Main

The influence of locoregional recurrences (LR) on the risk of breast cancer progression is not fully understood (Halverson et al, 1992; Willner et al, 1997; Koscielny and Tubiana, 1999; Elkhuizen et al, 2001; van Duijnhoven et al, 2011). Current evidence indicates worse outcomes following LR (Halverson et al, 1992; Willner et al, 1997; Elkhuizen et al, 2001; van Duijnhoven et al, 2011) or a worse prognosis associated only with specific LR tumour subtypes (Koscielny and Tubiana, 1999). It has been suggested that a shorter disease-free interval between the diagnosis of the primary tumour and a second lesion is a determinant of poor prognosis (Vicini et al, 2003; Komoike et al, 2006; Anderson et al, 2009; Botteri et al, 2010; Vichapat et al, 2011; Tanis et al, 2012; van Laar et al, 2013), and a short interval usually signals that distant metastasis has already occurred (van der Sangen et al, 2013). In one paper, the Early Breast Cancer Trialists Collaborative Group (EBCTCG) showed that in up to 25% of cases LR was likely associated with breast cancer death (Early Breast Cancer Trialists’ Collaborative G et al, 2011).

A model for the influence of metachronous contralateral breast cancer on prognosis indicated that some studies may have underestimated risk of death from contralateral tumours arising close to the time of diagnosis of the primary tumour (Vichapat et al, 2011). We applied a similar model to analyse breast cancer prognosis after the occurrence of LR.

Using a hospital-based cohort of 567 women diagnosed with LR and 4632 women who did not experience an LR, the current study aimed to assess breast cancer-specific death while taking into account LR as a time-dependent characteristic.

Materials and methods

Study population

From the Guy’s and St Thomas’ Breast Cancer Tissue and Data Bank, a total of 5199 women were identified as eligible for initial analysis as diagnosed with an invasive primary breast cancer between 1st January 1975 and 31st December 2007 – coded as C50 (based on ICD-10) in the Hospital’s Clinical Cancer Database. Tissue samples and prospectively collected clinical data on the diagnosis, treatment, and follow-up of all patients who were diagnosed and/or treated within the Breast Clinic at Guy’s and St Thomas’ NHS Trust were gathered. Trained data managers manually enter the clinical and pathological data into the master data set. For the purposes of this study, women with unknown menopausal status were excluded (n=202), as were those who were pregnant and/or lactating (n=14). Regional recurrences were identified through patient reporting of symptoms and/or clinical examination by a surgeon, and then confirmed using mammography or ultrasound scans. Types of LR included were disease in breast, ipsilateral axillary lymph nodes, internal mammary lymph nodes, and skin recurrences. A secondary analysis was repeated including only breast recurrences.

The following variables were retrieved from the clinical database: age at diagnosis, period of initial diagnosis (1975–1979, 1980–1984, 1985–1989, 1990–1994, and after 1995), and tumour characteristics (clinical size, pathological lymph node status, histopathological type and grade, oestrogen and progesterone receptor status), as well as treatment given for the initial cancer. Follow-up information, including date of recurrence, date of last contact, and cause of death, was obtained from medical records as registered in the Hospital’s Clinical Cancer Database. Cause of death was ascertained by national death certification obtained from the National Cancer Registry (Table 1). All data collection was carried out under NHS Research Ethics Committee approval (REC Number: 12/EE/0493).

Table 1 Descriptive characteristics by locoregional recurrence status, including women with a primary diagnosis between the years of 1975 and 2007
Full size table

Statistical analysis

We assessed how known clinicodemographic risk factors and LR effect risk of breast cancer-specific death using Cox proportional hazards regression analysis. All analyses were stratified by surgery type (mastectomy vs breast-conserving surgery (BCS)). Where cause of death was unknown, women who experienced distant metastases before death were considered to have died of breast cancer.

To investigate how time between primary diagnosis and LR affects breast cancer-specific death, we used LR as a time-varying covariate in a Cox proportional hazards model. To graphically illustrate the effect of time between primary diagnosis and LR on breast cancer-specific death, we used Poisson regression. These models estimate hazard curves for risk of breast cancer-specific death among women with and without LR whereby follow-up time is split as (i) time before development of LR and (ii) time after LR, using the logarithm of time at risk as an offset. Risk of breast cancer-specific death for women without LR was modelled through natural cubic splines with knots in quartiles of time-to-event. Risk of breast cancer-specific death for women with LR was similarly modelled through natural cubic splines with knots in tertiles of time-to-event with two additional restrictions: (i) risk of breast cancer-specific death because of LR was bound to be zero at the onset of LR; (ii) additional risk because of LR was bound not to be distinguished from the baseline risk after 20 years of follow-up. The risk because of LR was allowed to depend linearly on time between the two diagnoses.

These graphical analyses were not stratified by surgery type because there is good evidence to suggest that BCS is a feasible option for patients with tumours of up to 5 cm. Higher incidences of LR have been reported in the longer term, but this does not appear to influence prognosis (van Dongen et al, 2000; Litiere et al, 2012).

All analyses were performed using SAS software version 9.3.1 (SAS Institute Inc., Cary, NC) and R version 2.15.3 (R Foundation for Statistical Computing, Vienna, Austria).

Results

Demographics

Five hundred and sixty-seven women developed LR without a prior distant metastasis. The median interval between primary diagnosis and LR was 4.28 years (s.d. ±4.35). There was no statistically significant difference in age at primary diagnosis between women with (56.35 years (s.d. ±13.6)) and without LR (58.7 years (s.d. ±12.1)). There was no difference in family history of breast cancer between the two groups (Table 1).

Risk of breast cancer death: entire study population

Table 2 shows that women with a shorter time interval to LR had an increased risk of breast cancer-specific death compared with women with a longer interval, (e.g., the HR for women who underwent mastectomy in those who experienced an LR within 6 months of primary diagnosis was 10.56 (95% CI: 7.42–15.03), whereas the HR for LR occurring >10 years following primary diagnosis was 1.62 (95% CI: 0.97–2.68), when compared with a reference group of women who did not develop LR).

Table 2 Multivariate HRs and 95% CI for risk of BC death by surgery type
Full size table

Risk of BC death–LR subgroup

A multivariate analysis looking at the association between known clinicodemographic risk factors and breast cancer-specific death (Table 2) was repeated in the subgroup of women who experienced LR (Table 3). We observed decreased risk of BC-specific death as the time interval between primary tumour and LR increased. For instance, the HR for LRs occurring within 6 months of the primary tumour was 13.76 (95% CI: 7.08–26.74), when compared with a reference group where LRs occurred >5 years following primary diagnosis in women who underwent mastectomy.

Table 3 Multivariate HRs and 95% CIs for risk of BC death following local recurrence by surgery type
Full size table

Figure 1 demonstrates a sharp increase in the hazard of breast cancer-specific death for women diagnosed with LR soon after diagnosis of their primary breast tumour relative to later occurring LRs, as observed in the HRs (Table 3). The magnitude of this hazard declines continuously with time, but does not return to that of the level of women not experiencing an LR, even 20 years after the diagnosis of their primary diagnosis. These curves were replicated by year of primary diagnosis (1975–1989 and 1990–2007), by surgery type, and for breast recurrences only; all patterns remained consistent (Figure A1 in Appendix).

Figure 1
figure 1

Estimated hazards of breast cancer death from Poisson regression model. The blue curve represents the hazard of breast cancer death for women without recurrence. The dark red, bright red, dark orange, light orange, yellow, and black curves represent the hazard of breast cancer death for local recurrence diagnosed at different interval times (0.5, 1, 2, 5, 10, and 15 years). There were 5194 women who did not experience a locoregional recurrence (LR) (blue line), 34 women experienced LRs within 6 months, 85 women within 1 year, 108 women between 1 and 2 years, 174 women within 2–5 years, 103 women between 5 and 10 years, and finally, 62 women experienced LRs between 10 and 15 years following their primary diagnosis.

Full size image

Discussion

Women with LR had a considerably poorer prognosis with regard to breast cancer-specific death compared with those without LR, especially if the LR occurred close in time to primary diagnosis. The increase in risk after LR was still apparent 20 years postprimary diagnosis and at no point in time declined to the level of risk in women without LR at the corresponding time of follow-up.

Some studies reported a worse prognosis for women with LR compared with those without (Willner et al, 1997; Elkhuizen et al, 2001), whereas others found a similar prognosis for women with and without LR (Robinson et al, 1993). One study showed a 2-year survival of 53% following an LR, compared with 86% among non-relapsed women (Willner et al, 1997). A further study found that patients who experienced an LR within 2 years of BCS were at especially high risk of distant metastasis and breast cancer-specific death (Elkhuizen et al, 2001).

Although the cited studies have touched on the possible effect of time interval, our study benefits from both long-term and complete follow-up allowing for a more in-depth and precise investigation into the importance of the timing of the LR. In our study, short interval time was the strongest determinant of risk for breast cancer-specific death following LR. The increased elevation of the hazard ratio for LR closer in time to primary tumour may suggest that a rapidly occurring LR may be an indication of an undetected existing distant disease. This is also supported by the fact that women with a shorter interval time often rapidly develop metastasis after LR (Macmillan et al, 1996; Elkhuizen et al, 2001; Komoike et al, 2006).

Our findings combined with existing literature may indicate that the decline in risk of breast cancer-specific death following LR as the time interval increases could be because of a proportion of these tumours being second primaries, as opposed to genuinely recurring, active disease (Holmberg et al, 1988; Hartman et al, 2007). A proportion of the second primaries should be localised disease with a chance to cure – and especially so if the women were under regular mammography surveillance. Later occurring LR may also be less aggressive recurrences, as opposed to rapidly growing lesions. Another plausible explanation for the association between the interval and breast cancer-specific death is that two tumours occurring close in time indicate a depleted host defence mechanism (von Mensdorff-Pouilly et al, 2000; de Visser et al, 2006).

Strengths and limitations

A strength of our study is the long-term and complete follow-up of a cohort of breast cancer patients, in which it was possible to measure breast cancer-specific survival. However, there are limitations that should be addressed. Histopathological tumour size was not available for all patients, and clinically assessed tumour size was used in its place. Although this measurement is less accurate, it did allow for a further way in which to rank patients by disease severity. Information on the tumour characteristics of LR were not included, and thus we are unable to infer the significance of their contribution towards prognosis. Furthermore, we could not use the characteristics of the primary and the LR to distinguish true recurrences from second primaries. We have no reason to believe that the treatment protocol at Guy’s & St Thomas’ NHS Trust was different or in any way responsible for the higher estimates observed here of risk of breast cancer death following an LR compared with those observed in previous studies. Finally, additional information on treatment adherence and comorbidities may have confounded our observations. However, as all women were treated in the same Unit in the same Hospital, we are confident that we have included high-quality treatment data.

Conclusion

It remains difficult to distinguish between an LR and a new primary. However, our results of an increased risk of breast cancer-death after an LR/new primary suggest that LR acts as a marker of systemic disease, large tumour burden, or depleted host defence and that the interpretation of this marker varies with interval to the primary lesion. The clinically highly relevant impairment in prognosis calls for further research into the underlying mechanisms, and indicates that for at least a 15-year follow-up the occurrence of an LR, a careful diagnostic and therapeutic management, is needed to improve the prognosis of affected women. These data further informs the design of post-treatment surveillance programs.