Both genetic and environmental factors can predispose to rheumatoid arthritis (RA). HLA-DRB1 alleles featuring the shared epitope1 are found in most patients with RA, and genetic factors have an effect on disease progression.2 On the other hand, only an estimated 16% concordance of RA in monozygotic twins has been found,3 indicating that environmental factors also play an important role in the development of RA. Suggested predictors include smoking,4 level of education,5 and certain occupational exposures.6 7 Furthermore, it has been hypothesised that infectious agents, especially Epstein–Barr virus, could initiate the inflammatory process leading to RA.8

Overall, women have more than a twofold higher incidence of RA than men.9 10 This is mainly due to an increased risk in women during their reproductive years, when the incidence shows a female/male ratio of 5:1.9 This difference may partly be explained by hormonal factors.

A number of studies have shown that use of oral contraceptives (OCs), which contain hormones that are also raised during pregnancy, may protect against RA,11 and be associated with milder disease and less frequent hospital referrals,12 13 but some studies have failed to show any protective effect of OC use.14^–16 Methodological differences may explain these discrepancies, and a biological explanation has never been presented, leading to the hypothesis that OC use is more likely to be a confounder or a marker for other protective variables.

It has been known for a long time that amelioration of RA is often seen in pregnancy,17^–19 possibly because of increased immune tolerance and a shift towards Th₂ differentiation of T cells, driven by high concentrations of circulating sex hormones and endogenous corticosteroids.18

There is often a flare of the disease during the postpartum period, which is associated with a sudden fall in cortisol concentration and, during breast feeding, a high concentration of prolactin.20 It has been recognised that postpartum flares of the disease tend to be more pronounced in women who breast feed.21 22

In contrast, previous studies have suggested that extended breast feeding is associated with a reduced risk of future onset of RA.14 23 The underlying mechanisms are unknown, and there is limited information on the relative effect of breast feeding and other suggested predictors of RA, such as smoking4 and low level of formal education.5

The purpose of this study was to determine whether breast feeding and OC use affect the future risk of RA in a community-based cohort, with adjustment for confounders where applicable.

PATIENTS AND METHODS

This nested case–control study used information from the Malmö Diet and Cancer Study (MDCS), a community-based health survey performed between 1991 and 1996 in Malmö, Sweden. The MDCS included 30 477 subjects (18 326 women) aged 44–74. The main objective was to study the effect of diet on cancer incidence and mortality. Information on lifestyle factors, such as smoking and level of education, and previous and current health status was obtained using a self-administered questionnaire. Data were collected on reproductive factors, such as history of OC use (including number of years and when started), number of births, and whether each child was breast fed and for how many months. This survey has been extensively described previously.24

The MDCS cohort was linked to a community-based RA register,25 the local outpatient clinic administrative register, the National Hospital Discharge Register and the National Cause of Death Register. The medical records of patients identified through these sources were subjected to a structured review, and all women given a diagnosis of RA, according to the 1987 American College of Rheumatology (ACR) criteria for RA,26 after inclusion in the MDCS were included in our case–control study. Four female controls for every case, matched according to age and year of screening, who were alive and free of RA when the index person was given a diagnosis of RA, were selected from the MDCS survey population.

Patients were stratified according to OC use: never, 1–5 years of use, and >5 years of use. We also compared women who never used OCs with “ever users” and stratified them into groups depending on which decade they started OC use, from 1960 to after 1990, and by breast-feeding history as follows: never, total breast feeding for 1–12 months, total breast feeding for ⩾13 months. Non-responders were excluded from the analysis, with one exception: women who did not respond to the question “How many children have you given birth to?” were considered nulliparous.

Women who reported current daily smoking were compared with those who did not smoke. Level of formal education was stratified into five levels (elementary school only (⩽8 years of school), 9–10 years of school, 11–12 years of school, >12 years of school (but no university degree), and university degree).

Potential predictors were examined in χ² test and logistic regression models, taking into account the matched design of the study. Each case and the corresponding controls were assigned a group number, and this was entered into the logistic regression models as a categorical variable. To evaluate trends favouring a dose-dependent effect of breast feeding, the breast-feeding categories (no breast feeding vs 1–12 months vs ⩾13 months) were entered as an interval level variable. Multivariate models were used to adjust for potential confounders, such as smoking, level of education, and parity. Statistical significance was identified as p<0.05.

This study was approved by the regional research ethics committee for southern Sweden.

RESULTS

We identified 136 women with incident RA (table 1). The total follow-up for the cohort was 188 969 person-years, which gives an estimated incidence of 72/100 000 person-years. The mean age at RA onset was 63.3 years with a median duration of 5.5 years (range 1–13 years) from enrolment in the health survey to RA onset. These were compared with 544 age-matched female controls. OC use did not have any significant effect on the risk of RA onset (table 2). Longer reported OC use did not reduce the risk of RA (data not shown). Most OC users among the cases (81.7%) and the controls (84.5%) had started taking OCs between 1960 and 1970.

Although parity did not differ significantly between cases and controls (odds ratio (OR) 0.75, 95% CI 0.45 to 1.24), there was a trend towards a reduced risk of RA for each child born (OR 0.87, 95% CI 0.71 to 1.06). A history of breast feeding was less common in the RA group (table 1). A longer history of breast feeding was associated with a reduced risk of RA (OR 0.46 (95% CI 0.24 to 0.91) for women who had breast fed for ⩾13 months, and OR 0.74 (95% CI 0.45 to 1.20) for those who had breast fed for 1–12 months, compared with those who had never breast fed; fig 1) (p for trend 0.025; no breast feeding vs 1–12 months vs ⩾13 months). The trend was seen in both rheumatoid factor (RF)-positive and RF-negative women, but was stronger in the RF-negative group (table 2).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Risk of rheumatoid arthritis in women who reported total breast feeding for 1–12 months or ⩾13 months compared with no breast feeding (136 cases and 544 controls). Odds ratios with 95% CI are shown.

When the number of children and breast feeding were included as covariates in the same multivariate model, the results indicate that it is breast feeding, rather than the number of children, that is inversely associated with RA (table 3). The protective effect of longer breast feeding remained significant after adjustment for potential confounders such as smoking and level of education in multivariate models (table 4).

When the analysis was restricted to parous women, the estimated effect of long-term breast feeding was similar, although the confidence intervals were wider (OR 0.46 (95% CI 0.17 to 1.25) for women who had breast fed for ⩾13 months, and OR 0.78 (95% CI 0.33 to 1.83) for those who had breast fed for 1–12 months, compared with those who had never breast fed; fig 2). Fifty-one women (13 cases) had no reported history of breast feeding, although they had given birth to one or more children (median 2; range 1–4). Compared with these, nulliparous women (n = 129) tended to have a lower risk of RA (OR 0.48; 95% CI 0.13 to 1.72), although the difference did not reach significance and statistical power was limited for this comparison.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Risk of rheumatoid arthritis in women who reported total breast feeding for 1–12 months or ⩾13 months compared with no breast feeding. The analysis was restricted to parous women (106 cases and 445 controls). Odds ratios and 95% CI are shown.

DISCUSSION

Breast feeding is known to have multiple health benefits for the baby27^–30 and may protect mothers against breast cancer31 and ovarian cancer.32 In addition, we show that women who had breast fed for 13 months or more had a significantly reduced risk of developing RA. Furthermore, our results indicate that breast feeding may have a dose-dependent protective effect against RA, although the difference did not reach significance for women who had breast fed for 1–12 months.

Although it is difficult to separate the effect of breast feeding from that of childbirth, our data suggest that RA is inversely associated with long-term breast feeding, rather than with the number of children born. This should be confirmed in a larger study. Nulliparity has in previous studies been found to be a risk factor for RA,33 but not all studies have been able to confirm this relationship.11 34 We found no significant association between pregnancy and RA, suggesting that nulliparity did not confound our results. A Norwegian prospective study on risk of RA mortality23 showed that total time of breast feeding was associated with a decreased risk of RA, but this study had major limitations. RA cases were identified via the death register, and therefore long-term survivors and patients with RA who did not have a diagnosis of RA registered on the death certificate were excluded. Selection bias was thus a major problem with this early study. More recently, Karlson et al14 confirmed that there is a dose-dependent inverse correlation between breast feeding and RA. They showed a significant protective effect, adjusted for smoking, when total breast feeding exceeded 24 months, with a mean duration between breast feeding and RA onset of 25 years. These observations were based on the Nurses’ Health Study, a large cohort of women that included 674 incident cases of RA. Our study confirms and extends these findings. It shows that the negative association between long-term breast feeding and RA is not limited to one defined social group (ie, nurses), but extends to a mixed cohort from an urban area. A third prospective study16 of 158 cases of incident RA in Iowa found a modest non-significant protective effect of the number of children breast fed, after adjustment for age and smoking. As no data were available, the effect of the duration of breast feeding could not be studied.

Possible explanations for the protective effect of breast feeding include long-term immunomodulation, such as the development of progesterone receptors on lymphocytes, dysregulated hypothalamic–pituitary–adrenal axis, and differences in cortisol concentrations. Lankarani-Fard et al35 measured cortisol concentrations in postmenopausal women, and noted significantly higher concentrations in those who had breast fed. This requires further study.

Our findings may seem to contrast with the reported increased risk of RA during breast feeding.13 Oxytocin, one of the hormones that is raised in women who breast feed, is known to reduce cortisol concentrations,36 induce well-being, and lower blood pressure in the mothers.37 Prolactin, which also is increased during breast feeding, is a known immunostimulator,38 and high concentrations of prolactin are seen in patients with RA.39 40 Taken together, these findings suggest different short-term and long-term effects of breast feeding on the immune system and on susceptibility to RA.

In our study, OC use was not associated with a reduced risk of RA regardless of which decade it was started (from the 1960s to the 1990s) or its duration. A meta-analysis of nine studies concluded that OCs may prevent the progression of RA to severe disease by modifying the disease process, rather than having a protective effect.41 On the other hand, a population-based study did show a protective effect of OC use.11 Compared with that study, OC use was substantially more common in our sample. We cannot exclude the possibility that changes in the pattern of OC use explains these discrepancies. OC use could only have an effect in women who where sexually active when the pill became available in the 1960s, although this would probably not influence our results in any particular direction. As our study included women who were aged 45–73 when entering the study, we have mainly investigated previous, rather then current, OC use.

The strengths of our study include the community-based approach, the well-defined catchment area, and the comprehensive effort to identify incident RA cases using multiple sources. The estimated incidence of RA in this cohort of 72/100 000 person-years is slightly higher than recent findings in corresponding age groups in a population-based study from the UK (annual incidence 54–65 per 100 000 in women aged 50–70 and 36/100 000 in women overall).42 This suggests that we identified virtually all cases of incident RA in the cohort, indicating that our cases are likely to be a representative sample of patients with RA in the area. Data on predictors were collected before disease onset, which means that the effect of RA on lifestyle and recall bias could not influence our results. There may be recall errors in reported duration of breast feeding and duration of OC use, but there is no reason to believe that there is a differential effect between cases and controls. Limitations include the relatively low number of cases, especially when the analysis was limited to parous women. The number of women who gave birth but did not breast feed was too small for a reliable estimate of their risk compared with nulliparous women. Furthermore, our data are only from women with onset of RA in their 40s or later, and, because of the study design, the short-term effect of breast feeding could not be assessed. It is possible that unmeasured confounders influenced our analysis, in particular as there is a long lag period between the reported breast feeding and the development of RA.

A decline in incidence over time and a shift towards later onset of RA has been reported.43 Doran et al11 calculated that, on the basis of their data, the increased OC use can only explain a small portion of the decrease in RA. The breast-feeding rate in Sweden has increased steadily since the beginning of the 1970s, from less than 5% of women breast feeding for ⩾6 months to about 73%.44 A similar trend can be seen in the USA,45 and UNICEF launched a global programme, Baby Friendly Hospitals Initiative, in 1990 to promote breast feeding throughout the world.46 Whether there is a connection between the decreasing incidence of RA and higher rates of breast feeding cannot be concluded from the present data, but our findings suggest yet another reason for women to continue breast feeding.

In conclusion, our data show a decreased risk of RA in postmenopausal women with a history of long-term breast feeding. The effect was dose-dependent and remained significant after adjustment for smoking and level of education.