PATIENTS AND METHODS

Between 1991 and 1998, King's College School of Medicine, Guy's Hospital, and St Mark's Hospital, London, coordinated a collaborative programme to identify kindreds with at least two affected siblings with IBD. We identified 242 families with a total of 339 affected sibling pairs by patient interview, postal survey of clinicians, systematic search of hospital records, and advertisement through two UK patient organisations. The 339 affected sib pairs comprised 148 with CD only and 115 with UC. There were 76 sibling pairs where one had CD and the other UC. In the 242 pedigrees, there were a total of 658 affected relatives of which 541 were affected siblings. A breakdown of the structure of these families is shown in table 1.

Clinical details for each of the patients were obtained by personal interview, information acquired from clinicians, and review of hospital records. The diagnosis of IBD and classification as CD and UC were determined by standard diagnostic criteria.²⁸ The following information was also collected for each patient: date of birth, sex, age at onset of symptoms, age at diagnosis, presence or absence of extraintestinal disease manifestations, and smoking history (age at which smoking started, number of cigarettes smoked per day, and duration of smoking habit). For patients with CD, additional clinical information included disease localisation (ileal, colonic, or ileocaecal), history of small or large bowel resection, and disease type. The latter was inferred from the dominant clinical and pathological features as defined previously²⁹: in the cicatrising or fibrostenotic form, patients had clinical or radiological evidence of stenosis; in the perforating or fistulising form, patients had evidence of chronic intestinal fistulae or evidence of ileal or colonic perforation. Patients with the inflammatory form of CD had no evidence of stenoses or fistulae. For patients with UC, specific clinical information included extent of disease at the most recent investigation (either pancolitis or distal colitis defined by disease extending proximal or distal to the splenic flexure, respectively) and history of colectomy.

Smoking history was collected by personal interview in the patient's home (by SB, JL, and AM) and divided into two major categories similar to those used by Cottone et al and Lautenbach and colleagues.^30,31

Smoking habit was further subdivided into three groups.

• (B₁) “Smoker at diagnosis”—a patient who had smoked at least five cigarettes daily for at least a year at the time of diagnosis.

• (B₂) “Stopped smoking”—a patient who had given up smoking at least a year before diagnosis.

• (B₃) “Smoker post-diagnosis”—a patient who had only taken up smoking at any time after a diagnosis of IBD had already been established (table 2).

The extent of smoking history was measured in pack years, where 1 pack year constitutes smoking 20 cigarettes a day for one year.

The families were predominantly of European Caucasian origin: 13 families were Jewish, three African, four Asian, and two mixed Caucasian/African. In this familial cohort of patients, smokers were diagnosed with UC at a median age of 31 years, four years older than their non-smoking relatives (median age 27 years).

Study protocols were approved by the ethics committees at the main participating centres and at a further 18 regional hospitals located in London or the South of England. Written informed consent was obtained from all study participants.

Statistical analysis

Data management and statistical analysis were carried out using SPSS version 7.5 (SPSS Inc., Chicago, Illinois, USA). McNemar's test with Yates' continuity correction was used in the analysis of affected sib pairs discordant for either sex, diagnosis, or smoking habit at diagnosis. Comparisons for age at onset of disease were analysed by the Mann-Whitney test. All affected relatives with familial disease were stratified by diagnosis, sex, and smoking history. χ tests were used to assess the influence of smoking and sex on the type, site, and extent of disease. A logistic regression model was used to evaluate the independent effects of smoking with estimation of odds ratios (OR) and 95% confidence intervals (CI) for the relevant outcome variables; these included the presence or absence of the following: CD, UC, site of disease (ileal, ileocolonic, colonic, pancolitis, distal colitis), subtype of CD (perforating, stenosing, inflammatory), and gastrointestinal resection. These variables were selected for inclusion into the logistic regression model as they were those on which we had comprehensive information, and are standard variables for a study of IBD. While we also had comprehensive information about extraintestinal IBD disease manifestations, these occurred too infrequently to be useful in investigating the impact of smoking at diagnosis or at any time before diagnosis. Separate models were used for smoking at diagnosis (yes or no) and whether the patient had ever smoked (yes or no).

RESULTS

Different effects of smoking in females and males on the incidence of CD or UC

Comparison of our overall population of familial patients with IBD (n=658, of whom 48 were excluded because of incomplete clinical information, females 382 (58%), mean current age 46.1 years) showed that smoking at diagnosis was positively associated with CD (OR 3.55; 95% CI 2.50–5.02; p<0.001) and inversely with UC (OR 0.28; 95% CI 0.2–0.4; p<0.001). However, there was a more pronounced risk of CD (OR 5.44; 95% CI 3.33–8.88; p=<0.0001) and protection from UC (OR 0.18; 95% CI 0.11–0.3; p<0.0001) in 151 females smoking at diagnosis compared with 204 female non-smokers, whereas in 98 male smokers at diagnosis that were compared with 157 male non-smokers, these effects were smaller (smoking related CD risk OR 2.11; 95% CI 1.26–3.52; p=0.004; UC protection in smokers OR 0.47; 95% CI 0.28–0.79).

Smoking at diagnosis predisposes to the development of CD rather than UC in sib pairs discordant for disease type

To investigate the effects of smoking on the development of CD and UC in individuals with a similar genetic background, we examined 89 sib pairs with CD or UC who were discordant for smoking at the time of diagnosis (table 3). Of these, 23 sib pairs were also discordant for disease diagnosis, and 21 of these showed UC in the non-smoker and CD in the smoker (OR 10.5; 95% CI 2.6–92; p<0.0001). In only two sib pairs was this situation reversed. Even if we excluded 15 families contributing more than one sib pair, there were 17 sib pairs discordant for both smoking and disease type; of these, 15 had UC in the non-smoker and CD in the smoker (OR 7.5; 95% CI 1.74–67.6; p=0.0036). To examine the possible effects of sex differences in generating discordance, we also looked at 140 sib pairs discordant for sex at diagnosis (table 4); of these, 31 were also discordant for disease diagnosis but here the split was almost even (and not significantly different) between 13 UC females and CD males, and 18 CD females and UC males.

Paired analysis was also undertaken on the 42 CD and 24 UC sib pairs discordant for their smoking habits. For CD, smoking was associated with a greater tendency to develop small bowel disease (OR 3.3; 95% CI 0.86–18.8; p=0.09), and a reduced tendency to develop colonic disease (OR 0.78; 95% CI 0.43–4.1; p=0.6). No differences were seen for the influence of smoking on the subtype of CD or extent of UC, although this may reflect the small numbers of patients available for analysis in this group of two way discordant sib pairs. Indeed, in the overall group of familial patients, those who were smokers when diagnosed with CD had increased disease of the fibrostenosing type, and male smokers were predisposed to develop disease in the ileum (table 5).

Threshold for the environmental effect of smoking on the phenotype of IBD

To examine whether we could detect a lower threshold of cigarette smoking necessary to influence the development of CD, we compared the proportion of patients diagnosed with CD and UC as a function of when they had smoked and the number of pack years (fig 1, table 2). This showed that the increase in susceptibility to CD was associated with a very short smoking history at diagnosis: it was seen in those who had smoked 5 or fewer pack years at diagnosis which was maintained with a more extensive smoking history compared with non-smokers. (A single pack year constitutes smoking 20 cigarettes a day for one year.) However, those patients who gave up smoking at least a year before diagnosis were significantly less likely to develop CD, regardless of the amount they had smoked previously. It was clear that smoking at diagnosis was itself the risk factor for developing CD rather than the cumulative amount smoked prior to diagnosis.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Proportion of patients with familial inflammatory bowel disease diagnosed with Crohn's disease as a function of the cumulative pack years smoked. A pack year is defined as smoking 20 cigarettes a day for one year. Patients who were smokers at diagnosis were compared with former smokers who had given up at least a year before the diagnosis was made. The horizontal line represents the proportion of non-smokers diagnosed with CD (124 of 286 patients; 43.4%). Numbers above each bar give the total number of patients in each group from which the proportion with Crohn's disease was derived.

DISCUSSION

This study confirms that smoking is an important aetiological factor in the pathogenesis of IBD, and that it predisposes to CD, especially in women. However, the existence of sib pairs in which the smoker almost always has CD and the non-smoker UC shows that in some individuals who are closely related smoking can apparently influence whether the phenotype is UC or CD. In this familial cohort, smokers were diagnosed with UC at a median age of four years older than non-smokers, which is also consistent with tobacco usage inhibiting UC and enhancing CD pathogenetic mechanisms.

There are three important caveats in the interpretation that smoking cannot only exacerbate CD and heal UC but also (where the genetic risks are evenly balanced) result in a shift between the two forms. Firstly, while matched analysis of affected sibling pairs has the advantage that it directly compares patients with similar genetic and social susceptibility, our sib pairs were not genetically identical, and therefore gene differences between them could still influence the disease that develops: however, the strength of the effect of smoking in those pairs discordant for disease type and smoking was so strong that it is hard to see how the smoker could almost always have the CD genetic predisposition and the non-smoker the UC genetic predisposition by chance. Secondly, there are many examples of sib pairs discordant for smoking but not disease type (66 of 89 discordant for smoking), and so the effect was only seen in a proportion of cases: in other words, it is likely that genetic risks usually determine whether the disease is CD or UC, and only where these risks are evenly balanced between the two conditions can smoking determine the phenotypic outcome. Thirdly, in discordant sib pairs we could not distinguish whether the mechanism of this effect was due to exacerbation of the CD pathogenesis, protection from the UC pathogenesis, or both. To illustrate this point, one can consider the theoretical situation if enough identical twins were available, when one might be able to examine the strength of these different effects: (i) smoking exacerbates CD (smoker with CD, non-smoker disease free); (ii) smoking protects from UC (smoker disease free, non-smoker UC); (iii) pure shift of phenotype (smoker CD, non-smoker UC). In practice however, understanding the pathogenic mechanisms underlying tobacco usage are likely to come from animal models.

We were unable to detect a lower threshold (below 5 pack years) for the effects of smoking on the development of CD, and indeed it appeared to be effectively reversed regardless of previous exposure of tobacco within a year of giving up smoking. This suggests that the influences of tobacco on disease pathogenesis are on the immunopathogenic mechanisms at disease onset as opposed to the cumulative damage seen for example in atherosclerosis or chronic obstructive airways disease. Studies of the effectiveness of stopping smoking following diagnosis are confounded by the effects of medical intervention, and we cannot reliably assess this.

We found no evidence to indicate that recall bias may have influenced the results of this study. Although it is known that patients can under or over report their cigarette consumption, self reported smoking habit has been shown to exhibit good reliability and validity in cross sectional studies.^32,33 Indeed, compared with affected patients, relatives without IBD who were smokers had a greater cumulative tobacco consumption (in pack years per smoker). While we cannot discount a possible effect of smoking in suppression of colitis symptoms directly, there was no evidence of lead time bias (age of symptomatic onset to age of diagnosis), which was similar in both smokers and non-smokers who developed UC.

It is well recognised that smoking can be associated with development of systemic autoimmune disease,³⁴ including Goodpasture's syndrome, rheumatoid arthritis,^35,36 and Grave's disease, and increased severity of ophthalmopathy.³⁴ The mechanisms underlying the interaction between tobacco consumption and these diseases remain unclear and there are multiple effects of smoking on immune and inflammatory functions, including increased numbers of circulating and tissue neutrophils and lymphocytes and complex alterations of macrophage, B and T cell functional responses, depressed natural killer cell function, and increased complement activation.^37–39 In the oral mucosa, periodontal disease resulting from inflammatory responses against the dental microflora is also exacerbated by smoking.^36,40

The effects of cigarette smoke have been evaluated in the rodent model of IBD initiated by intestinal exposure to the haptens trinitrobenzene sulphonic acid (TNBS) or dinitrobenzene sulphonic acid (DNBS). In rats treated with either hapten, passive smoking by the animals exacerbated the resultant disease. Exacerbation of the TNBS effects was found to be associated with increased free radical production and inducible nitric oxide synthase activity⁴¹ whereas aggravation of DNBS colitis was reversed by hexamethonium, suggesting that a neural pathway is involved.⁴²

Chronic IBD is usually divided into CD and UC on the basis of clinical and pathological distinctions, although there are intermediate forms, shared genetic linkages, and families containing members with each condition. We have shown here that in some cases smoking can probably displace the phenotype of chronic IBD from UC to CD—an example of how environmental influences interact with the genetic background to influence not just whether there is disease but also the type of disease that develops. The observation has important implications for studies in the genetic and immunopathological mechanisms of IBD. Finally, the apparent protection that smoking offers against UC may be in part a shift of the cases towards CD.