Lessons from geographic mapping of IBD in France

The 2 articles published in the Journal by Nerich et al1 and Declercq et al2 are interesting examples of the cutting edge spatial/epidemiological techniques now available to researchers to explore the population distribution and the causes of complex diseases such as inflammatory bowel disease (IBD). These techniques include geographic mapping of disease patterns and spatial regression models which relate disease incidence to potential causes at the ecological level. Using these techniques, the authors have convincingly shown that IBD is not randomly distributed in France, but rather exhibits distinct spatial patterns and hot-spots, suggestive of potential environmental “causes” of the disease. Interestingly, the 2 studies using similar methods came to some different conclusions about incidence distribution and what these causes might be. As important, these 2 articles also highlight the significant methodological challenges of attempting to elucidate the causes of IBD in the context of a developed country using spatial/ecological methods. Since it appears that the causes of IBD are likely fully established in developed countries, as evidenced by stable and flat incidence rates for both Crohn's disease (CD) and ulcerative colitis (UC),3 with the potentially causal conditions for IBD widely distributed throughout most population groups (adoption of Western lifestyle), the resulting truncated levels of variability in both risk exposure and IBD incidence may make it challenging to statistically identify significant correlations between the 2. This issue is exacerbated by the cross-sectional methodology of spatial/ecological studies of IBD which can be blind to important temporal influences that may be driving IBD incidence across all populations groups.4 It is possible that the effect of long-term historical causes such as the adoption of aspects of the Western lifestyle may in fact dwarf the impact of causal influences observed across space in cross-sectional studies. Unfortunately, the historical data required to model the effect of temporal influences on IBD incidence are usually not available. Spatial/ecological studies are also very challenging to implement and interpret because the requisite data to undertake small-area analysis is simply not available at the scale required. The modifiable areal unit problem (MAUP) in medical geography is well known and refers to the fact that the size and the scale of the areal units used for spatial analysis can have significant impact on the spatial patterns and relationships observed in a study.5,–7 Existing census or administrative areas are often used in spatial/ecological studies because data are conveniently available at these levels. However, their large size can mask significant heterogeneities in population characteristics and can dilute the power and interpretability of predictive models.8 On the other hand, when disease rates are calculated for relatively rare conditions such as IBD at a small geographic scale, the resulting rates are often unstable and must be smoothed in order to facilitate visualization and analysis.9 Rates based on a small number of events can often be artificially elevated, with the result that generated spatial patterns are misleading. The technique of smoothing borrows data from neighboring geographic areas in order to estimate a more realistic rate for each region. It is important to note that there is considerable diversity in the methods used to smooth, visualize, and model disease patterns, making comparability between studies difficult and also making it difficult to compare their results to previous studies undertaken by others.7, 10 The differing scales and timeframes of the 2 studies may help explain why the 2 articles came to different conclusions. The very large spatial scale employed by both studies further exacerbates the issue of interpretation and comparison. The Nerich et al study was deployed at a broad spatial scale ≈10 times larger (average population of 98,455, range of 14,641–2,125,851) than the Declercq et al study (population range of 1500–212,000, average not reported), focused on the country of France as a whole as opposed to only a high-rate IBD northern region of the country, and employed a study timeframe of 3 years (2000–2002) as compared to 13 years in the Declercq et al study (1990–2003). While the coarse spatial scale used in both of these studies may be appropriate for visualizing and comparing regional differences in IBD rates (in fact, they show convincingly that there is nonrandom spatial variation in IBD rates), we would question whether the scale employed is really refined enough for defensible small-area modeling. In the Nerich et al study, for example, populations as large as 2,125,851 were used. Krieger et al8, 11 have convincingly argued that ecological modeling approaches should only be employed when geographic units having small heterogeneous populations are used, a condition clearly violated by both of these studies. Neither article identified this scale issue as a limitation. In a similar ecological modeling study that we undertook in Manitoba,12 where we found strong and consistent relationships between a range of ecological predictors and IBD incidence, we used spatial units having an average population size of only 2000 persons. Hence, the different scales used by these 2 studies may explain the discrepant conclusions about the relationship between IBD incidence, urban/rural residence, and socioeconomic status. This may be a clear case of a modifiable areal unit problem where analyses undertaken at differing spatial scales generate divergent results. This may have been further aggravated by the different timeframes and the different regional focus of the 2 studies. It is possible that there are unique disease dynamics operating in northern France as compared to the country as a whole. Because the 2 studies generated comparative rate ratios and not population-based rates, this makes their comparison to each other and to other published studies difficult, since it is not possible from either study to easily state what the range of observed IBD rates were across geographic areas. In comparing the maps of northern France (Declercq et al study) with overall rates for France as a whole (Nerich et al study), it is unknown if they found similar rates. Furthermore, in the Nerich et al article smoothed maps of CD were adjusted not only by age and gender, but also for a range of confounders including density of physicians, latitude, and the number of farmers. The map of UC presented in the same article appears not to have been smoothed at all and adjusted for only age and gender, again making visual comparison of these 2 maps challenging and direct comparison with the results of the Declercq et al study difficult. The Nerich et al study found a significant regional north/south trend in CD incidence consistent with previous studies from the US and Europe.13,–15 We do wonder, however, whether a description of such a gradient serves a useful purpose in helping to posit etiological questions. There have been significant questions raised about the significance of the north/south trend in light of the strong east–west gradient that is apparent at the regional level in Europe, and the high rates observed in southern countries such as New Zealand.16,–18 So, at face value, although it appears within France that there are higher rates in the north, this factor may not be so relevant in terms of specific latitude between areas of high versus low incidence. Rather, it is incumbent upon these 2 groups of investigators in France to pursue what is different about high versus low incidence areas and what is similar between the distinct high incidence areas and also between distinct low incidence areas. Any or all of water or soil microbial ecology, environmental chemicals, or local culture in terms of diet, marital patterns, ethnicity, and immigration should be explored. Since the critical exposure factor in IBD is unknown, the lag time between exposure and disease onset is unknown, which raises a critical uncertainty in estimating incidence in relation to area of residence. It would be important to try to determine whether the critical area of residence is the one lived in at time of symptom onset (or within 2 years from diagnosis) or the residence of early childhood. In conclusion, we would suggest that future analyses undertaken in France and elsewhere could maximize their contribution to understanding the etiology of IBD by building on the approaches used in these 2 studies, using more refined and consistently applied epidemiological and spatial methods where it is possible to do so. To facilitate comparison with other studies researchers should consider reporting population-based rates that have been minimally adjusted (age and sex at most) and declare their standard population. To more precisely model ecological correlates of IBD, attempts should be made to use smaller geographic areas in order to minimize the misclassification of populations by risk category. Key predictors such as urban/rural should be formally defined. When possible, a large range of environmental and social predictors should be integrated into predictive models. Temporal analyses should also be incorporated to try to identify long-term and life-course influences on IBD development, but this may only be possible in developing countries now adopting the Western lifestyle (if that proves to be a critical factor) where there are opportunities to prospectively collect changing exposure data over time. Finally, researchers should explore the use of cutting edge spatial techniques such as geographically weighted regression, which can model how the association between IBD incidence and its predictors may vary across space.19 This may be especially relevant in the French situation, where the relationship between IBD and its predictors appears to be different in northern France than in the rest of the country. Dr. Charles Bernstein is supported in part by a Research Scientist Award of the Crohn's and Colitis Foundation of Canada and holds the Bingham Chair in Gastroenterology. Dr. Bernstein has consulted to or served on advisory boards for Abbott Canada, Axcan Pharma, and Shire Pharmaceuticals Canada.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.