Study population and study design

UK Biobank is a national-based health project that aims to improve the diagnosis, treatment, and prevention of diseases such as cancers, diabetes, stroke, heart disease, osteoporosis, arthritis, eye diseases, dementia and depression [2]. A total of 502,650 participants aged between 39 to 71 years were enrolled in the study between 2006 and 2010 and they continue to be clinically followed up. Details can be found at http://www.ukbiobank.ac.uk/. In addition to the collection of biological samples (blood, saliva and urine), health, demographic and anthropometric data were collected in 22 UK assessment facilities across England, Wales and Scotland. Detailed physical / physiological measurements were further supported by the administration of questionnaires and eye examination. Many participants completed additional detailed questionnaires on work history, diet, and cognitive function. Anonymized data are now available to researchers across the world [2, 3]. Our study acquired data on the female cohort (273,467 female participants) from UK Biobank. The UK Biobank female cohort had a mean follow up time of 6.9 years (at 2016). Data on exposures were defined prior to the development of BC in cases or prior to the first assessment date in controls.

Statistical analysis

To assess associations between exposures and BC risk in the cohort, we computed relative risk (RR) and 95% confident intervals (95% C.I.) using a binomial generalised linear regression model. Regression analyses were performed for each independent variable and were adjusted for age, family history of BC in first degree relatives, and deprivation score. The independent variables list and description are presented in S2 Table.

All analyses were stratified by menopausal status: pre- and post-menopausal. The criteria for pre-menopausal were females aged ≤ 55 years old (according to the NHS the menopause age in the UK is between 40 to 55 years [10]) who reported that they still had periods and did not report a history of hysterectomy or bilateral oophorectomy, and menarche age ≥ 7 years old (the menarche age in the UK ranges from 7 to 20 years [11]). Post-menopausal females were defined as those who reported no longer having periods and did not report a history of hysterectomy or bilateral oophorectomy and their menopause age ≥ 40 years old. These criteria were employed to minimise inclusion of both pre-mature and the medically induced pre- or post-menopausal women. After further application of criteria, 61,903 participants were in pre-menopause group and 133,704 participants were in post-menopause group.

To compute BC incidence within the cohort, we used the STATA stptime command to obtain the overall person-time of observation and disease incidence rate. To calculate time for each participant, we subtracted the endpoint (either the date of cancer diagnosis or the end of the follow-up—January 1^st, 2016) with the date of study enrolment. Incidence rates were estimated for the whole cohort and pre- and post-menopausal separately. Moreover, population attributable fractions (PAF) were calculated using the punaf command [12] where the fraction was estimated compared to whole cohort and compared to the most significant subgroup associated with the BC. This was done to estimate how much risk could be eliminated by controlling that risk factor in both groups.

All statistical analysis was performed using STATA MP 14.1 software for Windows [13]. Results with 95% confident intervals not including 1 were considered as being statistically significant.

Reproductive factors and breast cancer

RRs for the reproductive factors and BC risk are presented in Table 3. For the pre-menopause group, menarche age as continuous variable showed a slight risk reduction (RR = 0.95, 95%CI; 0.90–1.00). When menarche age was grouped into >13 years old (as a reference group) versus ≤13 years old, a moderate increased risk was observed (RR = 1.23, 95% CI; 1.04–1.45). For the post-menopause group, age at menarche did not show any significant association with BC risk (confidence interval value included 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Relative risks of the reproductive factors based on the menopausal status.

https://doi.org/10.1371/journal.pone.0201097.t003

Parous women were at reduced BC risk in both pre- (RR = 0.76, 95% CI; 0.64–0.91) and post-menopausal women (RR = 0.82, 95% CI; 0.73–0.93) when compared to nulliparous women. The ‘number of children’ when treated as a continuous variable showed moderate decreased BC risk (pre-menopause group RR = 0.93, 95% CI; 0.86–0.99 and post-menopause group (RR = 0.90, 95% CI; 0.86–0.94). In contrast, increasing maternal age at live birth showed very slight increased BC risk in both pre- (2%) and post-menopausal women (1%). Further analysis was carried out in parous women to explore the association of age at live birth and BC risk. Age at first live birth as categorical variable (< 20 years old as the reference group, 20–24, 25–29, and ≥30 years old) showed that among pre-menopausal females, BC risk was almost double when they reported having had their first child at age ≥30 years old and at age 25–29 years as compared to women who reported having their first baby at age <20 years old (RR 1.94; 95% CI, 1.06–3.54 and RR = 1.88 with 95% CI; 1.04–3.42, respectively). This effect was not seen in post-menopausal females (all 95% CI values included 1). Both pregnancy termination history (ever versus none) and number of terminations were not significantly associated with BC development in both pre- and post-menopausal females (all 95% CI values included 1).

The Reproductive Interval Index (the difference between age at first child and the age of menarche) based on the interquartile range of the control group (low as reference group, moderate, high, and no children) only showed statistically significant increased risk in ‘high’ (RR = 1.42, 95% CI; 1.10–1.84) and ‘no children’ groups (RR = 1.53, 95% CI; 1.21–1.94) in pre-menopausal females. In post-menopausal group, only females reporting no children showed an increased risk of BC (RR = 1.33, 95% CI; 1.16–1.53) when compared to the low index group.

History of oral contraceptive (OC) pills used showed no association with BC risk in both pre- and post-menopause groups. Within the OC use group, however, OC duration showed a slight increased BC risk in pre-menopause women of 2% but not in post-menopausal women. Hormone replacement therapy (HRT) was not associated with risk of BC in pre-menopause UK females. In the post-menopause group, women who reported using HRT were at moderate significant increased risk (RR = 1.14, 95%CI; 1.04–1.26).

Women in both pre- and post-menopausal groups who reported having had mammograms were at increased risk of BC of 19% and 26%, respectively.

PAF were calculated for the modifiable risk factors only based on the menopause status (Table 4). Two fractions were estimated; the PAF among the studied population and the PAF among the sub-population (the exposed significant group) to evaluate how many cases could be avoided if a particular factor was eliminated. Among pre-menopausal females these modifiable factors were the strongest in reducing the BC risk. Giving birth at age <30 can eliminate about 44.6% of the BC cases in general population, and about 48.4% among females who had first children at age ≥30years old and about 46.9% of cases among females who had first children at age 25–29. Followed by low reproductive interval index with about 34.6% of BC cases can be eliminated among null-parous females and about 29.6% of BC cases can be eliminated among females with high index (>16.01). Being parous can eliminate only 9.2% of the cases without taking into consideration the number of children they gave birth to. Finally, having BMI ≥30 and WTH >0.85 can eliminate 70% and 66.2% of the cases among pre-menopausal women, respectively.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Population attributable fraction (PAF) among modifiable breast cancer risk factors according to the menopausal status.

https://doi.org/10.1371/journal.pone.0201097.t004

Among post-menopausal women; reducing BMI <30 can eliminate 8.3% among general population and 19.4% among obese females; being parous can eliminate 17.9% among null parous females; having more than one child can eliminate 21.1% % among females with <1 child; not using HRT can eliminate 12.5% of cancer cases among users.

The most effective preventative factors identified were giving birth at earlier age, having more than one child, reducing the reproductive interval index, and reducing weight.

A summary for the significant factors associated with development of BC among UK females is presented in S3 Table.

Age

For both pre- and post-menopausal groups, age is associated with increasing risk of developing BC. Age is a well-established risk factor for BC [16]. BC incidence increases with age during the reproductive years by the double in every 10 years up until the menopause [5, 15]. A potential explanation could be cells becoming more susceptible to environmental carcinogens and modification in the biological ageing which stimulates or allows tumour growth and metastasis [17].

Family history

Family history of BC is also a well-established risk factor. Our findings suggested that females with a first degree relative (sibling or mother) affected with BC were at high risk of developing BC. Regardless of menopause status, the estimated risks were higher in females who reported only their sibling(s) affected with BC as compared to females who reported only their mother affected with BC. The estimated risks were even higher when both mother and sister were affected with BC. Evidence of family history of BC in the first degree relatives and BC risk has been well documented by many studies with different study designs [14, 18]. The variation of reported estimated risks was due to family history nature such as affected age, number and type of the affected family members [19, 20]. It is known that BRCA1 and BRCA2 gene mutations are responsible for this strong association for cases diagnosed at young age [21, 22]. The stronger effect of family history among pre-menopausal females in this study suggested a component of familial BC [20]. Possible explanations to higher estimated risks observed in subjects with sibling affected include recall bias. With self-reported data, maternal history is more likely to be incomplete as compared to the sibling history. Another possibility is the confounder effect such as parity; mothers of subjects were obviously parous while sisters could be either parous or nulliparous. It is known that parity is a protective factor against BC hence if subject’s sisters were null-parous; one would expect to observe higher risk. Sisters are more likely to share the same or similar environmental factors than mother and a daughter. Finally, multiple family relatives having an early onset or bilateral cancer increases the risk even more [15].

Deprivation score

Deprivation score data was available for the dataset. Our result suggested that the most deprived females appeared to have lower BC risk compared to least deprived females in the UK Biobank cohort. Our cohort appeared to be mainly from least deprived districts like Bristol (8.8%), Leeds (8.9%), Newcastle (7.4%), and Nottingham (6.8%). Most deprived districts included Stockport (0.76%), Manchester (2.7%), and Birmingham (4.9) contributed less in this cohort. This sampling distribution could have an effect on the association direction between deprivation and BC.

Variables related to body size

Inverse associations were observed with BMI and waist to hip ratio in the pre-menopausal group. While among post-menopausal females, increased risks were reported. A Norwegian prospective study suggested a decreased risk of BC among overweight and obese females who had no family history of BC. Nevertheless once a female has a family history, that protection effect disappeared in both overweight and obese pre-menopausal females [23]. A meta-analysis conducted in 2012 showed no significant effect of BMI on the incidence of pre-menopausal BC [24]. Our results however suggested that risk was reduced even when family history of BC was present among pre-menopausal females. One study reported an estimation of 3% risk increase in BC for every 1 kg/m² in post-menopausal females [25], while another study reported that weight gains of 5–12 kg increases the post-menopausal BC risk by 50% and modest weight loss (5–10%) can decrease BC risk by 25–40% [26]. Furthermore, overweight and obesity are associated with poor prognosis and increased BC mortality [27]. BMI is a modifiable factor and can contribute to reduce the BC risk by 10.0% in pre- and 5.1% in post menopause women [28]. Our study confirmed a BC risk reduction of 8.3% if females reduced their BMI lower than 30 among general population but if obese females (BMI≥30) reduced their BMI to normal BMI range, a 19.4% of BC risk will be eliminated among post-menopaused females. Another way to assess central adiposity among individuals is by measuring WHR (waist to hip ratio). A systematic review on the relationship of WHR and BC concluded that 24% risk reduction was associated with small WHR in post-menopausal females. In contrast among pre-menopausal the effect was very little [29]. Another review suggested the same conclusion; pre-menopausal BC is not associated with WHR however, 1.4 to 5.4 times of BC risk was proven among post-menopausal females [30]. Our study showed BC risk reduction was associated with increased WHR up to 25.6% in pre-menopausal females but failed to prove any association with post-menopausal females. The findings on height and BC risk supported adult height being associated with BC risk in both pre- and post-menopausal groups. The EPIC cohort study [31] reported a positive association between height and post-menopausal BC (RR 1.10 with 95% CI 1.05–1.16). Furthermore, a meta-analysis of 159 prospective studies showed a pooled BC RR of 1.17 (95% CI = 1.15–1.19) per 10cm increase in height [32, 33]. Another pooled analysis also suggested positive association among post-menopausal females (RR = 1.07 with 95% CI: 1.03, 1.12) [34]. No association was reported in pre-menopausal females (RR 1.02 with 95% CI: 0.96, 1.10). Not all prospective studies confirmed the positive association. A register-based cohort study with 13,572 participants concluded no statistical evidence of association between height and BC risk [35]. Evidence from case-control studies was inconsistent. Our study showed an increased risk of 18% per 10cm increase in height among pre-menopausal and 23% per 10cm increase in height among post-menopausal. All the results mentioned previously were for standing height; we examined sitting height and found a BC risk association with sitting height. Taller sitting height is associated with 25.5% BC risk increase per each 10 cm increase in pre- and 37.0% in post-menopausal per 10 cm increase.

The relationship between height and BC suggests a protective effect among females with short stature rather than a continuous increased risk with the increasing of female’s height. One possible explanation is that short females would be exposed to lower levels of insulin like growth factor 1 (IGF 1) throughout childhood and adolescence. IGF-1 is considered to be a strong mitogen for BC cells and IGF-1 receptors are expressed in breast tumour tissues 10 folds higher than normal breast tissues [36, 37].

Reproductive factors

Our findings suggested protective effect of factors related to childbearing and having more children among pre- and post-menopausal females. Risk factors in pre-menopausal females were early menarche age (<13 years old), late age at first live birth (>25 years of age), high reproductive interval index, and increased duration of OC used were considered as risk factors for BC in pre-menopausal females. Factors such as nulliparous, high reproductive interval index and increased duration of OC used were risk factors in post-menopausal females.

Increased production of steroid hormone starts around the time of menarche and decreases significantly near the menopause [4]. Hormones produced by the ovary directly affect the breast function and development. Studies showed long period of hormonal exposure increases the risk to develop BC. Late menarche and early menopause are known to be protective factors as the period of hormonal exposure is reduced. Lengthening the reproductive years by an early menarche of one year has a stronger effect than delaying the menopause by one year [4]. The strength of menarche age and menopause age on BC development can be affected by BMI [38, 39]. The association between the BC and menopause age can be weaker among post-menopausal females with high BMI as seen in the meta-analysis [4]. Our results showed an evidence of BC risk reduction by late age of menarche but not by early age the menopause age as the previous studies even when BMI was adjusted for in the analysis. A meta-analysis of 120,000 BC cases and 300,000 controls done by a collaborative research group confirmed the existing association between early menarche and developing risk of BC. Extra risk is associated with lengthening female’s reproductive years by one year during menarche rather than lengthening one year at menopause [4]. The RR associated with early menarche was 1.05 (95% CI 1.04–1.06) and the RR associated with late menopause was 1.03(95% CI 1.03–1.03) [4].

Childbearing in a known protective factor against BC although other factors might help confound this protection, such as breast feeding [40]. Combination of both factors can help protect females even more. Unfortunately there were no data available on breastfeeding in our cohort and unable to assess this effect. In the case of parity, our results showed a significant evidence of risk reduction among both pre- and post- menopausal females with a stronger effect among pre-menopausal. Likewise, as the number of children increases, the protective effect increases. Our results suggested an elimination of 9.2% among pre- and 17.9% among post- BC risk associated with being a parous female while other study reported a lower yet an affective risk reduction of 13.3% for the same factor [41]. As the number of children increases, the attributed risk reduction increases accordingly with reduction of 5.2% among pre- and 5.4% among post-menopausal females [28]. Nevertheless, our results suggested a higher reduction among pre- (8.8%) and a lower reduction percentage among post- menopausal women (4.6%).

Termination of pregnancy, whether induced or natural did not appear to affect the BC risk. Thus, younger age at childbirth is a protective factor against BC and this was observed among pre-menopausal females with p values <0.05. Studies showed early pregnancy causes permanent morphological changes to the breast and makes it more resistant to carcinogenic changes [7]. Our study supported the elimination of 44.6% of BC risk if females in general had their first child in their 20s rather than ≥30 years old among pre-menopaused females. This reduction can reach up to 48.4% among females who had their first child at age of ≥30 if they had their first child in their twenties. Furthermore we explored the reproductive interval variable (duration between the menarche and first child) and the results supported evidence reported in the literature that as the duration increases the risk also increases. Long term hormonal exposure has been confirmed to be a risk for BC [15]. Our study showed a BC reduction of 14.9% in pre-menopausal women if they have reproductive interval of < 16 and this reduction can reach up to 29.6% if those females with reproductive interval of ≥16 had interval of 12 or less among pre-menopausal females.

Mammogram history suggested borderline significant increased risk in pre-menopausal women and no association in post-menopausal women. The mammogram itself per se is not a risk factor for BC but women who reported having had a mammogram were more likely to be diagnosed. Mammogram screening is proved to reduce the BC mortality by 29% among females aged between 50–69 years [42].

Hormone use

Oral contraceptive use is known to be a risk factor of BC and this risk rises with longer duration of use [43]. It has been proposed that using OC can activate breast tumours which are already present. Oestrogen is recognized as enhancing tumour growth, and with OC and later HRT use these hormones promotes the tumour growth even more [43]. Our findings suggested a positive association between BC and OC duration amongst pre- menopausal females only. Moreover, HRT users showed 14.1% more risk for developing BC among our cohort. Extensive evidence showed an increase in BC incidence in current HRT users and that risk returns to normal soon after use terminates. Combined oestrogen-progesterone therapy revealed higher risk compared to oestrogen only preparations including results from the Women Health Initiative study (WHI). Recent results from WHI found both oestrogen only and combined formulations convey greater risk for BC if the females started the HRT in less than 5 years after the menopause compared to longer gap [38, 44–47]. The study also carried out further analysis of HRT. Their results showed attenuated BC risk among obese females which is driven by hormonal adiposity of the breast. Endogenous oestrogen rises with the increase of the BMI among HRT non-users which increases the breast adiposity [38]. Another major study carried out in the UK (Million Women Study) identified that BC risk is associated with current use of HRT and the risk is considerably greater among combined oestrogen- progesterone users than other types of HRT [48]. According to our analysis stopping HRT can reduce the risk by 5.8% and by 12.5% risk among HRT users. The Million Woman Study estimated this figure to be 4.6% [41] and a more recent study has put this figure higher at 14.5% [28].

In conclusion, we carried out an analysis to confirm risk and protective factors and BC risk in the UK Biobank female cohort. The findings suggest that protective factors in women included reducing BMI, waist to hip ratio, increasing the numbers of births, having birth at an early age, minimising the use of oral contraceptive and HRT and their durations. Most of our findings are in keeping with evidence reported from the other UK large cohort studies such as the One Million Women and EPIC studies. Evidence from this large study can be further used in translational research such as prevention programmes. Our study has some strengths and limitations. The strengths of this study are: large nation-wide prospective population-based cohort with a follow up time of 9 years and a sizable number of incident cases (UK Biobank). Furthermore, to our knowledge, this is the first study investigating the effect of the anthropometric and reproductive factors with BC risk among the UK Biobank female cohort. The results of this study can be used to inform BC prevention strategies and be used to educate the public and form a basis for building risk prediction models for BC for the UK population. Additionally reproductive interval index is a new measure and only reported by our study using UK data. Estimation of the general PAF and the PAF of the subgroups for BC in the UK Biobank female cohort is novel. The attributable risks calculated for the modifiable factors can be translated into action to reduce BC incidence.

One of the study limitations is that the UK Biobank cohort is not the best representation of UK female population. A recent study investigated the sociodemographic characteristics of the UK biobank participants compared to normal UK population [49] found an evidence of “healthy volunteer” selection bias among the participants. UK biobank participants tend to be healthier, more educated and living in less deprived areas. This effect is common with other volunteer cohorts. Nonetheless, to overcome this limitation and to produce more generalizable associations it is very essential to use large sample size with high internal validity [50, 51]. Our study used a decent sample size and confirmed the expected associations which are similar to the published literature.

Another possible limitation is the lack of information such as breastfeeding history, ovarian cancer family history, BC onset of the family members and BC subtype (PR+, ER+, HER2+, triple negative). Some of the risk factors may affect BC subtype differently [52]. Finally, small sample size in some of the associations such as family history of breast cancer. There were only4 observations among pre-menopaused with both mother and sister family history which can affect the strength of the findings.