Long-Term Prognostication for 20 114 Women With Small and Node-Negative Breast Cancer (T1abN0)

Abstract Background Although small, node-negative breast cancer (ie, T1abN0) constitutes 20% of all newly diagnosed breast cancers, data on prognosis and prognostic factors are limited. Methods We conducted a population-based cohort study including 20 114 Swedish women treated for T1abN0 breast cancer from 1977 onward. Patient and tumor data were collected from Swedish breast cancer registries. Cohort subjects were followed through linkage to the Cause of Death Register. We calculated the cumulative incidence of breast cancer–specific and overall death and used Cox regression to estimate hazard ratios (HRs) and 95% confidence intervals (CIs). Results During a median follow-up of 9.1 years (range = 0-38), 915 women died of breast cancer and 5416 of any cause. The 10-, 20-, and 30-year cumulative incidences of breast cancer death were 3.4% (95% CI = 3.1% to 3.7%), 7.6% (95% CI = 7.1% to 8.2%), and 10.5% (95% CI = 9.6% to 11.4%), respectively. The multivariable hazard ratios and 95% confidence intervals of breast cancer death were 0.92 (95% CI = 0.88 to 0.97) for each additional calendar year of diagnosis, 4.38 (95% CI = 2.79 to 6.87) for grade 3 vs grade 1 tumors, 0.43 (95% CI = 0.31 to 0.62) for progesterone receptor–positive vs progesterone receptor–negative disease, and 2.01 (95% CI = 0.99 to 4.07) for HER2-positive vs HER2-negative disease. Women with grade 3 vs grade 1 tumors had a 56% increased risk of death from any cause (HR = 1.56, 95% CI = 1.30 to 1.88). Conclusions The risk of breast cancer death in T1abN0 disease continues to increase steadily beyond 10 years after diagnosis, has improved over time, and varies substantially by tumor characteristics.

Swedish women surgically treated for T1abN0 breast cancer from 1977 onward, followed for breast cancer death, all-cause mortality, and metachronous breast cancer.

Study Design and Data Sources
This is a population-based cohort study. Study participants were identified in Sweden's 6 regional breast cancer registries and the National Breast Cancer Registry and followed for outcomes through linkage to the Cancer Registry, the Total Population Register, and the Cause of Death Register. Linkage was conducted using the national registration number, a unique identifier assigned to all Swedish citizens.
Sweden is divided into 6 healthcare regions [residents in millions (12)]: North (0.9), Uppsala-€ Orebro (2.1), Stockholm-Gotland (2.4), West (1.9), South-East (1.1), and South (1.9). Before 2000, each region had a collaborative breast cancer group developing regional treatment guidelines. Since 2000, there are national treatment guidelines with regional adjustments. In 1977-1992, each regional group established a regional registry compiling information on patient and tumor characteristics and planned primary treatment of all newly diagnosed breast cancer patients; in the North region, only patients aged younger than 75 years were included in the registry. In 2007, the regional registries were discontinued because the National Breast Cancer Registry was established instead, which includes more than 95% of all newly diagnosed breast cancers patients in Sweden (13).
The Cancer Registry includes all newly diagnosed cancers in Sweden with more than 96% completeness (14). The Total Population Register includes virtually 100% of deaths and 91% of emigrations (15). The Cause of Death Register contains underlying and contributory causes of death with 96% completeness and with a disagreement of 6.9% for breast cancer as the underlying cause of death (16,17).
The study was approved by the Stockholm Regional Ethics Committee (2014/365-31/2), with jurisdiction for all participating sites. The study protocol has been registered at www.clinicaltrials.gov (NCT03390608).

Study Cohort
A flow diagram describing the assembly of the study cohort is available in Figure 1. Data were retrieved on all women (n ¼ 33 908) in the 6 regional breast cancer registries or the National Breast Cancer Registry who were surgically treated for invasive breast cancer with pathological tumor size 10 mm or smaller, any nodal status (N0, N1, NX), and no metastatic spread (M0). We excluded 1299 women for whom we could not verify a diagnosis of invasive breast cancer (International Classification of Diseases [ICD]-7: 170; ICD-8-9: 174; ICD-10: C50) in the Cancer Registry within an index period of 3 months prior to and 3 months after the date of diagnosis, 1481 women with a breast cancer diagnosis in the Cancer Registry prior to the index period, and 63 women with conflicting data and/or duplicates. We furthermore excluded women receiving neoadjuvant treatment (n ¼ 609), tumor size missing or 0 mm (n ¼ 217), bilateral or multiple breast tumors (n ¼ 3314), treated without surgery (n ¼ 78), unknown nodal status (NX) (n ¼ 4003), or positive nodal status (N1) (n ¼ 2730). The final study cohort included 20 114 women with T1abN0 tumors.
The prespecified primary outcome was breast cancer death (BCD), defined as breast cancer (ICD-7: 170; ICD-8-9: 174; ICD-10: C50) listed as the underlying cause of death in the Cause of Death Register (data available to December 31, 2014). Secondary outcomes included death from any cause (data available to July 30, 2016) and metachronous breast cancer (data available to December 31, 2014), defined as ipsilateral or contralateral breast cancer registered in the Cancer Registry at any date after the index period.

Statistical Analysis
Cohort subjects were followed from the date of diagnosis to the date of BCD, emigration, death from other causes, or end of follow-up (December 31, 2014), whichever occurred first. The cumulative incidence of BCD at different follow-up times was calculated using a competing-risks extension of the Kaplan-Meier estimator (18), with death from other causes considered a competing event. In analyses where any cause of death was considered the outcome, follow-up ended July 30, 2016. Of the 20 114 women in the study cohort, 2 had been diagnosed with breast cancer in January 2015. These 2 women were only included in the survival analysis where all cause death was considered the outcome. Cox proportional hazard regression was used to calculate hazard ratios (HRs) and 95% confidence intervals (CIs). All survival analyses were restricted to women with nonmissing data on relevant exposures and covariates (ie, no imputation was conducted).
"Simple" and "full" regression models were constructed. The simple models were adjusted for year of diagnosis, age at diagnosis, region, and whether the patient was included in the regional breast cancer registries or the National Breast Cancer Registry. The full models were further adjusted for tumor size, tumor grade, and ER status. We created a proxy for the intrinsic subgroups based on tumor grade, ER status, PR status, and HER2 status ( We conducted 2 prespecified sensitivity analyses: one where women diagnosed with metachronous breast cancer were censored at the time of diagnosis of the metachronous tumor (sensitivity analysis 1) and one where women with any prior cancer at study entry, except nonmelanoma skin cancer and cancer in situ of the cervix, were excluded (sensitivity analysis 2). The information available in the regional breast cancer registries, including the start date of the registry and the proportion of missing data, differed between regions and over time (Supplementary Tables 1-3, available online), with generally fewer variables and larger proportions of missing data in earlier years. The information available in the National Breast Cancer Registry, in contrast, is homogenous across regions and over time and has a low proportion of missing data (Supplementary  Tables 1-3, available online). We therefore conducted 2 additional sensitivity analyses that had not been prespecified in the study protocol; one restricted to women diagnosed from January 1, 2000, onward (sensitivity analysis 3) and one restricted to women in the National Breast Cancer Registry subcohort (sensitivity analysis 4). We also conducted a sensitivity analysis using the Fine and Gray subhazard method to account for competing events (sensitivity analysis 5).
Finally, we conducted the survival analysis using a landmark at 10 years, starting the follow-up at 10 years among women having survived and not being censored at that time. All analyses were conducted using Stata 15.1.

Discussion
The 10-year cumulative incidence of BCD was 3.4% in our cohort, in agreement with prior reports including a 4.0% estimate from Surveillance, Epidemiology, and End Results (19,20). For T1abN0 tumors, data on prognosis beyond 10 years are limited. A few publications have reported 20-year disease-free survival estimates of 70%-88% (20)(21)(22). In clinical decision-making, data on long-term prognosis in women not receiving adjuvant treatment are especially important. A Finnish study of 80 women diagnosed with T1abN0 tumors between 1945 and 1976 not receiving adjuvant systemic treatment reported 10-and 20-year BCD rates of 6% and 8%, respectively (4). A recently published Swedish study reported a 15-year breast cancer-specific survival rate of 93.7% among 1543 patients diagnosed between 1997 and 2002 with T1abN0 grade 1-2 tumors of whom 12% received adjuvant endocrine therapy and 0% adjuvant chemotherapy (5). Based on our data, the 10-year absolute risk of BCD among women not receiving adjuvant systemic treatment is around 6%; the observed 10-year cumulative incidence of BCD among those diagnosed between 1977 and 1989, a period during which adjuvant systemic treatment was recommended only in selected cases, was 5.5%. However, given the substantially improved prognosis over time, it is unclear whether this observed 10-year cumulative incidence of BCD is applicable for contemporary women. Our data also indicate that the risk of BCD continues to increase steadily beyond 10 years after diagnosis. This finding is corroborated by a recent Early Breast Cancer Trialists' Collaborative Group meta-analysis assessing 20-year prognosis among women with ER-positive tumors treated with 5 years of adjuvant endocrine treatment (10). In the meta-analysis, including only 5527 women, the 20-year risk of BCD in the T1N0 subgroup (ie, T1abcN0) was 12%, and in analyses starting follow-up at 5 years postdiagnosis, the risk of BCD after 15 years was 5% for T1abN0 and 8% for T1cN0 tumors. To the best of our   knowledge, before our study no such long-term follow-up data existed for ER-negative T1abN0 tumors. Our data suggest that standard prognostic factors in larger or node-positive breast cancer retain prognostic relevance in the T1abN0 subgroup. ER negativity, PR negativity, HER2 positivity, and higher tumor grade were associated with poorer prognosis, in agreement with results from most prior smaller studies (3,4,19,(23)(24)(25)(26)(27). Likewise, the 5-year cumulative incidence of BCD was 0.2% for luminal A tumors vs 3.4% for triplenegative tumors. Tumor grade in particular was a strong prognostic factor in our study. Women with grade 3 vs grade 1 tumors were more than 400% more likely to die from breast cancer and more than 50% more likely to die from any cause. Interestingly, the association between tumor grade and BCD was lower in the analysis restricted to women being alive 10 year after the diagnosis, suggesting that the prognostic utility of tumor grade diminishes over time.
In our study, women with ER-positive vs ER-negative tumors had a better 10-year prognosis but more similar longer-term prognosis, consistent with prior studies showing that late relapses are common in women with larger or node-positive ER-positive tumors (28). These results add further to the notion that T1abN0 tumors behave biologically as larger tumors. The high cumulative incidence of BCD among younger women is notable and in line with previous findings for both T1abN0 and larger tumors (20,24,29). We observed no difference in prognosis comparing T1aN0 with T1bN0, which is probably explained by a larger proportion of women with T1bN0 receiving active treatment, rather than tumor size not being a prognostic factor in the T1abN0 subgroup. We observed successively improved prognosis over time. The 10-year cumulative incidence of BCD was, for example, 2.5% among those diagnosed from 2000 to 2004 vs 5.5% among those diagnosed from 1977 to 1989. Several explanations are possible for this finding, including more and better adjuvant treatment over time. The improved prognosis over time can also be explained by other factors, such as earlier detection and active treatment of aggressive T1abN0 tumors, or by increasing overdiagnosis of indolent T1abN0 tumors because of mammography screening. The latter explanation seems unlikely, however, as we observed no appreciable difference in prognosis for tumors being screening detected vs not screening detected. Although we are unable to pinpoint the precise mechanisms, our data suggest that the changes over time in breast cancer screening, diagnosis, and treatment have led to improved prognosis not only for women with larger or nodepositive tumors but also for women with T1abN0 breast cancer. Key strengths of this study include its population-based design, large sample size, long-term follow-up, and the main outcome being BCD. Key limitations of these real-world data when assessing prognosis and prognostic factors are missing data and misclassification. We addressed the missing data issue by conducting sensitivity analyses restricted to women diagnosed from 2000 onward and, respectively, to women in the National Breast Cancer Registry subcohort, which yielded largely similar or even more pronounced relative risks. Registry data are prone to misclassification, and we conducted no medical chart review or re-review of pathological material. However, we have no reason to suspect that the key findings of this study are due to misclassification.
Indeed, any bias stemming from misclassification should primarily be at random and drive the associations toward the null. We did not focus this study on assessing treatment effects, which in these types of studies are prone to bias because of confounding by indication, selection bias, and lack of data on all relevant treatment selection factors.
It is important to note, however, that the relative treatment benefits of endocrine therapy and/or chemotherapy in randomized trials are independent of tumor size and stage (7)(8)(9)(10)(11), suggesting that the relative treatment benefits are similar also among women with T1abN0 tumors.
In conclusion, the data from this study suggest that the risk of BCD among women with T1abN0 breast cancer increases steadily with time since diagnosis, has improved over time, and varies substantially by tumor characteristics. For certain subgroups, for example, women with grade 3 tumors, the risk of BCD was considerable. The premise that prognosis for all women with T1abN0 breast cancer is sufficiently good to exclude adjuvant treatment irrespective of tumor features should be abandoned. Women with T1abN0 breast cancer should be considered for adjuvant therapy interventions, in particular those with aggressive tumor features or long life expectancy. Randomized clinical trials including patients with T1abN0 breast cancer are warranted.

Funding
This work was supported by grants from the Swedish Breast Cancer Association and the Swedish Society of Medicine (SLS-502451).

Notes
The role of the funder: The funders had no role in the design of the study; the collection, analysis, and interpretation of the data; the writing of the manuscript; and the decision to submit the manuscript for publication.

Disclosures:
The authors declare no potential conflicts of interest.