Background
The etiology of the inflammatory bowel diseases (IBD) has been extensively studied
1
, however, the environmental determinants of disease pathogenesis are not fully understood
2
3
. IBD is widely believed to be associated with industrialization of nations. This hypothesis is supported by the significant geographic variation in IBD with the highest incidence rate of IBD in North America and Europe
4
. Migrant studies have demonstrated that immigrants, and particularly their offspring, from low prevalent regions acquire a similar risk of IBD as the local population
5
. Furthermore, the incidence of IBD is steadily rising in several developing nations as they have become industrialized
6
7
. Within a country the incidence of IBD has been proposed to be higher in urban versus rural areas
8
9
.
Although numerous studies have investigated the association between urban environment and IBD, findings remain inconsistent. An urban association has been considered as far back as 1963, when Acheson and Nefzger suggested a positive relationship between urban areas and ulcerative colitis (UC)
10
. Many observational studies have subsequently shown an increase in UC and Crohn's disease (CD) incidence in more densely populated areas
8
11
12
13
14
15
. Numerous studies, however, have failed to find an association between urban exposure and IBD
4
16
, while others have shown an inverse association
17
. Establishing whether the risk of IBD is greater in urban environments is important because environmental exposures in urban societies are significantly different than those in rural areas. This information may direct research initiatives on specific environmental risk factors of IBD and establish distribution of IBD burden in society.
A systematic analysis of the association between living in an urban environment and IBD has yet to be conducted. Thus, we performed a systematic review and meta-analysis of all case–control and cohort studies that explored the association between residing in an urban region and IBD in order to determine whether the risk of CD and/or UC was increased in urban as compared to rural areas.
Methods
Search strategy
We conducted a systematic literature search using a predetermined protocol and in accordance with the quality of reporting meta-analyses of observational studies (MOOSE)
18
. We searched two computer-stored databases, Medline (1950-present) and Embase (Excerpta Medica Database; 1980-present) for studies describing the association between urban environment and IBD as of October 2009. The search strategy for Medline and Embase was conducted based on three themes. The first theme, the outcome measure, combined the exploded version of Medical Subject Headings (MESH) “inflammatory bowel disease” or “Colitis, Ulcerative” or “Crohn Disease”. The second theme, the exposure, combined the exploded version of MESH headings “urban population” or “urban health” or “rural population” or “rural health” or “geography”. The third theme combined exploded versions of MESH headings “risk factors” or “risk assessment” or “epidemiology” or “demography”. All the keywords were used to search the titles and abstracts. The search was not limited by language or human subjects to ensure capture of all appropriate papers. Abstracts from the American College of Gastroenterology for 2006, 2007, and 2008, and the American Gastroenterological Association for 2006 and 2007 were reviewed. The reference lists of relevant articles were also reviewed.
Selection criteria
Two reviewers (N.M. and I.S.) identified articles eligible for further review by performing an initial screen of identified abstracts and titles. Articles were eliminated in this initial screen if they were not observational or did not either investigate environmental risk factors for IBD in a case–control study or investigate epidemiology of IBD using incident cases. Studies that did not report original data and duplicated publications were also excluded. Full-text of the remaining articles were retrieved and systematically reviewed. Articles were considered for inclusion in the second screening if they reported a measure of association between urban environment and UC and/or CD. In both study designs, UC and CD were required to be reported separately for inclusion into the systematic review. Studies that did not report adequate information to calculate incidence rate ratio (IRR) or odds ratio (OR) with 95% confidence interval (CI) were excluded. Disagreement between reviewers was resolved by consensus with third party experts (DR and GK).
Data extraction
The outcome variable of interest was defined as the presence of UC and/or CD. The exposure variable of interest was residing in an urban versus rural environment. Urban and rural populations were not consistently defined across all studies. Several studies (n = 10 for UC; n = 11 for CD) did not define an urban and/or rural environment
12
13
16
17
19
20
21
22
23
24
25
26
27
28
29
30
31
. The studies that defined this exposure used definitions that varied and some papers stratified by multiple levels of exposure (e.g. urban, semi-urban, and rural)
24
32
33
34
35
36
. A priori we identified studies that defined urban as a population greater than 10,000
37
38
. Secondary variables extracted from the manuscripts included: study design (i.e. case–control or cohort); country of origin; publication year; timing of exposure; source of controls for case–control studies; and information on key indicators of study quality, using MOOSE
18
.
We extracted reported OR and IRR with 95% (CIs) or data enabling the calculation of these association measures. Both adjusted and unadjusted values were extracted; though, when available, the adjusted estimates were used.
Statistical analysis
Meta-analyses were initially conducted by combining cohort and case–control studies using IRRs, defined as the incidence rate of IBD in urban versus rural populations. Given the low prevalence of IBD, the OR would approximate the IRR under the rare disease assumption. Case–control and cohort studies were then analyzed separately using different measures of association. The IRR was used as measure of association for cohort studies and OR was used for case–control studies. The Test of Heterogeneity was performed using a Q statistic (5% level) and random effects models were used because of the presence of heterogeneity between studies. Stratified analyses were performed to explore factors that may explain heterogeneity between studies. Publication year was a priori stratified into three year categories: 1962–1988; 1989–1998; and 1999–2009. Similarly, stratified analyses and meta-regression were performed based on study design (i.e. case–control or cohort study), region of publication, timing of exposure, and source of controls in case–control studies. Sensitivity analyses were performed to exclude studies that did not include a definition of urban/rural region. Further sensitivity analysis was conducted on studies that used >10,000 people as a definition of urban environment
37
38
. Papers that stratified by multiple levels of exposure (e.g. urban, semi-urban, and rural) were analyzed as urban versus rural. The possibility of publication bias was assessed using the Begg tests.
Results
Literature search
The search strategy retrieved 6940 unique citations: 2964 from Medline and 3976 from Embase. Of these, 6434 citations were excluded after the first screening based on titles and abstracts, leaving 506 articles for full-text review (Figure
1). The observed agreement between reviewers for eligibility of articles was 97%, corresponding to a kappa statistic of 0.77. Upon full text review of 506 articles, 466 were excluded (see Figure
1 for rationale of exclusions), leaving 40 studies for final inclusion in the systematic review and meta-analysis. Of these 40 studies, 25 investigated the relationship between urban environment and UC and 30 investigated this relationship with CD. Included in our analysis were 7 case–control UC studies, 9 case–control CD studies, 18 cohort UC studies, and 21 cohort CD studies.
<p>Figure 1</p>Literature search results.
Literature search results.
![]()
Demographic and study quality characteristics
Characteristics of the 18 cohort studies for UC are shown in Table
1
19
20
21
23
24
32
33
34
35
39
40
41
42
43
44
45
. The reported incidence rate ranged from 0.97 to 15.6 per 100,000/year. A definition for urban and/or rural environment was provided for 13 cohort studies. Characteristics of the 7 selected case–control studies for UC are shown in Table
2
8
12
17
26
27
28
46
. Two of the studies had definitions for urban and/or rural environment. Characteristics of the 21 selected cohort studies for CD are shown in Tables
1
14
15
22
23
25
30
31
36
39
40
41
42
43
45
47
48
49
50
51
52
. The reported incidence rate of CD ranged from 0.51 to 15.6 per 100,000/year. A definition for urban environment was reported in 16 studies. Characteristics of the nine selected case–control studies for CD are shown in Table
2
8
13
16
17
27
29
46
53
. The studies were based in North America, Australia, Asia and Europe. The association between urban environment and IBD was not the primary outcome of interest in any of the cohort or case–control studies.
<p>Table 1</p>
Characteristics of cohort studies for ulcerative colitis and Crohn’s disease studies
Study characteristics and demographics
Study Quality Characteristics
Author (year)
Country
Study period
Number of cases
Incidence of UC (per 100,000)
Methods of diagnosis defined
Diagnosis based on recognized criteria
Urban/rural defined
Urban population >10,000
Time of exposure
Multiple levels of exposure
Ulcerative Colitis
Mate-Jimerez 1994
39
Spain
1981-88
111
3.16
Yes
Yes
Yes
Yes
At diagnosis
No
Ekbom 1991
40
Sweden
1965-83
2509
10.5
Yes
Yes
Yes
No
At diagnosis
No
Linden 1971
19
Finland
1967
223
4.76
No
N/A
No
No
At diagnosis
No
Probert 1992
32
UK
1972-80
192
5.33
Yes
Yes
Yes
No
At diagnosis
Yes
Probert 1992
32
UK
1981-89
211
7.29
Yes
Yes
Yes
No
At diagnosis
Yes
Niv 1990
33
Israel
1967-86
43
2.33
Yes
No
Yes
No
At diagnosis
Yes
Kildebo 1990
22
Norway
1983-86
179
13.2
Yes
Yes
No
No
At diagnosis
No
Gheorghe 2004
23
Romania
2002-03
163
0.97
Yes
Yes
No
No
At diagnosis
No
Sincic 2006
41
Croatia
2000-04
70
4.6
Yes
Yes
Yes
No
At diagnosis
No
Tsianos 1994
24
Greece
1982-91
61
4.18
Yes
Yes
No
No
At diagnosis
Yes
Ladas 2005
34
Greece
1990-94
56
10.6
Yes
Yes
Yes
Yes
At diagnosis
Yes
Moller 1971
20
Finland
1956-67
505
1.07
Yes
Yes
No
No
At diagnosis
No
Blanchard 2001
42
Canada
1987-96
1763
15.6
Yes
Yes
Yes
No
At diagnosis
No
Green 2006
15
Canada
1990-01
N/A
13.5
No
N/A
Yes
No
At diagnosis
No
Lakatos 2003
43
Hungary
1977-01
560
5.89
Yes
Yes
Yes
No
At diagnosis
No
Manousos 1996
44
Greece
1990-94
117
8.9
Yes
Yes
Yes
No
At diagnosis
No
Wigley 1962
35
New Zealand
N/A
132
N/A
Yes
No
Yes
Yes
At diagnosis
Yes
Latour 1996
45
Belgium
1993-94
36
3.5
Yes
Yes
Yes
No
At diagnosis
No
Crohn’s Disease
Kyle 1971
47
UK
1955-69
166
1.98
Yes
Yes
Yes
No
At diagnosis
No
Jayanthi 1992
48
UK
1972-80
161
2.89
Yes
Yes
Yes
Yes
At diagnosis
No
Jayanthi 1992
48
UK
1981-89
233
4.48
Yes
Yes
Yes
Yes
At diagnosis
No
Mate-Jimerez 1994
39
Spain
1981-88
57
1.61
Yes
Yes
Yes
Yes
At diagnosis
No
Ekbom 1991
40
Sweden
1965-83
1469
6.09
Yes
Yes
Yes
No
At diagnosis
No
Shivananda 1987
36
Netherlands
1979-83
54
3.90
Yes
Yes
Yes
No
At diagnosis
Yes
Kildebo 1989
22
Norway
1983-86
82
5.85
Yes
Yes
No
No
At diagnosis
No
Kyle 1992
49
Scotland
1955-88
856
5.65
Yes
Yes
Yes
No
At diagnosis
No
Manousos 1996
44
Greece
1990-94
37
3.00
Yes
Yes
Yes
Yes
At diagnosis
No
Gheorghe 2004
23
Romania
2002-03
85
0.51
Yes
Yes
No
No
At diagnosis
No
Sincic 2006
41
Croatia
2000-04
100
6.50
Yes
Yes
Yes
No
At diagnosis
No
Phavichitr 2003
25
Australia
1971-01
351
1.02
Yes
Yes
No
No
At diagnosis
No
Moum 1996
50
Norway
1990-93
225
5.86
Yes
Yes
Yes
No
At diagnosis
No
Blanchard 2001
42
Canada
1987-96
1765
15.6
Yes
Yes
Yes
No
At diagnosis
No
Green 2006
15
Canada
1990-01
N/A
14.8
No
N/A
Yes
No
At diagnosis
No
Lakatos 2003
43
Hungary
1977-01
212
2.23
Yes
Yes
Yes
No
At diagnosis
No
Nyhlin 1986
51
Sweden
1974-81
253
4.90
Yes
Yes
Yes
No
At diagnosis
No
Ruiz Ochoa 1984
30
Spain
1976-83
152
0.80
Yes
Yes
No
No
At diagnosis
No
Sedlack 1980
31
USA
1935-75
103
4.20
Yes
No
No
No
At diagnosis
No
Brandes 1983
52
Germany
1964-75
97
3.00
Yes
No
Yes
No
At diagnosis
No
Latour 1996
45
Belgium
1993-94
56
5.50
Yes
Yes
Yes
No
At diagnosis
No
<p>Table 2</p>
Characteristics of case–control studies for ulcerative colitis and Crohn’s disease studies
Study characteristics and demographics
Study quality characteristics
Author (year)
Country
No. of cases
No. of controls
Source of controls
Controls matched for:
Methods of diagnosis defined
Diagnosis based on recognized criteria
Urban/rural defined
Urban population >10,000
Time of exposure
Multiple levels of exposure
Ulcerative Colitis
Parrello 1997
26
Italy
509
657
Attending orthopedic and surgical clinics
Sex, age, year of diagnosis
Yes
Yes
No
No
Time of data collection
No
Feeney 2002
17
UK
137
137
Functional GI Disorders
Sex, age
Yes
Yes
No
No
Childhood (age 0–5)
No
Bernstein 2007
27
Canada
137
310
Manitoba Health Population Registry
Sex, age
No
N/A
No
No
N/A
No
Jiang 2007
28
China
177
177
Neighbours or colleagues
Sex, age
Yes
Yes
No
No
Childhood
No
Radon 2007
54
Germany
304
1481
Strabismus surgery
Age
No
N/A
No
No
Time of data collection
No
Martinez Salmeron 1994
46
Spain
63
63
Hospital
Sex, age
Yes
Yes
Yes
No
Time of data collection
No
Ekbom 1990
8
Sweden
164
328
Live births between 1924–1957 in Uppsala County
Date of birth, sex, and maternal age or parity
Yes
Yes
Yes
No
Time of birth
No
Crohn’s Disease
Feeney 2002
17
UK
139
139
Functional GI disorders
Sex, age
Yes
Yes
No
No
Childhood (age 0–5)
No
Bernstein 2007
27
Canada
235
310
Manitoba Health Population Registry
Sex, age
No
N/A
No
No
N/A
No
Ponsonby 2009
53
Australia
278
998321
Live births from 1983–1998 in Victoria
Age
Yes
No
Yes
No
Childhood (age 0–6)
No
Malekzadeh 2009
16
Iran
196
207
IBS treated by an expert gastroenterologist
Age
Yes
Yes
No
No
Childhood (before 16th birthday)
No
Radon 2007
54
Germany
444
1481
Strabismus surgery
Age
No
N/A
No
No
Time of data collection
No
Thompson 1998
29
England, Wales
291
1682
Same GP
Sex, age
No
N/A
No
No
N/A
No
Wurzelmann 1994
13
United States
322
262
Closest neighbour
Sex, age, race
No
N/A
No
No
Childhood (age 0–5)
Yes
Martinez Salmeron 1994
46
Spain
30
30
Hospital
Sex, age
Yes
Yes
Yes
No
Time of data collection
No
Ekbom 1990
8
Sweden
93
186
Live births between 1924–1957 in Uppsala County
Date of birth, sex, and maternal age or parity
Yes
Yes
Yes
No
Time of birth
No
Urban/rural environment and IBD
The pooled crude IRR for the 25 UC studies was 1.17 (1.03, 1.32) (Figure
2A). For the 18 cohort studies, the pooled crude IRR was 1.19 (1.03, 1.36) (Figure
2B). In 13 of the studies, a positive association between UC and urban environment was found, 7 of which were statistically significant. Heterogeneity across studies was observed (Q statistic, 95.14; P < .001). For the UC case–control studies, the pooled OR for the association between urban environment and UC was 1.06 (0.78, 1.45) (Figure
2C). Three out of seven studies found a positive association, and two were statistically significant. Heterogeneity was observed across studies (Q statistic, 19.32; P = .004). The pooled IRR for the 30 CD studies was 1.42 (1.26, 1.60) (Figure
3A). For the 21 cohort studies, the pooled crude IRR was 1.50 (1.30, 1.72) (Figure
3B). In 20 of the studies, a positive association between CD and urban environment was found, 10 of which were statistically significant. Heterogeneity across cohort studies was observed (Q statistic, 108.56; P < .001). For the 9 CD case–control studies, the pooled OR for the association between urban environment and CD was 1.26 (1.03, 1.53) (Figure
3C). Six out of the seven studies found a positive association, but only three were statistically significant. Heterogeneity was not observed across studies (Q statistic, 11.88; P = .157).
<p>Figure 2</p>Forest plot of the summary effect estimate with 95% CI studies that explored the relationship between urban environment and ulcerative colitis for both cohort and case–control studies (A); cohort studies (B); and case–control studies (C).
Forest plot of the summary effect estimate with 95% CI studies that explored the relationship between urban environment and ulcerative colitis for both cohort and case–control studies (A); cohort studies (B); and case–control studies (C).
![]()
<p>Figure 3</p>Forest plot of the summary effect estimate with 95% CI studies that explored the relationship between urban environment and Crohn’s disease for both cohort and case–control studies (A); cohort studies (B); and case–control studies (C).
Forest plot of the summary effect estimate with 95% CI studies that explored the relationship between urban environment and Crohn’s disease for both cohort and case–control studies (A); cohort studies (B); and case–control studies (C).
![]()
Sensitivity and stratified analyses
After excluding the UC cohort studies that did not define urban and/or rural environment the pooled IRR was 1.18 (1.09, 1.27), which did not differ from the crude pooled IRR (Table
3). The IRR lost significance 1.16 (0.9, 1.5) when we included only studies that defined urban as >10,000 people. Only cohort studies published after 1999 demonstrated an association between urban environment and UC (1.30; 95% CI: 1.10, 1.54). A stratified analysis was similarly performed for region in which the study was based, divided into European and non-European countries. The IRR for Non-European studies was statistically significant with pooled estimate of 1.13 (1.06, 1.22). Meta-regression was not statistically significant for year of publication, region of publication and European countries.
<p>Table 3</p>
Sensitivity and stratified analyses of pooled relative risk for cohort studies
Ulcerative colitis
Crohn’s disease
Stratified analysis
Meta-regression
Stratified analysis
Meta-regression
No. of studies
No. of cases
Pooled RR (95%CI)
Heterogeneity (p-value)
Coef.
P value
No. of studies
No. of cases
Pooled RR (95%CI)
Heterogeneity (p-value)
Coef.
P value
Crude
18
6901
1.19(1.03, 1.36)
<.001
21
6514
1.50(1.30, 1.72)
<.001
Definition of urban/rural environment
Urban/rural defined
13
5906
1.18(1.09, 1.27)
.19
16
5741
1.28(1.14, 1.43)
<.001
Urban defined as >10,000
3
299
1.16(0.90, 1.50)
.65
4
488
1.38(0.97, 1.97)
.072
Year of publication
1962-1988
3
860
0.96(0.60, 1.51)
<.001
0.16
.30
6
825
1.90(1.27, 2.85)
<.001
−0.095
.47
1989-1998
9
3459
1.23(0.95, 1.57)
<.001
9
3176
1.26(1.04, 1.53)
<.001
1999-2009
6
2612
1.30(1.10, 1.54)
.022
6
2513
1.49(1.18, 1.87)
.001
Region
European countries
15
5036
1.21(0.99, 1.46)
<.001
0.008
.98
17
4295
1.55(1.28, 1.88)
<.001
0.77
.75
Non-European countries
3
1895
1.13(1.06, 1.22)
.61
4
2219
1.39(1.18 1.62)
.12
European Countries
Northern
6
3819
0.97(0.75, 1.27)
<.001
−0.26
.086
9
3499
1.23(1.04, 1.45)
<.001
0.23
.04
Eastern
2
723
1.59(0.98, 2.59)
.013
2
297
2.30(0.54, 9.82)
<.001
Mediterranean
6
458
1.39(0.82, 2.36)
<.001
4
346
2.64(1.50, 4.62)
.02
Western
1
36
2.12(1.00, 4.50)
NA
2
153
1.72(1.17, 2.51)
.79
For the UC case–control studies, we failed to find a statistically significant association when stratifying by timing of exposure (Table
4). Meta-regression was performed for both timing of exposure and source of controls and no statistically significant associations were found.
<p>Table 4</p>
Sensitivity and stratified analyses of pooled odds ratio for case–control studies
Ulcerative colitis
Crohn’s disease
Stratified analysis
Meta-regression
Stratified analysis
Meta-regression
No. of studies
No. of cases
No. of controls
Pooled OR (95%CI)
Heterogeneity (p-value)
Coef.
P value
No. of studies
No. of cases
No. of controls
Pooled OR (95%CI)
Heterogeneity (p-value)
Coef.
P value
Crude
7
1491
3153
1.06 (0.78, 1.45)
.004
9
2028
1002618
1.26 (1.03, 1.53)
.16
Time of exposure
Time of data collection
4
1013
2511
1.18 (0.78, 1.80)
.006
-0.32
.42
2
474
1511
1.21 (0.58, 2.50)
.18
0.060
.89
Childhood
3
478
642
0.87 (0.65, 1.17)
.91
5
1028
999115
1.18 (0.82, 1.70)
.052
0-6 years
4
832
998908
1.12 (0.71, 1.77)
.026
0.16
.46
< 16 years
1
196
207
1.39 (0.74, 2.60)
NA
Source of controls
Population
2
314
487
1.02 (0.66, 1.58)
.16
0.035
.94
3
835
998893
1.40 (1.13, 1.73)
.37
-0.23
.32
Clinic/Hospital
5
1177
2666
1.06 (0.70, 1.60)
.003
6
1193
3725
1.09 (0.79, 1.51)
.11
After excluding CD cohort studies that did not define urban and/or rural environment, the pooled IRR was 1.28 (1.14, 1.43), which was similar to the crude IRR (Table
3). Restricting the urban definition to >10,000 people resulted in a loss of significance with an IRR of 1.38 (0.97, 1.97). The pooled IRRs were statistically significant across all 3 categories of publication year (Table
3). A stratified analysis was similarly performed for region in which the study was based, divided into European and non-European countries. The IRRs for both European and non-European studies were statistically significant with pooled estimates of 1.55 (1.28, 1.88) and 1.39 (1.18, 1.62), respectively. In meta-regression analysis only the European countries variable was statistically significant (Table
3).
For the CD case–control studies, we failed to find a statistically significant association when stratified by timing of exposure (Table
4). Meta-regression was performed for both time of exposure and source of controls and no statistically significant associations were found. The OR for population-based controls was 1.40 (1.13, 1.73) and for hospital/clinic-based controls was 1.09 (0.79, 1.51). Meta-regression was performed for both variables and no statistically significant result was found.
Publication bias
No publication bias was found; the Begg tests were not statistically significant for UC (z = −0.42, P = .675) or CD studies (z = 0.70, P = .487).
Discussion
Urbanization of society is an important risk factor for the development of IBD. This meta-analysis suggests that living in urban environments may increase the risk of developing CD and UC. Though, the strengths of association varied among the 40 studies due to heterogeneity between studies. The association between CD and urban environment persisted across a number of stratified analyses that explored clinical and study quality factors. In contrast, the association between UC and urban environment was weaker and less consistent upon sensitivity analyses. Publication bias was not observed suggesting that the association were likely not an artifact of unpublished studies.
Several theories may explain the increased incidence of IBD in urban societies. The Hygiene Hypothesis proposes that the lack of early childhood exposure to enteric pathogens with improved sanitation in urban cities increases the incidence of IBD
5
55
. The lack of exposure to enteric pathogens may lead to a greater susceptibility to develop an inappropriate immunologic response upon exposure to new antigens (e.g. gastrointestinal infection) later in life
56
. Other environmental risk factors of IBD that are more predominant in urban societies include smoking, lack of helminths exposure, and antibiotic use
57
58
59
. Additionally, a recent study demonstrated that air pollution in urban cities was associated with IBD in children
60
. Urban occupations such as driving and manufacturing have also been reported as risk factors for IBD
61
. While a cohesive hypothesis that explains the environmental determinants of IBD has not been proven, this meta-analysis supports further research exploring the environmental risk factors of urbanization and IBD.
Heterogeneity was observed in both the CD and UC cohort studies, whereas statistically significant heterogeneity was not observed for the UC and CD case–control studies. This inconsistency was likely due to the large overall study population in the cohort studies, leading to an overpowered heterogeneity statistical test. In contrast, the case–control studies were small in total study population. Alternatively, differences in heterogeneity between studies may be explained by the difference in the study design and by the intrinsic biases associated with cohort and case–control studies that explore environmental risk factors of IBD.
When studies were stratified by region the association between UC and urban environment persisted only in non-European studies, whereas the association between CD and urban environment remained significant in most regions. Regional differences may be due to the fact that studies originated from countries that differed by ethnicity, prevalence of IBD, and rates of IBD susceptibility genes
62
63
. Alternatively, the small number of studies in each region made appropriate and meaningful inferences challenging. Furthermore, data was lacking for low prevalent regions (i.e. developing world) and thus we could not explore the urban/rural effect in these regions.
We explored whether the year of publication contributed to the heterogeneity observed between studies. For UC, only publications after 1999 were significant after pooling. This time stratified finding may be due to study design differences in more recent publications or the changing pattern of UC diagnosis. In contrast, the urban/rural relationship persisted across all time strata for CD, which suggested a greater strength of association for CD over UC.
The definition of an urban environment was a source of heterogeneity across studies. The studies that clearly defined this exposure included population estimates of the urban and/or rural areas; however, the inconsistency or lack of definitions made study comparisons challenging. We a priori selected the definition of urban as more than 10,000 people because this definition is consistent with census of populations in many Countries (e.g. United Kingdom
37
and Canada
38
). When considering only studies that defined urban as >10,000, the association was no longer significant, which was likely due to the small number of studies with this definition (n = 2 for UC and n = 4 for CD). However, this finding may also reflect that the risk associated with urban society may be driven by greater population size (e.g. living in a metropolis) and/or by population density. Ideally, we would have investigated the potential for a dose response effect to explore whether the risk of IBD increased with increasing population sizes. However, few papers stratified their results by multiple levels of exposure; for example, Tsianos
24
and Ladas
34
stratified their analysis into urban, semi-rural, and rural. Future studies of the urban–rural association should explore the importance of a dose response effect, a threshold value of population size, and population density.
Heterogeneity between studies may have occurred due to the biases associated with cohort and case–control studies. For the case–control studies, a selection bias may have influenced the association if cases were selected by a different mechanism than controls. The controls were grouped into population- and hospital/clinic-based selection categories. For CD, only population-based controls demonstrated a significant association with urban environment; in contrast, the urban–rural association was insignificant in both control populations for UC. Additionally, inconsistency in timing of exposure (e.g. defining urban–rural status at time of diagnosis versus childhood) may have contributed to the heterogeneity between case–control studies. People in rural settings may have less access to health care leading to under diagnosis of IBD. Although the majority of the case–control studies were matched by age and gender, other potential confounders including socioeconomic status and smoking were often not considered. Finally, studies that used administrative databases to identify IBD patients likely introduced misclassification errors. Interpreting the results of meta-analysis should be cautious because pooling data does not address the intrinsic biases of observational studies.
Limitations in our systematic review should be considered. First, the number of studies included in the stratified analyses was small and may have been underpowered. Second, the quality of studies was not always optimal as was shown with the inconsistent definitions of urban/rural and time of exposure. Third, the included studies used population estimates as indicators for urban environment; however, other factors that contribute to an urban setting, such as socioeconomic characteristics, were not considered. Finally, due to the nature of observational studies, a temporal relationship could not be determined. Thus, urban environment cannot be conclusively established as a causal factor in IBD.
Author contributions
G K. Dr. Kaplan participated in conceiving the study idea, developing the study design, interpreting results, and writing of the manuscript. Dr. Kaplan confirms that he has had full access to all the data in the study and had final responsibility for the decision to submit for publication. Dr. Kaplan has seen and approved the final version of the manuscript and has no relevant conflicts of interest. I S S. Dr. Soon participated in conceiving the study and design, acquisition of data, analysis, interpretation of data, drafting manuscript, and statistical analysis. Dr. Soon has seen and approved the final version of the manuscript and has no relevant conflicts of interest. N M. Ms Molodecky participated in conceiving the study concept and design, acquisition of data, analysis and interpretation of data, drafting manuscript, and statistical analysis. Ms Molodecky has seen and approved the final version of the manuscript and has no conflicts of interest. D R. Dr. Rabi participated in conceiving the study study concept and design, analysis and interpretation of data, and critical revision of the manuscript for important intellectual content. Dr. Rabi has seen and approved the final version of the manuscript. Dr. Rabi has no conflicts of interest. W G. Dr. Ghali participated in analysis and interpretation of data, and critical revision of the manuscript for important intellectual content. Dr. Ghali has seen and approved the final version of the manuscriptand has no conflicts of interest. H B. Dr. Barkema participated in analysis and interpretation of data, and critical revision of the manuscript for important intellectual content. Dr. Barkema has seen and approved the final version of the manuscript, and has no conflicts of interest.