaska Native. There may have been systematic error introduced to our models due to missing information for non-respondents to both surveys (although the analysis weights did attempt to incorporate non-response), and individual question non-response. The number of respondents to both surveys was small, limiting the statistical power of our study to identify significant associations. As both surveys are on-going, further analysis using additional years of data could help to address some of these limitations.

Our sample size to compare differences in risk factors between non-Native and Alaska Native was also small so may not have had sufficient power to detect differences in many cases. Also, as we preformed many statistical tests and many of these statistical associations would be diminished or disappear after multiple comparison adjustment, some associations may have been found by chance and should be replicated in larger studies. As this was a hypothesis generating study with a limited sample size, we chose not to do multiple comparison testing [16], but follow-up studies with larger sample sizes conducted in Alaska that test a high number of predictors should use multiple comparison testing.