We describe temporal trends in breast cancer incidence by molecular subtypes in Scotland because public health prevention programmes, diagnostic and therapeutic services are shaped by differences in tumour biology.
Population-based cancer registry data on 72,217 women diagnosed with incident primary breast cancer from 1997 to 2016 were analysed. Age-standardised rates (ASR) and age-specific incidence were estimated by tumour subtype after imputing the 8% of missing oestrogen receptor (ER) status. Joinpoint regression and age–period–cohort models were used to assess whether significant differences were observed in incidence trends by ER status.
Overall, ER-positive tumour incidence increased by 0.4%/year (95% confidence interval (CI): −0.1, 1.0). Among routinely screened women aged 50–69 years, we observed an increase in ASR from 1997 to 2011 (1.6%/year, 95% CI: 1.2–2.1). ER-negative tumour incidence decreased among all ages by 2.5%/year (95% CI: −3.9 to −1.1%) over the study period. Compared with the 1941–1959 birth cohort, women born in 1912–1940 had lower incidence rate ratios (IRR) for ER+ tumours and women born in 1960–1986 had lower IRR for ER− tumours.
Future incidence and survival reporting should be monitored by molecular subtypes to inform clinical planning and cancer control programmes.
Breast cancer incidence is rising and it is the most common cancer among women worldwide.
There are relatively few population cancer registries that collect ER, PR and HER2 data, the key distinguishing markers for molecular subtypes of breast cancer. Recent analyses support divergent incidence trends by ER status in the United States, Denmark and Ireland, with ER+ breast cancer incidence increasing and ER− breast cancer incidence decreasing.
Within Scotland’s renowned, high-quality routine electronic health records, the Scottish cancer registry is an excellent resource to investigate temporal trends in cancer incidence. Data collection began for ER in 1997 and PR and HER2 in 2009, and so provides data almost a decade earlier than other UK national registries. While monitoring of breast cancer incidence in the United Kingdom is standard,
Here we report on breast cancer incidence trends in Scotland by ER and ER/HER2 combinations using several statistical methods: (1) age-standardised and age-specific incidence rates, which are typically used to report cancer statistics,
All primary invasive breast cancers (defined on the basis of the International Classification of Diseases, 10th revision code of C50) diagnosed in women aged 20+ years, between 1997 and 2016, were ascertained from the Scottish cancer registry held by Information Services Division (ISD) of NHS National Services Scotland. The Scottish cancer registry achieves 98% breast cancer case ascertainment and is over 99% complete.
Additional demographic and tumour data obtained were age at diagnosis, NHS Scotland regions (North, South East and West), tumour grade (grade I—well differentiated to III—poorly differentiated), tumour size (less than 10 mm, 10–20 mm and more than 20 mm), nodal involvement (yes or no), screen-detected tumour (yes or no) and the status of molecular markers ER, PR and HER2 (positive, negative or unknown). ER and PR status are measured using immunohistochemistry (IHC), and HER2 status was assessed using a combination of IHC with fluorescent in situ hybridisation for equivocal (2+) cases. Previous studies have noted that assessment of ER status reliability is high with an error rate below 5%.
Missing ER and ER/HER2 status were imputed conditioned on age and year of diagnosis, with the assumption that data were missing at random, using a validated method.
Joinpoint regression models were used to describe breast cancer incidence rates overall, by ER status and ER/HER2 combinations for all women in the cohort and for three age groups (20–49, 50–69 and 70+ years). Joinpoint models describe if changes in incidence trends occur and identify the time points at which a change is observed (referred to as joinpoints). The permutation test method, as described by Kim et al.,
APC models were fitted for age-standardised incidence of ER+ and ER− tumours. The APC model provides a unique set of best-fitting log10 incidence rates obtained by maximum likelihood estimators for period, age and cohort, which have been shown to provide similar rates to ASR, but allow investigation of differences by birth cohorts—with the middle cohort as referent—which are not investigated in ASR or joinpoint regression analysis. As a consequence of small numbers in some strata, we restricted these models to women aged 30–85 years and used 28 2-year age groups (from 30–31 to 84–85) and 10 2-year periods (from 1997–1998 to 2015–2016) of calendar year of diagnosis, which covered birth cohorts from 1912 to 1986. The net drift, similar to the EAPC and AAPC estimates, is reported with 95% confidence intervals (CI). Local drifts were also estimated and describe the annual percentage change for each age-specific rate over time.
Between 1997 and 2016, 72,217 women of 20 years of age or older were diagnosed with at least one invasive breast cancer in Scotland (Table Descriptive characteristics by ER status for all women with an invasive breast cancer diagnosed between 1997 and 2016 in Scotland. Characteristics ER− ER+ ER unknown % % % 11,726 [16] 55,144 [76] 5347 [8] Age at diagnosis <50 years 3196 (27) 10,550 (19) 695 (13) 50–69 years 5668 (48) 28,441 (52) 1580 (30) 70 years or older 2862 (24) 16,153 (29) 3072 (57) Grade I—well differentiated 195 (2) 8288 (15) 232 (4) II—moderately differentiated 1714 (15) 25,734 (47) 602 (11) III—-poorly differentiated 8308 (71) 14,586 (26) 642 (12) Unknown 1509 (13) 6536 (12) 3871 (72) Nodal status Uninvolved/negative 6194 (53) 29,400 (53) 869 (16) Involved/positive 4110 (35) 17,369 (31) 415 (8) Unknown 1422 (12) 8375 (15) 4063 (76) Tumour size Less than 10 mm 1017 (9) 6470 (12) 202 (4) 10–20 mm 3428 (29) 20,449 (37) 478 (9) More than 20 mm 4960 (42) 18,168 (33) 512 (10) Unknown 2321 (20) 10,057 (18) 4155 (78) PR statusa Negative 3803 (79) 3036 (12) <10 (<1) Positive 226 (5) 15,869 (62) <10 (<1) Unknown 764 (16) 6489 (26) 901 (99) HER2 statusa Negative 2761 (66) 18,709 (84) 36 (5) Positive 1210 (29) 2553 (11) 10 (1) Unknown 184 (4) 1129 (5) 725 (94) Brackets [] indicate row percentages and parentheses () indicate column percentages for that category. aDenotes markers that were recorded from 2009 to 2016, and the number of cases for those years = 31,099. Differences by known ER status for all characteristics were significantly different with
Tumour characteristics differed by ER status, with ER– tumours having characteristics associated with more advanced/aggressive disease. ER− tumours had higher grade, were larger and more likely to have positive lymph node status. The patterns of other molecular markers also differed by ER status, with ER– tumours more likely to be PR− and HER2+ than ER+ tumours. In contrast, ER+ tumours were more likely to be PR+ and HER2− than ER− tumours.
The combinations of ER/HER2 status after imputing for missing ER and HER2 status are shown in Fig.
Age-standardised incidence of ER+ tumours increased from 98 per 100,000 women in 1997 to 113 per 100,000 women in 2016 (Table Joinpoint regression analysis stratified by age groups and ER status from 1997 to 2016. ER status Age groups Rate in 1997 per 100,000 women Rate in 2016 per 100,000 women Change in rate from 1997 to 2016 per 100,000 women (%) Average annual percentage change (95% CI) Years before joinpoint EAPC (95% CI) for the period before joinpoint Years after joinpoint EAPC (95% CI) for the period after joinpoint Positive 20–49 41.9 52.1 10.2 (20%) 1.1% (0.7, 1.5) 10,550 11,083 50–69 192.3 237.4 45.1 (19%) 0.7% (0.2, 1.3) 28,441 29,758 1997–2011 1.6% (1.2, 2.1) 2011–2016 −1.8 (−3.7, 0.1) 70+ 235.9 234.5 −1.4 (−0.6%) 0.1% (−0.3, 0.5) 16,153 18,763 All ages 97.7 112.8 15.1 (13%) 0.4% (−0.1, 1.0) 55,144 59,604 1997–2012 1.2% (0.8, 1.5) 2012–2016 −2.2 (−4.7, 0.4) Negative 20–49 23.8 15.2 −8.6 (−36%) –2.2% (−3.9, −0.6) 3196 3358 1997–2001 −10.0% (−17.0, −3.0) 2001–2016 0% (−1.1, 1.2) 50–69 64.1 45.5 −18.6 (−29%) −1.6% (−2.5, −0.8) 5668 5931 70+ 71.8 41.2 −30.6 (-–43%) −2.4% (−4.2, −0.7) 2862 3324 1997–2003 −7% (−11.0, −2.0) 2003–2016 −0.3% (−1.9, 1.5) All ages 35.5 23.1 −12.4 (−35%) −2.5% (−3.9, −1.1) 11,726 12,613 1997–2000 −11% (−19.0, −3.0) 2000–2016 −0.7% (−1.5, 0) Joinpoint regression was performed using the estimated counts corrected for missing ER status, and analysis corrects for multiple testing using Bonferroni correction (see ‘Methods' section).
Women 50–69 years of age had the highest increases in ER+ incidence at a similar period as noted overall (Table ER-positive (
The results from APC models were consistent with those observed from joinpoint regression, with net drifts suggesting increases in the overall incidence of ER+ tumours by 0.8% per year (95% CI: 0.6–1.0%/year) from 1997 to 2016, and ER− tumour incidence decreasing by −1.4% (95% CI: −1.8 to −1.1%/year). After adjusting for period and cohort effects, local drifts showed that the highest increase in incidence of ER+ tumours was observed in women around 70 years of age (2% per year, 95% CI: 1.6–2.4%) (Supplementary Fig.
Compared with the women born in 1949, ER+ tumour incidence was higher among more recent birth cohorts. In contrast, ER− incidence was lower for more recent birth cohorts compared with the cohort born in 1949. CRRs compared with women born in 1949 ranged from 0.7 for women born in 1913 to 1.8 for women born in 1985 for ER+ tumours, and from 1.5 for women born in 1913 to 0.5 for women born in 1985 for ER− tumours (Fig. CRR describes the incidence rates for each birth cohort relative to the 1949 birth cohort.
This study demonstrates that, in Scotland, temporal trends of breast cancer incidence were distinct by molecular subtypes, with increases for ER+ and decreases for ER− tumours between 1997 and 2016. With respect to ER+ tumours, their incidence increased for all ages for the study period, but particularly among women of screening ages 50–69 years, with the largest increases occurring from around 1997 to 2011 followed by modest declines. In contrast, the incidence of ER− cancers decreased among all ages till the early 2000s. Finally, we noted cohort effects such that, in comparison with women born around 1950, women of older generations (those born in the 1910s–1940s) had a lower risk of ER+ tumours, whereas there was no significant evidence for cohort effects for ER− tumours. Further analysis of the incidence trends by subtype (as defined by ER/HER2 combinations) generally showed similar results to those observed by ER status only. ER+/HER2− (surrogate for luminal A) tumours followed the same pattern as all ER+ tumours. However, our findings suggest a significant increase in the rarer and more aggressive ER−/HER2− breast cancers among women 20–49 years of age, similar to recent increases noted in the United States that need careful future monitoring.
Consistent with reports from the United States, Denmark and Ireland,
Mammographic screening is likely to be an important contributing factor to the increased incidence of ER+ tumours we observed from 1997 to 2011. In Scotland, the breast screening programme was established in 1988 with full national coverage attained in 1991.
Yen and colleagues aimed to determine risk factors and molecular tumour markers that might be associated with screen-detected tumours using data on 1924 screen-detected and 1001 interval-detected cancer cases diagnosed in Sweden.
The strengths of our study are the high quality of the longitudinal data collected within the Scottish cancer registry, the first one in the United Kingdom that routinely started recording molecular marker data (ER status from 1997 and PR and HER2 status from 2009). Marker data can be used to monitor and describe incidence trends in the future and for other types of cancer that display heterogeneity. Further, monitoring breast cancer incidence by molecular subtypes can help the NHS allocate resources for treatment and prevention, and lead to the identification of high-risk groups of women for which to implement future prevention programmes and treatments.
A potential limitation of our study is imputation of ER status for 8% of the population and the assumptions used, which were that ER/HER2 data have the same chance of being missing among each cohort of patients by year and age at diagnosis. For this assumption to be wrong, there would have to be a confounder associated with ER status that would influence whether ER status was tested and recorded. This scenario seems unlikely in Scotland’s health service where guidelines are used to inform investigation and treatment. Missingness is more likely to reflect administrative omissions, and geographic uptake in reporting ER status. This assumption has been used in US, Denmark and Irish data.
Another limitation of our study is the absence of individual-level risk factor data, including participation in breast screening programmes in prior years to define interval breast cancers and stage data. However, in future studies, it should be possible to identify some key factors using linked data including detailed cohort data. The United Kingdom is renowned for its high-quality, longitudinal data and the ability to perform linkage studies using a unique identifier. Hence, we envision future analysis using the cancer registry linked to other datasets, including community prescription drug records, mammography imaging, maternity and hospital records to provide more detailed information on the role and patterns of key risk factors in breast cancer incidence trends. Another limitation of the study is the lack of mRNA expression assays for the classification of the molecular subtypes of breast cancer. In our study, markers measured by IHC are used as surrogates for the molecular subtypes, which are reasonably good proxies, but mRNA profiling data would be considered a gold standard for intrinsic-subtype classification.
In conclusion, incidence trends of breast cancer in Scotland differ by ER status, and are consistent with trends observed in other countries. It will be important to monitor whether ER+ tumour incidence stabilises or reduces over time. Additional data are needed to establish whether incidence of HER2+ tumours, which are ER−, remains low since their treatment involves monoclonal antibodies, such as trastuzumab and pertuzumab,
We thank NHS Scotland and Information Services Division for collection of data used for this analysis. We specifically would like to acknowledge Andrew Deas, Suhail Iqbal, Rita Nogueira and Ross Murdoch of NHS National Services Scotland. Pathologists Jeremy Thomas, Catherine Dhaliwal NHS Lothian and Mike Dixon for helpful discussions of these analyses. An earlier version of this work was deposited on a preprint server medRxiv at
Conception and design of the study: J.D.F., S.W., I.M.E. and S.B. Interpretation of data: all authors. Drafting of the paper: I.M.E., J.D.F. and S.W. Revised work and provided important intellectual content: all authors. Final approval of the paper: all authors.
Approval from the Public Benefit and Privacy Panel for Health and Social Care is a requirement for data access. Our project was approved by PBPP reference number 1718-0057.
Not applicable.
The data used in this study can be accessed through application to electronic Data Research and Innovation Service (eDRIS), a part of the Information Services Division of NHS Scotland.
S.M.B. holds shares in GlaxoSmithKline. Other authors declare no competing interests.
This project was funded by Wellcome Trust grant 207800/Z/17/Z.
Supplemental material_revision