Obesity in Canada

[Previous Page] [Table of Contents] [Next Page]

Appendix 3. Impact of behaviour and socioeconomic factors on the prevalence of obesity in the population: summary of methodology

Data sources and measures²⁰³

For this study, CCHS cycles 2000/01, 2003 and 2005 were pooled to obtain a total sample size of 283,097 adults aged 18 and over. This method combines the data at the record level, and the resulting file is treated as a sample from one large ”average” population from 2000 to 2005. Compared with other years, the method of data collection for the 2000/01 cycles was more often in person than by telephone.²⁰⁴

Body mass index (BMI) was computed according to self-reported height and weight as:

BMI = Weight (kg)/Height (m)²

BMI was grouped according to internationally accepted classifications as follows:⁷

• normal weight : BMI = 18.5 kg/m²-24.9 kg/m²
• overweight: BMI = 25 kg/m²-29.9 kg/m²
  • overweight I: BMI = 25 kg/m²-27.4 kg/m²
  • overweight II: BMI = 27.5 kg/m²-29.9 kg/m²
• obese: BMI ≥ 30 kg/m²

As this study focused exclusively on the risk of adult overweight and obesity, 7,322 participants (5,955 females and 1,367 males) with a BMI less than 18.5 kg/m² were excluded from the final study sample.

Explanatory variables of interest were set up as dichotomous variables and included the following: single/separated/divorced (Yes or No), total household income (lowest income quintile vs. highest income quintile), immigrant status (Y/N), visible minority status (Y/N), rural residence (Y/N), daily smoking (Y/N), physical inactivity (Y/N), low fruit and vegetable intake (Y/N) and high alcohol consumption (Y/N). See Table 7 for descriptive statistics.

Analysis

For all variables descriptive estimates, including proportions and 95% confidence intervals, were calculated separately by sex.

Poisson models were used to evaluate the simultaneous contribution of demographic, social and behavioural risk factors in the prediction of adult overweight and obesity. The Poisson distribution is used here to estimate dichotomous outcomes, since logistic functions tend to overestimate the cross-sectional prevalence of common diseases.^205-207 An additional benefit of using the Poisson model is found in the production of prevalence ratios, which better approximate the relative risk than do odds ratios. All analyses used bootstrap and probability weights rescaled to the 2001 Canadian population and were estimated using the BSWREG re-sampling procedure.

Population attributable risks (PAR) were derived from adjusted risk ratios (RR) using a conservative equation for potential confounding:²⁰⁸

Recently, the population impact number (PIN) has been proposed as a new measure to quantify and communicate the population burden of a risk factor – or, conversely, the potential number of disease events that may be prevented in a population through elimination of that risk factor – in a way that is easily conceptualized by policy-makers and the general public.^209-210 This measure may be applied to resource planning and the evaluation of public health interventions. The following formula for PIN, applicable to cross-sectional designs,⁸⁷ was used:

The computation of confidence intervals for PAR and PIN was based on a conservative Bonferroni inequality method.²¹¹ Stata 9.2 (Stata Corp, Texas Station) was used exclusively to conduct the analyses.

It should be noted that this study used nationally representative cross-sectional data to estimate the contribution of various factors to the population burden of overweight and obesity. The analysis does not explicitly define a cause-effect relation between the predictor variables (e.g., immigrant status, fruit and vegetable intake) and the outcome of interest (e.g., overweight or obesity). Rather, as with other studies that have similarly used cross-sectional data to explore the potential population impact of eliminating a risk factor of interest, the interpretation of present findings should be considered as suggestive of a relation between the two variables, after controlling for other covariates.⁸⁷ Additional studies using longitudinal, nationally representative Canadian data should be used to confirm the results described here.

Descriptive statistics

Table 7: Sample Size, Demographic Characteristics and Prevalence of Social Determinant and Health Behaviour Risk Factors by Sex, Canada 2000/01, 2003, 2005

	Females	Males
Sample size (N)	148,844	134,253
Weighted sample (N)	10,217,995	10,568,461
Age (mean, 95% confidence interval [CI])	44.4 (44.3-44.5)	43.6 (43.5-43.7)
Lowest income (%, 95% CI)	23.6 (23.2-24.0)	17.4 (17.1-17.7)
Middle income (%, 95% CI)	61.8 (61.4-62.3)	63.4 (62.9-63.8)
Highest Income (%, 95% CI)	14.6 (14.2-15.0)	19.2 (18.8-19.6)
Immigrant (%, 95% CI)	21.7 (21.3-22.1)	21.9 (21.5-22.3)
Visible minority (%, 95% CI)	14.2 (13.8-14.5)	14.5 (14.1-14.8)
Single/separated/divorced (%, 95% CI)	33.8 (33.4-34.2)	31.3 (30.9-31.7)
Rural residence (%, 95% CI)	17.7 (17.4-18.0)	18.8 (18.6-19.1)
Daily smoking (%, 95% CI)	23.9 (23.6-24.3)	28.3 (27.9-28.6)
Physical inactivity (%, 95% CI)	53.1 (52.7-53.5)	49.2 (48.8-49.7)
High alcohol use (%, 95% CI)	12.1 (11.8-12.4)	31.2 (30.8-31.6)
Less than 5 servings fruit & vegetable (%, 95% CI)	54.1 (53.7-54.6)	67.8 (67.4-68.3)
Normal weight (18.5 ≤ BMI < 25) (%, 95% CI)	54.0 (53.6-54.4)	40.7 (40.3-41.2)
Overweight I (25 ≤ BMI < 27.5) (%, 95% CI)	16.5 (16.2-16.8)	24.3 (23.9-24.6)
Overweight II (27.5 ≤ BMI < 30) (%, 95% CI)	10.7 (10.5-10.9)	17.2 (16.9-17.5)
Obesity (BMI ≥ 30) (%, 95% CI)	14.8 (14.5-15.1)	16.8 (16.5-17.1)

Source: R. Hawes and P. Stewart, unpublished manuscript prepared for the Public Health Agency of Canada; based on analysis of pooled 2000/01, 2003, and 2005 Canadian Community Health Surveys, Statistics Canada.

[Previous Page] [Table of Contents] [Next Page]