De novo and recurrent metastatic breast cancer – A systematic review of population-level changes in survival since 1995

Summary

Background

Advances in breast cancer (BC) care have reduced mortality, but their impact on survival once diagnosed with metastasis is less well described. This systematic review aimed to describe population-level survival since 1995 for de novo metastatic BC (dnMBC) and recurrent MBC (rMBC).

Methods

We searched MEDLINE 01/01/1995–12/04/2021 to identify population-based cohort studies of MBC reporting overall (OS) or BC-specific survival (BCSS) over time. We appraised risk-of-bias and summarised survival descriptively for MBC diagnoses in 5-year periods from 1995 until 2014; and for age, hormone receptor and HER2 subgroups.

Findings

We identified 20 eligible studies (14 dnMBC, 1 rMBC, 5 combined). Potential sources of bias in these studies were confounding and shorter follow-up for the latest diagnosis period.

For dnMBC, 13 of 14 studies reported improved OS or BCSS since 1995. In 2005–2009, the median OS was 26 months (range 24–30), a median gain of 6 months since 1995–1999 (range 0–9, 4 studies). Median 5-year OS was 23% in 2005–2009, a median gain of 7% since 1995–1999 (range -2 to 14%, 4 studies). For women ≥70 years, the median and 5-year OS was unchanged (1 study) with no to modest difference in relative survival (range: -1·9% (p = 0.71) to +2·1% (p = 0.045), 3 studies). For rMBC, one study reported no change in survival between 1998 and 2006 and 2007–2013 (median OS 23 months). For combined MBC, 76–89% had rMBC. Three of four studies observed no change in median OS after 2000. Of these, one study reported median OS improved for women ≤60 years (1995–1999 19·1; 2000–2004 22·3 months) but not >60 years (12·7, 11·6 months).

Interpretation

Population-level improvements in OS for dnMBC have not been consistently observed in rMBC cohorts nor older women. These findings have implications for counselling patients about prognosis, planning cancer services and trial stratification.

Funding

SL was funded in part by a National Health and Medical Research Council (NHMRC) Project Grant ID: 1125433. NH was funded by the NBCF Chair in Breast Cancer Prevention grant (EC-21–001) and a NHMRC Investigator (Leader) grant (194410). BD and SAP were funded in part by the NHMRC Centre of Research Excellence in Medicines Intelligence (1196900).

Research in context.

Evidence before this study

Breast cancer survival has improved worldwide since the mid-1990s, however survival trends for metastatic breast cancer (MBC) are not routinely reported by cancer registries and are not well described at the population level. We identified one prior systematic review of post-metastasis survival from MEDLINE using a combination of search terms for breast cancer, metastasis, survival and population-based for the search period 1/1/1995 to 12/4/2021. The authors reported median overall survival improved between 1990 and 2010 for both MBC occurring de novo and recurrent after an initial diagnosis of non-metastatic disease, but studies were not restricted to population-based cohorts.

Added value of this study

Our systematic review finding of sustained population-level improvements in overall survival and breast cancer-specific survival since 1995 for de novo MBC is consistent with prior evidence from more selected populations. Our finding that at least three-quarters of new MBC diagnoses represent recurrent disease, and survival gains have not been consistently observed for this group, is unique, and adds to evidence from trial and institution-based studies that postulate an ‘adjuvant therapy-related shortening of survival’ effect. We further report consistent evidence of no to modest survival improvement over time for patients aged 70 years and older, who make up around a third of new MBC diagnoses.

Implications of all the available evidence

This evidence is valuable for counselling patients with MBC about prognosis, planning cancer services, and establishes the importance of the stratification of MBC by de novo/recurrent status and age group for clinical trial design. Given our finding of limited population-level data on recurrent MBC, we advocate for cancer registries to routinely report on distant metastases to track population-level changes in characteristics and prognosis.

Alt-text: Unlabelled box

Introduction

Breast cancer (BC) mortality rates have decreased by approximately 40% since 1995,^1,2 with population-based cancer registries documenting improved survival worldwide.³ These survival gains correspond to major advances in BC multidisciplinary care, including the introduction of population BC screening programs for early diagnosis; adjuvant therapies to reduce the risk of distant metastasis; and new effective systemic therapies for metastatic BC (MBC). For early stage BC, there is consistent evidence that distant disease-free survival has improved over time.^4,5 However, survival trends for patients diagnosed with distant metastasis are not well described at the population level. While registries report on survival for de novo metastatic BC (dnMBC), comprising approximately 5% of incident BC,6, 7, 8 survival for recurrent MBC (rMBC) after an initial diagnosis of early BC is recorded less often, despite being much more common.

A systematic review and meta-analysis of post-metastasis survival estimated improvements in median overall survival (OS) between 1990 and 2010 for dnMBC (20 to 31 months) and rMBC (21 to 38 months) following no improvement in the prior decade.⁹ However, the analysis included both population-based and cohorts receiving treatment. Survival will differ at a whole-of-population level due to differences in baseline prognostic factors, treatment eligibility and access to services and clinical trials, in particular for older women. Further, survival gains may vary substantially for those with tumours that are hormone receptor or human epidermal growth factor receptor-2 (HER2) positive versus negative, as new and effective targeted therapies are introduced.

Documenting population-level changes in post-metastasis survival is important for understanding the full impact of changes in care on life span and the evolution of MBC prognosis over time. BC stakeholders have highlighted the importance of this evidence to inform individuals with MBC as well as cancer services planning and prioritising further research.^10,11 While clinical trials can provide the most valid estimates of treatment efficacy and prognosis for well-defined patient groups, population-level data are needed to assess the survival gains across an entire BC population reflecting the impact of BC services for the whole population.

This systematic review aims to assess: the extent survival has changed at a population-level for dnMBC and rMBC since 1995; and how survival has changed for age, hormone receptor and HER2 subgroups.

Methods

We performed a systematic review and report it according to the Preferred Reporting Items for Systematic Reviews and Meta-analyses guidelines.¹² We used the American Joint Committee on Cancer (AJCC) 8th edition definition of distant metastasis.¹³ We selected overall survival (OS) and BC-specific survival (BCSS) as primary outcomes. We assessed OS as a measure of actual survival (includes BC and non-BC deaths) most relevant to clinicians and patients¹⁴; and BCSS to estimate changes in MBC prognosis over time (BC death only, non-BC deaths censored). Changes in BCSS can be attributed to changes in BC diagnosis and treatment plus any changes in tumour prognostic factors unrelated to BC care. The review was not designed to assess a particular BC treatment or causal pathway. We assessed MBC cohorts since 1995, when tamoxifen was established as adjuvant therapy for those with hormone receptor-positive tumours, to capture potential changes in survival due to the introduction of treatment advances after 2000, such as trastuzumab for those with HER2-positive tumours.

Search strategy and study selection

We searched MEDLINE for peer-reviewed studies published between 1 January 1995 and 12 April 2021 using keywords and MeSH terms including: breast neoplasms/breast cancer AND metastatic/Stage IV AND survival/prognosis AND population-based/registries/time trend/temporal trend (Supplementary 1). Studies were eligible for inclusion if they: used a population-based data source to assemble cohorts of women with MBC by year of BC or MBC diagnosis; assessed survival after MBC diagnosis as OS, BCSS or relative survival for two or more cohorts since 1995, included at least one cohort with a diagnosis since 2000; reported median survival, 2-year or 5-year survival (or survival curves to extract these data); and were reported in English.

Studies that assessed a single time period were eligible if an additional publication using the same population-based data source reported on a second eligible time period. Studies that assessed a review-defined subgroup or specific distant metastatic sites were also included. To exclude superseded studies, where two studies reported survival for the same population and time periods, we included the study that reported the most complete data for the pre-specified outcomes. For studies reporting on the same survival outcome measure for the same population, we included the study that assessed the most recent time period. If two studies of the same population each reported data for different (non-overlapping) pre-specified outcomes (median survival, 5-year survival) or subgroups (eg. age or oestrogen receptor status), we included both studies in the review. Two investigators (KB, SL) reviewed search results independently by scanning titles and abstracts to identify potentially eligible articles. These were retrieved as full text articles to identify eligible studies. Discordant findings were resolved by discussion between the two investigators. We checked citations and reference lists of included studies to identify additional potentially eligible articles.

Data extraction

One investigator extracted study data into tables which were checked by a second investigator (KB, SL). Discordant findings were resolved by discussion. Extracted data included: number of women with MBC overall and within our study periods of interest (≥1995); population characteristics (age, hormone receptor status, HER2 status, adjuvant therapy use); and survival outcomes (OS, BCSS, relative survival) as median survival, 2-year and 5-year survival and hazard ratio (HR) for each MBC cohort and subgroups. If not reported, we extracted survival data from the survival curve where available using WebPlotDigitizer (https://automeris.io/WebPlotDigitizer/).

Assessment of risk of bias and applicability

One investigator assessed the risk of bias, which was checked by a second investigator (SL, NH) based on five domains: study participation; prognostic factor measurement (time periods, age, tumour receptor status); outcome measurement; confounding; and analysis method using criteria adapted from QUIPS 2013¹⁵ (Supplementary 2). ‘Risk of bias’ refers to the identification of a potential source of bias, not the magnitude. We rated risk of bias for each domain as: ‘high’ (potential source identified); ‘low’ (no potential source); or ‘unclear’ if inadequate information was reported for assessment. We rated the overall risk of bias for each study as high if one or more of the domains was rated as high; and low if all domains were rated as low risk. Studies without a high-risk rating but one or more domains with unclear ratings were classified as unclear for overall risk of bias. We assessed the applicability by appraising whether study participation, prognostic factor(s) and outcome measurement matched the review questions. We rated concerns about applicability as low, high or unclear, using the same approach. Discordant findings for risk of bias and applicability were resolved by discussion and consensus between the two investigators. In addition to the assessment of risk of bias and applicability, we considered precision of estimates and consistency of results between studies for judgements about the strength of the evidence underlying our conclusions.¹⁶

Data synthesis

We assessed survival following a dnMBC and rMBC diagnosis separately, using a descriptive approach. Where dnMBC and rMBC were not reported separately, we assessed them together as ‘combined’ MBC and reported%rMBC. For each study, we plotted median and 5-year survival for each time period reported to provide a visual summary of temporal trends. To avoid duplication of data for the same people, where studies reported on the same source population, we extracted data from the study that included the most recent time period and/or longest follow-up period for the survival outcome plotted. To summarise survival changes across studies quantitatively we reported the median and range for survival estimates for the following 5-year MBC diagnosis periods: 1995–1999, 2000–2004, 2005–2009, 2010–2014; and calculated the median absolute difference in survival estimates between these periods to describe changes over time. For this descriptive analysis, we included data from studies that reported survival in each 5-year period or in a period that covered at least 3 years of the period. We also summarised change over time from the study-estimated HR and 95% confidence interval (CI), if available.

We defined subgroups as follows: age at MBC diagnosis (<50, 50–69, ≥ 70 years), hormone receptor status (oestrogen receptor (ER)-positive and/or progesterone receptor (PR)-positive tumours, herein referred to as ER-positive; ER-negative and PR-negative tumours, herein referred to as ER-negative), HER2 status (HER2-positive, HER2-negative) and triple negative. The review was designed to inform our ongoing research program (research protocol: http://dx.doi.org/10.1136/bmjopen-2018-026414).¹⁷ The review protocol was not registered.

Role of the funding source

The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. SL and KB had access to the bibliographic search database. All authors had access to the included study data and all authors agreed with the final decision to submit for publication.

Results

We identified 20 eligible studies assessing dnMBC (14 studies), rMBC (1 study), or ‘combined’ dnMBC and rMBC without stratification (5 studies) (Figure 1, Supplementary 3). Studies assessed city, regional or country-wide populations in North America and Europe, grouping women by their MBC diagnosis date into one to eleven-year periods to estimate OS (16 studies), BCSS (4 studies) or relative survival only (3 studies). Fourteen studies included diagnosis periods commencing from 2005; and five studies included the period 2010–2014 (Table 1, Supplementary 4).

Figure 1.

PRISMA flow diagram of study selection

MBC=metastatic breast cancer.

Table 1.

Characteristics of included studies18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37

Study design characteristics

The 14 studies of dnMBC were conducted in the US (6 studies), Netherlands (4 studies), Germany (1 study), Italy (1 study), Sweden (1 study) and one study conducted in both the US and Germany (Table 1). Of these, five studies assessed age subgroups, including one study restricted to women <40 years.¹⁸ Four studies compared ER subgroups, including one study that reported survival stratified by ER status only.¹⁹ No studies of dnMBC assessed HER2-positive or triple-negative subgroups. Sample size ranged from 514²⁰ to 22,601.²¹ The six US studies used data held by the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) Program registries for 9 to 18 geographical regions (denoted as SEER-9, 13, 18). All six studies used data from at least 9 SEER registries in common, but time periods and eligibility criteria varied, such as: age limits^18,19,22,23; exclusion of participants with >1 primary cancer²¹; inclusion of invasive ductal or lobular tumours only²⁴ (Table 1).

The single study assessing rMBC included two diagnosis periods after 1995 (1998–2006 and 2007–2013, Munich Cancer Registry, N = 5700).²⁵ Five studies of combined MBC, from British Columbia, Canada and three regions in Sweden, included 76% to 89% rMBC (Table 1). One study assessed brain metastasis alone,²⁶ and one study was restricted to ER-positive HER2-negative tumours.²⁸ Other differences in eligibility criteria between studies of combined MBC were exclusion of: women aged >75 years²⁷; those with other cancer²⁷; or contralateral BC^27,28 (Table 1). One study assessed age subgroups²⁹; and two studies assessed ER and HER2 subgroups.^26,30

Risk of bias and applicability

Of 17 studies reporting adequate information for appraisal, 16 were assessed as having one or more potential sources of bias, most commonly: confounding (10 studies) eg. survival comparisons without consideration of differences in age at BC diagnosis over time; outcome assessment (9 studies) due to shorter follow-up time for women diagnosed in the most recent study period; or study participation (6 studies) eg. exclusion of older women (Supplementary 5). Five studies were assessed as high concern for applicability, most commonly due to no separate assessment of survival for rMBC for studies of combined MBC (4 studies, Supplementary 5).

Participant and treatment characteristics

Studies with no age restrictions reported at least one third of women were aged ≥70 at diagnosis (dnMBC 33–47%,^24,31,32 rMBC 36%,²⁵ combined MBC 34%²⁹ from the most recent diagnosis period). Five studies of rMBC and combined MBC reported an increased proportion of women receiving neo/adjuvant chemotherapy or adjuvant endocrine therapy or both since 1995 (Table 1).

MBC survival

De novo MBC

Thirteen of 14 studies of dnMBC reported improved survival (OS, BCSS or relative survival) since 1995, with 12 studies reporting 95% CIs for survival estimates and/or statistical tests for comparisons between time periods (Table 2, Figure 2A-D). Median of median OS was 20 months (range 18–26) for dnMBC diagnosed in 1995–1999; and 26 months (range 24 to 30) for dnMBC diagnosed a decade later in 2005–2009; a median OS gain of 6 months (range 0 to 9 months, 4 studies, Figure 2A).^20,31,33,34 Malmgren et al. (2020) provided high quality data for women with dnMBC <85 years and estimated median BCSS gains of 5 and 12 months in two US regions between 1999 and 2004 and 2005–2011 (p<0.001 for comparison of all time periods, Table 2, Figure 2B).²²

Table 2.

Changes in post-metastasis survival over time.

Author, year,MBC diagnosis periods	N		dnMBC or rMBC	Median survival, months (95% CI)	Survival (95% CI)					Adjusted hazard ratio (HR)(95% CI)	Co-variables for adjusted analysis
					2-year	5-year
Malmgren et al. (2020)22			dnMBC	Median BCSS*	2-yr BCSS*	5-yr OS		5-yr BCSS		HR BCSS
SEER 9 without Seattle-Puget Sound
1990–1998	3988		All	23	49%	16% (15–18)		19% (18–21)		1·00 (reference)	Age, race, ER+/PR+ status, surgery, XRT, chemotherapy
1999–2004	2958			26	54%	20% (19–21)		23% (21–24)		0·85 (0·80–0·91), p<0·001
2005–2011	4094			31	60%	23% (22–24)		26% (24–27)		0·72 (0·67–0·76), p<0·001
				p<0·001		p<0·001		p<0·001
Seattle-Puget Sound
1990–1998	608		All	21	44	18% (15–21)		21% (18–24)		1·00 (reference)
1999–2004	481			30	57	25% (21–28)		27% (23–31)		0·74 (0·63–0·87), p<0·001
2005–2011	707			42	66	32% (28–35)		35% (32–39)		0·57 (0·49–0·66), p<0·001
				p<0·001		p<0·001		p<0·001
Di Meglio et al. (2016)24			dnMBC	Median OS						HR OS*
1990–1992	NR		All	18						1·00 (reference)	SEER registry, age, race, marital status, tumour grade, ER+/PR+ status, histology
1993–1995				18*						1·07
1996–1998				19*						1·05
1999–2001				21*						1·00
2002–2004				23*						0·95
2005–2007				26*						0·92
2008–2011				28						0·85 (0·78–0·93)
2008–2011 vs 1990–1992				p<0·001
Per year 1990–2011										0·99 (0·98–0·99), p<0·05
Per 5 years										0·95 (0·93–0·97), p<0·05
Tao et al. (2016)34			dnMBC	Median OS*	3-year relative survival*
1995	NR		All	17	31
2000				20	34
2005				24	39
2009				25	39
Dawood et al. (2015)21			dnMBC		2-yr OS	2-yr BCSS				Odds ratio per year
1990–1995	4215		All		36·2%	40·1%				OS >2 years 1·06 (1·05–1·07), p<0·001 BCSS >2 years 1·06 (1·05–1·07), p<0·001	Age, race, marital status, tumour grade, MBC site, ER, PR status, surgery, XRT, inflammatory subtype, SEER registry
1996–2000	5108				40·1%	44·0%
2001–2007	13,278				44·2%	48·1%
					p<0·001	p<0·001
Holleczek et al. (2012)35						5-yr relative survival (SE), age standardised
SEER-13 USA			dnMBC
1993–1996	4926		All			20·2 (0·7)
1997–2000	5386					22·5 (0·7)
2001–2004	5626					25·2 (0·7)
2005–2008	6016					26·7 (0·7)
2005–2008 vs 1993–1996						6·5%, p<0·001
Saarland, Germany			dnMBC
1993–1996	228		All			20·3 (3·2)
1997–2000	262					22·1 (3·0)
2001–2004	271					19·0 (2·6)
2005–2008	242					23·7 (3·0)
2005–2008 vs 1993–1996						3·3%, p = 0·11
van der Meer et al. (2021)36		dnMBC				5-yr Relative survival (95% CI), age standardised
1989–1999	NR		All			12·3 (10·1, 14·8)
2000–2009						20·8 (18·0, 23·8)
2010–2016						33·2 (29·4, 37·2)
Vonderling et al. (2018)37			dnMBC		2-yr OS	5-yr OS
1989–1992	1990: 543		All		40%	14%
1993–1998	1995: 493				44%	16%
1999–2002	2000: 562				43%	17%
2003–2009	2005: 612				53%	22%
2010–2013	2010: 573				58%	NR
de Glas et al. (2015)33		dnMBC		Median OS*		5-year relative survival				HR OS	Age at diagnosis, ER/PR status, morphology, grade and number of metastatic sites
1995	NR		All	18		17				0·99 (0·99–1·00), p<0·001 per year 1990–2010
2000				17		16
2005				24		22
2010				25		21				Adjusted relative survival‡
										0·98 (0·98–0·99), p<0·001
Ruiterkamp et al. (2011)32			dnMBC	Median OS	2-yr OS*	5-yr OS*				HR OS
1995 – 1999	2688		All	17·0 (16·0–18·4)	41%	15%				1·00 (reference)	Age, primary tumour size, surgery, XRT, systemic therapy
2000 – 2004	2916			19·3 (18·1–20·4)	44%	17%				0·95 (0·89–1·01), p = 0·081
2005 – 2008	2427			23·4 (21·6–25·0)	49%	21%				0·83 (0·77–0·89), p<0·001
				p<0·001
Holzel et al. (2017a)31			dnMBC	Median OS*	2-yr OS*	5-yr OS /relative survival				HR OS	Age, primary tumour size, grade, ER+/PR+ status, lymph node involvement, ≥ 2 metastasis site, systemic therapy
1978–1987	376		All	25	52%	17·6% /18·9				1·10 (0·95–1·26)
1988–1997	875			26	54%	22·5% /24·7				1·05 (0·95–1·15)
1998–2006	1702			26	54%	24·7% /27·1				1·00 (reference)
2007–2013	1803			26	53%	20·5% /22·7				1·01 (0·93–1·10)
				p = 0·52						p = 0·77
Nordenskjold et al. (2019)23			dnMBC			5-yr relative survival*				5-year excess mortality rate ratio
1989–1993	224		All			1989–1991: 12 1992–1994: 12 1995–1997: 9 1998–2000: 20 2001–2003: 12 2004–2006: 20 2007–2009: 23				1.00 (reference)	Age
1994–1998	200									0.98 (0.79–1.21), p = 0.87
1999–2003	185									0.91 (0.73–1.13), p = 0.40
2004–2008	223									0.83 (0.67–1.03), p = 0.08
2009–2013	260									0.69 (0.55–0.86), p = 0.001
						p = 0.03
Cortesi et al. (2015)20			dnMBC	Median OS*	2-yr OS*	5-yr OS				HR OS
1990–1993	73		All	16	39%	11%				1·00 (reference)	Age, tumour grade, ER, PR status, Ki67, MBC site, chemotherapy, hormone therapy
1994–1997	82			21	46%	15%				0·76 (0·42–1·36), p = 0.357
1998–2001	114			21	42%	12%				0·62 (0·36–1·09), p = 0.097
2002–2005	126			21	46%	20%				0·58 (0·33–1·00), p = 0.050
2006–2009	119			30	51%	29%				0·53 (0·30–0·95), p = 0.043
						p = 0·012
Holzel et al. (2017b)25			rMBC	Median OS*	2-yr OS*	5-yr OS* /relative survival
1978 – 1987	1352		All	23	48%	20% /22·0
1988 – 1997	2170			21	45%	18% /22·6
1998 – 2006	3502			23	49%	21% / 17·6
2007 – 2013	3172			23	48%	21% /11·8
				p = 0.029
MBC and BC diagnosis within same study period						4-yr OS*				HR OS
1978 – 1987	1352			23	49%	26%				0·91 (0·82–0·99)	Age at BC diagnosis, primary tumour size, tumour grade, lymph node involvement, ER status
1988 – 1997	1429			18	39%	20%				1·11 (1·03–1·20)
1998 – 2006	1788			20	43%	22%				1·00 (reference)
2007 – 2013	1187			14	32%	16%				1·24 (1·13–1·35)
Le et al. (2020)28			rMBC	Median OS		5-yr OS
2003 – 2005	755		77·9%	24·4 (21·6–27·3)		18·1%
2007 – 2009	772		76·8%	23·1 (20·7–25·5)		17·7%
2011 - 2013	905		74·1%	23·1 (20·6–25·6)		17·3%†
Chia et al. (2007)27			rMBC	Median OS	2-yr OS	5-yr OS*				HR OS
1991–1992	423		73%	14·3	33%	13%				1·00 (reference)	Age at MBC diagnosis, tumour grade, ER status
1994–1995	561		82%	14·8	34%	14%				0·97 (0·85–1·11), p = 0·65
1997–1998	641		80%	18·5	44%	15%				0·84 (0·74–0·96), p = 0·011
1999–2001	525		78%	21·7	45%	NR				0·72 (0·61–0·84), p<0·001
				1997–1998 vs 1999–2001, p = 0·05
Sundquist et al. (2017)30			rMBC	Median OS (IQR)	2-yr OS	5-yr OS
1985 – 1989	124		76·6%	13 (4–31)	31%	10%
1990 – 1994	147		87·1%	16‡ (6–35)	38%	13%
1995 – 1999	160		85·6%	16 (7–36)	36%	9%
2000 – 2004	129		83·7%	20 (7–45)	43%	15%
2005 – 2009	152		81·6%	23 (6–48)	49%	17%
2010 – 2014	72		91·7%	33 (18–55)	64%	27%
				p = 0·009
Foukakis et al. (2011)29			rMBC	Median OS	2-yr OS	5-yr OS/Relative survival*				HR OS rMBC (dnMBC excluded)
1979–1984	899		84%	15·9	36·8%	11·4%/13				1·00 (reference)	Age at MBC diagnosis, primary tumour size, ER status, recurrence-free interval, systemic neo/adjuvant therapy, metastatic site (bone only, liver/brain)
1985–1989	1078		92%	15·1	36·1%	12·9%/14				0·99 (0·89–1·11), p = 0·90
1990–1994	1158		90%	14·5	34·6%	12·5%/14				1·00 (0·89–1·12), p = 1·00
1995–1999	1196		87%	16·1	36·2%	13·6%/15				1·00 (0·88–1·12), p = 0·95
2000–2004	1132		89%	15·3	37·7%	15·2%/17				0·94 (0·83–1·07), p = 0·34
				p = 0·12
Thulin et al. (2020), Brain metastasis26			rMBC	Median OS
1994 – 2004	45		80%	11·5
2005 – 2014	146		84%	7·2

BCSS = breast cancer specific survival; CI =confidence interval; dnMBC = de novo MBC; HR = hazard ratio; IQR = Interquartile range; MBC = metastatic breast cancer; OS = overall survival; SE = standard error; rMBC = recurrent MBC; XRT = radiation therapy.

* Extracted from published survival curve or figure; †Excludes women diagnosed after Aug 2012; ‡ Results reported in text differ from published table/figure.

Figure 2.

Survival over time for de novo MBC: (A) median overall survival; (B) median breast cancer-specific survival; (C) 5-year overall survival; (D) 5-year breast cancer-specific survival, relative survival

BCSS = breast cancer specific survival; MBC = metastatic breast cancer

Where more than one study reported OS from the same region, we selected the study that included the most recent study period.

The median probability of OS ≥5 years for dnMBC 1995–1999 was 16% (range 15–23%); and 23% (range 21–32%) for dnMBC 2005–2009, a median gain of 7% (range −2 to 14%, 4 studies, 5 cohorts, Figure 2C).^20,22,31,32 Malmgren et al. (2020) assessed both OS and BCSS and reported 5-year BCSS was 2–3 percentage points higher than 5-year OS in the two regions assessed (eg. SEER-9 without Seattle-Puget Sound 2005–2011: 5-year BCSS 26% (95% CI 24–27); 5-year OS 23% (95% CI 22–24)).²² Five-year BCSS for dnMBC increased by 3% to 8% between 1999 and 2004 and 2005–2011 across both regions assessed, similar to the 5-year OS gain of 3 to 7% during the same period (Figure 2C-D, p<0.001 for BCSS and OS comparison of all time periods).²² Another US study of overlapping SEER populations also reported similar 8% gains for 2-year OS (p<0.001) and BCSS (p<0.001).²¹ Three of four studies assessing relative survival for dnMBC reported 5-year relative survival gains in the US, Germany and the Netherlands between 1995 and 2009 (Figure 2D).^23,31,35,36 Two studies from the Netherlands that included dnMBC diagnosed from 2010 reported a further survival gain for this later period (2-year OS 2003–2009 53%, 2010–2013 58%, no statistical test reported³⁷; 5-year relative survival 2000–2009 20·8% (95% CI 18·0- 23·8), 2010–2016 33·2% (95% CI 29·4–37·2)³⁶ (Table 2).

Adjusting for prognostic factors that may change over time such as age and tumour characteristics (ER status, tumour grade and histology), two studies estimated a reduction in the hazard of all-cause death of 1% per year for dnMBC in the US (HR 0·99, 95% CI 0·98–0·99 per year between 1990 and 2011, p<0·05),²⁴ and the Netherlands (HR 0·99, 95% CI 0·99–1·00 between 1990 and 2010, p<0·001).³³ An additional three studies of dnMBC reported a statistically significant reduced hazard of all-cause^20,32 or BC death²² over time, although these studies were less applicable to our review question because they also adjusted for treatment factors such as surgery, radiotherapy, endocrine therapy and chemotherapy. In contrast, using data from the Munich Cancer Registry, Holzel et al. (2017a) did not find an ongoing OS or relative survival gain for dnMBC between study periods 1998–2006 and 2007–2013, nor evidence of reduced hazard for all-cause death after adjusting for age, primary tumour size, grade, ER status, lymph node involvement, multiple metastasis sites and primary systemic therapy (HR 1·01, 95% CI 0·93–1·10, Table 2).³¹

Two studies observed no to very small differences in OS and relative survival for women ≥70 years in three geographical regions between 1993/1995 and 2008 (5-year OS 11% for each period, 95% CIs not reported³²; difference in 5-year relative survival −1·9, p = 0.71 to +2·1%, p = 0.045, Supplementary 6–7).³⁵ Two additional studies reported relative survival did not improve for women ≥75 years in the Netherlands (relative excess risk per year 1990–2010, 1·01 (95% CI 1·00–1·02), p = 0·2;3³³ 5-year relative survival 2000–2009 21·1% (95% CI 17·7–24·8); 2010–2016 21·1% (95% CI 17·0–25·5).³⁶ Studies observed the largest survival improvements for women <50 years with more modest improvement for women 50–69 years (Supplementary 6–7).^32,35,36 A study of women aged <40 years reported a 24% increase in 5-year BCSS between 1995 and 1999 32·2%, (95% CI 26·7–37·9) and 2010–2015 56·5% (95% CI 50·1–62·4).¹⁸

Four studies reported longer survival for ER-positive tumours than ER-negative tumours in all study periods (Supplementary 8). Of these, three US studies of overlapping SEER regions observed OS and BCSS improved over time for ER-positive dnMBC.^19,21,34 Two of these studies reported 2-year²¹ and 5-year survival¹⁹ also improved over time for ER-negative dnMBC (percentage change per year, 5-year BCSS 1992–2006: ER-positive 2·3% (95%CI 1·0–3·6); ER-negative 3·6% (95% CI 1·3–5·9)).¹⁹ The third study reported median OS for ER-negative dnMBC remained stable at around 14 months between 1995 and 2009 (95% CI not reported).³⁴ Holzel et al. (2017a) observed improved median survival between the periods 1988–1997 and 1998–2006 for ER-positive (31, 34 months, respectively) and ER-negative (14, 15 months respectively) but not between 1998 and 2006 and 2007–2013 (ER-positive 30 months, ER-negative 13 months), statistical tests not reported (Supplementary 8–9).³¹

Recurrent MBC

For rMBC, using data from the Munich Cancer Registry, Holzel et al. (2017b) observed OS improved for rMBC diagnosed in 1998–2006 (median OS 23 months; 5-year OS 21%) compared to the prior period 1988–1997 (median OS 21 months; 5-year OS 18%); but no further improvement in 2007–2013 (median OS 23 months; 5-year OS 21%), p = 0.029 for comparison of all time periods, statistical tests were not reported for individual time period comparisons (Table 2, Figures 3A & 3B).²⁵ Taking into account both the period of initial BC diagnosis and period of rMBC diagnosis, women with BC and rMBC both diagnosed 2007–2013 had a 24% higher risk of all-cause death than women with BC and rMBC both diagnosed 1998–2006, adjusting for age at BC diagnosis, tumour size, grade, lymph node involvement and ER status (HR 1·24, 95% CI 1·13–1·35, Table 2).²⁵ As longer time to distant recurrence is associated with improved prognosis, a potential source of bias for this adjusted analysis is the 1998–2006 study period includes women with rMBC within 9 years of BC diagnosis; while the 2007–2013 period is limited to women with rMBC within 7 years.

Figure 3.

Survival over time for recurrent and combined (de novo and recurrent) MBC: (A) median overall survival, (B) 5-year overall survival.

Combined (de novo and recurrent) MBC

Estimates of median OS and 5-year OS from studies of combined MBC are illustrated in Figures 3A and 3B. Comparing MBC cohorts diagnosed between 1995 and 1999 and 2000–2004, two Swedish studies reported different temporal trends for median OS (Foukakis et al. (2011)²⁹ 1995–1999 16·1 months, 2000–2004 15·3 months, p = 0.12 for comparison of all time periods since 1979–1984, no potential source of bias identified; Sundquist et al. (2017)³⁰ 16, 20 months respectively, p = 0.009 for comparison of all time periods since 1985–1989, potential source of bias identified). For subgroup analysis by age, Foukakis et al. (2011) reported median OS for women aged ≤60 years at diagnosis of MBC improved from 19·1 months in 1995–1999 to 22·3 months in 2000–2004 (p<0.0001 for comparison of all time periods); with no improvement observed for women >60 years (12·7 months, 11·6 months respectively, and reduced survival for comparison of all time periods, p = 0.013 Supplementary 6–7).²⁹ A third study, Chia et al. (2007) that excluded women >75 years estimated a 3 month gain in median OS for MBC diagnosed in British Columbia, Canada between 1997 and 1998 (18·5 months) and 1999–2001 (21·7 months), p = 0.05.²⁷

Comparing MBC cohorts diagnosed after 2000, Sundquist et al. (2017) reported a 13 month gain in median OS between 2000 and 2004 (20 months) and 2010–2014 (33 months), p = 0.009 for comparison of all time periods.³⁰ In contrast, Le et al. (2020) assessed ER-positive, HER2-negative MBC from British Columbia and reported no improvement in median or 5-year OS across three study periods between 2003 and 2005 (median OS 24.4 months (95% CI 21.6–27.3) and 2011–2013 (median OS 23.1 months (95%CI 20·6–25·6)).²⁸ Thulin et al. (2020) reported a reduction in OS for brain metastasis (1994–2004 11·5 months; 2005–2014 7·2 months, statistical test not reported) with no difference in age at BC diagnosis.²⁶ Le et al. (2020)²⁸ and Sundquist et al. (2017)³⁰ did not assess or adjust for age as a potential confounder.

For subgroup analysis by ER and HER2 status: Sundquist et al. (2017) assessed post-metastasis survival by tumour receptor status pre- versus post-2000.³⁰ Median OS remained at 10 months for triple negative tumours but improved for other subgroups with the largest improvement observed for HER2-positive tumours (median OS 14 to 29 months; 5-year OS 2% to 31%, between 1996 and 1999 and 2000–2014, statistical test not reported Supplementary 8–9).

Discussion

This systematic review provides strong evidence that OS and BCSS following de novo MBC have continued to improve at a population-level since 1995 with a relative reduction in the risk of all-cause death of 1% per year across different populations. In contrast, the population-level survival gains observed for cohorts of women with recurrent and combined (recurrent and de novo) MBC before 2000 were not consistently observed in more recent cohorts, despite the introduction of new therapies for metastatic disease. These findings have implications for all stakeholders in advanced breast cancer, including cancer clinical services and surveillance systems, as discussed below in our interpretation of the evidence.

All included studies were conducted in North America or Europe highlighting the gap in evidence for post-metastasis survival in low- and middle-income countries. Further, our finding of variation in rMBC survival between regions demonstrates the importance of region-specific data. For high-income countries with similar health systems and available treatments, our review also provides valuable evidence about the current population characteristics of MBC. From included studies, we estimate at least three-quarters of new MBC diagnoses at a population level are recurrences following an initial diagnosis of early BC (study range 76–89%). Our review indicates that at least one third of women are 70 years or older at MBC diagnosis (de novo or recurrent). This finding is important given our review shows consistent evidence for both de novo and recurrent MBC of no to modest survival improvement over time for this older age group. Consequently, there is a widening gap in prognosis between younger versus older women with recent survival gains largely restricted to women under 70 years.

From the limited data available for subgroup analysis by ER and HER2 status, we observed population-level survival gains for MBC for both ER-positive and ER-negative subtypes. However, survival remained higher for ER-positive versus ER-negative tumours at each time period. One Swedish study provided evidence that the introduction of targeted therapies for HER2-positive MBC (of whom few rMBC received adjuvant HER2-targeted therapy) has translated to substantial population-level benefits with the median survival for this group similar to that for ER-positive MBC after 2000.³⁰ The same study did not find evidence of improved survival for triple negative MBC.

For dnMBC, our finding of improved survival since 1995 across 13 of 14 studies is consistent with Caswell-Jin et al's estimate from a meta-analysis of an 8 month improvement in median OS from 2000 to 2010.⁹ The conflicting finding from one study of no survival gain from 1998 to 2006 to 2007–2013³¹ may be explained in part by changes in the prognostic characteristics of the study dnMBC population over time that were not shared by other studies. For example, the proportion of women aged ≥70 years at dnMBC diagnosis increased from 36·4% to 46·9% during this period, whereas other studies did not observe an increase in age at dnMBC over time. Holzel et al's (2017a) finding of a survival improvement for the subgroup assembled with similar age and tumour prognostic characteristics (6 month gain in median survival) support this explanation; while their age-adjusted analysis did not, but was potentially limited by shorter follow-up time for the final cohort.³¹ Other included studies of dnMBC were also appraised as having one or more potential sources of bias. Nevertheless, the consistency of results for OS and BCSS across these studies support conclusions that advances in the diagnosis and management of de novo MBC have translated to incremental population-level survival gains.

Our review findings for rMBC and combined MBC (which is mostly comprised of rMBC) are more challenging to interpret. The variation in survival trends between populations is an important finding in itself. Evidence of no or limited OS gain in study periods after 2000 do not align with Caswell-Jin et al.’s (2018) review estimate of a 12 month improvement in median OS for rMBC between 2000 and 2010,⁹ possibly reflecting the different review approaches. Caswell-Jin et al. included trial and single institution populations, while our review focused on changes in rMBC survival in the entire MBC population, which includes older patients and those with co-morbidities who may not be eligible for clinical trials and may not tolerate standard treatment protocols.

Together with evidence of the increased use of adjuvant therapy since 1995, our finding of no to limited improvement in prognosis for rMBC in some populations provides some support for the postulation from clinical trials of an ‘adjuvant therapy-related shortening of survival’ (ATRESS) effect in pre-treated patient cohorts.^38,39 Recent institution-based cohort studies of MBC treatment also provide evidence of this effect.^40,41 Proposed causal pathways include the prevention of treatment sensitive metastases in pre-treated patients, leading to an apparent shift to more aggressive metastases as treatment non-sensitive tumours dominate; and potentially adjuvant therapy induces treatment resistance.³⁸ At a molecular level, evidence of genomic differences between rMBC and treatment-naive dnMBC point to biological differences requiring further investigation to help elucidate the mechanisms for differences in prognosis.⁴²

Our descriptive approach for evidence synthesis does not provide pooled estimates of changes in survival from MBC. However, as the diagnosis periods, exclusion criteria, and survival measures all varied between studies, only a descriptive approach was possible. The two major limitations for addressing our review questions were the paucity of published population-based contemporary data for rMBC; and the methodological limitations of the studies included in our review. The most common potential sources of bias in our included studies were: comparisons of survival over time without adjustment for age at BC diagnosis; and a shorter follow-up time, and thereby potentially more unreliable survival estimate, for the most recent study cohort. For ageing populations, the median age at BC diagnosis can be expected to increase,⁴³ increasing the risk of both BC and non-BC death. Other sources of bias were the exclusion of older women and those with prior non-breast cancers, two groups which make up a sizeable proportion of MBC populations. Most studies assessed OS without separate assessment of BCSS. While OS is important as a measure of actual survival, changes over time reflect changes in survival from both cancer and non-cancer causes.¹⁴ Thus, assessment of BCSS is also needed to assess the impact of changes in BC care without the potential for confounding by changes in comorbidities over time. Even so, we found that where both were reported, BCSS gains were proportionally similar to OS gains. No studies assessed BCSS for our subgroup analysis of post-metastasis survival for women ≥70 years. However, we believe our finding of no to very limited survival gain for this age group is more plausibly explained by limited benefit from advances in BC therapy over time than an increase in non-cancer mortality.

Our findings have implications for clinicians, cancer service planning and the design of clinical trials. Clinicians can use this evidence to inform prognostic discussions for patients with a new diagnosis of MBC. Our finding of consistent evidence of survival improvement over time for women with de novo MBC adds to evidence from more selected BC populations and may provide hope to women diagnosed today as they are likely to live longer than published survival times which represent women often diagnosed 10 years earlier. For cancer service planning, our findings highlight the importance of tracking the prognosis and care needs of women with rMBC, as distinct from dnMBC. While the clinical benefits of adjuvant therapies to reduce the risk of rMBC and improve BC survival are well established^4,5; our findings have important clinical implications for those with a diagnosis of rMBC. As new adjuvant therapies are introduced, the characteristics of the rMBC population are expected to change. For example, as practice has changed to incorporate HER2-targeted therapies in the adjuvant setting, most women with HER2-positive rMBC will have previously received HER2-targeted therapies. This contemporary population may be more resistant to further HER2-targeted therapy than a population such as assessed by Sundquist et al. (2017)³⁰ where many did not receive HER2-targeted therapy in the adjuvant setting. The lack of survival improvement over time for older women will help inform decisions and timing on palliative care and appropriate accessible services. For clinical trial design, our findings establish the importance of stratifying randomisation by de novo/recurrent MBC status and by age group; and the need for more trials in older women to identify optimal treatment strategies.

For research implications, given rMBC represents the large majority of MBC diagnoses and thus clinical load, our finding that most population-based studies of MBC assess dnMBC alone represents an important evidence gap, and highlights the limitations of current reporting of MBC data by cancer registries. Further research is needed to understand the extent new adjuvant therapies may shift the biological profile of recurrent MBC toward more treatment resistant disease. Most MBC trials of new therapies are conducted before the treatment is tested in the adjuvant setting, thus observational data are essential to assess how new adjuvant therapies change the characteristics of rMBC. To provide high quality evidence for these investigations, we advocate for cancer registries to include and validate notifications of distant metastasis in women with an initial diagnosis of early BC. Routine reporting of rMBC by cancer registries with population-based record linkage to treatment data and death registrations would allow critical post-marketing surveillance of new therapies and outcomes in the real-world setting.

In conclusion, this review presents strong evidence of population-level improvements in OS and BCSS following dnMBC since 1995. In contrast, we found limited and conflicting evidence of changes in survival following rMBC, which represents the majority of new diagnoses of MBC. In addition, we identified no to modest survival improvement over time for women aged 70 years and older, who make up around a third of the MBC population. These findings establish the importance of the stratification of MBC by de novo/recurrent status and age group for clinical trial design; and suggest that routine reporting on distant metastases by cancer registries would support future research on the impact of new adjuvant therapies on the characteristics and prognosis of rMBC.

Declaration of interests

SJL received grant funding for research support from the National Health and Medical Research Foundation (NHMRC). BEK received funding for: advisory board participation (Roche and Gilead); an educational presentation (Novartis); meeting registration fees (Novartis). SAP received funding for post-market surveillance research (AbbVie). JB received funding for conference attendance (Novartis). NH received grant funding for research support from: the National Breast Cancer Foundation (NBCF) and NHMRC; and honoraria for journal editorial work (Elsevier).

All remaining authors declare no competing interests.