Cohabitation, infection and breast cancer risk

For 50 years, the effect of age at first birth (AFB) has been thought to explain the strong association between breast cancer risk and age at first marriage (AFM), which was first reported in 1926. The independent effects of AFM, AFB and number of sexual partners adjusted for parity and other risk factors were estimated in reanalysis of a large international case‐control study conducted in 1979 to 1982 (2274 breast cancers, 18209 controls) by unconditional logistic regression. Respective AFB and AFM breast cancer odds ratios (ORs) for ≥31 years relative to ≤18 years were 3.01 (95% CI 2.44‐3.71; P(trend) < .0001) and 3.24 (95% CI 2.62‐4.01; P(trend) < .0001) in univariate analyses. Among married parous women, these ORs fell to 1.38 (95% CI 0.98‐1.95; P(trend) < .03) for AFB and 1.70 (95% CI 1.17‐2.46; P(trend) < .002) for AFM when fitted together in multivariate analysis including other risk factors. A similar adjusted OR for AFM ≥ 31 years relative to ≤18 years was seen among married nulliparous women (OR 1.71, 95% CI 0.98‐2.98; P(trend) < .001). AFM (a surrogate for age at starting prolonged cohabitation) is thus strongly associated with breast cancer risk. This suggests an effect of close contact. Identifying the (probably infective) mechanism might lead to effective prevention of breast cancer. The independent effect of AFB is smaller and could be due to residual confounding.


| METHODS
The WHO Collaborative Study of Neoplasia and Steroid Contraceptives (CSNSC) was carried out in 1979 to 1982 and included 2760 cases of breast cancer and 18 381 controls aged <65 years. The methods have been described previously. 7,8 The 12 participating centres were in 10 countries: Australia, Chile, China, Colombia, East Germany, Israel, Kenya, Mexico, the Philippines and Thailand (three centres). Stata 15.1 was used to calculate odds ratios (ORs) for invasive breast cancer by unconditional logistic regression. Table 1 shows univariate ORs for each variable adjusted for age (5-year age groups) and centre (n = 12). Subsequent analyses were  Table 2. In married parous women, AFM and years from first intercourse to marriage were fitted jointly (Table 3). Table 4 shows multivariate regression results among married parous women fitting AFM, AFB, parity, duration of breast feeding and number of partners (Model 1) and the effect of omitting each variable (Models 2-6). For each fitted variable, the significance of the difference in trend among married parous women between women aged 50 to 64 and below age 50 years and between more and less developed countries (Table S1) was assessed by including an interaction term. For trend analyses, each categorical variable was refitted as continuous retaining other variables as categorical. Table 5 shows estimates of the increase in OR per year for AFM, AFB and interval from marriage to first birth (IMFB) in married parous women when these variables were fitted individually and in pairs. All significance levels are two-sided. Table 1 shows univariate ORs adjusted only for age and centre for each variable. The trend in OR is negative for parity, duration of breastfeeding and number of sexual partners and positive for AFB, AFM, AFSR and higher SES (P < .0005 for all these trends). The OR is significantly elevated for current OC use of ≥5 years' duration (heterogeneity P = .0002). Few women had five or more alcoholic drinks per week, and alcohol consumption, which showed no effect on risk (P(trend) = .9), was not considered further. Subsequent analyses were adjusted by including age, centre, OC use and SES in all models categorised as in Table 1. The analyses in Table 2, which are restricted to nulliparous women and adjusted for variables not related to childbirth, show a highly significant trend with increasing AFM in married nulliparous women (P(trend) < .001) and a weak and nonsignificant trend with increasing AFSR (P(trend) = .13), which is further reduced when these variables are fitted jointly (P (trend) = .6).

| RESULTS
The independent contributions of AFM and time from first sexual relationship to marriage in married parous women are shown in Table 3. The trend with AFM remains significant after adjustment for all variables (P(trend) = .002), but time from first intercourse to marriage has no independent effect (P(trend) = .8). AFSR was therefore omitted in subsequent analyses. The independent effects of AFM, AFB, parity, duration of breast feeding and number of sexual partners were investigated in married parous women by fitting all five variables ( but were very imprecise due to small numbers and are uninformative on the joint effects of AFM and AFB (data not shown). In the fully adjusted analysis restricted to married parous women (Table 4, Model 1), the quantitative effect of each variable is lower than that in the univariate analyses in Table 1. Fully adjusted significance levels for trend in married parous women are .001 for AFM, .03 for AFB, <.0001 for parity, .02 for number of partners and .04 for duration of breast feeding. These trends are quantitatively similar in more developed vs less developed countries and below age 50 vs 50 to 64 years (Table S1: P > .3 for all differences between trends). Correlations between these variables and IMFB in married parous controls are shown in Table S2. There was a strong correlation between AFM and AFB (correlation = 0.84) but a negative and much weaker correlation between AFM and IMFB (correlation = −0.07). AFB equals AFM plus IMFB, so the independent effects of these variables were estimated by fitting each pair ( Table 5). The OR for IMFB showed little or no effect when fitted alone (P(trend) > .9) or jointly with AFM (P(trend) > .3) but decreased sharply with increasing IMFB (3.9% per year, P(trend) = .003) when fitted together with AFB.

What's new?
Early investigations of breast cancer risk factors identified potential associations with age at first marriage (AFM) and age at first birth (AFB). Here, reanalysis of data from the WHO Collaborative Study of Neoplasia and Steroid Contraceptives (1979-82) suggests that AFM, a surrogate for age at beginning cohabitation, is strongly associated with breast cancer risk in parous and nulliparous women. The effect of AFB, a long-accepted breast cancer risk factor, appears to be due largely, and perhaps entirely, to confounding. This surprising conclusion warrants further study, along with investigation of underlying infective mechanisms as possible targets for breast cancer prevention.   Table 2 shows the marked effect of AFM on breast cancer risk in married nulliparous women (Table 2: P(trend) < .001) unconfounded by AFB and parity, and a weak trend with AFSR (P(trend) > .1) that disappears (P(trend) = .6) when AFM and AFSR are fitted together. Table 3 shows similar results in married parous women after adjusting for other variables including AFB and parity. There is a consistent and significant trend with increasing AFM (last row, Table 3; P(trend) = .002), and no evidence that time from first intercourse to marriage has any independent effect (P(trend) = .8). AFSR was therefore ignored in subsequent analyses. Table 4 shows results in married parous women for AFM, AFB, parity, duration of breast feeding and number of partners in the fully adjusted multiple regression (Model 1). The fully adjusted ORs for AFM (P(trend) = .001) are almost identical to those shown in Table 3, with the only difference between the analyses being exclusion of AFSR from the model. Adjusted ORs for AFB also increase (P(trend) = .03) but are lower than for AFM at each age.
The reduction in risk with increasing numbers of sexual partners is statistically significant (P(trend) = .002) but weak (OR relative to 1 partner = 0.85 for 2 partners, 95% CI 0.73-0.99; 0.78 for >2 partners, 95% CI 0.66-0.93). The estimated effects of breast feeding and high parity are similar to estimates based on pooled analysis of the worldwide data. 9 Table 4 also shows the effect of omitting each of these factors from the regression (Models 2-6). OR estimates for number of partners are almost identical in all models. Parity and duration of breast feeding are highly correlated (Table S2:  We conclude that age at marriage (ie, beginning prolonged cohabitation) is a cause or correlate of an important risk factor in both nulliparous and parous women, and that the estimated effect of AFB is substantially inflated if AFM is not adjusted for.
AFM and AFB are strongly correlated (  negligible) independent effect of IMFB and hence of AFB. IMFB and hence AFB would be associated with increased risk if less fertile women were at higher risk and took longer to conceive, but no association between delay in conception and subsequent breast cancer risk was seen in British women trying to conceive. 10 AFM equals AFB minus IMFB, so if AFM is the relevant risk factor and neither AFB nor IMFB has any independent effect the joint analysis of AFB and IMFB excluding AFM would be expected to show the strong positive trend with AFB and strong negative trend with IMFB seen in the right-hand column of Table 5. Evidence from larger studies is needed to estimate the independent effect of AFB (and ages at subsequent births 11 ) more precisely, but the strong negative trend with IMFB when fitted jointly with AFB confirms that AFM is an important risk factor.

| Origin of the assumption that the effect of AFM is due to confounding by AFB
The association of breast cancer with late AFM was first reported by Lane-Claypon in 1926, 1 but in 1970 McMahon and colleagues 4 concluded from their analysis of the SCS that any effect of AFM is entirely accounted for by the effect of AFB. They wrote: The pooled data for all centres do suggest lower risks for nulliparous women married under the age of 25 years than for those married later. However, relative to the trend in risks associated with age at first birth, that with age at marriage is weak. In addition, the deficit of cases observed among nulliparous women first married under the age of 20 years is confined to 2 centres. If these are excluded the trend disappears. We have no explanation for the appearance of this feature in these two centres. In view of the relatively small change in risk associated with it and its limitation to 2 of the 7 centres, we conclude that early marriage is not associated with reduction in risk of cancer of the breast, unless it is associated with early confinement. 4 Despite the weakness of this rationale, the role of AFM seems never to have been reconsidered. In a study investigating whether late AFB reflects difficulty in conceiving, and hence some hormonal aspect affecting breast cancer risk, about 80% of the difference in AFB between cases and controls was accounted for by differences in their ages at starting regular sexual activity. 10 An effect of cohabitation was not considered, however. AFM was even used as a surrogate for AFB in a recent reanalysis of two early studies. 12 Our data suggest the opposite conclusion: that some effect of cohabitation substantially influences breast cancer risk, and that most of the unadjusted effect of AFB is due to its correlations with AFM, parity and other factors (AFB ≥ 31 years vs ≤18 years: unadjusted OR 3.01 in Table 1, fully adjusted OR 1.38 in Table 4, Model 1).
In the original SCS report, the effect of AFM in parous women was not analysed, 4 but evidence of an effect in parous women in the SCS data from Wales 5 prompted one of us (L.K.) to ask Dimitrios Trichopoulos, one of the original SCS investigators, to examine the joint effects of AFM and AFB in parous women in the SCS. He died before this analysis was completed and we no longer have access to the SCS data, but the results of the preliminary analysis of the independent effects of AFB and AFM, fitted jointly but unadjusted for parity and the other factors in Table 4, are similar to the CSNSC results analysed in the same format (Table S3). Both studies show a marked trend with increasing AFM (P < .001) and a weaker although still significant trend with increasing AFB (P < .001). A long-standing hypothesis was thus discarded inappropriately, as the study generally thought to have shown that AFM has no independent effect on breast cancer risk in fact suggests a marked effect. The protective effect of high parity may have made a protective effect of young AFB seem plausible, perhaps contributing to the failure by cancer epidemiologists over the last half century to examine the joint effects of AFB and AFM in other studies; but in fact there are no strong grounds for expecting AFB to affect breast cancer risk merely because high parity is protective. The risk of ovarian cancer also declines with increasing parity but is unaffected by AFB. 13 AFM is strongly related to breast cancer risk in the only three datasets in which we could estimate its effects independent of or adjusted for AFB (nulliparous and parous women in the CSNSC and the overall results of the SCS). Our results imply that the marked effect of AFB unadjusted for AFM seen in these data and in every other study was greatly inflated by failure to adjust for AFM. Pending joint analysis of AFM and AFB in other studies we conclude that there is no published evidence that AFB is an important risk factor for breast cancer. The majority of unmarried women were virgins in the CNSNC but cohabitation is now common, so age at beginning prolonged cohabitation rather than AFM must be adjusted for in more recent studies.

| The effects of early and later cohabitation on breast cancer risk
Nulliparous women (Table 2) were reanalysed with the unmarried, 82% of whom were virgins, as the reference group. The risk is lower in those who marry before age 25 (OR 0.82 95% CI 0.63-1.06) but higher in those who marry after age 27 (OR 1.50, 95% CI 1.12-2.03), suggesting that early marriage is protective and late marriage increases breast cancer risk compared to unmarried women. However, beginning cohabitation could be slightly protective even after age 30 if women who will remain single are atypical in other ways that reduce their risk, perhaps having a pattern of interpersonal contacts different from women who will marry. Conversely, it is also possible that women who remain single are at greater risk for other reasons and cohabitation increases the risk slightly when it begins before age 19 and more when it begins later.
The lower risk in women with more than one partner suggests a protective rather than a carcinogenic effect, but pending discovery of the underlying mechanism(s) this ambiguity may be difficult to resolve.

| Possible explanations of the effect of cohabitation
The evidence that earlier age at beginning cohabitation is associated with reduced breast cancer risk seems strong and consistent,  16 while another concluded that they are unlikely to play any significant role. 17 Earlier exposure to an oncogenic infection might reduce its long-term effect through a mechanism analogous to the reduced risk of paralysis following T A B L E 5 Married parous women: odds ratios (ORs) for age at first marriage (AFM), age at first birth (AFB) and interval from first marriage to first birth (IMFB) adjusted for age group, centre, socioeconomic status, oral contraceptive use, parity, duration of breast feeding and number of sexual partners early poliovirus infection, but such a pattern has not been shown for any infective cause of human malignancy. Highly infectious sexually transmitted agents are unlikely to be involved. The risk of cervical cancer, which is caused almost entirely by sexually transmitted human papillomaviruses, is strongly associated with early age at first intercourse and increases steeply with increasing numbers of sexual partners. In contrast, early age at first intercourse has no detectable effect on breast cancer risk, and the OR is slightly lower for two sexual partners than for one but shows little further reduction for three or more partners (OR for >2 vs 2 partners 0.91, 95% CI 0.75-1.11, P = .4). This suggests a protective effect requiring prolonged contact with a partner. Recent findings suggest that more complex infective mechanisms, including effects of the microbiome on the immune system, may also affect tumorigenesis. 18,19 The epidemiology of breast cancer in relation to menarche, childbirth and menopause as well as OC use and hormone replacement therapy is plausibly accounted for by hormonal effects. An influence on differentiating mammary cells via circulating hormone levels was proposed to explain the strong protective effect of high parity, 20 and Pike and colleagues 21 suggested that changes in breast cancer incidence associated with menarche, birth and menopause are consistent with concomitant changes in breast stem cell division rates. Both gut and mammary microbiota may influence breast cancer risk by hormonal mechanisms. 22 A doserelated protective immunological response to sperm 23,24 has been suggested, but an effect of the microbiome, perhaps through anticancer immune responses, 19 may be more plausible. Correlates of prolonged cohabitation include convergence of gut microbiota between couples and increased diversity of gut microbiota compared with those who live alone. 25 Microbiota at other sites, including the breast, could also be relevant.
We hope that our results will encourage others with relevant data to examine the effects on breast cancer risk of age at beginning cohabitation, number of long-term relationships, parity and ages at first and subsequent births. Confirmatory evidence that age at beginning prolonged cohabitation influences risk and accounts for much of the effect of age at childbirth would be important whatever the mechanism. If infection were involved, transmission must occur within families and through other social contacts as well as between couples. The resulting variation in risk in the general population could be greater than the variation associated with age at beginning cohabitation, and might even account for a substantial proportion of the nongenetic familial risk. However, contacts before menarche when the breast has not developed or in childhood while the immune system is still evolving could be less important than later exposures. If an underlying infective process were identified, it might be targeted to reduce breast cancer incidence.