Geographical variation of Crohn's disease residual incidence in the Province of Quebec, Canada

Abstract

Background: Crohn's disease (CD) is clinically expressed as a chronic affection of the gastrointestinal tract currently known to have a multifactorial etiology involving a complex pathophysiological host response modulated by genetic susceptibilities, demographic determinants and environmental factors. With more than 20 cases per 100,000 person-years, the province of Quebec, Canada is among regions of the world with highest reported occurrence of CD in relation to other places where comparable estimates are available. This ecological study was designed to provide a medium-scale spatial exploration of CD incidence after accounting for the influence of known population and regional determinants. Health records of consulting patients in southern Quebec were compiled from 1995 to 2000 and used to estimate age and sex standardized rates per health area (n = 156). Various statistical models taking into account the regional effect of Jewish ethnicity, aboriginal ancestry, material deprivation, prescription for oral contraceptives, reportable enteric infection incidence, smoking as well as latitude and longitude locations were fitted.

Results: The final regression model presented a coefficient of determination of 22.8% and there was evidence of an eastern trend in the residual incidence (p = 0.018). Overall, the smoothed residual incidence presented a heterogeneous spatial pattern with evidence of patches (multiple health areas) of high, low and contrasting values. Health areas with most extreme incidence residuals where also distributed over the whole province including one area in the metropolitan area of Montreal and others in surrounding areas.

Conclusions: These findings suggest that known populational and regional factors derived through census information only explain a limited fraction of the geographical variation of CD incidence and lead to speculate that the effects of these factors may be incompletely captured (imperfect construction of proxy variables) or that other important factors remain unmeasured. In this view, markers of genetic profiles of homogeneous sub-populations, and other factors linked to agroenvironmental microbial exposure should be further investigated. Once accounting for known factors, it would also be worth comparing adjacent geographical areas demonstrating abrupt changes in residual incidence rates to further explore effect linked to regional factors from those resulting from various reporting systems.

Figure 1

Crohn's Disease Standardized Cumulative Incidence (1995 - 2000). Incidence rates were transformed using empirical Bayesian smoothing function (EBS) and represent age and sex directly standardized cumulative measures. The classification is based on a standard deviation bi-directional method.

Figure 2

Crohn's Disease Residual Incidence (1995 - 2000). Values represents empirical Bayesian smoothed (EBS) incidence residuals from a regression model adjusting for the regional effect of immigration, population of Jewish ethnicity, population of Aboriginal ancestry, incidence of enteric diseases and socio-economic status. Positive values reflect a larger observed incidence than predicted by the model (under-estimation) and negative values reflect a smaller observed incidence than the one predicted by the model (over-estimation). Patches of areas with lower values (i.e. regions A and E), higher values (i.e. regions B and F) and with discordant values (i.e. regions D and C) are illustrated.

Figure 3

Local Indicators Of Spatial Autocorrelation (LISA) Of CD Residual Incidence. Significant negative spatial association (Low/Low) is shown in grey while significant positive association is colored green (High/High). Areas that had non-significant LISA values are represented with no color shading.

Figure 4

Health Areas With Most Extreme Incidence Residuals Of CD (1995 - 2000). Health areas in Quebec with largest residuals according to a regression model accounting for known regional risk factors. Residual values were smoothed using an empirical Bayes (EBS) method. EBS-residuals larger and smaller than two standard deviations from the mean (zero) were tagged as "largest residuals". Areas in green reflect locations of largest residuals corresponding to an overestimation of the statistical model and areas in grey reveal locations of largest residuals corresponding to an underestimation of the model.