Feasibility and utility of mapping disease risk at the neighbourhood level within a Canadian public health unit: an ecological study

Abstract

Background: We conducted spatial analyses to determine the geographic variation of cancer at the neighbourhood level (dissemination areas or DAs) within the area of a single Ontario public health unit, Wellington-Dufferin-Guelph, covering a population of 238,326 inhabitants. Cancer incidence data between 1999 and 2003 were obtained from the Ontario Cancer Registry and were geocoded down to the level of DA using the enhanced Postal Code Conversion File. The 2001 Census of Canada provided information on the size and age-sex structure of the population at the DA level, in addition to information about selected census covariates, such as average neighbourhood income.

Results: Age standardized incidence ratios for cancer and the prevalence of census covariates were calculated for each of 331 dissemination areas in Wellington-Dufferin-Guelph. The standardized incidence ratios (SIR) for cancer varied dramatically across the dissemination areas. However, application of the Moran's I statistic, a popular index of spatial autocorrelation, suggested significant spatial patterns for only two cancers, lung and prostate, both in males (p < 0.001 and p = 0.002, respectively). Employing Bayesian hierarchical models, areas in the urban core of the City of Guelph had significantly higher SIRs for male lung cancer than the remainder of Wellington-Dufferin-Guelph; and, neighbourhoods in the urban and surrounding rural areas of Orangeville exhibited significantly higher SIRs for prostate cancer. After adjustment for age and spatial dependence, average household income attenuated much of the spatial pattern of lung cancer, but not of prostate cancer.

Conclusion: This paper demonstrates the feasibility and utility of a systematic approach to identifying neighbourhoods, within the area served by a public health unit, that have significantly higher risks of cancer. This exploratory, ecologic study suggests several hypotheses for these spatial patterns that warrant further investigations. To the best of our knowledge, this is the first Canadian study published in the peer-reviewed literature estimating the risk of relatively rare public health outcomes at a very small areal level, namely dissemination areas.

Figure 1

Box plot of SIR variation at the DA level for male lung cancer and prostate cancer raw SIRs, BYM smoothed SIRs and household income adjusted, BYM smoothed SIRs. The length of the rectangular box represents the interquartile range (25^th percentile to the 75^th percentile), the line in the box represents the median value, the filled circles within the box represent the mean value, the lower whisker extends to the first quartile minus 1.5 times the interquartile range, the upper whisker extends to the upper quartile plus 1.5 times the interquartile range, the unfilled circles represent those data points that are beyond the upper whisker, and the dashed line at 1.0 represents the provincial average. The y-axis scale was reduced to improve the visibility of the box plot, and resulted in the removal of three extreme values from LungSIR and one extreme value from ProstateSIR.

Figure 2

Wellington-Dufferin-Guelph Health Unit location.

Figure 3

2000 Average household income by 2001 DA, WDG. Moran's I: 0.46 (p < 0.001).

Figure 4

Male lung cancer incidence 1999-2003 (337 observed cases), WDG, full Bayesian smoothing, by 2001 DA. Total: 331 DAs; Overall SIR: 0.79 (95% CI: 0.70-0.90); WinBUGS fracspatial: 0.94 (95% CI: 0.44-1.00). Indirectly standardized incidence ratios calculated for all ages, using Ontario age-specific rates, 1999-2003. Full Bayesian smoothing using the BYM model [13]. Excludes 3.1% of male lung cancer cases with missing or invalid residential postal code at diagnosis.

Figure 5

Male lung cancer incidence 1999-2003 (333 observed cases), WDG, full Bayesian smoothing and adjusted for average household income quintiles, by 2001 DA. Total: 331 DAs; Overall SIR: 0.91 (95% CI: 0.80-1.01); WinBUGS fracspatial: 0.61 (95% CI: 0.01-1.00). Indirectly standardized incidence ratios calculated for all ages, using Ontario age-specific rates, 1999-2003, adjusted for average household income quintiles (2001 Census DAs). Full Bayesian smoothing using the BYM model[13]. Excludes 3.1% of male lung cancer cases with missing or invalid residential postal code at diagnosis and 1.2% of cases due to suppressed income data for the 2001 Census.

Figure 6

Prostate cancer incidence 1999-2003 (735 observed cases), WDG, full Bayesian smoothing, by 2001 DA. Total: 331 DAs; Overall SIR: 0.90 (95% CI: 0.82-0.98); WinBUGS fracspatial: 0.90 (95% CI: 0.46-1.00). Indirectly standardized incidence ratios calculated for all ages, using Ontario age-specific rates, 1999-2003. Full Bayesian smoothing using the BYM model[13]. Excludes 1.4% of prostate cancer cases with missing or invalid residential postal code at diagnosis.

Figure 7

Prostate cancer incidence 1999-2003 (723 observed cases), WDG, full Bayesian smoothing and adjusted for average household income quintiles, by 2001 DA. Total: 331 DAs; Overall SIR: 0.90 (95% CI: 0.82-0.98); WinBUGS fracspatial: 0.92 (95% CI: 0.51-1.00). Indirectly standardized incidence ratios calculated for all ages, using Ontario age-specific rates, 1999-2003, adjusted for average household income quintiles (2001 Census DAs). Full Bayesian smoothing using the BYM model[13]. Excludes 1.4% of prostate cancer cases with missing or invalid residential postal code at diagnosis and 1.6% of cases due to suppressed income data for the 2001 Census.