Problem Formulation

  • Health status is often categorized based on body mass index as: healthy weight (18.5 - 24.9), overweight (25.0 - 29.9), and obese (30.0 or higher) using international standard for adults.
  • Obesity is a chronic health condition, and associated with an arrey of health consequences including physical, social, and emotional at individual level.
  • Research also lacks an explicit focus on how unhealthy risk varies by geographical contexts.
  • Our objective is to investigate the variation in overweight and obesity risks adjusted for personal characteristics across different health regions in Canada.

Data

  • Aggregated data for trend analysis was downloaded from Statistics Canada website.
  • Individual-level data came from Canadian Community Health Survey (2013-14). Permission to use survey data was obtained from Research Data Centre located at the University of Saskatchewan.
  • Outcome variable is health status with three categories: healthy weight, overweight, and obese. Healthy weight will be used as a reference category.
  • Explanatory variables at individual-level are age, sex (binary: male and female), education (categorical: < secondary, secondary, post-secondary ED, and post-secondary cert.), physical activity (categorical: active, moderate active, and inactive), Smoking (binary: yes and no), and SES (categorical: lowest, lower middle, upper middle, and highest).

Statistical Methods

  • The variation in overweight and obesity risk had been explored through graphing the percentage of overweight and obesity people in each health region over the duration 2003 - 2014 as trend analysis. The provincial average percentages as well as ranges had been compared with corresponding national statistic.
  • The risk of overweight and obesity had intrinsically been categorized into broad sources: individual and beyond-indidividual (possible sources are environmental, spatial, and so on).
  • A multinomial multilevel (at health regions level) model had been employed as:
    Health status = β0 + Age*β1 + Female*β2 + (< Secondary)*β3 + Secondary grad*β4 + Post-secondary*β5 + Moderate*β6 + Inactive*β7 + Smoker*β8 + Lowest*β9 + Lower middle*β10 + Upper middle*β11 + U
  • While individual-level variation had been modelled as a function of personal characteristics, variation due to factors beyond-individual factors had been incorporated as a random effect (U) at health regions level.
  • The random effect specification is expected to address the collective influence of all factors beyond individual-level.

Instruction

  • 'Trend' tab presents overweight and obesity trends over the period 2004 - 2014 at health region level as well as provincial level.
  • You can choose the health status (overweight or obesity), a province, and minimum or maximum cut-off ponts to show data on the graphs. All graphs will be updated in real time based on your choice.
  • Estimated results have been presented in the 'Statistical Model', 'Obesity Risk', and 'Overweight Risk' tabs.
  • You can zoom-in or zoom-out using the + or - button located on the top-left corner.
  • While hovering the mouse over the map will show the estimated relative risk for one health status, relative risk for other health status will pop-up on click.

Individual Characteristics

  • The results are based on information from 111,084 adults (18 years or older) living in 117 health regions across Canada.
  • While the statistical model was estimated in SAS, this web-app had been developed and deployed on the server using several packages of R including shiny, dplyr, leaflet, plotly, and sf.
  • Compared with healthy weights, the overall risks are 29% higher (overweight) and 55% lower (obesity) after adjustments for personal characteristics included into the model and random variation among health regions.
  • Female have 40% lower risk, on average of overweight and obesity than male.
  • The secondary school graduates are at maximum risks compared to participants with post-secondary education.
  • Inactive people have a two-fold higher risk of obesity.
  • Individuals with lowest SES had a higher risk of obesity but lower risk of overweight compared with highest SES.
  • Estimated effects of different individual characteristics have been compared in the right panel.
  • Estimated risks of overweight and obesity based on random-effect specification has been categorized into four groups using quantile distribution and mapped in the 'Obesity Risk' and 'Overweight Risk' tabs.

Estimated Risks

Discussion

  • Obesity risks had been distributed all over Canada; although, risks are comparatively higher in northern parts of most provinces.
  • High risk areas - both overweight and obesity risks are high- are mainly from Atlantic Canada (NL and NS) and prarie regions (AB, SK, and MB).
  • Low risks health regions are mostly located in central Canada and Westren Canada.
  • The random effects estimate (maps) are interpreted as residual area-specific risks conditional on fixed effects included in the model; similarly, the fixed effect estimates (graph) are covariate effects adjusted for the area-specific random effects.
  • A moderate size (4.72% for obesity and 1.50% for overweight) of observed variations are attributable to differences among health regions.
  • The unexplained risks can be thought of as due to factors not included in the analysis, i.e., spatial effect or area-level factors (e.g., population density, neighborhood characteristics).
  • There is a possibility of respondent bias due to the use of self reported data; however, such bias - if any - should be non-differential.
  • The multinomial multilevel model is essential for separate risk assessment of overweight and obesity in order to identify current hardest hit communities for immediate intervention.