Ethnic and socioeconomic trends in breast cancer incidence in New Zealand

Data

The datasets were created through linking records from the New Zealand Cancer Registry (a population based cancer register which collects information on all malignant tumours except basal and squamous cell skin cancers, with mandatory notification since 1994) and records from a 5 yearly census of population and dwellings. Five closed cohorts were created of the New Zealand usual resident population (all ages) on census night 1981, 1986, 1991, 1996, 2001, followed up for incident cancer(s) until the subsequent census or in the case of the 2001 cohort, until 31 December 2004 (the most recent data available at the time of the study's record linkage). Cohorts were created using probabilistic record linkage software (QualityStage) to anonymously link census and cancer register records within a geographic area (meshblock or census area unit) on sex, date of birth, ethnicity, and country of birth. Details on the linkage methods, process and outputs are detailed elsewhere [27, 28].

Briefly, 73.2% (1981-86) to 81.7% (2001-04) of eligible cancer registry records were linked to a census record, with 95.2% (1981-86) to 96.9% (2001-04) of these linked census-cancer records estimated to be true links. To adjust for underestimation of rates using the linked datasets, and to correct for any linkage bias whereby the percentage of eligible cancer records linked varied by socio-demographic characteristics, we calculated weights for strata of age, sex, ethnicity, region, small area deprivation, and time since census. All analyses presented in this report use these weights.

Approval was granted for this project under the Statistics New Zealand Data Integration Policy [29], and the Wellington Ethics Committee granted ethics approval for CancerTrends (Ref 04/10/093).

Variables

Analyses

For the analyses presented here, data were restricted to adults aged 25 and over who were in their usual residence on census night. All analyses were conducted in SAS v8.2 or v9.

Person time was censored after development of the first breast cancer. Incidence rates, rate ratios and rate differences (and 95% confidence intervals) were calculated after direct standardisation of the cohorts to the age structure of the WHO World Standard population [36], and to the ethnic structure of the 2001 New Zealand population for trends by income (because ethnicity may act as a confounder on the income - breast cancer relationship). All analyses were performed on complete data only, with no imputation of missing data.

Statistical tests of trend were conducted for rates, and rate differences, and of the log transformed rate ratios. All measures were also calculated for all five cohorts pooled [28].

Ethnic trends

Figure 1 and Table 2 show age standardised breast cancer rates for Māori, Pacific, Asian and European/other women in each time period. Rates for Māori women increased seventy percent in the time period studied, compared to about 50% for European/other (p for trend both < 0.01). Incidence rates in Pacific and Asian women increased by 25% and 80% respectively.

Both the relative and absolute differences between Māori and European/other women increased between 1981 and 2004 (Table 2). Māori women had 7% higher rates than European/other women in 1981-86, and 24% higher rates in 2001-4 (SRR 1.07 in 1981-86 and 1.24 in 2001-04), and the absolute difference increased from 9 to 40 per 100,000 women. By contrast Asian and Pacific women had consistently lower breast cancer rates than European/Other women and the differences between Pacific and Asian and European/Other women did not change markedly over time.

Socioeconomic trends

Figure 2 shows age and ethnicity standardised breast cancer incidence rates for low, medium and high income women for each cohort. A gradient with income is clearly seen in 1986-91, 1996-01 and 2001-04, with high income women having consistently higher rates, medium income women intermediate rates, and low income women the lowest rates. Over time, rates increase by about 50% in all income groups.

Ethnic differences and trends

Breast cancer rates in European women in New Zealand, which rose from 114 to 170/100,000 women per year between 1981-86 and 2001-04 (for women 25+), are comparable to breast cancer rates in European women in other developed countries. For example, in the US non-Hispanic white women have an annual age standardised rate of 189/100,000 for women 25+, (based on 14 SEER registries, 1998-2002; using the same WHO World Standard) [2]. In Scotland women 25+ have an annual age standardised rate of 152/100,000 (1998-2002; using WHO World Standard) [2].

The incidence pattern for breast cancer amongst Pacific women is consistent with our knowledge of breast cancer risk factor distribution in this group. Pacific women are more likely than European women to have children, to commence child bearing earlier and to have more children [37]. Lower alcohol consumption [38] will also contribute to the lower risk of breast cancer amongst Pacific women. The higher rates seen in younger Pacific women (Table 3) may be related to the cancer promoting effect of pregnancy. A similar pattern is seen in the US, with young black women having higher rates than young white women (and older black women having lower rates than older white women), which is thought to relate to earlier childbearing amongst black women [15]. Age at first birth data in New Zealand is only collected for "nuptial" births and is not reported by ethnic group. However based on age specific fertility rates [39] it is apparent that although Pacific women have similar early fertility rates to Māori women, both groups have age specific fertility rates more than double those of European women up to age 25, and so differences in early parity may explain the differences in breast cancer incidence seen between young Pacific and European women.

Rates of breast cancer amongst Asian women were consistently lower than those of European and Māori women. This "Asian" group is diverse both in terms of country of origin and time in New Zealand, and so is likely to be diverse in terms of reproductive patterns and other risk factors for breast cancer, making the results for Asian women hard to interpret. However despite small numbers, the relatively low rates seen are consistent with the low rates seen in Asian countries. For example IARC's Cancer Incidence in Five Continents IX reports age standardised rates per 100,000 for those aged 25+ of 52 in India, 67 in China, 45 in Korea, and 56 in Malaysia for the period 1998-2002 (based on available cancer registry data, standardised to WHO World Standard) [2]. The ASR of 126 per 100,000 (95% CI 110-142) found amongst New Zealand Asian women in 2001 - 2004 is considerably higher than rates in Asian countries, which is likely to reflect changes in risk factor distribution following migration.

The high incidence of breast cancer amongst Māori women is however not easily explained, as Māori women have a more favourable (known) risk factor profile than European women. Māori women tend to start having children younger and have more children compared to European women [37]. Māori women have lower rates of HT use [40], and similar rates of oral contraceptive use, alcohol consumption and physical activity compared to European women [38, 41]. Screening is not contributing to the higher incidence seen, as Māori women have lower uptake of screening [42]. All of this risk factor information suggests that European/Other women should have higher rates than Māori, yet it is Māori who consistently do so. One partial explanation for the observed higher rates for Māori is post-menopausal obesity among Māori, however it is unlikely to be a full explanation for higher post-menopausal rates.

The timing of menarche has previously been postulated as a possible reason for the difference in breast cancer rates between Māori and European women [43]. However the evidence to support differences in the timing of menarche and menopause between Māori and non-Māori women is inconclusive and any differences in timing are small [24, 40, 44]. The prevalence of other postulated breast cancer risk factors such as shift work [45], exposure to particular environmental pollutants [46], or Vitamin D levels [47] may differ between ethnic groups but is unlikely to explain the large differences identified here.

Several recent studies have suggested that cigarette smoking initiation shortly after menarche and before first full term pregnancy may increase the risk of breast cancer [48–50]. Early initiation (mean age 11 in 2007) [51] and very high rates of tobacco use amongst young Maori women [52] may be important for breast cancer risk, although the effect of early initiation of smoking on breast cancer risk does not appear to be large enough to explain the patterns seen.

There is some evidence that there are differences in tumour biology between ethnic groups. For example, there are differences in tumour grade and estrogen and progesterone receptor (ER and PR) positivity between black, Hispanic and white women in the US [53]. In New Zealand, the potential of genetic differences between ethnic groups has not been thoroughly explored and findings in relation to receptor positivity status have been inconsistent [54, 55]. Regardless, while differences in tumour biology may be important with respect to survival differences between women with breast cancer in different ethnic groups [56], it is unclear whether they are important for explaining differences in incidence. Moreover, such genetic differences will not account for rapid changes in rates over time in any group, and could only account for diverging ethnic group trends over time in the presence of some environmental exposure that is also changing over time and interacting with a varying genetic predisposition by ethnicity.

Indigenous Hawaiian women also have surprisingly high rates of breast cancer given their distribution of risk factors, and it has been postulated that genetic polymorphisms in the sex steroid and gonadotropin metabolism pathways, causing high endogenous hormone levels, are responsible for their high rates [58]. Dietary factors and high insulin-like growth factor levels are also mentioned as possible factors [57]. Indigenous Hawaiians also have higher breast density than European Hawaiians [58]. It is possible that there may be similarities between the drivers of breast cancer rates in Māori and indigenous Hawaiian women, but more research is needed to elucidate potential mechanisms in both groups. However it is not obvious why Māori and Hawaiian women would have such different breast cancer rates from other groups of Polynesian women from the Pacific.

The above discussion has been orientated mainly at trying to explain differences between ethnic groups' rates on average, but our study also finds a more rapidly increasing Māori breast cancer rate than European/Other. (The Asian rate also increased rapidly, but was prone to greater statistical imprecision - hence we focus on just Māori and European/Other trends.) This divergence over time may be due to changes in the factors which are driving the underlying differences between Māori and European/Other rates (although as noted above these factors are currently unknown). Alternatively, this change may be due to changes in the distribution of established risk factors amongst the Māori population, overlying the already higher risk for Māori. For example there is some evidence that Māori fertility rates are dropping more quickly than among European women, with a convergence on similar fertility patterns [37, 59]. Rapid changes in postmenopausal obesity rates amongst Māori women may also be an important factor in this change.

Finally, it is important to note that breast cancer mortality differences between Māori and non-Māori remain more significant [26] and are more likely to be amenable to intervention than incidence differences, and so should remain a central concern of researchers and policy makers.

Socioeconomic trends

The pattern of higher breast cancer rates amongst women of higher socioeconomic status seen here is consistent with that found internationally [60–62], and is likely to be primarily due to differences in reproductive behaviour. Socioeconomic differences in reproductive behaviour, including differences in parity, age at first birth, childlessness, and the duration of breast feeding, are seen in almost all settings internationally [12]. In New Zealand there are significant fertility differentials by level of education, with women with the highest level of education having the fewest children, and this difference is greatest in women born more recently indicating an increasing socioeconomic differentiation in reproductive behaviour [63]. It is possible that over time this will result in widening of differentials in breast cancer rates between socioeconomic groups in New Zealand. Screening uptake and HT use have also been socioeconomically patterned, with higher uptake amongst better off women [64], and this will contribute to the difference in incidence seen. HT use in New Zealand declined markedly following the publication of the results of the Women's Health Initiative trial in 2002 [65], and this is likely to have reduced the difference in the last time period by socioeconomic status.

Strengths and Limitations

This study is a series of five short-duration cohort studies of the entire New Zealand population, followed up for breast cancer over a 24 year period. By using data from the entire New Zealand (census night) population as well as data from New Zealand's population based cancer registry, it was possible to overcome numerator/denominator bias and misclassification of ethnicity and socioeconomic group. Self identified ethnic affiliation from census forms was used to define the numerator and denominator populations, and therefore produce more reliable estimates of differences between ethnic groups. Individual measures of socioeconomic position were also able to be used rather than relying on area-based measures. Considering trends over 24 years also provides a good basis for understanding evolving differences between population groups.

However the study is not without limitations. Statistical imprecision becomes a problem when stratifying by ethnicity or socioeconomic position. This is particular problem for the Asian and Pacific groups where numbers are small. It was not possible to link all cancer records back to the census, and so weighting for linkage bias was undertaken; we are confident that this will have eliminated most bias due to misclassification of outcome. The Cancer Registry Act, which came into effect in 1995, required mandatory registration of all cancers, and there may have been a small artefactual increase in breast cancer registrations following its introduction, but this is not thought to have a had a major effect on breast cancer rates and probably impacted ethnic groups and income groups in a similar manner.

Household income is presented as an indicator of individual socioeconomic status. While education measures socioeconomic status closer to the time of reproductive choices which are important for breast cancer risk, more proximal exposures such as post menopausal obesity, screening, and HT use are also important for breast cancer risk and so both measures are relevant. Education also lacks discriminatory power in older women in the earlier cohorts, as the majority of these women fall into the category of no qualifications. As similar results were found for both education and household income a decision was made to only present the results for household income in this paper.

Finally, the lack of good information about risk factor distribution in the population, particularly in the relevant time period, makes it difficult to fully interpret the differences found between ethnic and socioeconomic groups; rather, these breast cancer rates over time might provide clues about unmeasured risk factors back in time.