Introduction

Breast cancer, especially postmenopausal, is the most occurring cancer in women worldwide and the second leading cause of female cancer death [1]. In Western Europe, one in eight women develops breast cancer during her lifetime, of whom more than 75 % after the age of 50 [2]. The high burden of disease and associated treatment costs makes postmenopausal breast cancer a major public health issue. Not only incidence rates differ according to menopausal status, but effects of some risk factors are also modified by menopausal status. For example, overweight has no or even a small protective effect in premenopausal women, while it increases risk after menopause [3].

Several established risk factors for postmenopausal breast cancer are not, or rather difficult, to modify when the age of 40 has been reached, e.g. age at menarche, parity, age at first child birth and duration of breastfeeding. As lifestyle is modifiable, it provides an opportunity for primary prevention. Overweight and obesity, physical inactivity, alcohol consumption, smoking and low dietary fibre intake are all associated with an increased breast cancer risk after menopause [4–7] and are still present and modifiable at a later age.

The potential impact of preventive measures can be assessed by computing the population attributable fraction (PAF). This fraction represents the proportion of cases in a population that could be prevented if exposure to a causal factor had not occurred [8].

This research is the first to describe the situation for the Netherlands regarding exposure to lifestyle-related risk factors and breast cancer occurrence. We computed individual and combined PAF estimates for the above five lifestyle-related risk factors for the Netherlands, a country with one of the highest incidence rates of breast cancer worldwide [1].

Methods

PAF calculations

The PAF was calculated for four age categories (50–60, 60–70, 70–80, >80 years) for each of the five risk factors individually using the formula [9, 10]: PAF = 1−1/(P1*RR1 + … + Pn*RRn), where P is the prevalence of each exposure, for each exposure level of the risk factor (1 to n), see Table 2 for the different levels of exposure. For example, risk factor BMI has three exposure levels: <25 (reference), 25–30 and >30 kg/m2. The prevalence is quantified as the percentage of women that is exposed to the risk factor of all middle-aged women. The prevalence is quantified as the percentage of the total population of middle-aged women of women that is exposed to the risk factor. The RR is the relative risk of breast cancer for the risk factor of interest, for each exposure level specific (Table 1). For example, the RR for BMI < 25 kg/m2 is 1, being the reference, for 25–30 kg/m2 is 1.15 and for >30 kg/m2 is 1.33.

Table 1 Estimated relative risks for five lifestyle-related risk factor and breast cancer
Full size table

We defined postmenopausal breast cancer as all invasive breast malignancies in women aged 50 years or older. A latency period of 10 years between exposure to the hazardous lifestyle and breast cancer occurrence was assumed. Exact information about the true latency period between different exposures and clinical breast cancer presentation is not available. It is however generally accepted that this latency period is about 10 years, which we and others [11] used for our present study.

Therefore, prevalence rates were taken from the years 2000–2001, and 1997 for dietary fibre consumption, of women aged 40 years and older and related to breast cancer occurrence in women of 50 years and older in the year 2010.

To estimate an overall PAF for each risk factor, we first calculated age-specific PAFs for each age category of exposure (40–50, 50–60, 60–70 and 70+). We, therefore, multiplied the risk factors RR by the prevalence of exposure in each age category. Second, we calculated the number of preventable or attributable cases per age category in 2010 (in women aged 50 and over) by multiplying the age-specific PAFs by the number of incident invasive breast cancer cases in 2010 in the corresponding age category. In the third step, the number of attributable cases in each age category was summed over all ages and divided by the total number of invasive breast cancers diagnosed in 2010 in women aged 50 and over. By this method, we incorporated that the prevalence of exposure and the number of invasive breast cancers vary across age categories.

To estimate the PAF of postmenopausal breast cancer for five risk factors combined, summing of the five separate PAFs would lead to an overestimation of the attributable proportion of cases because women may be exposed to more than 1 risk factor. The following multiplicative formula was proposed which, under the assumption of independent exposures and effects, considers the overlap between risk factors within individuals [12]: PAF (joint risk factors) = 1− (1 − PAFx_ 1 )* (1 − PAFx_ 2 )*… (1 − PAF x_n ), where x_ 1 to x_n refers to the different risk factors being the five lifestyle-related risk factors in our current analysis.

We used a 20,000-fold Monte Carlo simulation to derive 95 % confidence intervals (95 % CI) for the PAF estimates for each risk factor and joint. Monte Carlo simulation uses random sampling according to a specified data distribution taking into account the precision of each RR and prevalence estimate. RRs and prevalence rates were independently sampled in each Monte Carlo trial from a lognormal distribution (based on a literature-derived RR estimate with 95 % CI) and a beta distribution, respectively [13]. Analyses were performed using R statistics software, version 3.0.2.

Risk factors and relative risks

We considered lifestyle-related—thus potentially modifiable—risk factors for postmenopausal breast cancer with sufficient scientific proof for a causal association (i.e. judged by the World Cancer Research Fund as ‘probable’ or ‘convincing’ causally related [6], or with a large body of evidence based on other scientific literature [4, 5] ). Furthermore, we evaluated risk factors that are currently present in middle-aged women in the Netherlands and only those which can be modified at a later age.

We derived RRs adjusted for confounding factors from meta-analyses [4–7] (see Online Resource 1 for more information). For each risk factor, a theoretical optimum level of exposure was defined and used as the reference level, with a corresponding RR of one. Reference exposures were zero where possible (e.g. zero units of alcohol intake per day), or when this was physiologically impossible, the advised level by (inter)national health guidelines was taken (e.g. a BMI < 25 kg/m2) (see Table 1).

For overweight/obesity (defined by BMI), physical activity, alcohol and fibre intake, a continuous RR was obtained from the literature, assuming a log-linear association between exposure and risk increase [4, 6, 7]. To match these continuous RRs with categorised risk factor prevalence rates, we calculated new categorical RRs based on the literature-derived continuous RR. These categorical RRs were combined with the mean exposure level within each risk factor category, as observed from the population exposure rates (for an example see footnote Table 1).

Prevalence of exposure

Age-specific prevalence rates of risk exposure were derived from large national surveys or registration databases in 1997 [14] and 2000–2001 [15–17]. Detailed information about these surveys is available in the online supplement.

Results

Prevalence rates

Table 2 presents the prevalence rates of exposure to lifestyle-related risk factors in women >40 years of age in the Netherlands in 2000–2001 and 1997. Of these women, on average 51 % were overweight/obese, which increased with age from 40 to 56 % in the ages 40–50 and >70 years, respectively. On average 55 % were estimated to be less active than prescribed by physical activity guidelines (i.e. 5 days/week 30 min of moderate intensity physical activity). Non-adherence to the national activity guideline also modestly increased with age (i.e. 53 % in 40–50 years, and 58 % in >70 years). Alcohol was regularly consumed by on average 75 % of women. Consumption was less prevalent in older than younger women (61 % in >70 years, versus 84 % in 40–50 years). Of all women, an average of 42 % reported to be currently smoking, or smoked in the past, which decreased with an increasing age (54 % in women aged 40–50 and 28 % in women aged >70 years). Dietary fibre intake was below the recommended level in on average 97 % of women, being lowest in women aged 40–50 (85 %).

Table 2 Prevalence rates of risk factor exposure among Dutch women per age category (in 2000–2001)
Full size table

Population attributable fraction of postmenopausal breast cancer

The estimated PAFs for the separate and combined risk factors are presented in Table 3. PAFs varied across age categories, as a result of the above-described differences in prevalence rates. Overweight/obesity had the highest PAF of 8.8 % (95 % CI 6.3–11.3) (on average for all age categories). The PAF increased with age, from 7.3 % in ages 50–60, to maximum 10 % in women >70 years. Alcohol consumption had the second highest overall PAF of 6.6 % (95 % CI 5.2–8.0). This PAF decreased with age from 7.4 % in 50–60 years to 3.9 % in >80 years. Physical activity had an average PAF of 5.5 % (95 % CI 4.0–7.0), ranging from 4.9 % in ages 50-60, to 7.8 % in women >80. Smoking had an average PAF of 4.6 % (95 % CI 3.3–6.0), which was highest in younger women (i.e. 5.6 % in ages 50–60), and decreased with age (2.9 % in ages >80). Low-fibre intake had a PAF of 3.2 % (95 % CI 1.6–4.8) for all age categories, which was highest in younger women (i.e. 3.7 %, ages 50–60).

Table 3 Population attributable fraction (PAF) for five lifestyle-related risk factors and postmenopausal breast cancer
Full size table

Combined, these risk factors accounted for an estimated 25.7 % (95 % CI 24.2–27.2) of all 10,367 postmenopausal breast cancer cases in the Netherlands in 2010 [2]. This implies 2,665 excess cases due to these five risk factors (see Table 3).

Discussion

Our results imply that approximately one out of four postmenopausal breast cancer cases in women aged >50 years in 2010 was attributable to lifestyle factors as present at age 40 and older. Overweight/obesity (8.8 %) contributed the most, followed by alcohol consumption (6.6 %), physical inactivity (5.5 %), smoking (4.6 %) and suboptimal dietary fibre intake (3.2 %). These estimates were based on comprehensive and up-to-date literature and matched with detailed prevalence rates of risk factor exposure in the Netherlands.

Estimations of the attribution of these modifiable lifestyle risk factors to postmenopausal breast cancer have not been described for the Netherlands previously. Furthermore, in this research, we replicated the results of other western European countries of population attributable risks of lifestyle-related risk factors for breast cancer.

Strengths of our study include detailed data on prevalence of risk factor exposure, allowing us to use continuous RRs that ensured little loss of information. In addition, we used RRs which were derived from recent meta-analyses [4–7] evaluating multiple studies with risk estimates that were adjusted for several confounders, including lifestyle-related risk factors. Furthermore, Monte Carlo simulations were performed to compute 95 % confidence intervals for the PAF estimates, incorporating imprecision in RRs (defines by the literature derived 95 % confidence intervals of the RR estimates) and prevalence rates (including the most detailed prevalence rates available for levels of exposure, for example, for alcohol we used prevalence rates per each glass/day also for the exposure levels >4 glasses/day).

However, there are also some limitations. We cannot rule out possible residual confounding which could have influenced our PAF estimates. However, since the literature-derived RRs incorporated in the meta-analyses usually are adjusted for most important confounders, it is unlikely that remaining unmeasured confounders influenced the results considerably. Simulation studies show that estimates which are corrected for major confounders are affected minimally after additional correction for more possible confounders [18]. Nevertheless, measuring lifestyle habits in a valid way is difficult due to measurement errors in assessing the confounders.

Prevalence rates were based on self-reported exposure. Misclassification (most likely due to underreporting of exposure) may have led to an underestimation of our PAFs. Also, the prevalence rates were measured in a subsample of people, wherein response rates were high (60 %) but not 100 %. Therefore, also participation bias may have affected the results. Furthermore, we included exposure to risk factors from age 40 on only, while it is also likely that not only short-term, but also life-long exposure to lifestyle-related risk factors, or exposure during a critical period of life (e.g. between menarche and first childbirth) contributes to a higher breast cancer risk [19]. However, there is still much uncertainty around the latency period and which period in life is most influential.

In comparable research, hormone replacement therapy (HRT) is often included as a risk factor. Although RRs of 1.10 to 1.66 have been described for current HRT use [20, 21], we did not include this factor in our analysis. In 2001, the estimated prescription of HRT in women > 40 in the Netherlands was 5.6 % and dropped to 2.4 % in 2004 [22]. Currently, prescriptions are close to zero [23]. As shown by the Million Women study, the increased risk of breast cancer caused by HRT almost disappears after 5 years of cessation [21], meaning that HRT use (past and current) barely influences breast cancer incidence in the Netherlands anymore.

Attributable fractions of modifiable risk factors for all age breast cancer have been estimated for several countries in Europe, reaching up to 25 % in the UK and Germany [24, 25]. However, different sets of risk factors were considered, making results difficult to compare.

Regarding the whole of Europe, Soerjomataram et al. [26], estimated the number of excess cases, i.e. avoidable breast cancer cases, by comparing a countries all-ages incidence rate to the lowest incidence rate in a European country (the baseline incidence rate). For the Netherlands, they estimated around 30 % of all age breast cancer to be avoidable, which was comparable to their estimates for other Western and Northern European countries, but much higher than estimates for Eastern (i.e. Czech Republic, Romania, Lithuania; up to approximately 5 %) and Southern Europe (i.e. Spain, Portugal; up to approximately 15 %). The authors speculate that this higher incidence rate could be caused by over-diagnosis due to extensive screening programmes and higher exposure to reproduction-linked risk factors. Even though these estimates cannot be directly compared to our PAF numbers, as they used a different methodology, it gives us an idea about the Dutch situation in proportion to the rest of Europe with regard to avoidable cancer cases. And although their number refers to all age breast cancer, it will largely refer to postmenopausal breast cancer as most cases occur after age 50.

We included five lifestyle-related risk factors for postmenopausal breast cancer for which a large body of evidence is available and that occur with substantial prevalence rates in middle-aged women in the Netherlands.

Fibre intake and smoking are not, or seldom, considered when estimating PAFs for breast cancer. Since there is emerging strong evidence that these factors increase breast cancer risk, we included these factors and recommend including them in future studies. A recent Canadian study that included smoking as a risk factor reported a PAF of 3–4 % based on prevalence rates of risk facture exposure in the years 1994–2006 [27].

Overweight and obesity, alcohol consumption and physical inactivity are often included in other studies. Considering these three factors, we estimate a combined PAF of around 20 %. Similar results were found for neighbouring countries. Parkin et al. estimated that 17 % of all breast cancer cases, irrespective of age, in 2011 were attributable to these factors in the UK [24]. Barnes et al. estimated a PAF of 21 % for Germany in 2010 [25]. However, we observed some differences for the separate risk factors. PAF estimates for a BMI > 25 kg/m2 vary from 2.5 % in Germany [25], to 5.6 % in France [28] and 8.7 % in the UK [29], the latter being comparable to our estimate (8.8 %). The attribution of overweight/obesity has previously been computed for the Netherlands. Bergstrom et al. estimated a PAF of 6.3 % based on a 42 % exposure rate in the years 1993–1996, and similar RRs as we used [30]. Since the prevalence of overweight/obesity is still increasing in the Western world, the PAF is doing so concordantly.

For alcohol consumption, similar PAFs, ranging from 6.4 to 9.4 %, are described in adjacent countries [25, 28, 31]. However, PAFs for alcohol consumption differ in other developed countries as the US and Australia, where PAFs reach up to maximum 3 % [27, 32, 33]. Consumption of alcohol by European women is rather high; 75 % of Dutch women >40 years drink on a regular basis.

For physical inactivity, mainly higher PAF estimates than ours (5.5 %) were reported in Europe, of around 10–14 % [25, 28, 34], except for the UK (3.4 %) [35]. Numbers in the U.S. even rise up to 16 % [36]. Differences in prevalence rates largely explain this variation, i.e. in the U.S., 78 % of women were considered physically inactive, versus 56 % in the Netherlands. Another explanation why estimates vary greatly could lie in the fact that PAFs are sensitive to differences in risk category definitions with their accompanying RR [37]. Due to the great difficulty of measuring activity levels and determining proper risk categories, other definitions for physical inactivity and RRs are used in literature. Also, we did not incorporate intensity of activities.

In the Netherlands, incidence of breast cancer is among the highest worldwide. We estimated that approximately 25 % of postmenopausal breast cancer is associated with lifestyle behaviour at age 40 years. Reproductive factors and hormones will be associated with another proportion of cases, but these are less modifiable. Still, there is a substantial proportion of cancers that seem to occur at random [38]. However, we should also not exclude the possibility of yet undetected exposures, such as naturally occurring estrogens in the environment; or other chemicals with estrogenic function.

Often, success rates of lifestyle modifying programmes are limited. Therefore, for the Netherlands, a 25.7 % reduction in postmenopausal breast cancer incidence would be the maximum to be achieved, rather than realistic. However, these estimates may help motivating women as well may they inform policy makers about which risk factors should be addressed first.

To conclude, our results imply that one in four postmenopausal breast cancer cases in the Netherlands in 2010 is attributable to five strongly associated lifestyle-related risk factors. These risk factors are excess body weight, an inactive lifestyle, alcohol consumption, smoking and low dietary fibre intake.