The geography of overweight in Quebec: analyzing and visualizing spatial inequalities using second-level residuals.

Lebel, Alexandre ; Riva, Mylene ; Pampalon, Robert 等

During the last decade, there has been a strong research interest in the influence of local context on health, and on overweight and obesity more specifically. (1-6) Because health inequalities are never fully clarified by ordinary (single-level) statistical models, and part of these inequalities might be explained by the specific context in which one lives, (7,8) multilevel models are increasingly used to investigate the social determinants of health. (9-11) These models allow for the assessment of variation in health across small areas as a function of both composition (characteristics of the individuals) and context (characteristics of the group) of these areas. Consideration of second-level residuals obtained from multilevel models is particularly interesting because it allows for the identification of 'outliers', (12-15) i.e., areas with better/worse than expected health outcomes. When individual and area-level variables do not entirely explain the spatial distribution of a health phenomenon, this unexplained variation is not necessarily due to similar spatial processes across space. Rather, it might suggest that specific areas are significantly different from others. Although second-level residual analysis has the potential to contribute to our understanding of spatial inequalities in health, very few studies have employed this method. (12-15)

In a previous multilevel study, we explored the geography of overweight in the province of Quebec, Canada. (16) Findings showed a complex geography, suggesting important differences in the socio-economic and environmental determinants of overweight, especially between men and women and between rural and urban areas.

Using and extending these results, the purpose of this paper is to statistically describe the regional variation in overweight and to geographically visualize areas in the province of Quebec with above- and below-average levels of overweight. We critically discuss how the analysis of second-level residuals can enhance the understanding of the spatial distribution of health outcomes and its role in formulating hypotheses about the pathways linking place to health and the relevance of such analysis for public health, public policies and interventions.

METHOD

A full description of methods and analyses are presented in our earlier paper, (16) but are summarized here. Data on socio-economic characteristics and health-related behaviours of a sample of 20,449 adults are from the Canadian Community Health Survey Cycle 2.1 (CCHS 2.1, 2003). Respondents also reported on their height and weight, from which their Body Mass Index (BMI) was computed; overweight was defined as a BMI [greater than or equal to] 25. To investigate the geography of overweight, 51 spatial units were purposely created by grouping dissemination areas according to area deprivation, rurality and urban sprawl which are contextual factors reported to be associated with overweight (6,17-19) (for further details on the design of the spatial units, please refer to this methodological supplement: ftp://ftp.crad.ulaval.ca/Pub/crad/Lebel_2008_Area.pdf).

Statistical analysis

Data were analyzed using multilevel models, as they allow for the assessment of variation in overweight across areas as a function of both composition and context of these areas and further account for the non-independence of the within-group observations and quantify the magnitude of variation between and within groups.

First, multilevel logistic regression models, estimated separately for men and women, are used to obtain estimates of between-area variance in overweight (null or unadjusted model) and to measure individual and area-level factors associated with increased likelihood of reporting overweight (adjusted model). The adjusted model contains information on socio-economic status and health-related behaviours at the individual level and information on rurality and deprivation (social and material) at the area level. The partition of the variation in overweight between individuals and between areas is measured using the intra-class correlation coefficient (ICC), which indicates the relative importance of the contextual effect, (20) and the overall predicted probability (OPP), which provides an estimate of the average probability of observing a specific health indicator across all area units. (21)

Then, second-level residuals are computed for the unadjusted and adjusted models to give, respectively, the unadjusted estimated area-level proportion of overweight and the estimated area-level proportion of overweight for the reference person in the fully adjusted models. The reference person is aged between 18 and 24, single, with a high income and a university diploma, eats less than five fruits or vegetables per day, smokes, is physically active, and lives in an urban area that is materially and socially privileged. These individual characteristics were chosen based on our previous models in which they were significantly associated with lower odds of overweight. (16) For each of the 51 areas, the difference in risk of overweight compared to the average risk was estimated by dividing the area's probability on the adjusted OPP.

Finally, results are presented cartographically, allowing visualization of the spatial distribution of the unexplained variance in overweight across areas.

[FIGURE 1 OMITTED]

RESULTS

Multilevel analyses

Table 1 presents results of the unadjusted and adjusted logistic multilevel models for men and women. Most demographic, socioeconomic and lifestyle indicators are associated in the expected way with overweight. Significant (p<0.05) between-area variation in obesity remains after controlling for these individual characteristics, rurality and area deprivation. The ICC of the adjusted models indicates that about 1.08% of the variation among men and 1.59% of the variation among women can be attributed to differences between areas. Although the ICC for the logistic model is hard to interpret in epidemiologic terms, (22) here it can nonetheless be interpreted to suggest that contextual effect appears stronger for women. In the unadjusted model, the area-level OPP of overweight was 55.8% for men and 42.8% for women, which correspond to the estimated prevalence for adults in the province of Quebec. (23) In adjusted models, the OPP decreased by 25% for men and by 34% for women, suggesting a reduction in risk of overweight among people enjoying privileged socio-economic conditions and a predominantly healthy lifestyle.

Second-level residual analyses and cartography

Area-level residual log odds ratios of fully adjusted models were computed into a probability and the associated confidence intervals were used to identify areas that significantly differed from the fully adjusted multilevel model. This allows identifying areas presenting significantly (p<0.05) lower or higher levels of overweight and the extent to which each area differs from what is expected for the 'reference person'. Results are presented in Figure 1 for men and Figure 2 for women; area probabilities with significantly higher levels of overweight are identified in black, and those with significantly lower levels appear in grey. The geography of overweight appears to follow different regional patterns for men and women, with significant differences in 15 of 51 areas among men, and 17 of 51 among women.

Living in the city centre of Montreal, the largest city in the province, is associated with the province's lowest risk for both men (-22.5%) and women (-38.0%); lower risks of overweight are also observed for people living in Sherbrooke city. Living in (most but not all) middle-size towns south of Montreal is associated with a 5.1% greater risk of being overweight among men only. Women living in the north shore area of Montreal appear to report higher levels of overweight. In the Gatineau area (fifth largest city), men living either in urban or rural areas closer to the urban core present a higher risk of overweight at close to 8%; for women, this probability reaches 28.5%. In the Trois-Rivieres area, greater risk of overweight is observed among women, but not men. Two rural areas located near the city of Quebec show lower than expected levels of overweight for both women and men.

DISCUSSION

The purpose of this paper is to illustrate and discuss how second-level residual analysis can enhance the interpretation of multilevel area effects on health. Results show that contextual effects on overweight are not distributed evenly between areas, as indicated by different geographic patterns between regions and between men and women. These observations can point to new research questions and hypotheses addressing how the context of different places relates to overweight and obesity levels.

[FIGURE 2 OMITTED]

We observed that men living in most middle-size towns close to Montreal, and women living in a similar location north of Montreal, are at greater risk of being overweight. Perhaps this illustrates the influence on body weight of commuting by private motorized transport. (19) Yet this pattern is not observed for women living in the south of Montreal, nor for those living on the outskirts of the city of Quebec, the second largest city in the province, although commuting to work by means of private motorized transport is as likely for this area. In the Montreal city centre and in the city of Sherbrooke, the proportions of overweight are lower than expected for both men and women; the specific set of built environment features, resources and services in these regions may be associated with lifestyles among the population that are involved in the aetiology of obesity.

Lower than expected levels of overweight were observed in rural areas on the outskirts of Quebec city. These results differ from those of other North American studies which reported a greater prevalence of overweight and obesity in rural than urban areas (17-19,24,25) but may highlight, as observed elsewhere, (18,26-28) the existence of inequalities in health between rural areas. Further studies should be conducted to unravel inequalities in overweight between rural areas.

Distinct spatial patterns of overweight were observed for women and men, as well as for regions, indicating that the processes linking place to health may differ by gender and between regional contexts. This highlights the idea that social inequalities in health may be territorialized, or that the specific interaction occurring between socio-economic, demographic, political, and physical contexts may bring about distinct health patterns. Future studies should attempt to disentangle the differences in contextual influence on overweight and obesity among men and women while taking into account place idiosyncrasies.

Limitations

Results of this study should be interpreted in light of several limitations. The cross-sectional design does not allow for the determination of causal effects. Self-reported height and weight as provided by the CCHS to derive the body mass index is also a well-known source of potential bias. (29) The small number of contextual indicators is another limitation. As the province of Quebec encompasses a very large area (1,667,926 [km.sup.2]) with most of its population living in urban settings, data availability is unfortunately not the same all over the territory and this precludes the construction of reliable contextual indicators with the same precision for each area unit. Future studies should attempt to use individual travel information and consider regional mobility patterns for studying the geography of overweight.

CONCLUSION

The observed inequalities between and within regions, as well as between men and women, which were not fully explained by individual or area-level characteristics, are likely to be important in defining actions in public health and public policy at regional and local levels. Many conceptual models of overweight and obesity suggest multi-scale and multi-dimension interventions, (23,30) yet most intervention programs in the province are of an informational/educational nature and often target behaviour change in 'at-risk' individuals. As Quebec's public health intervention is also administered at the regional level, some initiatives have been undertaken at a more local scale (Quebec-en-forme and Kino-Quebec are good examples). However, local interventions are rare, sparse, and mostly based on socio-economic observations; they rarely integrate physical activity and nutrition, and are not supported by a global framework.

Promoting a multi-scale and multi-dimensional framework for public health interventions is not an easy task and must be based on reliable monitoring data. Second-level residuals analysis has the potential to provide important information for the formulation and implementation of regional and local health promotion and prevention strategies by illustrating that contextual effects in relation to overweight (or any health outcome) might be linked to different area-level factors in different environments.

Acknowledgements: This research could not have been carried out without the financial support of the Quebec Population Health Research Network (QPHRN), the Quebec Inter-University Centre for Social Statistics (QICSS), and the Fonds de la recherche en sante du Quebec (FRSQ). The Institut national de sante publique du Quebec (INSPQ) and the Natural Science and Engineering Research Council of Canada (NSERCC) also contributed to this study. Finally, we thank Denis Hamel for his statistical advice and Pierre Masson for his revision of an earlier version of this paper.

Conflict of Interest: None to declare.

Received: July 3, 2009

Accepted: November 19, 2009

REFERENCES

(1.) Cummins S, Macintyre S. Food environments and obesity - neighbourhood or nation? Int J Epidemiol 2006;35(1):100-4.

(2.) French SA, Story M, Jeffery RW. Environmental influences on eating and physical activity. Annu Rev Public Health 2001;22:309-35.

(3.) Giskes K, Kamphuis CB, van Lenthe FJ, Kremers S, Droomers M, Brug J. A systematic review of associations between environmental factors, energy and fat intakes among adults: Is there evidence for environments that encourage obesogenic dietary intakes? Public Health Nutr 2007;10(10):1005-17.

(4.) Kim D, Subramanian SV, Gortmaker SL, Kawachi I. US state- and county-level social capital in relation to obesity and physical inactivity: A multilevel, multivariable analysis. Soc Sci Med 2006;63(4):1045-59.

(5.) Kim S, Popkin BM. Commentary: Understanding the epidemiology of overweight and obesity - A real global public health concern. Int J Epidemiol 2006;35(1):60-67.

(6.) Ross NA, Tremblay S, Khan S, Crouse D, Tremblay M, Berthelot JM. Body mass index in urban Canada: Neighborhood and metropolitan area effects. Am J Public Health 2007;97(3):500-8.

(7.) Diez Roux AV. The study of group-level factors in epidemiology: Rethinking variables, study designs, and analytical approaches. Epidemiol Rev 2004;26:104-11.

(8.) Subramanian SV. The relevance of multilevel statistical methods for identifying causal neighborhood effects. Soc Sci Med (1982) 2004;58(10):1961-67.

(9.) Papas MA, Alberg AJ, Ewing R, Helzlsouer KJ, Gary TL, Klassen AC. The built environment and obesity. Epidemiol Rev 2007;29:129-43.

(10.) Pickett KE, Pearl M. Multilevel analyses of neighbourhood socioeconomic context and health outcomes: A critical review. J Epidemiol Community Health 2001;55(2):111-22.

(11.) Riva M, Gauvin L, Barnett TA. Toward the next generation of research into small area effects on health: A synthesis of multilevel investigations published since July 1998. J Epidemiol Community Health 2007;61(10):853-61.

(12.) Boyle PJ, Gatrell AC, Duke-Williams O. Do area-level population change, deprivation and variations in deprivation affect individual-level self-reported limiting long-term illness? Soc Sci Med 2001;53(6):795-99.

(13.) Chaix B, Merlo J, Chauvin P. Comparison of a spatial approach with the multilevel approach for investigating place effects on health: The example of healthcare utilisation in France. J Epidemiol Community Health 2005;59(6):517-26.

(14.) Chaix B, Merlo J, Subramanian SV, Lynch J, Chauvin P. Comparison of a spatial perspective with the multilevel analytical approach in neighborhood studies: The case of mental and behavioral disorders due to psychoactive substance use in Malmo, Sweden, 2001. Am J Epidemiol 2005;162(2):171-82.

(15.) Wiggins RD, Joshi H, Bartley M, Gleave S, Lynch K, Cullis A. Place and personal circumstances in a multilevel account of women's long-term illness. Soc Sci Med 2002;54(5):827-38.

(16.) Lebel A, Pampalon R, Theriault M, Hamel D. The geography of overweight in Quebec: A multilevel perspective. Can J Public Health 2009;100(1):18-23.

(17.) Huot I, Paradis G, Ledoux M. Factors associated with overweight and obesity in Quebec adults. Int J Obesity 2004;28(6):766-74.

(18.) Pampalon R, Martinez J, Hamel D. Does living in rural areas make a difference for health in Quebec? Health & Place 2006;12(4):421-35.

(19.) Frank LD, Andresen MA, Schmid TL. Obesity relationships with community design, physical activity, and time spent in cars. Am J Prev Med 2004;27(2):87 96.

(20.) Snijders T, Bosker R. Multilevel Analysis. London, UK: Sage Publication, 1999.

(21.) Raudenbush SW, Bryk A. Hierarchical Linear Models: Applications and Data Analysis Methods. Newbury Park, CA: Sage, 2002.

(22.) Merlo J, Chaix B, Ohlsson H, Beckman A, Johnell K, Hjerpe P, et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: Using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health 2006;60(4):290-97.

(23.) Mongeau L, Audet N, Aubin J, Baraldi R. L'exces de poids dans la population quebecoise de 1987 a 2003. Institut national de sante publique du Quebec, editor. 2005; 24.

(24.) Belanger-Ducharme F, Tremblay A. Prevalence of obesity in Canada. Obes Rev 2005;6(3):183-86.

(25.) Eberhardt MS, Pamuk ER. The importance of place of residence: Examining health in rural and nonrural areas. Am J Public Health 2004;94(10):1682-86.

(26.) ICIS. How Healthy Are Rural Canadians? 2006;1-205.

(27.) Mitura V, Bollman RD. The health of rural Canadians: A rural-urban comparison of health indicators. Rural and Small Town Canada Analysis Bulletin 2003;4(6):1-21.

(28.) Riva M, Curtis S, Gauvin L, Fagg J. Unravelling the extent of inequalities in health across urban and rural areas: Evidence from a national sample in England. Soc Sci Med 2009;68(4):654-63.

(29.) Tjepkema M. Obesite chez les adultes au Canada: Poids et grandeur mesures. Statistique Canada, editor. 2005;1-36.

(30.) Swinburn B, Egger G, Raza F. Dissecting obesogenic environments: The development and application of a framework for identifying and prioritizing environmental interventions for obesity. Prev Med 1999;29(6 Pt 1):563-70.

Correspondence: Alexandre Lebel, Centre de recherche en amenagement et developpement, Pavillon Felix-Antoine-Savard, 2325 rue des Bibliotheques, local 1612, Universite Laval, Quebec, QC G1V 0A6, Tel: 418-650-5115, ext. 5704, Fax: 418-643-5099, E-mail: alexandre.lebel.1@ulaval.ca

Alexandre Lebel, MSc, [1,2] Mylene Riva, PhD, [3] Robert Pampalon, PhD, [2] Marius Theriault, PhD [1]

Author Affiliations

[1.] Centre de recherche en amenagement et developpement, Universite Laval, Quebec, QC

[2.] Institut national de sante publique du Quebec, Quebec, QC

[3.] Department of Geography, Institute of Hazard and Risk Research, Wolfson Research Institute, Durham University, UK

Table 1. Results of Unadjusted and Adjusted Multilevel Models:
Association between Overweight and Individual, Demographic,
Socio-economic, and Lifestyle Characteristics, As Well As
Area-level Characteristics

                                                Men

                                    Unadjusted     Adjusted Model
Demographic and                       Model
Socio-economic Characteristics                       OR      95% CI

Age (years)
  18-24                                             1.00
  25-44                                             2.48   1.99-3.10 *
  45-64                                             2.74   2.03-3.71 *
  65+                                               2.14   1.66-2.75 *
Marital status
  Single                                            1.00
  Previously married                                1.13   0.89-1.42
  Union                                             1.50   1.23-1.83 *
Disposable income
  High income                                       1.00
  Low income                                        1.05   0.89-1.24
  Missing income                                    1.06   0.83-1.37
Education level
  University                                        1.00
  College or high school                            1.43   1.23-1.66 *
  Less than high school                             1.54   1.22-1.95 *

Lifestyle
Fruits or vegetables
  Eat less than 5/day                               1.00
  Eat 5 or more/day                                 0.79   0.68-0.91 *
Smoking status
  Current smoker                                    1.00
  Former smoker                                     1.75   1.48-2.06 *
  Never-smoker                                      1.62   1.22-2.16 *
Leisure physical activity level
  Active                                            1.00
  Moderately active                                 1.15   1.00-1.32
  Somewhat active                                   0.99   0.77-1.29
  Sedentary                                         1.01   0.85-1.19
Occupational physical activity
  Physically difficult                              1.00
  Usually standing                                  0.86   0.70-1.07
  Usually seated                                    1.08   0.83-1.40

Area-level Characteristics
Rural                                               1.17   1.02-1.33 *
Materially deprived                                 0.86   0.74-1.01
Socially deprived                                   0.93   0.80-1.08

Variance Estimation
Coefficient (std error)           0.041 (0.013) *   0.036  (0.008) *
Intraclass correlation
  coefficient                     1.23%             1.08%
Overall predicted probability    55.75%            20.26%

                                              Women

                                    Unadjusted     Adjusted Model
Demographic and                       Model
Socio-economic Characteristics                       OR      95% CI

Age (years)
  18-24                                            1.00
  25-44                                            2.12    1.63-2.76 *
  45-64                                            3.28    2.45-4.38 *
  65+                                              2.85    2.02-4.03 *
Marital status
  Single                                           1.00
  Previously married                               0.95    0.76-1.20
  Union                                            1.03    0.87-1.22
Disposable income
  High income                                      1.00
  Low income                                       1.30    1.09-1.54 *
  Missing income                                   1.06    0.89-1.25
Education level
  University                                       1.00
  College or high school                           1.54    1.19-1.99 *
  Less than high school                            2.39    1.91-2.98 *

Lifestyle
Fruits or vegetables
  Eat less than 5/day                              1.00
  Eat 5 or more/day                                1.07    0.94-1.21
Smoking status
  Current smoker                                   1.00
  Former smoker                                    1.29    1.06-1.56 *
  Never-smoker                                     1.32    1.00-1.74 *
Leisure physical activity level
  Active                                           1.00
  Moderately active                                1.14    0.96-1.35
  Somewhat active                                  1.48    1.23-1.78 *
  Sedentary                                        1.61    1.35-1.92 *
Occupational physical activity
  Physically difficult                             1.00
  Usually standing                                 1.03    0.77-1.37
  Usually seated                                   1.17    0.89-1.55

Area-level Characteristics
Rural                                              0.96    0.83-1.10
Materially deprived                                1.16    0.98-1.37
Socially deprived                                  0.93    0.77-1.12

Variance Estimation
Coefficient (std error)           0.088 (0.037) *  0.05    (0.018) *
Intraclass correlation
  coefficient                     2.61%            1.59%
Overall predicted probability    42.80%            8.50%

* Significant at p < 0.05

All of the table except "Overall predicted probability" was
originally published in: Lebel A, Pampalon R, Theriault M, Hamel
D. The geography of overweight in Quebec: A multilevel
perspective. Can J Public Health 2009;100(1):18-23. Reproduced
with permission.