Meta-analysis contraceptive - Gut · The included studies could be classified as cohort or case-controls. Although each design offers weaknesses and strengths, the RR estimates ofeachstudywerein

Gut 1995; 37: 668-673

Meta-analysis of the role of oral contraceptiveagents in inflammatory bowel disease

P G Godet, G R May, L R Sutherland

AbstractNumerous epidemiological studies havebeen performed to determine factors thatmight contribute to the development ofinflammatory bowel disease. Although therole oforal contraceptive agents in Crohn'sdisease (CD) and ulcerative colitis (UC)have been assessed, most studies were ofsmall sample size and characterised by lowstatistical precision. A meta-analysis wasperformed to increase the statistical powerand to investigate the association betweenthe use of oral contraceptives and thedevelopment ofCD and UC. The study wasbased on a search of a Medline databasefrom 1975 to October 1993 and a review ofreference lists from published articles,reviews, symposia proceedings, andabstracts from major gastrointestinalmeetings. All studies specifically designedto evaluate this association were selected.The combined results ofnine studies - twocohort studies (30 379 unexposed and 30 673exposed patients) and seven case-controlstudies (482 CD, 237 UC, and 3198controls) - which satisfied our selectioncriteria were evaluated. The pooled rela-tive risk (adjusted for smoking) associatedwith oral contraceptive use was 1.44 (1.12,1.86) for CD and 1.29 (0.94, 1.77) for UC.These results suggest modest associationsbetween the use of oral contraceptives andthe development of CD and UC. As theseassociations are weak, non-causalexplanations for the findings cannot beeliminated.(Gut 1995; 37: 668-673)

Keywords: inflammatory bowel disease, oralcontraceptives, meta-analysis.

GI Research Group,University of Calgary,Calgary, Alberta,CanadaP G GodetG R MayL R Sutherland

Correspondence to:Dr P G Godet, PhysiologieDigestive, H6pital Notre-Dame, 1560 Sherbrooke Est,Montreal, Quebec, CanadaH2L 4M1.

Accepted for publication21 April 1995

Numerous epidemiological studies have beenperformed to determine factors which mightcontribute to the development of Crohn'sdisease (CD) and ulcerative colitis (UC).Although many factors have been suggested,including smoking, family history, geographiclocation, socioeconomic status, diet, and oralcontraceptive agents, only the first two factorshave a firmly established association.l1 Manycase reports have been published since 1970suggesting a possible causal association betweenthe use of oral contraceptives and a Crohn's likedisease which was reversible after stopping oralcontraceptives.5-l0 Furthermore, histopatholog-ical studies have proposed a pathogenic modelin which CD might be caused by multifocalinfarction in which smoking and oral contracep-tives act as potentially thrombogenic agents.1' 12

Recent epidemiological studies have assessedthe role of oral contraceptives in CD or UC, orboth.13-27 Most of these studies, however, wereof small sample size and characterised by lowstatistical power. We therefore performed ameta-analysis to increase the statistical power toaddress this important clinical question. Ourprimary objective was to investigate the associa-tion between the use of oral contraceptives andthe development of CD and UC.

MethodsWe performed a meta-analysis, based on asearch in English and French language using aMedline database under the list terms 'Crohn'sdisease', 'ulcerative colitis', or 'inflammatorybowel disease' and 'oral contraceptive agents'from 1975 to 1993. We also reviewed referencelists from published articles, reviews, symposiaproceedings, and abstracts from major gastro-intestinal meetings. We selected all case-control and cohort studies specifically designedto evaluate this association. An acceptablecomparison group had to be formed by either anon-user population (cohort studies) or byinpatients, outpatients, or neighbours of thecases (case-control studies). The identificationof women as never, ever, or current users wasnot required for this analysis, although most ofthe studies provided this information. Norestrictions were imposed for the method bywhich information was obtained (by mail,phone, or interview in all studies) or thenumber of subjects in the comparison group.

While the use of prevalent cases maygenerate a recall bias,28 studies using thismethodology have taken precautions tominimise bias (for example, including onlypatients with disease onset within three or fouryears of interview21 or by confirming dataobtained by questionnaire with data fromreview of the charts of patients and controls27).The use of incident cases would be preferablebut inflammatory bowel disease is notcommon enough to allow a single investigatorsuch an approach.

Statistical methodsTo pool relative risks (RR) from severalstudies, we used a meta-analytic method29 thatyields a weighted average of the log RR fromthe individual studies, in which the weightdepends on the inverse of the variance of thelog RR.29-31 This approach gives larger studiesgreater weight in the summary measure. Toconvert confidence intervals (CI) from the dif-ferent studies into estimates of the variance ofthe log RR to the log scale, we assumed that

668

on July 4, 2020 by guest. Protected by copyright.

http://gut.bmj.com

/G

ut: first published as 10.1136/gut.37.5.668 on 1 Novem

ber 1995. Dow

nloaded from

Meta-analysis of the role of oral contraceptive agents in inflammatory bowel disease

the 95% CI was obtained by adding and sub-tracting 1.96 times the standard error of thelog RR. We divided the interval length by 3.92to obtain an approximate standard error.

Although we combined the results fromcohort and case-control studies when possible,we evaluated separately the effect of oralcontraceptives for CD and UC. We distin-guished between studies that were adjusted forsmoking and those which were not. Althoughthe results will be reported separately for sub-groups, we tested for homogeneity29 for eachsubgroup. Potentially, heterogeneity may

result when results from studies giving ratesadjusted only for smoking are combined withstudies adjusted for smoking, diet, and educa-tion. The statistical tests for homogeneity gave

p values greater than 0.30 for CD studiesadjusted for smoking, for CD studies un-

adjusted for smoking, and for UC studiesadjusted for smoking. We thus assumed thatexisting heterogeneity between studies was

small enough to be reasonably ignored.Homogeneity has not been tested for the twoUC studies unadjusted for smoking becausethe value of the pooled risk estimate was

unclear.We recalculated the RRs whenever raw

data were provided. None of the recalculatedestimates differed appreciably from thosereported in the original research.

Results

EXCLUDED STUDIESTable I shows a summary of the characteristicsof excluded studies. Two studies13 14 specifi-cally addressed recurrence not development ofCD and were excluded. The Ramcharanstudy'5 did not report separate numbers forCD and UC. The Rhodes,16 Entrican,17 andLichtarowicz18 studies did not include appro-priate control groups.

INCLUDED STUDIESOf the reported studies (Tables II and III) ninehad sufficient data to meet the eligibility criteria.The seven case-control studies20 2 23-27

gathered information about 719 cases of

inflammatory bowel disease (482 CD and237 UC) and 3198 controls, while the two

cohort studies'922 followed 32673 users and30 379 women who did not use oral contra-ceptives.The CIs for the RR estimate in seven of the

eight reported studies of CD included unity.The estimated RRs ranged from 0.7 to 2.5.The pooled RR for cohort and case-controlstudies adjusted for smoking was 1.44 (1.12,1.86). The pooled estimate, unadjusted forsmoking, was 1-68 (0.97, 2.88).For UC, in the four reported studies, the

CIs for the RR again included unity. The RRsranged from 0.7 to 2.4. The pooled adjustedRR was 1-29 (0.94, 1-77). The pooled esti-mate, unadjusted for smoking, as 1.68 (097,2 88).The RRs for each study and the pooled

estimates are displayed in figures 1 and 2. Wedid not show graphically the unadjusted esti-mated for UC because of the inconsistency ofthese results for only two studies. Summariesof all the studies included are presented in theAppendix.

DiscussionIn this report we reviewed nine studies thatinvestigate a possible link between oral contra-ceptive use and the development of CD andUC. The studies were characterised by smallsample size, selection of different comparisongroups, and non-uniformity in adjusted vari-ables, although existing heterogeneity betweenstudies was small enough to be reasonablyignored. For the cohort studies, users and non-

users were recruited from a population attend-ing family practice clinics. For case-controlsstudies, cases included were inflammatorybowel disease in- or outpatients and controlsselected as non-relatives of inflammatorybowel disease cases. The purpose of oralcontraceptive use was mainly for contra-ception. Combining results from selectedstudies seems appropriate.

In general, the reported estimates of an

association between the use of oral contra-ceptives and either CD or UC are remarkablyconsistent in terms of the direction of the find-ings - an increased RR of CD and UC in oral

TABLE I Summary ofstudies excludedfrom the meta-analysis

NumbersAuthors (CD, UC) Comments

Sutherland et al' 3 87 Retrospective/case-control; association between need of second surgery in CD in OC users andnon-users; adjusted for smoking, age, site.

Wright et all'4 162 Descriptive data on women with CD. Proportion of users and non-users in ileal, ileocaecal,and colonic CD groups. No control group.

Ramcharan et all 5 107, 165 Prospective, cohort study examining IBD incidence in OC users v never and ever users.Cannot separate the 2 specific RRs for CD and UC.

Rhodes et all 6 100 OC use in consecutive outpatients with IBD. No controls (matching between colonic CD andUC). Higher OC use rate in colonic CD than in UC patients.

Entrican et all 7 75 OC use in CD and UC patients with adequate data among randomly selected charts. Nocontrol (OR users: non-users in both diseases).

Lichtarowicz et all 8 146 Retrospective study on menopause and contraception in women with CD. No control (% ofusers in CD: 62% is compared with the general population rate of OC use in 24-28 year oldmarried women: 51/%)

Tedesco et all0 5Conri et al9 10Husson et al8 1 Case reportsBourdais et al7 1Heron et al5 2

CD=Crohn's disease, UC=ulcerative colitis, OC=oral contraceptives, IBD=inflammatory bowel disease, RR=relative risk,OR=odds ratio.

669

on July 4, 2020 by guest. Protected by copyright.

http://gut.bmj.com

/G

ut: first published as 10.1136/gut.37.5.668 on 1 Novem

ber 1995. Dow

nloaded from

Godet, May, Sutherland

TABLE II Cohort studies included in the meta-analysis

Study Sample size RR (95% CI) Comments

Vessey, 198619 17 032 women (9653 users) CD: 1.9 (0.7, 4.7)* Smoking status assessed at entry only. Trend(Oxford, UK) attending family planning CD: 1-5 (0-6, 3.8) suggesting a positive association between

clinics. Annual clinic, home, UC: 2-5 (1-2, 5 1)* duration of OC use and incidence of CD.mail, or phone questionnaire UC: 2-1 (1-0, 4.2) Former users do not have increased relative risk

for both diseases.Logan, 198922 46000 women (23 000 OC users) CD: 1-5 (0-8, 3-1) Smoking status assessed at entry only. Former

(Nottingham, attending general practician UC: 1-4 (0.9, 2.4) users do not have increased risk for bothUK) clinics. Bi-annual visits. diseases.

*Unadjusted for smoking. CD=Crohn's disease; UC=ulcerative colitis; OC=oral contraceptives.Pooled RR for CD: 1-51(0-86, 2 63) and UC: 1-63 (1-06, 2 63).

contraceptive users, even when smoking iscontrolled. These increases are modest and inmost reports, the individual CI did not reachstatistical significance. This meta-analysisshows evidence of a modest increase of the RRfor CD and UC in oral contraceptive users.

These results suggest modest associationsbetween oral contraceptive use and thedevelopment of CD and UC, but these associ-ations are so weak that we cannot eliminatenon-causal explanations.The included studies could be classified as

cohort or case-controls. Although each designoffers weaknesses and strengths, the RRestimates of each study were in the same range.Cohort studies can potentially elucidatetemporal relationships between an exposurefactor (oral contraceptive) and a disease(inflammatory bowel disease) and allow directmeasurement of the incidence of disease inexposed and unexposed subjects. They do notseem to be well suited to evaluate this specificRR, however, because inflammatory boweldisease is rare. While, in practice, prospectivecohort studies eliminate selection bias, theaccuracy of risk measurement can be biased byeither loss at follow up (restricted to 0.3% inVessey study and unknown in Logan study) or

by random misclassification of an exposurefactor. For example, in both cohort studiessmoking information was collected only at theentry. As we do not have infornation on thedistribution of changes in this habit during thefollow up period in the two groups (users andnon-users), 'crude' RR and 'adjusted' RR can

either overestimate or underestimate the truerelative risk. Case-control studies are more effi-cient than cohort studies for rare diseases and

common exposure factors such as inflammatorybowel disease and oral contraceptive use. Onlytwo case-control studies21 25 had a sample sizewith sufficient power (80%) to detect a signifi-cant increase in RR. Although misclassificationbias may be of less concern, case-control studiesare more prone to selection and recall bias.29

Bias may also have been present in thereports. In all studies using prevalent cases,

recall bias may skew the RR in favour of theexposure factor. Most authors were aware ofthis bias and selected only recently diagnosedinflammatory bowel disease or double checkedthe information collected in the two groups.The selection of inpatients as either cases or

controls could also have introduced a bias.32Inflammatory bowel disease inpatients (cases)have more severe disease then their counter-parts who are outpatients.23 Furthermore,they may be more prone to take oral contra-ceptives than the general inflammatory boweldisease population to prevent a pregnancy thatmight affect their clinical status.1 Oral contra-ceptive use in the control group formed byinpatients20 23 compared with a sample drawnfrom the general population may be different.These trends can underestimate or over-

estimate the true RR. The use of a secondcontrol group drawn from general populationmay give a relative degree of assurance that therisk estimate using an inpatient control groupis not biased.32 In the Calkins study,20 the RRfor the control group formed by inpatients waslower than the RR for the control group drawnfrom general population.The studies of Lashner et al24 26 were of

matched case-control design, using friends ascontrols. Since the selection of controls should

TABLE III Case-control studies included in the meta-analysis

Sample size

Study Cases Controls RR (95% CI) Comments

Lesko, 198523 57 CD 2189 Patients with infection, CD: 1-7 (1-0, 3.2) Trend for a positive association between duration of OC(North American cities) (inpatients) trauma use and RR for CD. Trend for a negative association

between time elapsed since last OC use and RR for CD.Person, 199325 365 IBD 390 Individuals from a ran- CD: 1-7 (0.9, 3.2)* Trend for an increased incidence of colonic involvement

(Stockholm, (inpatients) dom general population UC: 1-7 (0-8, 3.3)* compared to ileal and ileocaecal CD in OC users. NoSweden) sample stratified for age preferential site for UC.

Calkins, 198620 67 CD, 32 UC Matched neighbours (71 for CD: 1-6 (0-6, 4.6)* Size ofUC group is too small to be interpreted properly.(Baltimore, USA) (inpatients) CD, 38 for UC) UC: 0.5 (0-1, 3.0)*

Lashner, 198924 51 CD 51 Neighbours and friends CD: 0.7 (0.3, 1-6)(Chicago, USA) (outpatients) (paired data)

Lashner, 199026 46 UC 46 Neighbours and friends UC: 0.7 (0.3, 1-8) No dose-response effect. Smoking had no confounding or(Chicago, USA) (outpatients) (paired data) interaction effect. Low power.

Katschinski, 199327 90 CD 90 Individuals matched from CD: 2-5 (1-6, 6.6) Positive association between OC use and CD only for(Essen, Germany) (outpatients) age from a random sample non-smokers.

(city population)Sandler, 199221 184 CD, 100 UC 276 Matched neighbours CD: 1-3 (0-8, 2.0) Positive association between OC use and CD or UC only

(North Carolina, USA) UC: 1.1 (0-6,. 1-8) for smokers. Trend for an increased risk of colonic CDin OC users. No dose response-effect.

*Unadjusted for smoking. CD=Crohn's disease; UC=ulcerative colitis; OC=oral contraceptives. Pooled RR for CD: 1-48 (1-15, 1-90) and UC: 1.11 (0-76, 1-63);including Lashner studies24 26. Pooled RR for CD: 1 59 (1-25, 2.03) and UC: 1-25 (1.20-1-63); excluding Lashner studies.

670

on July 4, 2020 by guest. Protected by copyright.

http://gut.bmj.com

/G

ut: first published as 10.1136/gut.37.5.668 on 1 Novem

ber 1995. Dow

nloaded from

Meta-analysis of the role of oral contraceptive agents in inflammatory bowel disease

! *

I:II

I1 I*

m*

1]a

1 m

X1 -

ml1

0 1 2 3

Figure 1: (A) Individual and pooled relative risks (RR) (9Crohn's disease adjustedfor smoking. (B) Individual andjdisease unadjustedfor smoking. *Including Lashner study24

be independent of thestratum of matchirconcerned that thecontrols might bedefence the authors xthe use of friend nor:firming the well kncsmoking and the de)study dataset. We elecresults with andIncluding or excludithe recalculated RR i

or UC changed only i

A major drawbacis publication bias.reviewers to favour stihas been documentedcorrection for this bi;significant associatiomceptive use and CD cA thorough search foiaddressing the specimade in this meta-anthis seems to be the r

this problem, althou~have eliminated it.

Vessey (19)

Logan (22)

Sandier (21)Lashner (24)

Pooled RR

When assessing the association between oralcontraceptive use and inflammatory boweldisease, smoking must be considered as an effectmodifier and a confounding factor. Smoking isprotective in UC and is a positive risk factor forCD. Separation ofrisk estimates by CD and UCstrata is thus mandatory. Smokers are morelikely to take oral contraceptives than non-smokers'2 and oral contraceptive users are morelikely to smoke.22 Therefore, smoking may con-found the effect of oral contraceptive on thedevelopment ofinflammatory bowel disease andrisk estimates must be adjusted for smoking

I when data are analysed. The effect ofthe adjust-4 5 6 7 ment for smoking was different among each

individual study. Adjusting for smoking andother factors either increased CD,2' 2527lowered,23 or did not change (UC)2' theestimates of RR. As most of these studies gaveestimates of RR, adjusted for factors other thansmoking, the real effect ofthis adjustment on thedirection of the RR is unclear.We were unable to show a dose-response

4 5 6 7 effect (change in RR for developing CD and UC

5% confidence intervals) for associated with change in exposure duration to

5ooled relative risks for Crohn's oral contraceptive) or a difference between RRtfexcluding Lashner study. estimates for ileal, ileo-colonic, or colonic CD.

The data on these variables were either absent,study factor within each inadequate (important proportions of unknown

ng factors,33 we were patients), without evidence of any statistical sig-use of friend nominated nificance, or conflicting between studies.inappropriate. In theirnoted that they validatedninated controls by con- Appendix)wn association betweenvelopment of CD in the STUDIES INCLUDEDcted to present the pooledwithout these studies. 1 Cohort studiesng the Chicago dataset, Purpose: The two cohort studies examined thefor developing either CD incidence of inflammatory bowel disease inminimally. users and non-users of oral contraceptives, andSk in any meta-analysis in smokers and non-smokers.

' 35 The tendency ofudies with positive results 1.1 Vessey 1986.19 Prospective study from[previously.36A potential Oxford, UK; includes 17032 women (9653as would tend to put the users), aged 25-39 years.n (between oral contra- Lost to follow up: 03%/O per year (other than)r UC) closer to the null. death, emigration).r registration of all studies Excluded: inflammatory bowel disease diag-fic association has been nosed before beginning of oral contraceptivetalysis. To our knowledge use.most pgh we

m

I_ m

_ 1

1 2 3 4 5Figure 2: Individual and pooled relative risks (RR) (95% confidence intervals) forukerative colitis adjustedfor smoking. *Including Lashner study26; texcluding Lashner

study.

ractical solution to RR for CD unadjusted for smoking: 1.88cannot be sure to (0.74, 4 75).

RR for CD adjusted for smoking: 1.49(0.58, 3.85).RR for UC unadjusted for smoking: 2-50

(1.24, 5 06).RR for UC adjusted for smoking: 2.05

(1.00., 4.20).Comments: adjusted RRs fell short of statisti-

cal significance. Smoking status was assessed atentry only. The data are insufficient to assess a

specific disease site association in both diseases.There is a trend suggesting a positive associationbetween duration of oral contraceptive use andincidence of CD only. Former users (>2 years)do not have increased RR for either disease.

1.2 Logan 1989.22 Prospective study fromNottingham, UK. Includes 23000 oral

A

Vessey (19)

Logan (22)Lesko (23)

Katschinski (27)Sandier (21)Lashner (24)

Pooled RR

B

Persson (25)Calkins (20)

Pooled RR

671

1 lb -5 A /

on July 4, 2020 by guest. Protected by copyright.

http://gut.bmj.com

/G

ut: first published as 10.1136/gut.37.5.668 on 1 Novem

ber 1995. Dow

nloaded from

Godet, May, Sutherland

contraceptive users and 23 000 non-users,aged 15-45 years.No data on those lost to follow up.RR for CD adjusted for smoking: 1.51

(0.75, 3.05), ever smokers v never smokers.RR for UC adjusted for smoking: 1-44

(0-85, 2 45), ever smokers v never smokers.Comments: all RRs fell short of statistical

significance, although the trend suggests amodest association for both diseases. Formerusers did not have an increased risk for bothdiseases oral contraceptive use was defined ascontinuous use for more than six monthsbefore beginning of symptoms. Smoking wasassessed only at recruitment.

2 Case-control studies2.1 Lesko 1985.23 Prospective study fromseveral North American cities examiningprimarily the association between oral contra-ceptive use and inflammatory bowel disease.

Cases: 57 inpatients with CD (84% severe),82% diagnosed with 3 months of interview.

Controls: 2189 inpatients with infection,trauma.

Excluded: gastrointestinal infection, pelvicinflammatory disease, UC.RR estimate given by the author: 1.7 (10,

3.2); adjusted for age, race, religion, education,cigarette use, year of interview, geographicregion, number of previous admissions.Comments: data are not adequate to study

smoking interaction. The study suggests thatformer oral contraceptive users are not atincreased risk. There is a positive trend but nota significant association between duration oforal contraceptive use and RR (p=027).There is a negative trend but not a significantassociation between time elapsed since last oralcontraceptive use and RR. Patients take oralcontraceptive for different purposes (contra-ception, menstrual problems, sexual difficul-ties, endometriosis, and infertility). Forty twoper cent of cases and 51% of controls were 40years old or over and four per cent refused.

2.2 Persson 1993.25 Prospective study fromStockholm, Sweden, examining the associationbetween inflammatory bowel disease on onehand and oral contraceptive use, physicalactivities, psoriasis, childhood gastrointestinalinfections, family history, and having a bird asa pet on the other.

Cases: 365 new cases ofinflammatory boweldisease/inpatients, aged 15-79 years.

Control: 390 randomly selected from popu-lation register stratified by age.

Excluded: outpatients and inpatients whosemedical charts could not be reviewed within 4years of diagnosis.

Multiple logistic regression RR estimatesgiven by the author: RR for CD= 1.7 (09, 3.2)(adjusted for age only; RR for UC= 1-7 (0-8,3.3) (adjusted for age only).Comments: a user was defined as oral con-

traceptive use during the last 6 months. Noinformation was given on oral contraceptiveuse duration or the time elapsed since last use.The study suggests an increased incidence for

colonic involvement compared with ileal andileocaecal involvement for CD patients butthere is overlap between the three specific con-fidence intervals. For UC there is no preferen-tial site. Eighty per cent of patients respondedto the questionnaire.

2.3 Calkins 1986.20 Prospective study fromBaltimore examining the association of oralcontraceptive use and the incidence of inflam-matory bowel disease.

Cases: 66 inpatients from 1977 to 1979,aged 10-60 years

Control: (a) 67 inpatients matched for age,sex, race, discharge data for CD; 32 for UC.(b) 71 neighbourhood controls matched forage, sex, race, neighbourhood of residence forCD; 38 for UC.OR for CD: 1.63 (0.58, 4.57) unadjusted

(neighbour control).OR for UC: 0.52 (0.09, 3.01) unadjusted

(neighbour control).Comments: this study has not been pub-

lished as a full size report. ORs are not adjustedfor smoking. There are different purposes fororal contraceptive use (contraception, men-strual difficulties, hormone replacement). Wedo not know the age distribution of the popu-lation studied. There is a weak but not signifi-cant trend toward an increased RR in Crohn'sdisease for oral contraceptive users. The size ofthe ulcerative colitis group is too small to beinterpreted properly.

2.4 Lashner 1989 and 1990.24 26 Two differentprospective studies with the same design fromChicago, examining oral contraceptive use inCD (1989) and UC (1990) incident cases vcontrols (paired data).

Cases: outpatients from 1981, aged 20-50years.

Controls: friends of cases in neighbourhoodmatched for age, race, sex, religion, residence,marital status, and parity.

Fifty one pairs (cases and controls) for CDand 46 pairs for UC.

Excluded: hysterectomy, contraindication tobe exposed to oral contraceptive.OR given by the author: 0.73 (0.34, 1-59)

current users.OR given by the author: 0.70 (0.27, 1-83)

current users.

Comments: the OR estimates given by theauthor for the possible confounding effect ofsmoking computed with three different condi-tional regression models for the duration of useare not different from 1. Authors conclude thereis no evidence that oral contraceptive are a riskfactor for either diseases because: (1) overall ORis not different from unity, (2) there is no asso-ciation demonstrated by stratifying the data,and (3) there is no dose response effect. Thesestudies are externally validated by the finding ofa positive and a negative associations forsmoking on one hand and CD and UC, respec-tively, on the other. Studies did not show aninteraction or a confounding effect between oralcontraceptive use and smoking. The minimalOR for oral contraceptive use if statistical

672

on July 4, 2020 by guest. Protected by copyright.

http://gut.bmj.com

/G

ut: first published as 10.1136/gut.37.5.668 on 1 Novem

ber 1995. Dow

nloaded from

Meta-analysis of the role of oral contraceptive agents in inflammatory bowel disease 673

significance is achieved (ot error of 5%, C errorof 20%) is around 2-8 for both studies.2.5 Katschinski 1993.27 Retrospective andprospective study from Essen, Germany; exam-ining an association between CD and oral con-traceptive use and smoking.

Cases: 90 outpatients attending clinic from1986-88, among whom 61% were aged <25years at disease onset.

Controls: 90 matched for age and randomlyselected among city population.

Excluded: women with hysterectomy orthose unable to take oral contraceptive.RR given by the author: 2 5 (1I6, 6 6)

adjusted for smoking.Comments: this study gives evidence of an

association between oral contraceptive useand CD only for non-smokers. There is noevidence of a duration effect or a preferential site(non-significant trend toward ileal CD in oralcontraceptive users). The study shows clearly aninteraction effect ofsmoking and oral contracep-tive use in inflammatory bowel disease. Eighty to88% responded to the questionnaire.

2.6 Sandler 1992.21 Retrospective study fromNorth Carolina examining the associationbetween oral contraceptive use and inflamma-tory bowel disease as part of a larger study ofpotential environmental factors.

Cases: 184 CD, 100 UC from the Crohn'sand Colitis Foundation of America; mean ageCD 28 years, UC 33 years.

Controls: 276 neighbours matched for sex,age (+/-5 years), race: included some friends.

Excluded: onset of inflammatory boweldisease symptoms before menarche (n= 18),missing index date (n= 16), missing informa-tion on oral contraceptive use (n= 18).

Estimate of RR given by the author: 1 30(085, 2 01) for CD, adjusted for smoking.

Estimate of RR given by the author: 1 08(0-63, 1 85) for UC, adjusted for smoking.Comments: there is interaction between oral

contraceptive use and smoking for inflamma-tory bowel disease. There is an associationbetween oral contraceptive use and inflamma-tory bowel disease only in smokers (inverse rela-tionship compared with Katschinski report).There is a trend toward an increased risk ofcolonic CD in oral contraceptive users (ORs notadjusted for smoking and 14% of unknown siteof disease in this specific group). For UCpatients, the OR is not affected by adjusting forsmoking or by stratifying for disease site. Thereis no evidence of a dose-response effect. Seventyfour point six per cent of cases and 95 3% ofcontrols responded to the questionnaire.Dr Patrick Godet is a member of the Gastroenterology Unit ofNotre-Dame Hospital, affiliated centre to the Universite deMontreal, Montreal, Quebec, Canada. This work has beenfunded by the Crohn's and Colitis Foundation of Canada.

1 Sandler RS, Golden AL. Epidemiology of Crohn's disease.JT Clin Gastroenterol 1986; 8: 160-5.

2 MendeloffAI, Calkins BM. The epidemiology of idiopathicinflammatory bowel disease. In: Kirshner JB, Shorter RG,eds. Inflammatory bowel disease. Philadelphia, Lea andFabinger, 1988.

3 Albert MB, Ginsberg AL. Oral contraceptives and Crohn'sdisease. Gaseroeneerologfy 1986; 99: 1097-8.

4 Delavierre P, Bourdais JP, Hureau J, Vayre P. Lea accidentsintestinaux au cours de l'emploi des contraceptifs oraux.Semaine des Hopitaux de Paris 1976; 52: 1225-8.

5 Heron HC, Khubchandani IT, Trimpi HD, Sheets JA,Stasik JJ. Evanescent colitis. Dis Colon Rectum 1981; 24:555-61.

6 Kilpatrick ZM, Silverman JF, Betancourt E, Farman J,Lawson JP. Vascular occlusion of the colon and oral con-traceptives. N EnglJ Med 1968; 278: 438-40.

7 Bourdais JP, Delavierre P, Vayre P, Hureau J, Letailleur M.Maladie de Crohn ileo-colique apres emploi d'un contra-ceptif oral. Archives Franfaises des Maladies de l'AppareilDigestif 1976; 65: 155-60.

8 Husson J. Ileite (Maladie de Crohn probable) et oestro-progestatifs. Semaine des H6pitaux de Paris 1981; 57:1750-1.

9 Conri C, Schmitt B, Bache-Gabrielsen P, de Mascarel A,Moreau F. Maladie de Crohn colique et contraceptionorale. Semaine des Hopitaux de Paris 1979; 55: 1733-5.

10 Tedesco FJ, Valpolicelli NA, Moore FS. Oestrogen andprogesterone associated colitis: a disorder with clinicaland endoscopic features mimicking Crohn's disease.Gastrointest Endosc 1982; 26: 247-9.

11 Wakefield AJ, Sawyerr AM, Hudson M, Dhillon AP,Pounder RE. Smoking, the oral contraceptive pill, andCrohn's disease. Dig Dis Sci 199 1; 36: 1147-50.

12 Wakefield AJ, Dhillon AP, Rowles PM, Sawyerr AM, PittiloRM, Lewis AAM, Pounder RE. Pathogenesis of Crohn'sdisease: Multifocal gastrointestinal infarction. Lancet1989; ii: 1057-62.

13 Sutherland LR, Ramcharan S, Bryant H, Fick G. Effect ofcontraceptive use on reoperation following surgery forCrohn's disease. Dig Dis Sci 1992; 37: 1377-82.

14 Wright JP. Factors influencing first relapse in patients withCrohn's disease. Jf Clin Gastroenterol 1992; 15: 12-6.

15 Ramcharan S. The Walnut Creek contraceptive drug study.A prospective study of the side effects of the oral contra-ceptives. Washington, DC: DHEW Publications, 1981:74-562.

16 Rhodes JM, Cockel R, Allan RN, Hawker PC, Dawson J,Elias E. Colonic Crohn's disease and use of oral contra-ception. BMJ 1984; 288: 595-6.

17 Entrican JH, Sircus W. Chronic inflammatory boweldisease, cigarette smoking, and use of oral contraceptives.BMJ 1986; 292: 1464.

18 Lichtarowicz A, Norman C, Calcraft B, Morris JS, RhodesJ, Mayberry J. A study of menopause, smoking, and con-traception in women with coeliac disease. Q J Med 1989;72: 267, 623-31.

19 Vessey M, Jewell D, Smith A, Yeates D, McPherson K.Chronic inflammatory bowel disease, cigarette smoking,and use of oral contraceptives: findings in a large cohortstudy of women of childbearing age. BMJ 1986; 292:1101-3.

20 Calkins BM, Mendeloff AI, Garland C. Inflammatorybowel disease in oral contraceptive users. GastroenterologCy1986; 91: 523-4.

21 Sandler RS, Wurzelmann JI, Lyles CM. Oral contraceptiveuse and the risk of inflammatory bowel disease.Epidemiology 1992; 3: 374-8.

22 Logan RFA, Kay CR. Oral contraception, smoking andinflammatory bowel disease - findings in the RoyalCollege of General Practitioners oral contraception study1989; 18: 105-7.

23 Lesko Sm, Kaufman DW, Rosenberg L, Helmrich SP,Miller DR, Stolley PD, Shapiro S. Evidence for anincreased risk of Crohn's disease in oral contraceptiveusers. Gastroenterology 1985; 89: 1046-9.

24 Lashner BA, Kane SV, Hanauer SB. Lack of associationbetween oral contraceptive use and Crohn's disease: acommunity-based matched case-control study.Gastroenterology 1989; 97: 1442-7.

25 Persson PG, Leijonmarck CE, Bemell 0, Hellers G,Ahlbom A. Risk indicators for inflammatory boweldisease. IntJ Epidemiol 1993; 22: 268-72.

26 Lashner BA, Kane SV, Hanauer SB. Lack of associationbetween oral contraceptive use and ulcerative colitis.Gastroenterology 1990; 99: 1032-6.

27 Karschinski B, Fingerle D, Scherbaum B, Goebell H. Oralcontraceptive use and cigarette smoking in Crohn'sdisease. Dig Dis Sci 1993; 38: 1596-600.

28 Hennekens CH, Buring JE. Case-Control studies. In:Mayrent SL, ed. Epidemiology in medicine. Boston: LittleBrown 1987: 132-77.

29 Greenland S. Quantitative methods in the review ofepidemiologic literature. Epidemiologic Rev 1987; 9: 1-30.

30 Halvorsen K. Combining results from independent investi-gations: meta-analysis in medical research. In: Baylar JC,Mosteller F, eds. Medical uses of statistics. Boston: N EnglJ Med 1992: 393-416.

31 DerSimonian R, Laird N. Meta-analysis in clinical trials.Controlled Clin Trials 1986; 7: 177-88.

32 West DW, Schuman KL, Lyon JL, Robison LM, Allred R.Differences in risk estimations from a hospital and a pop-ulation-based case control study. IntJ Epidemiol 1984; 13:235.

33 Flanders WD, Austin H. Possibility of selection bias inmatched case-control studies using friend controls. Am JEpidemiol 1986; 124: 150-3.

34 Naylor DC. Two cheers for meta-analysis: problems andopportunities in aggregating results of clinical trials. CanMed AssocJf 1988; 138: 891-5.

35 O'Brien PC. Meta-analysis: its role in medical research andin assessment of the association between low levels ofcholesterol and excess mortality. Mayo Clin Proc 1993; 68:891-93.

36 Dickersin K. The existence of publication bias and riskfactors for its occurrence.3JAMA 1990; 263: 1385-9.

on July 4, 2020 by guest. Protected by copyright.

http://gut.bmj.com

/G

ut: first published as 10.1136/gut.37.5.668 on 1 Novem

ber 1995. Dow

nloaded from