Smoking, gender and rheumatoid arthritis–epidemiological clues to etiology

Joint Bone Spine 70 (2003) 496–502 www.elsevier.com/locate/bonsoi

Original article

Smoking, gender and rheumatoid arthritis–epidemiological clues to etiology Results from the behavioral risk factor surveillance system > Eswar Krishnan * Clinical Research Center of Reading, 401 Buttonwood St., West Reading, PA 19611, USA Division of Immunology, Department of Medicine, Stanford University, 1000 Welch Road, Suite 203, Palo Alto, CA 94304, USA Received 4 February 2003; accepted 4 June 2003

Abstract Objective.– This study was undertaken to confirm and extend our earlier observation that gender is a biological effect modifier of smoking–rheumatoid arthritis (RA) relationship in a diverse national survey sample in the United States. Methods.– Smoking history of 644 cases of RA and 1509 geographically matched general population controls were compared using weighted logistic regression. Results.– There were 644 respondents with RA (cases) and 1509 geographically matched controls. Cases were significantly younger, less educated, more likely to be single and female than controls. Among cases 57% were smokers while among controls 49% smoked. Among women, after adjusting for age, hysterectomy had an age adjusted odds ratio 1.45, (95% CI 0.99–2.10) and menopause an adjusted odds ratio 1.18 (95% CI 0.99–2.10) were associated with smoking. In univariable analysis ever-smoking was associated with increased risk of RA (odds ratio 1.34, 95% CI 1.0–1.81). Among the strata of smokers, there was an increasing gradient of risk with increasing exposure to smoking (P = 0.041). In separate multivariable models, smoking increased the risk in men (odds ratio 2.29, 95% CI 1.35–3.90) while in women the risk was not elevated (odds ratio 0.98, 95% CI 0.67–1.42). After adjusting for the statistically significant interaction both female gender (odds ratio 2.30, 95% CI 1.39–3.83) and having ever smoked (odds ratio 2.31, 95% CI 1.36–3.94) emerged as significant risk factors for RA. Conclusions.– Gender interacts with smoking in by an unknown mechanism to lead to differential risk of RA. © 2003 Éditions scientifiques et médicales Elsevier SAS. All rights reserved. Keywords: Rheumatoid arthritis; Smoking; Interaction; Risk; Effect modification; Gender

1. Introduction Smoking is known to be associated with production of rheumatoid factor [1–3]; rheumatoid factor production, in turn, often precedes development of clinical rheumatoid arthritis (RA) [3,4]. Although epidemiological studies have suggested that smoking leads to increased risk of RA in men [5,6], few studies have focused on the interaction between smoking and gender in the pathogenesis of RA. Recently our group formally presented a case for such an interaction and proposed that gender specific factors, especially menstrua>

This data was presented as a poster in the 65th Annual Scientific Meeting of the American College of Rheumatology at San Francisco, October 2001. * Corresponding author. E-mail address: [email protected] (E. Krishnan). © 2003 Éditions scientifiques et médicales Elsevier SAS. All rights reserved. doi:10.1016/S1297-319X(03)00141-6

tion may modify the immunological milieu resulting in some form of “protection” for women. If this hypothesis is true, such an effect should be observable in diverse populations. We used data from the behavioral risk factor surveillance system (BRFSS), an ongoing state-based survey in the United States to test the hypothesis that gender acts as a biologic effect-modifier in the smoking–RA association. 2. Methods 2.1. Data collection The BRFSS is a collaborative project of the centers for disease control and prevention (CDC), and US states and territories. The BRFSS, administered and supported by the Behavioral Surveillance Branch of the CDC, is an on-going data collection program designed to measure behavioral risk

E. Krishnan / Joint Bone Spine 70 (2003) 496–502

factors in the adult population 18 years of age or over living in households. Random digit dialing is used to select households and within each household, one adult is randomly selected for a computer assisted telephone interview. Details of the survey methodology have been described in the web site http://www.cdc.gov/brfss/ (accessed 3 August 2002). In 1999, an arthritis questionnaire module was administered in seven states (Florida, Georgia, Louisiana, Mississippi, Missouri, Nebraska, Oklahoma, and West Virginia). A series of questions were asked to identify respondents with RA. Respondents were first asked if they had pain, stiffness and swelling in or around joints any time in the previous 12 months and if so whether the symptoms were present on most days every month. Further questions were not asked if the answer to these questions were negative. More specific questions about the type of arthritis were posed to those who replied yes to the above questions. Cases were subjects who responded affirmatively to the question “Have you been told by your physician that you have arthritis?” and subsequently chose RA as their diagnosis from a verbal menu of osteoarthritis/degenerative arthritis, rheumatism, RA, Lyme disease, and others in that order. All the respondents who reported RA were included in this analysis. The respondents who denied any joint symptoms were used as a source for controls. We used a computerized geographically stratified random sampling algorithm [7] to derive approximately three controls per subject. Thus, each case had three geographicarea matched general population controls. Three types of smoking questions were asked (1) have you smoked at least 100 cigarettes in your entire life? (ever-smoking) (2) On the average, about how many cigarettes a day do you now smoke? (3) Do you now smoke cigarettes every day, some days, or not at all? Clinical variables like disease duration, age at onset etc. were unfortunately not available for this study. 2.2. Data validation Several validation studies have been performed on the BRFSS data. Individual level reliability estimates for demographics, chronic conditions, and risk factors were high (Kappa values 0.82–1.00) [8]. Comparison of pooled BRFSS estimates of smoking with national estimates from National Heart Lung Blood Institute survey, National Survey of Personal Health Practices and Consequences and National Health Interview Survey showed that BRFSS estimates are very close to national estimates for all of these risk factors (within 1–2% points) [9–11]. A bibliography of validation studies for the socio-demographic variables is available from http://www. cdc.gov/brfss/mvr.htm (accessed on 3 August 2002).

497

adults in the household and individual decision to participate or not. Although 95% of the US households have telephones, coverage ranges from 87 to 95% across states surveyed for this study and varies for subgroups as well. For example, persons living in the South, minorities, and those in lower socio-economic groups typically have lower telephone coverage. Refusals to respond to arthritis question were rare (<2%). To account for the non-response, unequal probability of selection, clustering of observation, and stratification, post-stratification weights were prepared by the CDC for analysis. We used the SVY suite of command functions in [13] for our analysis. A weighted logistic regression model was used to study the effects of socio-demographic covariates on the dichotomous dependent variable, RA. In addition to calculating odds ratios stratified for potential predictors, we tested for trends across continuous variables using covariates as continuous variable where available. Student’s t-test and Pearson’s v2 test were used to compare means and proportions across categories. In this report, the term primary interaction has been used synonymously with multiplicative effect modification [14] and it signifies modification of the effect of a predictor variable on the outcome variable by a third variable. For example, if the association between a predictor A and outcome B is statistically different in different strata of C, a third factor, then the factor C is termed an effect modifier of the relationship between A and B. We examined our data for significant interaction by two complementary methods. First, the smoking–RA risk estimate was stratified by the potential effect-modifiers namely age, gender, and socio-economic variables. We then performed formal statistical tests for primary interaction by generating interaction terms (gender/ever-smoking) [15] and using them as covariates in addition to the main effects (gender and ever-smoking) as independent variable. 3. Results 3.1. Description of cases and controls 3.1.1. Socio-demographic variables We identified a total of 644 persons with RA (cases) and 1509 persons with no joint symptoms (controls). Fig. 1 illustrates the age distribution of cases and controls. Table 1 shows the comparison between the socio-demographic and smoking characteristics of cases and controls. Cases were more likely than controls to be younger, female, less well educated, of lower annual income, obese, and living alone.

2.3. Statistical analysis Data from the BRFSS are collected using a complex survey design named disproportionate stratified sample [12]. The probability of being selected for interview is thus not equal for all the individuals in the population but is determined by a combination of telephone availability, number of

3.1.2. Smoking Overall, cases were more likely than controls to report having ever smoked (57 vs. 49%, P = 0.05). The proportion of smokers was higher in cases than controls in men (77 vs. 65%, P = 0.002) but not in women (42 vs. 38%, P = 0.15). On an average cases smoked more cigarettes per person compared to

498

E. Krishnan / Joint Bone Spine 70 (2003) 496–502

Fig. 1. Violin plot of the age distribution between cases and controls. The violin plot shows the median, inter-quartile range (light horizontal band and dark vertical band, respectively), and range (thin vertical line). The curve on the sides provides the approximate frequency distribution at each age.

controls. When non-smokers were excluded, there was no difference in the number of cigarettes smoked per day between cases and controls (18.6 and 19.4, P = 0.25). Although among smokers, men overall smoked more number of cigarettes per day than women (24 and 20, respectively, P < 0.001), there was no differences in the number of cigarettes smoked between cases and controls in men and women. 3.1.3. Reproductive factors Overall 1028 women respondents had attained menopause at the time of the interview. Cases were less likely to have attained menopause than controls (66% and 78%, respectively, P < 0.001). Cases (49%) were more likely than controls (42%) to have had a hysterectomy (P = 0.005). Cases who had reported having a hysterectomy were more likely to be relatively younger than controls (mean age 66 and 62 years, respectively, P < 0.001). Data on current estrogen use were available only on 284 of the 1387 women in the study. The median age of the women using estrogen was 63 years (inter-quartile range 55–68 years).

3.2. Univariable logistic regression We performed a univariable logistic regression with cases and controls as the dichotomous outcome including all the subjects (Table 2). Most notably, higher education and higher income were associated with substantial reduction of risk of RA while gender was not a risk factor. Ever-smoking cigarettes was associated with an odds ratio of 1.34 (95% CI 1.0–1.81). There was a significant trend of risk reduction for RA for ex-smokers and never smokers compared to current smokers (P = 0.04). However, the number of cigarettes per day did not influence the risk of RA. Female gender was not associated with statistically significant increased risk of RA in the univariable analysis where men and women were pooled. In an age adjusted logistic regression model where only women were included, we examined the risk for reproductive factors on RA (Table 3). Among women, having attained menopause and having had hysterectomy were associated with increased odds for RA. Current estrogen use was not associated with risk of RA.

Table 1 Socio-demographic factors of cases and controls Characteristic Mean age (years) (S.D.) Proportion of men (%) Proportion of Caucasians (%) Proportion with education >12 years of school (%) Proportion with income >$35 000 (%) Proportion ever smoked? (%) Proportion currently smoking (%) Mean number of cigarettes per day (S.D.) Proportion living alone (%)

Controls 1509 63 (12) 37 88 42 49 49 20 3.5 (8.2) 44

Cases 644 60 (14) 32 85 33 37 57 24 4.6 (10) 49

P value <0.001 0.01 0.08 <0.001 <0.001 0.05 0.13 0.01 0.03

E. Krishnan / Joint Bone Spine 70 (2003) 496–502

499

Table 2 Univariate logistic regression models for risk of developing RA Characteristic Age 25–35 35–44 45–54 55–64 65–74 >75 Gender Female Male Race Non-Caucasian Caucasian Education High school or less At least some college Annual income <$35 000 >$35 000 Social status Lives with someone else Lives alone Smoker? No Yes Levels of smoking * Current daily Former smoker Never smoked

Controls ( n )

Cases ( n )

Odds ratio

95% CI

17 105 245 398 452 292

22 87 119 156 163 97

0.53 0.31 0.38 0.27 0.23 0.35

0.22–1.25 0.13–0.71 0.16–0.88 0.12–0.62 0.09–0.56 0.12–1.05

946 543

441 203

1.00 0.86

0.63–1.17

181 1325

95 549

1.00 0.79

0.49–1.22

870 639

429 215

1.00 0.53

0.39–0.71

766 743

408 236

1.00 0.56

0.41–0.77

851 658

331 313

1.00 1.26

0.94–1.7

784 725

303 341

1.00 1.34

1.0–1.81

291 434 784

160 181 303

1.00 0.89 0.70

0.60–1.34 0.48–1.00

* P value for trend 0.04.

3.3. Multivariable analysis To test the hypothesis that the relationship between smoking and risk of RA is influenced by gender, we fitted separate regression models for men and women. These models included all of the covariates in the overall multivariable model. Ever-smoking was significant risk factor for men but not for women (Table 4). There was a trend of decreasing risk for ex-smokers (P < 0.05) and non-smokers compared to current smokers among men but not among women. We subsequently tested for primary interaction between the eversmoking variable and age, gender, education and income and found that there was significant interaction between smoking

and gender in the risk for RA (P = 0.02). Further parallel analyses were performed by stratifying age, education, income, menopausal status, current estrogen use and hysterectomy status, and examining the smoking–gender relationship in each of the strata. We did not find any additional significant effect modifiers of the smoking–RA relationship. To demonstrate the extent to which smoking–gender interaction can lead to biased results when data from men and women are pooled, we fitted two multivariable models. In the first model (Table 5), after adjusting for age, gender education and income, ever-smoking conferred an odds ratio of 1.38 (1.02–1.90). Intensity of smoking (average number of cigarettes smoked as well as frequency of smoking) did not

Table 3 Age adjusted odds ratios for reproductive factors (women only) Characteristic Menopause No Yes Had hysterectomy No Yes Current estrogen use No Yes

Controls

Case

OR

95% CI

207 739

152 289

1.18

0.58–2.40

548 391

222 219

1.45

0.99–2.10

45 141

22 76

0.98

0.39–2.48

500

E. Krishnan / Joint Bone Spine 70 (2003) 496–502

Table 4 Multivariable logistic regression models for risk factors of RA for men and women Odds ratio Multivariable model for men (N = 763) Age (years) 0.96 Caucasian race 1.82 Education (years) 0.67 Income 0.56 Living alone 0.80 Ever-smoking 2.29 Multivariable model for women (N = 1384) Age (years) 0.97 Caucasian race 0.79 Education (years) 0.71 Income 0.76 Living alone 1.35 Ever-smoking 0.98

affect the risk of RA. We then fitted a second multivariable model after adjusting for the effect of interaction between smoking and gender. In this (final) model (Table 5), both female gender and smoking were significant risk factors for RA, and the magnitude of the risk similar to previous prospective and retrospective studies.

4. Discussion The results of our study substantiate our earlier report and hypothesis that gender is an effect modifier of smoking–RA relationship [16]. Our observation of the relationship between smoking and the risk of RA is in line with previous population based case-control [6,17–19], cohort [5,20–23] and twin studies [24]. Our risk estimates for men and women are consistent with the previous studies [5,6]. Among studies done on women-only cohorts, the presence of association between smoking and RA depended on whether the study

95% CI

P value

0.94–0.99 0.65–5.11 0.52–0.86 0.29–1.08 0.45–1.43 1.35–3.90

0.001 0.25 0.002 0.08 0.45 0.002

0.95–0.98 0.47–1.32 0.59–0.85 0.50–1.15 0.91–2.00 0.67–1.42

<0.001 0.37 0.001 0.20 0.14 0.90

subjects were pre- or post-menopausal. In the Iowa Women’s Health Study that looked at post-menopausal women, a dosedependent effect of smoking was evident [21] (relative risk of RA for current smokers vs. never smokers 2.0). Data from the prospective Nurses Health Study did not show a relationship between smoking and RA [20] while the Women’s Health Study showed relatively modest effect size (relative risk 1.39 and 1.49 for all RA and seropositive RA, respectively), [20] compared to the Iowa study. In our study, we found that smoking was a preferential risk factor for RA in men. Although we did not find a dose–response in terms of number of cigarettes and risk of RA, the data did show a significant decreasing trend in risk of RA from current smokers to never smokers. One of the explanations for these observations made consistently in different populations in prospective and retrospective studies could be that the way in which smoking triggers the immunological cascade leading to RA is modulated differently due to differences in hormonal milieu in

Table 5 Multivariable logistic regression models before and after adjusting for the effect of gender–smoking interaction Odds ratio Multivariable model unadjusted for smoking–gender interaction (N = 2147) Age (years) 0.97 Female gender 1.32 Caucasian race 0.99 Education (years) 0.69 Income 0.67 Living alone 1.12 Ever-smoking 1.30 Multivariable model adjusted for smoking–gender interaction (N = 2147) Age (years) 0.97 Female gender 2.30 Caucasian race 1.01 Education (years) 0.69 Income 0.68 Living alone 1.13 Ever-smoking 2.31

95% CI

P value

0.96–0.98 0.97–1.80 0.63–1.57 0.59–0.80 0.47–0.96 0.81–1.55 0.96–1.74

<0.001 0.08 0.97 <0.001 0.03 0.48 0.09

0.95–0.98 1.39–3.83 0.064–1.59 0.60–0.80 0.48–0.96 0.82–1.56 1.36–3.94

0.000 0.001 0.971 0.000 0.028 0.444 0.002

E. Krishnan / Joint Bone Spine 70 (2003) 496–502

men and women. More specifically, the hormonal milieu in menstruating women could act as biological effect modifier and block the smoking–rheumatoid factor–RA pathway. As a corollary, within women, end of menstruation may be thought to some how remove the “protection” from the effects of smoking. This would explain the observation in many other studies that smoking is related to RA in men and pre-menopausal women preferentially. Thus, the epidemiological interaction we have observed is consistent with an underlying biological effect modification. In women with RA, association between reproductive variables like age at menarche, progestin use [25], oral contraceptive use [22,26–29], termination of pregnancies [30], lactation [31,32] and a short fertile period [33] have been reported. A role for the hormone prolactin has been proposed [28]. Most of these reproductive variables were not available for our analysis. There were only 33 cases and 64 controls who were menstruating at the time of the interview. Being menopausal or having had hysterectomy was associated with odds ratios greater than 1 but these did not reach statistical significance. Interestingly, however, a test for direct primary interaction between smoking and menopausal status was almost significant with a P value of 0.06. The population based selection of cases and controls, the large number of cases, utilization of weighting to offset the differential selection probabilities of cases and controls are the major strengths of this study. The main limitation of this study as in the Women’s Health Study reported by Karlson et al. [23] is that the diagnosis of RA is by self-report. In other studies the positive predictive values of self-reported RA diagnosis have generally been modest [34,35]. A case validation study of RA in BRFSS is underway. Respondents with severe OA are more likely to report erroneously that they have RA and respondents with very mild RA are unlikely to reach specialist attention to enable correct diagnostic labeling. Although misclassification bias could lead us to underestimate the magnitude of smoking–RA association, but it does not explain the differential risk estimates for men and women. Another explanation for our findings could be that a unmeasured factor, say out-door exposure to a putative agent that may be associated with men and smoking and thus confound the smoking–RA relationship. Many of our respondents were from the southern USA where agriculture is a major occupation for men. Thus the possibility that smoking may be marker of exposure to another unmeasured factor can never be discounted in epidemiological studies including ours. Our data set did not have sufficient clarity in terms of different occupations to permit an analysis for such confounding. Another limitation of our study is the lack of information on the duration, stage medication, etc. on respondents with RA. While availability of such information would have increased the yield of our analyses, their absence do not diminish the validity of our findings. We found, similar to others [18,20], a self-report of having ever smoked was a better marker for risk for RA than a self-report of current smoking. This result is not unexpected. Respondents

501

with RA who continue to smoke are more likely to have experienced higher disability, mortality, and institutionalization rates, and therefore, under-represented in our sample. The “ever-smoking” variable is thus likely to be a better discriminator between those exposed to tobacco and those who were not. Our study highlights the need for further studies on smoking, RA, and in addition, their effect modification by gender. We believe that future studies where detailed smoking data is collected prospectively, with specific attention to temporal sequence of smoking and onset of RA, will provide leads for further etiologic research on the role of hormones, tobacco products, and immune abnormalities. Our results also suggest that pooled summary of risk estimates for smoking on RA should be interpreted with caution and that interactions should be explored in future epidemiologic studies on smoking and RA.

References [1] [2]

[3] [4]

[5] [6]

[7] [8]

[9]

[10]

[11]

[12] [13] [14]

[15] [16]

Albano SA, Santana-Sahagun E, Weisman MH. Cigarette smoking and rheumatoid arthritis. Semin Arthritis Rheum 2001;31:146–59. Jonsson T, Thorsteinsson J, Valdimarsson H. Does smoking stimulate rheumatoid factor production in non-rheumatic individuals? APMIS 1998;106:970–4. Tuomi T, Heliovaara M, Palosuo T, Aho K. Smoking, lung function, and rheumatoid factors. Ann Rheum Dis 1990;49:753–6. Aho K, Heliovaara M, Maatela J, Tuomi T, Palosuo T. Rheumatoid factors antedating clinical rheumatoid arthritis. J Rheumatol 1991;18: 1282–4. Heliovaara M, Aho K, Aromaa A, Knekt P, Reunanen A. Smoking and risk of rheumatoid arthritis. J Rheumatol 1993;20:1830–5. Uhlig T, Hagen KB, Kvien TK. Current tobacco smoking, formal education, and the risk of rheumatoid arthritis. J Rheumatol 1999;26: 47–54. Wessie J. dm46: enhancement to the sample command. Stata Technical Bull 1997;37:6–7. Brownson RC, Jackson-Thompson J, Wilkerson JC, Kiani F. Reliability of information on chronic disease risk factors collected in the Missouri Behavioral Risk Factor Surveillance System. Epidemiology 1994;5:545–9. Marks JS, Hogelin GC, Gentry EM, Jones JT, Gaines KL, Forman MR, et al. The behavioral risk factor surveys: I. State-specific prevalence estimates of behavioral risk factors. Am J Prev Med 1985; 1:1–8. Gentry EM, Kalsbeek WD, Hogelin GC, Jones JT, Gaines KL, Forman MR, et al. The behavioral risk factor surveys: II. Design, methods, and estimates from combined state data. Am J Prev Med 1985;1: 9–14. Bradstock MK, Marks JS, Forman MR, Gentry EM, Hogelin GC, Trowbridge FL. The behavioral risk factor surveys: III. Chronic heavy alcohol use in the United States. Am J Prev Med 1985;1:15–20. Levy SL, Lemeshow S, editors. Sampling of populations: methods and applications. New York: Wiley; 1991. STATA program. Version 7.0. Stata Corporation; 2001. Greenland S, Rothman K. Concepts of interaction. In: Rothman K, Greenland S, editors. Modern epidemiology. Philadelphia: Lippincott, Williams and Wilkins; 1998. p. 329–42. Clayton D, Hills M. Statistical models in epidemiology. Oxford, UK: Oxford University Press; 1993. Krishnan E, Sokka T, Hannonen P. Smoking–gender interaction and risk for rheumatoid arthritis. Arthritis Res 2003;5:R158–62.

502

E. Krishnan / Joint Bone Spine 70 (2003) 496–502

[17] Voigt LF, Koepsell TD, Nelson JL, Dugowson CE, Daling JR. Smoking, obesity, alcohol consumption, and the risk of rheumatoid arthritis. Epidemiology 1994;5:525–32. [18] Symmons DP, Bankhead CR, Harrison BJ, Brennan P, Barett EM, Scott DG, et al. Blood transfusion, smoking, and obesity as risk factors for the development of rheumatoid arthritis: results from a primary care-based incident case–control study in Norfolk, England. Arthritis Rheum 1997;40:1955–61. [19] Hutchinson D, Shepstone L, Moots R, Lear JT, Lynch MP. Heavy cigarette smoking is strongly associated with rheumatoid arthritis (RA), particularly in patients without a family history of RA. Ann Rheum Dis 2001;60:223–7. [20] Hernandez Avila M, Liang MH, Willett WC, Stampfer MJ, Colditz GA, Rosner B, et al. Reproductive factors, smoking, and the risk for rheumatoid arthritis. Epidemiology 1990;1:285–91. [21] Criswell LA, Merlino LA, Cerhan JR, Mikuls TR, Mudano AS, Burma M, et al. Cigarette smoking and the risk of rheumatoid arthritis among post-menopausal women: results from the Iowa Women’s Health Study. Am J Med 2002;112:465–71. [22] Vessey MP, Villard-Mackintosh L, Yeates D. Oral contraceptives, cigarette smoking and other factors in relation to arthritis. Contraception 1987;35:457–64. [23] Karlson EW, Lee IM, Cook NR, Manson JE, Buring JE, Hennekens CH. A retrospective cohort study of cigarette smoking and risk of rheumatoid arthritis in female health professionals. Arthritis Rheum 1999;42:910–7. [24] Silman AJ, Newman J, MacGregor AJ. Cigarette smoking increases the risk of rheumatoid arthritis. Results from a nationwide study of disease–discordant twins. Arthritis Rheum 1996;39:732–5. [25] Koepsell TD, Dugowson CE, Nelson JL, Voigt LF, Daling JR. Noncontraceptive hormones and the risk of rheumatoid arthritis in menopausal women. Int J Epidemiol 1994;23:1248–55.

[26] Brennan P, Bankhead C, Silman A, Symmons D. Oral contraceptives and rheumatoid arthritis: results from a primary care-based incident case–control study. Semin Arthritis Rheum 1997;26:817–23. [27] Brennan P, Silman AJ. An investigation of gene–environment interaction in the etiology of rheumatoid arthritis. Am J Epidemiol 1994;140: 453–60. [28] Brennan P, Ollier B, Worthington J, Hajeer A, Silman A. Are both genetic and reproductive associations with rheumatoid arthritis linked to prolactin? Lancet 1996;348:106–9. [29] Hannaford PC, Kay CR, Hirsch S. Oral contraceptives and rheumatoid arthritis: new data from the Royal College of General Practitioners’ oral contraception study. Ann Rheum Dis 1990;49:744–6. [30] Carette S, Surtees PG, Wainwright NW, Khaw KT, Symmons DP, Silman AJ. The role of life events and childhood experiences in the development of rheumatoid arthritis. J Rheumatol 2000;27:2123–30. [31] Brennan P, Silman A. Breast-feeding and the onset of rheumatoid arthritis. Arthritis Rheum 1994;37:808–13. [32] Brun JG, Nilssen S, Kvale G. Breast feeding, other reproductive factors and rheumatoid arthritis. A prospective study. Br J Rheumatol 1995;34:542–6. [33] Reckner Olsson A, Skogh T, Wingren G. Comorbidity and lifestyle, reproductive factors, and environmental exposures associated with rheumatoid arthritis. Ann Rheum Dis 2001;60:934–9. [34] Star VL, Scott JC, Sherwin R, Lane N, Nevitt MC, Hochberg MC. Validity of self-reported rheumatoid arthritis in elderly women. J Rheumatol 1996;23:1862–5. [35] Kvien TK, Glennas A, Knudsrod OG, Smedstad LM. The validity of self-reported diagnosis of rheumatoid arthritis: results from a population survey followed by clinical examinations. J Rheumatol 1996;23: 1866–71.