The authors have declared that no competing interests exist.
Current address: Department of Clinical Microbiology, Hvidovre Hospital, Hvidovre, Denmark
Current address: Institute of Preventive Medicine, Bispebjerg and Frederiksberg Hospital, Copenhagen, Denmark.
Crohn’s disease (CD) has traditionally been associated with weight loss and low BMI, yet paradoxically obesity has recently been suggested as a risk factor for CD, but not for ulcerative colitis (UC). We therefore hypothesized that the relation between BMI and CD is U shaped.
To conduct a large population-based prospective cohort study of BMI and later risk of IBD, taking age at IBD diagnosis into account.
A cohort of 74,512 women from the Danish National Birth Cohort, with BMI measured pre-pregnancy and 18 months after delivery, was followed for 1,022,250 person-years for development of IBD, according to the Danish National Patient Register. Associations were tested by Cox regression.
Overweight subjects (25≤BMI<30 kg/m2) had the lowest risk of CD, whereas obesity (BMI≥30kg/m2) increased the risk of CD at all ages, and low BMI (BMI<18.5kg/m2) associated with CD diagnosed at age 18-<40 years. Hence, using normal weight subjects as the reference, adjusted HRs for risk of developing CD (at age 18-<40 years) were 1.8(95%CI, 0.9–3.7) for underweight, 0.6(0.3–1.2) for overweight, and 1.5(0.8–2.7) for obese individuals (pre-pregnancy BMI). HRs were greater for BMI determined 18 months after delivery. Splines for CD risk according to waist:height ratio confirmed a U-shaped relationship with CD occurring <40 years, and a linear relationship with CD diagnosed at age 40+. There was no relationship between BMI and risk of UC.
For the first time, we demonstrate that both high BMI and low BMI are risk factors for CD. Underweight may be a pre-clinical manifestation of disease being present many years before onset with obesity being a true risk factor. This raises the question as to whether there may be two distinct forms of CD.
Traditionally Crohn’s disease(CD) has been associated with weight loss and underweight, which are common in newly diagnosed subjects with CD particularly those diagnosed at a young age[
In an initial case control study, underweight and obese subjects were over-represented at diagnosis. Obese subjects tended to be diagnosed at older ages and the underweight mainly at younger ages. In subjects with UC there was no association[
There is a growing awareness that there may be a prolonged pre-clinical phase to Crohn’s disease[
The Nurse’s Health Study(NHS)[
We aimed to extend the follow-up time of the DNBC cohort and conduct detailed analyses to investigate whether obesity is independently associated with future risk of CD, whether low BMI is independently associated with future development of CD, and whether the relationship between BMI and risk of development of CD is truly U-shaped. Taking advantage of the fact that the DNBC unlike the NHS includes a high proportion of Montreal A2(age at diagnosis 17-<40) patients as well as A3(age at diagnosis 40+), we speculated that a U-shaped relationship would be most clearly manifested in the A2 group, as opposed to the A3 group.
The study population consisted of women participating in the DNBC recruited by general practitioners in early pregnancy during 1996–2002. The women were interviewed in 16th (interview I) and 32nd (interview II) week of gestation and 6(interview III) and 18 (interview IV) months after delivery (
In order to calculate pre-pregnancy BMI and BMI 18 months after delivery, questions concerning height (“How tall are you?”) and weight (“What was your weight before the pregnancy?”) from interview I and IV, respectively, were used. Self-reported heights and weights have been shown to be reliable[
Waist:height ratio was determined in a limited number of women 18 months after birth using self-reported waist measurement[
Linkage to registers were done using a 10-digit unique personal identification number registered in the Danish Civil Registration System[
Confounding variables chosen a priori were obtained from the DNBC. These included alcohol consumption per week prior to pregnancy (0 units, 1–7 units, ≥8 units), socio-occupational status at pregnancy (based on the woman’s current or most recent occupation within the last six months, or if in school, on type of education; 1. Long education or leaders in large companies; 2. Middle long education or leaders in small companies; 3. Short education, vocational, or under education, 4. Unskilled, other work, or receiving unemployment benefits; 5. On state welfare; variable defined previously [
Hazard ratios (HRs) with 95% confidence intervals (CIs) were calculated using Cox regression. Women were followed (using pre-pregnancy BMI from interview I) or 18 months after delivery (using BMI from interview IV) until CD or UC development, emigration, death, or end of follow-up, 31st of December, 2014 with the woman’s age as underlying timescale. Follow-up started from the 6 months post-pregnancy interview. The women only contributed with time while registered as living in Denmark (i.e. women moving away from Denmark for a time period but later moved back contributed with time in both periods living in Denmark). Similar analyses were performed for the two different BMI exposures. Alcohol, socio-occupational status and parity were accounted for by including them as strata variables in the Cox regression whereas smoking and risk factors were adjustment variables fixed at baseline. Exercise was excluded as a confounder in the final results since it showed an effect of less than 3% on estimates, did not seem related to either risk of disease or to BMI unlike the other covariates, and to increase power.
Cubic splines restricted to be linear in the tails were constructed to visualize a potential linear or U-shaped relationship between risk of IBD, BMI and in a limited number of subjects waist:height ratio. The splines were estimated for both the total cohort and for age-groups (with 40 years of age as cutting point). No violation of the proportional hazards assumption was found evaluated by the empirical score process[
We additionally divided the women into either smokers or no indication of smoking to evaluate whether smoking had an effect on estimates especially in the underweight category.
The main cohort of 74,512 women (the sub-cohort of women with BMI 18 months measurements consisted of 48,262 women) were followed for a median of 13.9 years (lower-upper quartile: 12.7–15.0). In
3302 (4,4) | 50,776 (68,1) | 14,403 (19,3) | 6031 (8,1) | |
2095 (4,3) | 32,473(67,3) | 9933 (20,6) | 3729 (7,7) | |
30,0 (27.1:33.3) | 30.8 (28.1:33.9) | 30.5 (27.7:33.6) | 30.3 (27.4:33.3) | |
0 units | 555 (16.8) | 5428 (10.7) | 2101 (14.6) | 1274 (21.1) |
1–7 units | 2466(74.7) | 39604(78.0) | 11122(77.2) | 4421 (73.3) |
8+ units | 281 (8.5) | 5744 (11.3) | 1180 (8.2) | 336 (5.6) |
Long edu/leaders in large companies | 680 (20.6) | 11990 (23.6) | 2303 (16.0) | 620 (10.3) |
Middle long edu/leaders in small companies | 879 (26.6) | 16987(33.5) | 4344 (30.2) | 1536 (25.5) |
Short edu/vocational/under edu | 1314(39.8) | 17787(35.0) | 6187 (43.0) | 2869 (47.6) |
Unskilled/other work/unemployment benefits | 305 (9.2) | 3081 (6.1) | 1250 (8.7) | 791 (13.1) |
On state welfare | 124 (3.8) | 931 (1.8) | 319 (2.2) | 215 (3.6) |
0 children | 1618 (49.0) | 25613 (50.4) | 6691 (46.5) | 2784 (46.2) |
1 child | 1219 (36.9) | 17360 (34.2) | 5269 (36.6) | 2213 (36.7) |
2+ children | 465 (14.1) | 7803 (15.4) | 2443 (17.0) | 1034 (17.1) |
Smoking reported in interview I, II, or III | 1252(37.9) | 13729(27.0) | 3919 (27.2) | 1612 (26.7) |
No smoking reported in interview I, II, or III | 2050(62.1) | 37047(73.0) | 10484(72.8) | 4419 (73.3) |
Yes | 246 (7.5) | 3693 (7.3) | 1146 (8.0) | 525 (8.7) |
No | 3056(92.5) | 47083(92.7) | 13257(92.0) | 5506 (91.3) |
Yes | 4 (0.1) | 106 (0.2) | 65 (0.5) | 53 (0.9) |
No | 3298(99.9) | 50670(99.8) | 14338(99.5) | 5978 (99.1) |
Yes | 2427 (73.5) | 37496 (73.8) | 10847 (75.3) | 4503 (74.7) |
No | 875 (26.5) | 13280 (26.2) | 3556 (24.7) | 1528 (25.3) |
Yes | 161 (4.9) | 1636 (3.2) | 556 (3.9) | 320 (5.3) |
No | 3141(95.1) | 49140(96.8) | 13847(96.1) | 5711 (94.7) |
Yes | 71 (2.3) | 676 (1.4) | 91 (0.7) | 32 (0.6) |
No | 2995 (97.7) | 47139 (98.6) | 13515 (99.3) | 5638 (99.4) |
2233(67.6) | 30923(60.9) | 9346(64.9) | 4107(68.1) | |
1-<120mins/week | 555(16.8) | 10714(21.1) | 2869(19.9) | 1139(18.9) |
120-<240min/week | 329(10.0) | 5998(11.8) | 1523(10.6) | 552(9.2) |
240-<420min/week | 129(3.9) | 2332(4.6) | 479(3.3) | 156(2.6) |
420+mins/week | 56(1.7) | 809(1.6) | 186(1.3) | 77(1.3) |
Years_since_entry | Outcome_type | ||
---|---|---|---|
Frequency | CD(column percent) | UC(column percent) | Total |
11(8.03) | 39(8.71) | 50 | |
6(4.38) | 28(6.25) | 34 | |
12(8.76) | 52(11.61) | 64 | |
13(9.49 | 32(7.14) | 45 | |
8(5.84) | 47(10.49) | 55 | |
11(8.03) | 23(5.13) | 34 | |
10(7.30) | 28(6.25) | 38 | |
12(8.76) | 29(6.47) | 41 | |
7(5.11) | 33(7.37) | 40 | |
6(4.38) | 24(5.36) | 30 | |
12(8.76) | 26(5.80) | 38 | |
10(7.30) | 29(6.47) | 39 | |
6(4.38) | 27(6.03) | 33 | |
8(5.84) | 21(4.69) | 29 | |
5(3.65) | 10(2.23) | 15 | |
137 | 448 | 585 |
During 1,022,250 years of follow-up, 137 women (0.2%) developed CD and 448 (0.6%) developed UC.
No of persons | PYRFU |
No of cases in total | No of cases < 40 years | No of cases ≥ 40 years | Unadjusted HR | Adjusted HR |
Unadjusted HR < 40 years | Adjusted HR < 40 years |
Unadjusted HR ≥ 40 years | Adjusted HR |
|
---|---|---|---|---|---|---|---|---|---|---|---|
BMI<18.5 | 3302 | 44982.3 | 12 | 11 | 1 | 2.1 (1.1–3.9) |
1.4 (0.7–2.9) | 2.5 (1.3–4.7) |
1.8 (0.9–3.7) |
0.8 (0.1–5.8) | 0.0 (0.0-) |
18.5≤BMI<25 | 50776 | 690743.2 | 85 | 63 | 22 | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) |
25≤BMI<30 | 14403 | 195682.4 | 23 | 14 | 9 | 0.9 (0.6–1.5) | 0.8 (0.5–1.4) | 0.8 (0.4–1.4) | 0.6 (0.3–1.2) | 1.5 (0.7–3.3) | 1.5 (0.7–3.2) |
BMI≥30 | 6031 | 81580.4 | 17 | 12 | 5 | 1.7 (1.0–2.8) | 1.5 (0.9–2.6) | 1.5 (0.8–2.8) | 1.5 (0.8–2.7) |
2.1 (0.8–5.6) | 1.6 (0.6–4.9) |
BMI<18.5 | 2095 | 26645.3 | 7 | 7 | 0 | 2.1 (0.9–4.6) |
1.6 (0.7–3.7) |
2.8 (1.2–6.3) |
2.1 (0.9–5.0) |
0.0 (0.0-.) | 0.0 (0.0-.) |
18.5≤BMI<25 | 32473 | 410895.7 | 51 | 36 | 15 | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) |
25≤BMI<30 | 9933 | 125630.2 | 9 | 2 | 7 | 0.6 (0.3–1.2) | 0.6 (0.3–1.1) | 0.2 (0.0–0.7) | 0.2 (0.0–0.7) | 1.6 (0.6–3.9) | 1.5 (0.6–3.7) |
BMI≥30 | 3729 | 46880.7 | 12 | 9 | 3 | 2.0 (1.1–3.8) |
1.9 (1.0–3.7) |
2.0 (1.0–4.2) |
1.9 (0.9–4.0) |
2.0 (0.6–6.8) | 1.9 (0.5–6.9) |
BMI<18.5 | 3302 | 44982.3 | 25 | 18 | 7 | 1.2 (0.8–1.9) | 1.1 (0.7–1.7) | 1.2 (0.8–2.0) | 1.1 (0.7–1.9) | 1.3 (0.6–2.7) | 1.0 (0.4–2.3) |
18.5≤BMI<25 | 50776 | 690743.2 | 308 | 212 | 96 | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) |
25≤BMI<30 | 14403 | 195682.4 | 85 | 59 | 26 | 1.0 (0.8–1.2) | 0.9 (0.7–1.2) | 1.0 (0.7–1.3) | 1.0 (0.7–1.3) | 1.0 (0.7–1.6) | 0.9 (0.5–1.4) |
BMI≥30 | 6031 | 81580.4 | 30 | 20 | 10 | 0.8 (0.6–1.2) | 0.7 (0.5–1.1) | 0.8 (0.5–1.2) | 0.7 (0.5–1.2) | 1.0 (0.5–1.9) | 0.7 (0.4–1.5) |
BMI<18.5 | 2095 | 26645.3 | 13 | 8 | 5 | 1.0 (0.6–1.8) | 1.0 (0.6–1.7) | 0.9 (0.4–1.8) | 0.8 (0.4–1.7) | 1.4 (0.5–3.4) | 1.4 (0.5–3.4) |
18.5≤BMI<25 | 32473 | 410895.7 | 194 | 130 | 64 | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) | 1 (Ref.) |
25≤BMI<30 | 9933 | 125630.2 | 46 | 26 | 20 | 0.8 (0.6–1.1) | 0.7 (0.5–1.0) | 0.6 (0.4–1.0) | 0.6 (0.4–1.0) | 1.1 (0.6–1.8) | 1.0 (0.6–1.7) |
BMI≥30 | 3729 | 46880.7 | 20 | 13 | 7 | 0.9 (0.6–1.4) | 0.8 (0.5–1.3) | 0.8 (0.5–1.4) | 0.7 (0.4–1.3) | 1.1 (0.5–2.4) | 0.9 (0.4–2.2) |
# Patient years of follow-up.
@ Adjusted for smoking, appendectomy, socio-economic status, parity, diabetes, oral contraceptive pill use, proton-pump inhibitor, vegetarian diet, and alcohol consumption. Exercise was excluded as it had no association with risk of disease or BMI and merely reduced power.
* p<0.05 v overweight category.
This is reflected in the splines which are presented in
For pre-pregnancy BMI for the CD spline p = 0.2777 for UC 0.4146. For 18 months after delivery for CD p = 0.0354 and for UC p = 0.6359. 95% Cis are displayed.
In
Compared to the overweight lowest risk group, all differences with low BMI and obesity were significant with the exception of obesity and low BMI pre-pregnancy and both low and high BMI in the 40+ age group.
The probability that the splines for the under and over 40 years of age are different pre-pregnancy = 0.0575, and for the 18 months post delivery p = 0.0618. F or UC the respective probabilities were p = 0.4989 and p = 0.3395. The spline age 18-<40 and 95% Cis are shaded and the spline for age 40+ and 95% Cis are unshaded.
There was no significant difference in time to diagnosis by BMI category (
Number of cases unadj. | Number of cases adj. | Median years (lower-higher quartile) | |
---|---|---|---|
BMI<18.5 | 12 | 9 | 7.0 (2.8–10.0) |
18.5≤BMI<25 | 85 | 81 | 6.9 (3.7–10.3) |
25≤BMI<30 | 23 | 21 | 6.9 (2.7–11.3) |
BMI≥30 | 17 | 16 | 5.1 (3.3–7.6) |
Age < 40 years | 100 | 93 | 5.5 (2.7–8.1) |
Age ≥ 40 years | 37 | 34 | 10.4 (8.2–12.8) |
BMI<18.5 | 25 | 23 | 7.2 (2.1–10.1) |
18.5≤BMI<25 | 308 | 293 | 6.5 (3.1–10.2) |
25≤BMI<30 | 85 | 80 | 4.9 (2.5–8.9) |
BMI≥30 | 30 | 27 | 5.4 (2.8–10.3) |
Age < 40 years | 309 | 294 | 4.4 (2.3–7.4) |
Age ≥ 40 years | 139 | 129 | 10.6 (8.0–12.7) |
We determined whether a history of smoking may explain the association between low BMI and risk of CD by performing an analyses restricted to non-smokers and smokers. For pre-pregnancy BMI, the HRs for CD in non-smokers were 2.28 (0.97–5.33), 0.90 (0.48–1.70) and 1.96 (1.02–3.79) for underweight, overweight and obese relative to normal weight subjects, as opposed to 1.65 (0.70–3.91), 1.00 (0.51–1.95), 1.29 (0.55–3.05) in smokers. Likewise, for BMI 18 months after delivery the respective HRs for CD were 1.71 (0.52–5.60), 0.62 (0.26–1.48) and 1.63 (0.68–3.89) in non-smokers and 2.16 (0.73–6.35), 0.49 (0.15–1.66), 2.68 (1.07–6.71) in smokers. Tests for interaction between BMI and smoking for CD were non-significant, p = 0.91 and p = 0.09 for UC for pre-pregnancy BMI. There was no evidence of interaction with oral contraceptive use either p = 0.95 for CD and 0.21 for UC.
Waist measurements were obtained in 21,014 subjects at 18 months after delivery covering 32 incident cases. For CD, unadjusted HRs for successive quartiles of waist:height ratio were 1.54 (95% CI, 0.60–3.97) for the first quartile, the second quartile was the reference group, 0.57 (0.17–1.93) and 1.42 (0.54–3.73) for the fourth quartile. Adjusted the respective values (there were 2 fewer cases) were 1.41 (0.53–3.75), 1, 0.53 (0.16–1.83) and 1.10 (0.40–3.00). The corresponding spline shown in
For CD spline p = 0.0256 and for UC p = 0152. Test for trend for CD p = 0.9032 and for UC 0.059.
This large population-based cohort study suggests an association between BMI and risk of developing CD, but not UC. Risk of CD was lowest in the overweight suggesting a U-shaped relationship between both pre-pregnancy and 18 after delivery BMI and risk of CD, although findings were not statistically significant after full adjustment relative to normal weight. When the analysis was restricted to patients aged 18-<40 years at diagnosis, the magnitude of the U-shaped pattern was accentuated, mainly because of an increase in the HR associated with low BMI, again with some attenuation after adjustment but remaining significantly greater than the HR for overweight BMI at both time points. For the over 40s at diagnosis the scarcity of subjects with a low BMI prior to diagnosis is notable, there being an apparent non-significant rise in HR with increasing BMI.
The spline of a subset of women with waist:height ratios confirmed a statistically significant U-shaped relationship in line with that observed for BMI. Splines for the 18 month after delivery BMI confirmed a U shaped relationship, whilst the spline for pre-pregnancy although supportive, was not statistically significant. Splines comparing the association of BMI with risk of CD in the 18-<40s and over 40s confirmed with borderline statistical significance that the risk of developing CD according to BMI is different in the two age groups with a U-shaped relation in the former and a linear association in the latter.
A careful comparison with Nurses Health Study (NHS) is warranted as there are similarities, but also some illuminating differences. Our findings mirror those from the NHS[
The finding of a lack of association between low BMI with risk of CD solely in the over 40s could be spurious. It may be a chance result of the smaller number of cases (only ¼ of cases were aged 40+). Alternatively, it could be that the longer time to development of IBD in those subjects who developed it over the age of 40 meant that low BMI was not evident at baseline, with a reduced BMI emerging closer to the manifestation of the disease. The time to development of CD was longest in the underweight category and shortest in the obese category argues against this as an explanation. Our previously reported studies also support this explanation: low BMI at diagnosis was mainly apparent in younger subjects[
The strength of this study is that it is a prospective large population based cohort study with index cases straddling the conventional Montreal classification between A2 and A3, although a sizeable proportion of patients do develop Crohn’s disease at an even earlier age. The cases were well characterized and required at least two clinical contacts to be included as a case. It is possible that some of the cases were misclassified, but this would only have weakened our findings. Another strength is the broad range of risk factors for IBD which were controlled for. Furthermore, the subjects represent the full social strata of Danish society, unlike the NHS.
An important weakness is the relatively small number of cases yielding p values in some instances only just or short of statistical significance particularly after adjustment. This is in the nature of prospective cohort studies for relatively uncommon diseases such as IBD. However, associations were consistent when looked at in a number of different ways. Furthermore the attenuation of effect sizes after adjusting for confounders was modest in the underweight and virtually non-existent for the obese. That our findings for BMI were confirmed on analyzing waist:height ratio adds confidence that the findings have not arisen by chance.
Another weakness of this study is that it is restricted to women who became pregnant. Nevertheless, this represents a sizeable minority of the population whose study should inform us about CD in general. Underweight and obesity can be associated with sub-fertility, and hence the extremes of BMI may be under-represented in this cohort, but if anything this potential source of bias would be expected to diminish the effect sizes observed.
Whilst we were able to control for all the main risk factors for CD we were unable to control in detail for diet apart from vegetarianism. However in the NHS controlling for dietary components did not have any effect on risk estimates. Also some of the risk factors controlled for relate to pre-pregnancy or pregnancy exposure eg oral contraceptive use and exercise, with no information regarding use post pregnancy, whereas others such as smoking status were collected both during and after pregnancy. Again, in the NHS controlling for these risk factors had no effect.
It would have been informative to have the full Montreal classification for these subjects, but this was not available. It would be of particular interest to determine whether the site of the GI tract affected differed with increasing BMI and increasing age at diagnosis. Colonic location for instance becomes more common with increasing age[
BMI is a relatively poor measure of the visceral adiposity which is believed to be important in the metabolic and systemic effects of obesity. More specifically in the context of CD, visceral fat area which does not necessarily correlate with BMI, is the best marker of post-operative recurrence[
A number of mechanisms have now been recognized whereby obesity could contribute to risk of CD. These include alterations in gut permeability[
Whilst the current study and the NHS do not support confounding of the association of obesity by standard risk factors, the most likely being lack of fibre intake and exercise, some other feature of dietary intake such as fat consumption is a possibility. We cannot discount this.
It is less likely that low BMI is causally related to risk of CD, but rather a sub-clinical early manifestation of the disease longer than the clinical prodrome, which is usually less than 2 years[
Many factors related to CD pathogenesis have been proposed as playing a role in this weight loss, including the anorectic and catabolic effects of pro-inflammatory cytokines[
This study provides confirmation of the association between obesity and risk of CD and for the first time show that low BMI can precede the development of CD by many years. This finding requires replication in other cohorts and in men. The possibility that there may be two forms of CD—a classical skinny type and an obese phenotype is raised, requiring further investigation into differing pathogenesis and treatment. If weight loss occurs many years before diagnosis early markers of disease are required possibly allowing intervention before clinical presentation. This would offer the hope of preventing the irreversible intestinal damage frequently present when first diagnosed.