Study population and location
Details of the study population and methodology have been presented elsewhere [2, 3]. The study population consisted of Inuit adults aged 18 years or older who resided in northern Canada and were included in the International Polar Year Inuit Health Survey in 2007 and 2008. The survey took place in the late summer and fall of 2007 and 2008 in 33 coastal communities through the use of the Canadian Coast Guard Ship Amundsen, and in 3 non-coastal communities using land teams. The communities included all communities in three regions of northern Canada: Nunavut, Nunatsiavut, and the Inuvialuit Settlement Region of the Northwest Territories. Homes were assigned numbers and randomly selected using a computerized random digit assignment or table. Inuit adults in randomly selected homes were invited to participate. Pregnant women were not eligible to participate. Research personnel included trained bilingual interviewers, community research assistants, nurses, lab and other technicians, and quality control staff. Consent forms, an informative DVD, and all questionnaires were in Inuit languages and dialects and English. The study was developed in a participatory approach with members of the steering committees of 3 jurisdictions: Inuvialuit Settlement Region, Nunavut Territory, and Nunatsiavut Region, the new land claim area in northern Labrador. The McGill Faculty of Medicine Institutional Review Board issued a certificate of Ethical Acceptability. Informed consent was obtained from each participant.
The cross-sectional study for adults consisted of interviewer-administered questionnaires. The International Physical Activity Questionnaire (IPAQ) short form was used to assess participants’ physical activity during the previous seven days  in which metabolic equivalent (MET) intensity levels were taken from the comprehensive compendium of physical acivity . In an earlier analyses of the current study population, walking MET scores were found to be significantly associated with an at-risk BMI whereas MET scores for moderate and vigorous-intensity activity were not found to be significantly related with an at-risk BMI  possibly reflecting inadequacies of the short form 7-day IPAQ for capturing habitual intake or participant reporting errors. IPAQ short form was chosen for the current study due to community concerns regarding research burden to participants.
Interviewers were trained in administering the 24-hour recall based on an adaptation of 5-stage multiple-pass recall interviewing technique developed by the United States Department of Agriculture (USDA) designed to maximize an individual’s recollection of foods consumed the previous day, beginning at midnight (00:01) and ending at midnight (24:00) [13, 14].The recall queries involved a quick list, time and occasion, forgotten foods and a detailed cycle review. Description of amounts and portion sizes were aided by portion size model kits (Santé Québec). Dietary recall data were entered using CANDAT Software (Godin London) and nutrient composition of foods was determined using the 2007b Canadian Nutrient File  with supplemental information on nutrient composition from an in house (School of Dietetics and Human Nutrition) food file for foods not represented in the Canadian Nutrient File . Data entry of recalls was double verified. When recalls did not include recipes for mixed dishes, northern recipe default values obtained through regional contacts and detailed dietary recalls were used. A total of 74 dietary recalls were invalidated due to either insufficient information (n = 28); participant deciding to not continue with the recall (n = 17); or misunderstanding regarding the length of fasting (n = 29) (i.e., participants were requested to fast for eight hours).
Intakes of nutrients and total energy were estimated using data collected by the 24 hour recall. Due to logistical constraints a repeat dietary recall was not collected. As food frequency questionnaire data were available for past-year traditional food intake and past-month beverage consumption, we evaluated correlations between the 24-hr traditional food and high-sugar drink data with that of the FFQ data. The correlation coefficient comparing high-sugar drink consumption noted in the 24-hr recall and the past-month beverage FFQ was statistically significantly related with an r of 0.45, P < 0.0001. For traditional food consumption, the correlation coefficient between the past 24-hr and the past-year FFQ intake data was 0.30 (P < 0.001).
Trained nurses measured weight using a Tanita instrument (Tanita Corporation Tokyo, Japan) and height using a stadiometer. Participants were asked to remove shoes and socks and a standard clothing weight was subtracted (0.4 kg) for all participants.
Coding of variables
Walking was based on calculated MET scores using an energy requirement of 3.3 and the continuous variable used in analyses. Obesity was defined as BMI ≥ 30 and overweight as BMI ≥ 25 and < 30 kg/m2 and an at-risk BMI was defined as BMI ≥ 25 kg/m2.
Dietary factors were evaluated as nutrient densities (nutrient in kcalories/total kcalories) and total kilocalories was also entered as an additional covariate. High-sugar drinks were defined as sodas, juices, sports drinks, and punches containing >25% of total energy as sugar and excluded diet sodas and real fruit juices. High-sugar drink intake was not normally distributed with approximately 30% consuming no high-sugar beverages, thus high-sugar drink consumption was divided into three categories based on approximate tertiles: non-consumer, low (1–15.5% of E), and high (>15.5% of E) consumer, where non-consumers were used as the reference category. Likewise, a high percent consumed no soft drinks, thus this variable was evaluated as percent soft drink consumers (yes vs no) in which diet soda consumers were excluded.
High-fat foods were defined as foods containing >40% of total energy as fat and included both market food and traditional food sources of fat. The traditional food variable was composed of traditional foods from the predominantly meat-based Inuit diet, which included various sea mammals, land animals, fish, birds, and plants and excluded any beverages or bannock. Percent of total energy intake from saturated fatty acids (SFA) included both traditional and market food sources of saturated fatty acids. To calculate the potential underreporting of dietary intake, the ratio of reported energy intake (EI) to basal metabolic rate (BMR) was calculated using the FAO/WHO/UNU equation and a value below 1.5 was used to identify underreporting .
A Canadian version of the USDA Healthy Eating Index (HEI) was used in this study based on Canadian age and sex specific dietary recommendations (maximum score = 100). HEI scores were calculated in this study and compared among overweight/obese versus normal weight individuals by gender. The HEI assesses how diet adheres to dietary recommendations in which the Canadian modifications are based upon Canada’s Food Guide to Healthy Eating and the Nutrition Recommendations for Canadians. The Canadian HEI has nine sub scores each worth a maximum of 10 points, with the exception of the fruits and vegetables sub score which is worth 20 points, for a total maximum of 100 points. The Canadian HEI measures diet quality by assessing number of servings from the main food groups, intake of dietary fats, sodium, as well as diet variety .
Socioeconomic (SES) characteristics were similar between overweight and obese participants . Also, dietary characteristics did not differ between overweight and obese individuals in multivariable logistic regression. Thus, the overweight and obese groups were combined into one at-risk category for analyses. Logistic regression analyses were performed using at-risk BMI (BMI > =25 kg/m2 ) as the dependent variable and each nutrient was evaluated separately in models as the independent variable. With the exception of fiber and HEI, all models used the nutrient density approach (% E); total kilocalories was entered as a covariate. In the evaluation of dietary associations with an at-risk BMI, we did not control for SES indicators because SES is a strong determinant of diet quality and one means by which SES may contribute to obesity. Dietary analyses adjusted for age, sex, survey region (Nunavut, Nunatsiavut Region, and Inuvialuit Settlement Region), walking (continuous IPAQ MET score), current smoking, and past-year alcohol consumption (yes vs no). Additional multivariable modeling adjusted for the aforementioned variables in addition to important nutrients. Cluster analyses were used for logistic regression modeling since there were, on average, 1.38 participants per home. All statistical analyses were performed using STATA version 10.0 statistical software package. Statistical significance was declared at P < 0.05.