Articles

Long-term outcomes for neoadjuvant versus adjuvant

chemotherapy in early breast cancer: meta-analysis of
individual patient data from ten randomised trials
Early Breast Cancer Trialists’ Collaborative Group (EBCTCG)*

Summary
Background Neoadjuvant chemotherapy (NACT) for early breast cancer can make breast-conserving surgery more Lancet Oncol 2018; 19: 27–39
feasible and might be more likely to eradicate micrometastatic disease than might the same chemotherapy given after Published Online
surgery. We investigated the long-term benefits and risks of NACT and the influence of tumour characteristics on December 11, 2017
http://dx.doi.org/10.1016/
outcome with a collaborative meta-analysis of individual patient data from relevant randomised trials. S1470-2045(17)30777-5
See Comment page 2
Methods We obtained information about prerandomisation tumour characteristics, clinical tumour response, surgery,
*Full list of members in the
recurrence, and mortality for 4756 women in ten randomised trials in early breast cancer that began before 2005 and appendix (pp 19–24) or at
compared NACT with the same chemotherapy given postoperatively. Primary outcomes were tumour response, https://www.ctsu.ox.ac.uk/
extent of local therapy, local and distant recurrence, breast cancer death, and overall mortality. Analyses by research/ebctcg
intention-to-treat used standard regression (for response and frequency of breast-conserving therapy) and log-rank Correspondence to:
methods (for recurrence and mortality). EBCTCG Secretariat, Medical
Research Council Population
Health Research Unit, Nuffield
Findings Patients entered the trials from 1983 to 2002 and median follow-up was 9 years (IQR 5–14), with the last Department of Population
follow-up in 2013. Most chemotherapy was anthracycline based (3838 [81%] of 4756 women). More than two thirds Health, Oxford OX3 7LF, UK
(1349 [69%] of 1947) of women allocated NACT had a complete or partial clinical response. Patients allocated NACT bc.overview@ndph.ox.ac.uk

had an increased frequency of breast-conserving therapy (1504 [65%] of 2320 treated with NACT vs 1135 [49%] of See Online for appendix
2318 treated with adjuvant chemotherapy). NACT was associated with more frequent local recurrence than was
adjuvant chemotherapy: the 15 year local recurrence was 21·4% for NACT versus 15·9% for adjuvant chemotherapy
(5·5% increase [95% CI 2·4–8·6]; rate ratio 1·37 [95% CI 1·17–1·61]; p=0·0001). No significant difference between
NACT and adjuvant chemotherapy was noted for distant recurrence (15 year risk 38·2% for NACT vs 38·0% for
adjuvant chemotherapy; rate ratio 1·02 [95% CI 0·92–1·14]; p=0·66), breast cancer mortality (34·4% vs 33·7%; 1·06
[0·95–1·18]; p=0·31), or death from any cause (40·9% vs 41·2%; 1·04 [0·94–1·15]; p=0·45).

Interpretation Tumours downsized by NACT might have higher local recurrence after breast-conserving therapy than
might tumours of the same dimensions in women who have not received NACT. Strategies to mitigate the increased
local recurrence after breast-conserving therapy in tumours downsized by NACT should be considered—eg, careful
tumour localisation, detailed pathological assessment, and appropriate radiotherapy.

Funding Cancer Research UK, British Heart Foundation, UK Medical Research Council, and UK Department of
Health.

Copyright © The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license.

Introduction the local (and, by implication, disseminated) tumour to

Neoadjuvant chemotherapy (NACT)—ie, chemotherapy
begun before breast cancer surgery—was introduced in
the 1970s,1 aiming to downstage locally advanced
(inoperable) disease and make it operable. NACT was
subsequently extended to operable (early) breast cancer,
mainly to allow breast-conserving surgery, and is now
widely used, particularly for large tumours.2–4 Further­
more, NACT might be somewhat more likely to eradicate
micrometastatic disease than might chemotherapy
delayed until after surgery.
NACT might mitigate the hypo­ thesised stimulatory
effect of surgery on occult disease5 and reduce tumour
cell shedding during surgery. NACT might also provide
useful in-vivo information about the chemosensitivity of
different chemotherapy regimens, helping to guide
subsequent drug selection.6,7 Conversely, by delaying
surgery, NACT might increase the risk of metastatic
spread, particularly for chemoresistant tumours.
Several randomised trials8–17 have compared NACT
with the same chemotherapy given postoperatively.
Interpretation of these trials is complicated, however, as
the frequency of breast-conserving surgery often differed
between groups because of tumour shrinkage after
NACT. In certain trials,14,15 some good responders to
NACT did not receive surgery, and high frequencies of
local recurrence with NACT in these trials have been
attributed to omission of definitive local therapy. Any
such differences in the extent of surgery confound

www.thelancet.com/oncology Vol 19 January 2018 27

 Articles
For more on trial identification
and data checking see
https://www.ctsu.ox.ac.uk/
research/ebctcg/prisma-ipd-
statement-for-ebctcg.pdf
28 www.thelancet.com/oncology Vol 19 January 2018
Research in context
Evidence before this study
The Early Breast Cancer Trialists’ Collaborative Group’s ongoing
extensive searches of bibliographic databases, including
MEDLINE, Embase, the Cochrane Library, and meeting abstracts
up to March 2017, identified 16 trials that compared
neoadjuvant chemotherapy (NACT) with the same
chemotherapy postoperatively. Meta-analyses of published
reports indicate that NACT reduced the frequency of
mastectomy but did not affect mortality. Interpretation is
complicated, however, as the use of breast-conserving surgery
often differed between groups because of tumour shrinkage by
NACT. In certain trials, some patients with a good response did
not receive surgery. Hence, women allocated NACT retained
more breast tissue than did those allocated adjuvant
chemotherapy, and higher local recurrence frequencies in some
neoadjuvant trials than in others have been attributed to
omission of definitive local therapy.
Added value of this study
We did a meta-analysis of individual patient data from trials
that compared NACT with the same chemotherapy given
postoperatively. We assessed effects of patient and tumour
characteristics on tumour response, extent of local therapy,
local and distant recurrence, breast cancer death, and overall
mortality. This individual patient data meta-analysis, involving
comparisons of the efficacy of NACT with that of
adjuvant chemotherapy.18,19 Another complication is that
investigations of the influence of tumour characteristics
on outcome need to use prerandomisation data, as
analyses by postsurgical characteristics would be
substantially biased by downstaging.20 To investigate
such issues in more detail than was possible in reviews18,19
of published data, we did a patient-level meta-analysis of
the trials that directly compared any NACT regimen
with the same regimen begun postoperatively.
Methods
Study design and participants
We sought data from all randomised trials in early
(ie, operable) breast cancer that began before 2005 and
compared NACT with the same chemotherapy begun
after surgery (ie, standard adjuvant chemotherapy).
NACT always started before surgery, although in some
trials, some of the chemotherapy in the NACT group
was given postoperatively, whereas all chemotherapy in
the control group had to be postoperative (so trials such
as NSABP B-2721 were ineligible). Trial identification
and data checking were as reported previously22,23 and
conformed to the Preferred Reporting Items for
Systematic Review and Meta-Analyses (Individual
Patient Data).24 For every woman, we requested
information from the trial’s principal investigator or
another appropriate member of their research group
4756 women in ten trials, found that the frequencies of clinical
response and breast-conserving therapy were higher for smaller,
higher-grade, and oestrogen receptor-negative and
progesterone receptor-negative tumours, and for one trial using
anthracycline and taxane chemotherapy. Although responders
to NACT had lower distant recurrence and breast cancer
mortality than did non-responders, when responders and
non-responders were combined, distant recurrence and breast
cancer mortality were similar for NACT and adjuvant
chemotherapy. Local recurrence was, however, higher with
NACT than with adjuvant chemotherapy, which persisted for
10 years after treatment and was not confined to trials in which
surgery could be omitted after response to NACT.
Implications of all the available evidence
NACT is as effective as adjuvant chemotherapy in reducing the
risk of distant recurrence and death from breast cancer.
However, NACT is associated with higher local recurrence than
adjuvant chemotherapy, which could be at least partly
explained by wider use of breast-conserving therapy after NACT
than with postoperative chemotherapy. Strategies to mitigate
the increased local recurrence after breast-conserving therapy in
tumours downsized by NACT should be considered—eg, careful
tumour localisation, detailed pathological assessment, and
appropriate radiotherapy.
about patient and tumour characteristics, treatments,
dates of any local recurrence (breast, chest wall, or
regional nodes), distant recurrence, contralateral breast
or other second primary cancer, and date last known to
be alive or date and underlying cause of death. To avoid
bias, we sought data for tumour characteristics recorded
before randomisation since these characteristics can be
altered by neoadjuvant treatment. We also requested
tumour response after NACT (assessed mostly by
palpation and mammography). To investigate the
influence of NACT on extent of surgery, we sought
details of surgery planned at randomisation and surgery
actually done. When planned surgery was unknown, it
was inferred from clinical tumour size. Patient-level data
for radiotherapy were unavailable.
Statistical analysis
A detailed description of the statistical methods has been
previously published.22 Primary outcomes assessed were
tumour response (complete response [no clinical
evidence of disease after NACT], partial response
[≥50% reduction in initial size], or stable or progressive
disease [<50% reduction, no change, or increased tumour
size]), extent of local therapy (mastectomy, lumpectomy
[either with or without radiotherapy], and radiotherapy
alone), local and distant recurrence, breast cancer death
(via subtraction of the log-rank statistics of death without
recurrence from those of overall survival23), and overall
 Articles

mortality (death from any cause). Analyses were by
intention to treat and are of first isolated local recurrence
(site not generally available), any distant recurrence
(irrespective of previous local or contralateral recurrence),
breast cancer mortality, and all-cause mortality.
We treated deaths from new cancers of unknown primary
sites as breast cancer deaths. When no recurrence was
reported before breast cancer death, we assumed distant
recurrence to have just preceded it. We took deaths from
an unknown cause without recorded recurrence to be
non-breast cancer deaths. Comparisons of NACT
response and frequency of breast-conserving therapy
used regression models,25 accounting for tumour size
and trial. We stratified log-rank analyses by trial,
follow-up year, age at entry (<35 years, 35–44 years,
45–54 years, 55–69 years, and ≥70 years), and
prerandomisation clinical nodal status (N0 or other).
In such analyses, if a log-rank statistic (o−e) has
variance v, then, defining z=(o−e)/√v and b=(o−e)/v, the
event rate ratio (RR; NACT vs control) is estimated as
exp(b) with SE=(RR−1)/z. RRs and confidence limits
for RR are derived from those for b (by normal
approximations). To test for a trend between n strata
(eg, of age) in the effects of treatment, we supposed that
stratum number s (s=1,2,…,n) has log-rank statistics
(o–e) and v (with grand total over all strata O–E and V).
We defined m, the mean stratum number, to be the sum,
one term per stratum, of sv/V and define T to be the
sum, one term per stratum, of (s–m)(o–e).26 The variance
of T, var(T), is then the sum, one term per stratum, of
(s–m)²/v. The trend test statistic (ie, the change from one
stratum to the next in the log of the event RR) is then
T/var(T), which has variance 1/var(T). Tests of whether
two trends are the same involve subtraction of the
corresponding trend test statistics from each other.
A χ² statistic on one degree of freedom (χ²₁) for testing of
whether some quantity Q differs significantly from zero
is given by Q/var(Q). A χ² test (on n–1 degrees of
freedom) for heterogeneity can be obtained by subtracting
(O–E)²/V from the sum of the separate values, one per
stratum, of (o–e)²/v. For analyses by regression, we
estimated RRs by maximum likelihood; tests for trend
and heterogeneity were by likelihood ratio.
Associations between baseline variables and outcome
used prerandomisation values. Only two trials provided
pathological response data, so correlations of charac­
teristics with response to NACT use clinical response
data (available for eight of ten trials). Subgroup analyses
compare outcomes in trials in which all women
allocated NACT were, or were not, scheduled to receive
breast surgery and in women whose initially planned
local treatment was mastectomy or breast-conserving
therapy (lumpectomy with or without radiotherapy or
radiotherapy alone). Sensitivity analyses assess the
potential effect27 on local recurrence of competing
events (distant recurrence and death without
recurrence) and of omission of trials with only first
recurrences recorded. p values of 0·05 or less are
described as significant. Analyses used Stata 13.1 and
R 2.13.2.
Data sharing
Procedures for data access are available online. For data access procedures see
https://www.ndph.ox.ac.uk/
about/data-access-policy
Role of the funding source
The funders of the study had no role in study design,
data analysis, data interpretation, or writing of the
report. The secretariat had full access to all the data in
the study. The writing committee had final responsibility
for the decision to submit for publication.
Results
Individual patient data were available from ten8–17 of
16 eligible trials identified and from 4756 (91%) of the
5250 women in total (table 1, appendix pp 2, 6, 17). Trial
entry year for participants was 1983–2002, median
follow-up was 9 years (IQR 5–14), with the last follow-up
in 2013, and median age was 49 years (43–57). 1604 deaths
occurred, including 248 (15%) without recurrence. Of the
4756 women included in the analysis, 3838 (81%) were in
trials of regimens that included an anthracycline, one of
which (902 women) also gave a taxane.13 Four trials
(918 women) used MMM (mitoxantrone, methotrexate,
and mitomycin-C)11,12 or CMF (cyclophosphamide,
methotrexate, and fluorouracil)8,17 as NACT; in these
trials, some chemotherapy in those allocated NACT was

Trials Women (n) Deaths (n)† Median years per woman Woman-years by years since entry
(n)* (IQR) (thousands)
<10 10–19 ≥20 Total
No anthracycline or taxane8,11,12,17‡ 4 918 315 7·0 (4·2–9·3) 6·0 0·8 0·2 7·0
Anthracycline, no taxane9,10,14–16 5 2936 1163 10·2 (4·9–15·4) 22·1 7·7 <0·1 29·8
Anthracycline and taxane13 1 902 126 7·9 (5·0–10·7) 6·5 0·5 0 7·0
Total 10 4756 1604 8·6 (4·8–13·7) 34·6 9·0 0·2 43·7

*Data are missing for six small trials that randomised about 500 women, so they were not included in this analysis (appendix p 17). †Includes 1356 deaths with recurrence,
72 of unknown cause without recurrence, and 176 of known cause without recurrence. ‡In these trials, women allocated to the neoadjuvant group completed their
chemotherapy after surgery.

Table 1: Trials of neoadjuvant versus adjuvant chemotherapy that began by 2005

www.thelancet.com/oncology Vol 19 January 2018 29

 Articles

given after surgery (table 1, appendix pp 3–4). No patients RR 1·28 [95% CI 1·06–1·55], p=0·01; figure 3A). However,
received trastuzumab. the RRs for local recurrence in these two sets of trials
Across all trials, NACT was associated with substantial were not significantly different (heterogeneity p=0·19).
tumour response (table 2), moderately increased use of Between-trial RRs for local recurrence ranged from
breast-conserving therapy in the NACT group compared 0·67 (95% CI 0·24–1·91) to 4·59 (1·19–17·8), but this
with the adjuvant chemotherapy group (figure 1), and an apparent heterogeneity was not significant (χ²10=11·8;
absolute increase in 15 year local recurrence of p=0·30; figure 4A). RRs for local recurrence also did
5·5% (95% CI 2·4–8·6; 21·4% for NACT vs 15·9% for not differ significantly between the three classes of
adjuvant chemotherapy), corresponding to a RR of chemotherapy used in these trials (figure 4, appendix
1·37 (95% CI 1·17–1·61; p=0·0001; figure 2A). The p 13), between trials in which chemotherapy in the NACT
incidence of local recurrence was significantly higher group was or was not completed after local therapy
with NACT than with adjuvant chemotherapy in years (appendix p 13), or between use or not of tamoxifen
0–4 (RR 1·35 [95% CI 1·11–1·64]; p=0·003) and (figure 5, appendix p 13).
5–9 (1·53 [1·08–2·17]; p=0·02), with few local recurrences We noted no significant differences between NACT and
after year 10. Sensitivity analyses indicated no substantial adjuvant treatment in 15 year distant recurrence (38·2% for
influence of competing risks from other breast events on NACT vs 38·0% for adjuvant chemotherapy; RR 1·02
the RRs for local recurrence (appendix p 18). [95% CI 0·92–1·14]; p=0·66), breast cancer death (34·4%
As anticipated,18,19 the absolute increase in 10-year local vs 33·7%; 1·06 [0·95–1·18]; p=0·31), or death from any
recurrence with NACT was largest in the two trials14,15 in cause (40·9% vs 41·2%; 1·04 [0·94–1·15]; p=0·45;
which, after NACT, many women did not have breast figure 2B, C, D). The RRs for these three outcomes did not
surgery (13·3% [95% CI 5·5–21·1]; 33·7% for NACT vs differ significantly between any subgroups of trials,
20·4% for adjuvant chemotherapy; RR 1·62 [95% CI including those for which use of surgery was or was not
1·20–2·19], p=0·002; figure 3B). In the other eight dependent on response to NACT, those using different
trials,8–13,16,17 surgery was scheduled irrespective of response types of chemotherapy, or those using or not using
to NACT, and the absolute increase in 10 year local tamoxifen (figure 3 C–F and figure 4B, appendix pp 7, 13).
recurrence was 3·2% (95% CI 0·6–5·8; 15·1% vs 11·9%; Three trials8,9,16 collected only first recurrence and death
rather than all events; however, sensitivity analyses
omitting these trials had no material effect on distant
Clinical response
recurrence estimates (appendix p 18). Mortality from
Complete* Partial† Stable or Unknown Total
causes other than breast cancer was no different between
progressive
disease‡ the NACT and adjuvant chemotherapy groups
(appendix p 7).
Planned breast-conserving therapy
Information about clinical tumour response was
Breast-conserving 215 (96%) 256 (90%) 119 (77%) 211 (81%) 801 (87%)
available for 1947 (82%) of 2387 patients allocated
Mastectomy 10 (4%) 30 (10%) 35 (23%) 48 (19%) 123 (13%)
NACT; 546 (28%) of 1947 had a complete response,
Unknown 0 0 0 2 (NA) 2 (NA)
803 (41%) of 1947 had a partial response, and
Total response§ 225/665 (34%) 286/665 (43%) 154/665 (23%) 261 (NA) 926 (100%)
598 (31%) of 1947 had stable or progressive disease
Planned mastectomy
(table 2, appendix p 8). The clinical tumour response to
Breast-conserving 75 (60%) 121 (41%) 30 (12%) 26 (36%) 252 (33%)
NACT affected surgical treatment decisions: more
Mastectomy 49 (40%) 175 (59%) 231 (88%) 47 (64%) 502 (67%)
women with a complete response had breast-conserving
Unknown 0 1 (NA) 2 (NA) 11 (NA) 14 (NA)
therapy (452 [83%] of 544) than did those with a partial
Total response§ 124/684 (18%) 297/684 (43%) 263/684 (38%) 84 (NA) 768 (100%)
response (541 [68%] of 799) or no response
Unknown planned therapy (246 [42%] of 588). Consequently, we noted an
Breast-conserving 162 (83%) 164 (76%) 97 (56%) 28 (49%) 451 (70%) imbalance by treatment group in the extent of surgery:
Mastectomy 33 (17%) 53 (24%) 76 (44%) 29 (51%) 191 (30%) although breast-conserving therapy was initially
Unknown 2 (NA) 3 (NA) 8 (NA) 38 (NA) 51 (NA) intended for equal numbers of patients in each group,
Total response§ 197/598 (33%) 220/598 (37%) 181/598 (30%) 95 (NA) 693 (100%) actual use of breast-preserving therapy (including no
All women surgery) was 1504 (65%) of 2320 in the NACT group
Breast-conserving 452 (83%) 541 (68%) 246 (42%) 265 (68%) 1504 (65%) versus 1135 (49%) of 2318 in the adjuvant chemotherapy
Mastectomy 92 (17%) 258 (32%) 342 (58%) 124 (32%) 816 (35%) group excluding patients with unknown surgeries
Unknown 2 (NA) 4 (NA) 10 (NA) 51 (NA) 67 (NA) (p<0·0001; figure 1, appendix pp 9, 10).
Total response§ 546/1947 (28%) 803/1947 (41%) 598/1947 (31%) 440 (NA) 2387 (100%) Figure 6 shows proportions of women with complete
clinical response according to patient and tumour
Data are n (%) or n/N (%). NA=not applicable. *No clinical evidence of disease. †≥50% reduction in tumour size.
‡<50% reduction or increase in tumour size. §Percentages are of those with a known response. characteristics. Complete response decreased with
increasing clinical tumour size (trend p<0·0001) and was
Table 2: Local therapy, planned versus done, in women allocated to neoadjuvant chemotherapy, by
higher with oestrogen receptor (ER)-negative biopsies
clinical response
than with ER-positive biopsies (p<0·0001) and with

30 www.thelancet.com/oncology Vol 19 January 2018

 Articles

Number given BCT/women Ratio of BCT rates neoadjuvant:adjuvant Rate ratio of BCT
(95% CI)
Allocated Allocated
neoadjuvant adjuvant
Age at entry (years) (χ21=2·8; p=0·09)
<45 453/712 383/761 1·20 (1·11−1·30)
45−54 539/822 415/817 1·22 (1·13−1·31)
≥55 512/786 337/740 1·38 (1·27−1·51)
Clinical nodal status (χ21=6·3; p=0·01)
Negative 917/1338 730/1307 1·19 (1·12−1·25)
Positive 461/776 342/793 1·36 (1·24−1·49)
Unknown 126/206 63/218 2·11 (1·67−2·66)
Clinical tumour size (χ =43·1; p<0·0001)
2
1

1−19 mm 301/382 299/402 1·04 (0·97−1·12)

20−49 mm 1061/1547 746/1551 1·41 (1·33−1·50)
≥50 mm 124/337 71/323 1·68 (1·31−2·15)
Unknown 18/54 19/42 0·71 (0·43−1·15)
Biopsy ER and PR status (χ22=4·5; p=0·10)
ER+ and PR+ 226/343 150/345 1·51 (1·31−1·73)
ER+ and PR− 54/91 25/98 2·28 (1·57−3·31)
ER− and PR− 278/407 165/406 1·61 (1·42−1·84)
Unknown 946/1479 795/1469 1·13 (1·08−1·20)
Biopsy grade (χ21=2·5; p=0·12)
Well 32/66 17/60 1·67 (1·05−2·67)
Moderate 186/303 84/317 2·32 (1·90−2·84)
Poor 117/189 46/195 2·66 (2·02−3·50)
Unknown 1169/1762 988/1746 1·13 (1·08−1·18)
Biopsy grade and ER status (χ23=13·7; p=0·003)
Well or moderate, ER+ 141/237 77/234 1·78 (1·45−2·19)
Poor, ER+ 45/84 22/92 2·30 (1·53−3·47)
Well or moderate, ER− 76/128 21/138 3·86 (2·55−5·83)
Poor, ER− 71/104 24/103 2·85 (1·97−4·11)
Other or unknown 1171/1767 991/1751 1·15 (1·10−1·21)
Type of neoadjuvant chemotherapy (χ21=44·9; p<0·0001)
No anthracycline or taxane 329/446 292/439 1·07 (0·99−1·16)
Anthracycline, no taxane 891/1436 696/1438 1·24 (1·16−1·31)
Anthracycline and taxane 284/438 147/441 1·95 (1·68−2·25)
Planned local therapy (χ22=121·7; p<0·0001)
Radiotherapy only 167/200 144/190 1·10 (1·00−1·22)
Lumpectomy 634/724 623/735 1·03 (0·99−1·08)
Mastectomy 252/754 73/757 3·48 (2·74−4·43)
Unknown 451/642 295/636 1·53 (1·38−1·68)
Total 1504/2320 (64·8%) 1135/2318 (49·0%) 1·28 (1·22−1·34)

0·2 1·0 4·0

Lower frequency with NACT Higher frequency with NACT

Figure 1: BCT rate ratios

Numbers with BCT or mastectomy after chemotherapy. Excludes local therapy unknown (67 patients with NACT and 51 with adjuvant chemotherapies).
BCT=breast-conserving therapy. ER=oestrogen receptor. PR=progesterone receptor. NACT=neoadjuvant chemotherapy.

poorly differentiated tumours than with well or
moderately differentiated tumours (trend p=0·001, even
after allowance for high-grade tumours tending to be ER
negative), and was higher in the one trial13 that combined
anthracycline and taxane therapy than in the other trials
(p<0·0001). Age, nodal status, and planned local therapy
did not affect response.
The proportion of women having breast-conserving
therapy in various different subgroups in the NACT and
adjuvant chemotherapy groups are shown in figure 1.
The strongest predictors of the effect of NACT on breast
conservation frequency were tumour size, planned local
therapy, and type of chemotherapy (all p<0·0001). The
effect of NACT on surgery de-escalation was most
apparent among women with large (20–49 mm or
≥50 mm) tumours; we noted little effect of NACT on
breast conservation frequency in women with small
(<20 mm) tumours. As expected, women with
mastectomy originally planned were more likely to have
lesser surgery than were those with breast-conserving

www.thelancet.com/oncology Vol 19 January 2018 31

 Articles

A B
60 4756 women, 635 events 60 4756 women, 1562 events
15 year loss 5·5% (95% CI 2·4−8·6) 15 year loss 0·2% (95% CI −3·1 to 3·5)
RR 1·37 (95% CI 1·17−1·61) RR 1·02 (95% CI 0·92−1·14)
Log-rank p=0·0001 Log-rank p=0·66
50 50
NACT
Adjuvant Neoadjuvant

Distant recurrence at any time (%)

38·2%
40 40
33·8%
Local recurrence (%)

Adjuvant
38·0%
30 30 32·4%
Neoadjuvant 24·9%
21·4%
17·9%
20 20 23·5%

12·1%
Adjuvant
10 13·2% 15·9% 10
9·0%

0 0
0 5 10 15 0 5 10 15
Local recurrence crude rates (events per woman-years) Distant recurrence at any time crude rates (events per woman-years)
and log-rank analyses and log-rank analyses
Years 0−4 Years 5−9 Years 10−14 Years ≥15 Years 0−4 Years 5−9 Years 10−14 Years ≥15
Neoadjuvant 2·58 (245/9493) 1·43 (79/5528) 0·93 (26/2784) 2·16 (16/740) 5·69 (568/9983) 2·58 (162/6291) 1·49 (50/3351) 1·44 (14/974)
Adjuvant 1·95 (185/9477) 0·96 (54/5618) 0·69 (19/2769) 1·42 (11/772) 5·44 (535/9840) 2·54 (157/6187) 1·84 (60/3270) 1·74 (16/919)
Rate ratio 1·35 (1·11−1·64) 1·53 (1·08−2·17) 1·29 (0·70−2·38) 1·11 (0·48−2·57) 1·07 (0·94−1·21) 0·99 (0·79−1·24) 0·80 (0·55−1·18) 0·75 (0·35−1·61)
(95% CI) from 30·4/102·0 13·6/31·8 2·7/10·3 0·6/5·4 16·5/251·5 −0·6/75·5 −5·7/25·8 −1·9/6·7
(O−E)/V

C D
60 4756 women, 1329 deaths 60 4756 women, 1604 deaths
15 year loss 0·7% (95% CI −2·7 to 4·1) 15 year gain 0·3% (95% CI −3·2 to 3·8)
RR 1·06 (95% CI 0·95−1·18) RR 1·04 (95% CI 0·94−1·15)
Log-rank p=0·31 Log-rank p=0·45
50 50
Adjuvant
41·2%
Neoadjuvant
40 40
Breast cancer mortality (%)

34·4%
All-cause mortality (%)

31·9% Neoadjuvant
28·5% 40·9%
30 Adjuvant 30
33·7% 31·0%

26·6% 19·4%
18·1%
20 20

17·8%
16·2%
10 10

0 0
5 0 10 15 0 5 10 15
Time since trial entry (years) Time since trial entry (years)
Breast cancer mortality crude rates (events per woman-years) Any death crude rates (events per woman-years)
and log-rank analyses and log-rank analyses
Years 0−4 Years 5−9 Years 10−14 Years ≥15 Years 0−4 Years 5−9 Years 10−14 Years ≥15
Neoadjuvant 3·90 (412/10 567) 2·82 (191/6785) 1·93 (69/3570) 1·24 (13/1050) 4·22 (446/10 567) 3·51 (238/6785) 2·91 (104/3570) 3·52 (37/1050)
Adjuvant 3·49 (364/10 432) 2·81 (190/6771) 2·19 (78/3559) 1·18 (12/1014) 3·86 (403/10 432) 3·56 (241/6771) 3·20 (114/3559) 2·07 (21/1014)
Rate ratio 1·12 (0·97−1·30) 1·03 (0·84−1·27) 0·88 (0·63−1·21) 0·90 (0·41−1·97) 1·09 (0·95−1·25) 0·98 (0·81−1·17) 0·90 (0·68−1·18) 1·69 (0·95−2·99)
(95% CI) from 20·5/179·6 2·8/91·6 −4·8/36·6 −0·7/6·2 16·7/196·6 −2·7/112·2 −5·6/51·3 6·1/11·7
(O−E)/V

Figure 2: Effect of neoadjuvant versus adjuvant chemotherapy on recurrence and mortality

Local recurrence (A), distant recurrence (B), breast cancer mortality (C), and death from any cause (D). Three trials recorded causes of any deaths but only the first
breast cancer event. Hence, for these trials, distant recurrence includes the first distant recurrence as the first event and death from breast cancer. Error bars are
95% CIs. NACT=neoadjuvant chemotherapy. O–E=observed minus expected. RR=rate ratio. V=variance of O–E.

surgery originally planned. NACT with an anthracycline with other regimens. In women with node-positive
and taxane combination was also associated with disease, the rate ratio for BCT was higher than for those
substantially more surgery de-escalation than was NACT with node-negative disease (p=0·01). Despite the high

32 www.thelancet.com/oncology Vol 19 January 2018

 Articles

A Surgery commonly used B Surgery less commonly used*

60 4094 women, 453 events 662 women, 182 events
10 year loss 3·2% (95% CI 0·6−5·8) 10 year loss 13·3% (95% CI 5·5−21·1)
RR 1·28 (95% CI 1·06−1·55) RR 1·62 (95% CI 1·20−2·19)
50
Log-rank p=0·010 Log-rank p=0·002
Neoadjuvant
NACT 33·7%
40
Local recurrence (%)

Adjuvant
24·5%
30

Neoadjuvant
20 15·1%
10·0% Adjuvant
10 Adjuvant 15·6% 20·4%
8·0% 11·9%
0
0 5 10 0 5 10
Local recurrence crude rates (events per woman-years) Local recurrence crude rates (events per woman-years)
and log-rank analyses and log-rank analyses
Years 0−4 Years 5−9 Years ≥10 Years 0−4 Years 5−9 Years ≥10
Neoadjuvant 2·12 (173/8164) 1·21 (57/4721) 0·85 (22/2577) 5·42 (72/1329) 2·73 (22/806) 2·11 (20/947)
Adjuvant 1·73 (142/8200) 0·94 (45/4778) 0·55 (14/2547) 3·37 (43/1277) 1·07 (9/840) 1·61 (16/994)
Rate ratio (95% CI) 1·26 (1·00−1·58) 1·35 (0·91−2·01) 1·32 (0·65−2·69) 1·63 (1·12−2·39) 2·30 (1·13−4·70) 1·15 (0·58−2·27)
from (O−E)/V 17·3/75·2 7·3/24·3 2·1/7·5 13·2/26·8 6·3/7·5 1·1/8·2
C Surgery commonly used D Surgery less commonly used*
60 4094 women, 1238 events 662 women, 324 events
10 year loss 1·8% (95% CI −1·4 to 5·0) 10 year gain 1·5% (95% CI −6·4 to 9·4)
RR 1·05 (95% CI 0·94−1·18) Adjuvant
RR 0·92 (95% CI 0·73−1·15)
50 43·5%
Distant recurrence at any time (%)

Log-rank p=0·38 Log-rank p=0·46

40 Neoadjuvant 32·1%
32·2%
Neoadjuvant
30 42·0%
24·1%
Adjuvant
30·4% 29·5%
20
22·0%
Figure 3: Time to recurrence
10 and breast cancer mortality
Local recurrence for surgery
0 commonly used (A) and less
0 5 10 0 5 10 commonly used (B), distant
Distant recurrence at any time crude rates (events per woman-years) Distant recurrence at any time crude rates (events per recurrence for surgery
and log-rank analyses woman-years) and log-rank analyses
commonly used (C) and less
Years 0−4 Years 5−9 Years ≥10 Years 0−4 Years 5−9 Years ≥10
Neoadjuvant 5·50 (470/8545) 2·29 (121/5284) 1·30 (40/3078) 6·81 (98/1438) 4·07 (41/1007) 1·92 (24/1248) commonly used (D), and
Adjuvant 5·07 (431/8496) 2·32 (122/5262) 1·77 (54/3050) 7·74 (104/1344) 3·78 (35/926) 1·93 (22/1139) breast cancer mortality for
Rate ratio (95% CI) 1·12 (0·97−1·28) 0·98 (0·75−1·26) 0·74 (0·49−1·13) 0·87 (0·65−1·17) 1·05 (0·66−1·68) 0·91 (0·50−1·66) surgery commonly used (E)
from (O−E)/V 22·8/205·9 −1·4/58·0 −6·6/21·8 –6·3/45·6 0·8/17·5 –1·0/10·7 and less commonly used (F).
Heterogeneity by surgery use:
E Surgery commonly used FF Surgery less commonly used*
local recurrence p=0·19,
60 4094 women, 1055 deaths 662 women, 274 deaths
distant recurrence p=0·29, and
10 year loss 2·7% (95% CI −0·3 to 5·7) 10 year gain 2·1% (95% CI −5·5 to 9·7)
RR 1·09 (95% CI 0·97−1·24) RR 0·93 (95% CI 0·73−1·19) breast cancer mortality
50 Log-rank p=0·15 Log-rank p=0·56 p=0·24. Error bars are 95% CIs.
Adjuvant
NACT=neoadjuvant
Breast cancer mortality (%)

35·4%
40 chemotherapy. O–E=observed
Neoadjuvant minus expected. RR=rate ratio.
27·5% V=variance of O–E. Three trials
30
19·7% recorded causes of any deaths
Neoadjuvant but only the first breast cancer
18·3% 33·3%
20 Adjuvant event. Hence, for these trials,
24·8% distant recurrence includes the
10 15·6% 16·6% first distant recurrence as the
first event and death from
breast cancer. *Includes
0
5 0 10 0 5 10 Institut Bergonié Bordeaux14
Time since trial entry (years) Time since trial entry (years) (in NACT group, 33% had
Breast cancer mortality crude rates (events per woman-years) Breast cancer mortality crude rates (events per woman-years) radiotherapy alone) and
and log-rank analyses and log-rank analyses Institut Curie S615 (in NACT
Years 0−4 Years 5−9 Years ≥10 Years 0−4 Years 5−9 Years ≥10 group, 51% had radiotherapy
Neoadjuvant 3·95 (356/9007) 2·48 (139/5610) 1·65 (53/3206) 3·59 (56/1560) 4·43 (52/1175) 2·05 (29/1414) alone; in adjuvant
Adjuvant 3·37 (301/8944) 2·49 (141/5662) 2·01 (65/3237) 4·23 (63/1488) 4·42 (49/1109) 1·87 (25/1336) chemotherapy group,
Rate ratio (95% CI) 1·19 (1·01−1·39) 1·03 (0·81−1·31) 0·82 (0·57−1·17) 0·81 (0·56−1·18) 1·03 (0·69−1·53) 1·03 (0·60−1·79)
from (O−E)/V
46% had radiotherapy alone)
26·4/152·1 2·1/67·4 −6·0/30·0 −5·8/27·6 0·7/24·2 0·4/12·9
trials.

www.thelancet.com/oncology Vol 19 January 2018 33

 Articles

A Events/women Neoadjuvant events Ratio of annual event rates neoadjuvant:adjuvant

Preoperative vs postoperative Allocated Allocated Log-rank Variance
chemotherapy* neoadjuvant adjuvant O−E of O−E
Trials where preoperative chemotherapy included neither anthracyclines nor taxanes
BCCA Vancouver17 1C600M40F600 10/42 3/39 3·2 2·1
St George’s London¹² 4Mz7M35Mit7 11/53 8/50 1·5 4·3
RMH London 11 4Mz7M35Mit7 7/155 8/150 −1·4 3·5
Austrian BCSG VII⁸ 3C600×2M40×2F600×2 16/142 12/137 1·6 6·8
Austrian BCSG VII⁸ 3C600×2M40×2F600×2 11/74 4/76 3·8 3·5
Subtotal 55/466 (11·8%) 35/452 (7·7%) 8·8 20·2
1·55 (1·00−2·39); p=0·05

Trials using anthracyclines but not taxanes

IB Bordeaux14 3E50Vc1000M20; 3Mit10Tt20Vd4000 39/134 18/138 11·1 13·6
Institut Curie S615 4F500A25C500 75/200 50/190 9·5 28·9
NSABP B-189 4A60C600 124/760 105/763 9·0 55·1
NCI Bethesda16 5F400×3Fol500×3A15×3C600 2/26 1/27 0·1 0·5
EORTC 1090210 4F600E60C600 33/350 30/348 3·9 14·9
Subtotal 273/1470 (18·6%) 204/1466 (13·9%) 33·5 113·0
1·35 (1·12−1·62); p=0·002
Trials using anthracyclines and taxanes
ECTO Italy13 4E60P200; 4C600×2M40×2F600×2 38/451 30/451 4·9 16·4
Subtotal 38/451 (8·4%) 30/451 (6·7%) 4·9 16·4
1·35 (0·83−2·19); p=0·22
Total 366/2387 (15·3%) 269/2369 (11·4%) 47·2 149·5
99% CI 95% CI 1·37 (1·17−1·61); p=0·0001
Heterogeneity between three subtotals: χ22=0·3; p=0·85
Heterogeneity within subtotals: χ28=11·5; p=0·18
Heterogeneity between 11† trials: χ210=11·8; p=0·30

B
Trials where preoperative chemotherapy included neither anthracyclines nor taxanes
BCCA Vancouver¹⁷ 1C600M40F600 20/42 20/39 0·4 8·0
St George’s London¹² 4Mz7M35Mit7 19/53 17/50 0·6 8·0
RMH London 11 4Mz7M35Mit7 37/155 41/150 −5·0 17·8
Austrian BCSG VII⁸ 3C600×2M40×2F600×2 37/142 25/137 5·6 14·3
Austrian BCSG VII⁸ 3C600×2M40×2F600×2 23/74 23/76 1·2 10·3
Subtotal 136/466 (29·2%) 126/452 (27·9%) 2·8 58·4
1·05 (0·81−1·36); p=0·72

Trials using anthracyclines but not taxanes

IB Bordeaux14 3E50Vc1000M20; 3Mit10Tt20Vd4000 66/134 65/138 0·7 30·7
Institut Curie S615 4F500A25C500 97/200 96/190 −7·1 43·1
NSABP B-189 4A60C600 256/760 264/763 −5·4 122·7
NCI Bethesda16 5F400×3Fol500×3A15×3C600 3/26 7/27 −1·5 2·1
EORTC 1090210 4F600E60C600 139/350 122/348 11·1 59·3
Subtotal 561/1470 (38·2%) 554/1466 (37·8%) −2·2 258·0
0·99 (0·88−1·12); p=0·89

Trials using anthracyclines and taxanes

ECTO Italy13 4E60P200; 4C600×2M40×2F600×2 97/451 88/451 7·8 43·2
Subtotal 97/451 (21·5%) 88/451 (19·5%) 7·8 43·2
1·20 (0·89−1·62); p=0·23

Total 794/2387 (33·3%) 768/2369 (32·4%) 8·4 359·5

1·02 (0·92−1·14); p=0·66
99% CI 95% CI
Heterogeneity between three subtotals: χ =1·4; p=0·50
2
2 0·5 1·0 2·5
Heterogeneity within subtotals: χ28=8·2; p=0·42
Neoadjuvant better Adjuvant better
Heterogeneity between 11† trials: χ210=9·5; p=0·48

Figure 4: Rate ratios for the effect of neoadjuvant versus adjuvant chemotherapy on recurrence by trial
(A) Local recurrence. (B) Distant recurrence. Three trials recorded causes of any deaths but only the first breast cancer event. Hence, for these trials, distant recurrence includes the first distant recurrence
as the first event and death from breast cancer. The appendix (pp 3–4) contains a full description of each trial’s chemotherapy regimen. A=doxorubicin (adriamycin). BCCA=British Columbia Cancer
Agency. BCSG=Breast Cancer Study Group. C=cyclophosphamide. E=epirubicin. ECTO=European Cooperative Trial in Operable Breast Cancer. EORTC=European Organisation for Research and Treatment
of Cancer. F=fluorouracil. Fol=folinic acid. IB=Institut Bergonié. M=methotrexate. Mit=mitomycin-C. Mz=mitoxantrone. NCI=National Cancer Institute. NSABP=National Surgical Adjuvant Breast and
Bowel Project. O–E=observed minus expected. P=paclitaxel. Tt=thiotepa. Vc=vincristine. Vd=vindesine. *Chemotherapy regimens given preoperatively in those allocated neoadjuvant and
postoperatively in those allocated adjuvant chemotherapy. The number of cycles, agents, and drug doses (in mg/m) per cycle are given. †The Austrian BCSG VII trial8 has two entries to take into
account the two postoperative chemotherapies given to both randomised groups (appendix pp 3–4).

34 www.thelancet.com/oncology Vol 19 January 2018

 Articles

frequency of clinical response in patients with
ER-negative and poorly differentiated tumours, ER status
and tumour grade were not associated with frequency of
breast-conserving therapy, although after accounting for
grade, ER-negative women did appear to have higher
breast-conserving frequencies. Age was not associated
with the freqeuncy of breast-conserving therapy.
If the increased local recurrence in the NACT groups
(figure 3) is due to de-escalation of local therapy from
mastectomy to breast-conserving therapy, the RR for the

Events/women Ratio of annual event rates Rate ratio

neoadjuvant:adjuvant (95% CI)
Allocated Allocated
neoadjuvant adjuvant

Age at entry (years) (χ21=0·2; p=0·68)

<45 150/729 108/785 1·49 (1·16−1·92)
45−54 127/849 101/830 1·23 (0·94−1·60)
≥55 89/809 60/754 1·41 (1·02−1·95)
Clinical nodal status (χ21=0·2; p=0·66)
Negative 172/1349 135/1314 1·26 (1·00−1·58)
Positive 136/796 104/801 1·36 (1·05−1·76)
Unknown 58/242 30/254 1·95 (1·26−3·03)
Clinical tumour size (χ21=0·2; p=0·62)
1−19 mm 60/384 39/405 1·62 (1·06−2·48)
20−49 mm 233/1576 177/1559 1·28 (1·05−1·56)
≥50 mm 67/345 49/327 1·38 (0·92−2·07)
Unknown 6/82 4/78 1·25 (0·16−9·86)
Biopsy ER and PR status (χ22=0·1; p=0·96)
ER+ and PR+ 60/351 37/347 1·72 (1·14−2·59)
ER+ and PR− 20/100 16/98 1·72 (0·81−3·65)
ER− and PR− 76/425 41/413 1·85 (1·25−2·72)
Unknown 210/1511 175/1511 1·20 (0·98−1·48)
Biopsy grade (χ21=3·9; p=0·05)
Well 5/67 1/60 2·52 (0·45−14·04)
Moderate 54/312 26/321 2·16 (1·37−3·40)
Poor 15/198 20/196 0·90 (0·44−1·83)
Unknown 292/1810 222/1792 1·30 (1·09−1·55)
Biopsy grade and ER status (χ21=0·1; p=0·76)
Well or moderate, ER+ 27/244 14/235 1·98 (1·04−3·78)
Poor, ER+ 8/89 10/92 0·92 (0·34−2·53)
Well or moderate, ER− 31/131 13/141 2·58 (1·38−4·84)
Poor, ER− 7/108 10/104 0·97 (0·35−2·73)
Other or unknown 293/1815 222/1797 1·31 (1·09−1·56)
Type of neoadjuvant chemotherapy (χ22=0·3; p=0·84)
No anthracycline or taxane 55/466 35/452 1·55 (1·00−2·39)
Anthracycline, no taxane 273/1470 204/1466 1·35 (1·12−1·62)
Anthracycline and taxane 38/451 30/451 1·35 (0·83−2·19)
Planned local therapy (χ22=3·3; p=0·20)*
Radiotherapy only† 75/200 50/190 1·39 (0·96−2·00)
Lumpectomy 108/726 96/737 1·14 (0·86−1·52)
Mastectomy 125/768 79/762 1·66 (1·24−2·21)
Unknown 58/693 44/680 1·26 (0·84−1·90)
Period of follow-up (woman-years) (χ21=0·1; p=0·72)
Years 0–1 124/4421 97/4371 1·26 (0·96−1·65)
Years 2–4 121/5072 88/5106 1·45 (1·10−1·91)
Years 5–9 79/5528 54/5618 1·53 (1·08−2·17)
Years ≥10 42/3524 30/3542 1·22 (0·75−2·00)
Total 366/2387 (15·3%) 269/2369 (11·4%) 1·37 (1·17−1·61)
p=0·0001

0·5 1·0 2·5

Neoadjuvant better Adjuvant better

Figure 5: Local recurrence rate ratios

For lumpectomy versus mastectomy, χ²1=3·3; p=0·07. ER=oestrogen receptor. PR=progesterone receptor. *408 women with missing data had planned local therapy
imputed (appendix p 9). †Refers to Institut Curie S615 (appendix p 9).

www.thelancet.com/oncology Vol 19 January 2018 35

 Articles

% (CR/total) CR rate ratio Rate ratio for clinical

CR (95% CI)
Age at entry (years) (χ21=0·7; p=0·40)
<45 29·8 (164/550) 1·04 (0·90−1·21)
45−54 29·0 (196/675) 1·03 (0·94−1·13)
≥55 25·8 (186/722) 0·90 (0·78−1·03)
Clinical nodal status (χ21=0·8; p=0·37)
Negative 28·6 (326/1138) 0·97 (0·88−1·07)
Positive 27·2 (168/617) 1·04 (0·93−1·16)
Unknown 27·1 (52/192) 1·14 (0·57−2·26)
Clinical tumour size (χ21=37·5; p<0·0001)
1−19 mm 34·6 (117/338) 1·27 (1·09−1·48)
20−49 mm 29·7 (390/1312) 0·99 (0·91−1·07)
≥50 mm 13·3 (35/264) 0·47 (0·35−0·64)
Unknown 12·1 (4/33) 0·66 (0·28−1·58)
Biopsy ER and PR status (χ22=21·8; p<0·0001)
ER+ and PR+ 31·9 (83/260) 0·77 (0·64−0·92)
ER+ and PR− 26·6 (21/79) 0·71 (0·50−1·02)
ER− and PR− 35·5 (117/330) 1·24 (1·09−1·41)
Unknown 25·4 (325/1278) 0·73 (0·49−1·08)
Biopsy grade (χ21=10·3; p=0·001)
Well 20·9 (14/67) 0·58 (0·36−0·91)
Moderate 36·0 (108/300) 0·93 (0·80−1·07)
Poor 44·6 (83/186) 1·15 (0·98−1·34)
Unknown 24·5 (341/1394) 1·08 (0·66−1·76)
Biopsy grade and ER status (χ23=16·9; p=0·0007)
Well or moderate, ER+ 31·4 (74/236) 0·76 (0·63−0·92)
Poor, ER+ 34·9 (29/83) 0·88 (0·66−1·17)
Well or moderate, ER− 37·2 (48/129) 1·04 (0·84−1·29)
Poor, ER− 52·9 (54/102) 1·32 (1·11−1·56)
Other or unknown 24·4 (341/1397) 0·88 (0·53−1·48)
Type of neoadjuvant chemotherapy (χ21=49·0; p<0·0001)
No anthracycline or taxane 18·5 (69/373) 0·36 (0·23−0·55)
Anthracycline, no taxane 26·0 (293/1125) 0·64 (0·47−0·87)
Anthracycline and taxane 41·0 (184/449) 1·13 (1·01−1·25)
Planned local therapy (χ21=0·3; p=0·57)
Radiotherapy only No response information
Lumpectomy 33·8 (225/665) 1·02 (0·91−1·14)
Mastectomy 18·1 (124/684) 0·97 (0·78−1·20)
Unknown 32·9 (197/598) 0·90 (0·64−1·25)
Total 28·0 (546/1947)
0·2 1·0 6·0
Clinical CR

Figure 6: Clinical complete response rate ratios

Three trials are excluded, as individual responses are not available; 440 women have missing clinical response data. CIs are group specific.25 Rate ratios are scaled such
that, within each category, their inverse variance-weighted sum is 1—ie, ratios are with respect to the mean CR. The appendix (p 6) contains data available for each
trial. CR=complete response. ER=oestrogen receptor. PR=progesterone receptor.

effect on local recurrence of allocation to NACT should be
greatest in women for whom mastectomy was originally
planned. 252 (33%) of 754 women converted from planned
mastectomy to breast-conserving therapy. However,
although the RR for local recurrence among all women
planned to have a mastectomy was 1·66 (95% CI 1·24–2·21)
compared with 1·14 (0·86–1·52) for women with
lumpectomy planned, the two-tailed test for heterogeneity
was not significant (p=0·07; figure 5, appendix pp 11, 12).
Heterogeneity between the RRs for local recurrence
was lower across all other tumour characteristics than it
was for RRs in the subgroup by planned local therapy
(figure 5); although the p value for the trend in RR with
biopsy grade was 0·05, this p value could have been a
chance finding given that it was the most extreme from
many subgroup analyses. Despite surgery de-escalation
being more common in larger tumours than in smaller
tumours, and in the trial combining anthracycline and
taxane13 than in trials of other regimens, the proportional
increases in local recurrence did not vary significantly by
tumour size or chemotherapy regimen (figure 5).
RRs also did not differ by age, nodal status, ER or

36 www.thelancet.com/oncology Vol 19 January 2018


progesterone receptor status, or period of follow-up.
Patient-level data for radiotherapy were not available, and
trial-level data for radiotherapy intent and practice were
incomplete (appendix p 2), so the effect of radiotherapy
on local recurrence cannot be studied. Radiotherapy was
scheduled for most women who had breast-conserving
surgery and actual use of radiotherapy was more frequent
in the NACT than in the adjuvant therapy groups
(appendix p 2). The RRs for distant recurrence and breast
cancer mortality did not vary by any tumour factor
measured, type of chemotherapy, timing of chemotherapy
use in the NACT group, type of planned local therapy, or
period of follow-up (appendix p 13).
As expected, distant recurrence and breast cancer
mortality were substantially lower in complete
responders than in non-responders (appendix p 14).
However, women who had a complete clinical response
after NACT had a frequency of local recurrence similar to
that of partial responders or non-responders. Ordering of
trials by the percentage of patients with a complete
clinical response to NACT did not reveal any significant
trend of improved recurrence or breast cancer mortality
RRs in trials with a higher frequency of response
(appendix p 16). No patterns emerged between trials in
complete response when considering the year that the
trial started or the frequency of breast-conserving therapy
within a trial (appendix p 15).
Discussion
In early breast cancer, high frequencies of complete or
partial clinical response can be achieved with NACT,
which can lead to a higher frequency of breast-conserving
therapy than with adjuvant chemotherapy. However, we
found NACT to be associated with a higher frequency of
local recurrence than was the same chemotherapy
started after surgery. Reassuringly, the increase in local
recurrence was not associated with any significant
increase in distant recurrence or breast cancer mortality.
More than two thirds of patients receiving NACT
responded, with more than a quarter achieving complete
clinical response, despite some trials using old
chemotherapy regimens and four administering some of
the chemotherapy postoperatively. The one regimen that
included both anthracycline and taxane had the highest
frequency of complete response. Within trials, response
was more common in women with small, ER-negative
and progesterone receptor-negative, or high-grade
tumours, as measured before randomisation, but was
little affected by age, nodal status, or planned local
therapy.8–17 As expected, use of NACT was associated with
an increase in the use of breast-conserving therapy.
An increase in the use of breast-conserving therapy in
women who responded well to NACT and who would
otherwise have had mastectomy is a likely explanation
for the increase in local recurrence in patients allocated
NACT. As anticipated,18,19 the absolute increase in local
recurrence was greatest in the two trials14,15 in which
www.thelancet.com/oncology Vol 19 January 2018 37
Articles
surgery could be avoided completely in the event of a
complete clinical response to NACT. This apparent
heterogeneity of effect was, however, not significant, and
NACT appeared to also have resulted in some increase in
local recurrence in the aggregated results from the eight
other trials. Hence, the increased local recurrence with
NACT is not wholly explained by omission of surgery.
Other unexamined factors might also have contributed to
the increased local recurrence with NACT. For example,
after NACT, tumour localisation can be difficult28 and
response patterns can be heterogeneous,29 making
surgery technically more difficult than without use of
NACT. Differing use of radiotherapy or axillary surgery
in the NACT group might also have contributed to
the higher local recurrence, although patient-level
information about this factor was not available. Trial
reports indicate that radiotherapy was scheduled for
most women who had breast-conserving surgery and
that actual use of radiotherapy was, if anything, more
frequent in the NACT than in the adjuvant therapy
groups. Thus, lesser use of radiotherapy after NACT than
that without NACT is unlikely to explain this increase in
local recurrence. Indeed, even with radiotherapy, local
failure is higher after breast-conserving surgery than
after mastectomy without radiotherapy.30
Our finding of an overall increase in local recurrence in
the trials using optimal local treatment is at odds with a
meta-analysis18 based on published data rather than
individual patient data, but this discrepancy could be
because the meta-analysis included comparisons that were
confounded by differing background systemic therapy.
Tumour response is predictive of lower distant
recurrence and death than an absence of tumour response.
Compared with all women randomly allocated to adjuvant
chemotherapy, outcomes were better for those with a
complete clinical response after NACT than for those with
a partial response and far better than for those with little
or no response to NACT. However, even in trials with high
frequencies of complete response, NACT was not
significantly better than adjuvant chemotherapy with
respect to distant recurrence or breast cancer mortality.
This finding could be because tumour characteristics that
are associated with higher response—such as smaller
tumour size—are also associated with lower distant
recurrence and are balanced between the NACT and
adjuvant groups by random­ isation. In each trial, both
groups eventually receive the same chemotherapy, so any
differences between trials in the efficacy of chemotherapy
regimens will apply to both groups. The CTNeoBC study6
reported similar findings in that efficacy as assessed by
high pathological complete response at the trial level did
not correlate well with long-term efficacy.
A limitation of our meta-analysis is that we have not
been able to assess reliably whether presurgical systemic
therapy is more effective at eradicating micrometastatic
disease than the same chemotherapy administered after
recovery from surgery because of the confounding
 Articles
effect of differences in the extent of surgery between
women allocated NACT and those allocated adjuvant
chemotherapy. A large trial with the same surgery and
radiotherapy in the NACT and adjuvant chemotherapy
groups could assess this question. At present, we cannot
exclude the possibility that NACT does moderately
reduce distant recurrence compared with the same
chemotherapy given postoperatively, but that this
benefit was obscured by an increase in local recurrence
due to less extensive surgery after NACT than in
patients who did not receive NACT. Trials of radiotherapy
after surgery indicate that substantially decreasing local
recurrence does also decrease breast cancer mortality,
with about one breast cancer death prevented for every
four local recurrences prevented.30 The small,
non-significant excess of breast cancer mortality in
patients allocated NACT is consistent with this risk
ratio, so could therefore be due to the increase in local
recurrence, but it could equally well be a chance finding.
The main aim of NACT in contemporary practice is to
reduce the extent of breast surgery, thereby making
breast conservation feasible in women who would
otherwise need mastectomy. In the time since the trials
in this meta-analysis were done, pathology reporting,
surgery, and radiotherapy have improved, and more
effective systemic neoadjuvant regimens have been
introduced than were available when these trials took
place. These changes should increase the likelihood of
successful downstaging to allow conservative surgery in
current and future practice. But, although improvements
in treatment mean local recurrence risk should be lower
than in these trials, our findings indicate that tumours
downsized by NACT might continue to be associated
with higher local recurrence risk after breast-conserving
surgery than might tumours of the same dimensions
in women who have not received NACT. Strategies
to mitigate the increased local recurrence after
breast-conserving therapy in tumours downsized by
NACT should be considered—for example, careful
tumour localisation, detailed pathological assessment,
and appropriate radiotherapy.31 Prospective randomised
trials would also help to establish the optimal clinical
management in this context.
NACT allows more breast-conserving therapy than
does adjuvant chemotherapy and provides information
about an individual patient’s response to a particular
chemotherapy regimen. However, it appears to be no
better than postoperative adjuvant treatment at reducing
breast cancer mortality and, perhaps as a consequence of
a reduction of the extent of surgery, NACT is associated
with moderately increased local recurrence risk, which
persists for at least 10 years.
Contributors
The EBCTCG secretariat (C Boddington, R Bradley, J Burrett, M Clarke,
C Davies, L Davies, D Dodwell, F Duane, V Evans, L Gettins, J Godwin,
R Gray, S James, H Liu, Z Liu, E MacKinnon, G Mannu, P McGale,
T McHugh, P Morris, H Pan, R Peto, S Read, C Taylor, Y Wang, and
Z Wang) identified trials, obtained datasets, and had full access to them.
38 www.thelancet.com/oncology Vol 19 January 2018
P McGale, R Gray, and R Peto designed and carried out the analyses with
computing assistance from Y Wang and Z Wang. P McGale, D Dodwell,
R Gray, G von Minckwitz, R Peto, and C Taylor drafted the report with
advice from S Anderson, J Bergh, L Gianni, C Lohrisch, and
K I Pritchard. All writing committee members contributed to revising
the report. Interim analyses were discussed by the steering committee
and trialists who supplied data for the analysis.
Writing committee
P McGale, D Dodwell, R Gray, G Mannu, R Peto, C Taylor, Y Wang,
Z Wang, M Clarke, C Davies, R Bradley, J Braybrooke, H Pan (Medical
Research Council Population Health Research Unit, Nuffield
Department of Population Health, University of Oxford, Oxford, UK);
C Lohrisch (Breast Cancer Outcomes Unit, British Columbia Cancer
Agency, Vancouver, BC, Canada); L Gianni (Istituto Nazionale Tumori,
Milan, Italy); S Anderson (Department of Biostatistics, University of
Pittsburgh Graduate School of Public Health, Pittsburgh, PA, USA);
K I Pritchard (Sunnybrook Odette Cancer Centre, Toronto, ON, Canada;
J Bergh (Karolinska Institutet and University Hospital, Stockholm,
Sweden); G von Minckwitz (German Breast Group, Neu-Isenburg,
Germany).
Groups (and key trialists) contributing data
Istituto Nazionale per lo Studio e la Cura dei Tumori, Milan, Italy
(P Valagussa, L Gianni, G Bonadonna); European Organisation for
Research on the Treatment of Cancer (J Bogaerts, J van der Hage,
C J H van de Velde); Austrian Breast and Colorectal Cancer Study
Group, Vienna, Austria (L Solkner, M Gnant, R Jakesz); Center for
Cancer Research of the National Cancer Institute, Bethesda, MD,
USA. (S Steinberg, D Danforth, J Zujewski); Royal Marsden Hospital,
Sutton, London, UK (T Powles, S Ashley, H Ford, M Makris);
St George’s Hospital, London, UK (J-C Gazet, C Coombes,
R Sutcliffe); National Surgical Adjuvant Breast & Bowel Project,
Pittsburgh, PA, USA (S Anderson, J Costantino, J Bryant, N Wolmark,
E Mamounas, B Fisher); Fondation Bergonié—Comprehensive Cancer
Centre of South West France, Bordeaux, France (L Mauriac,
S Mathoulin); Cancer Control Agency of British Columbia, BC,
Canada (J Ragaz, I Olivotto); Institut Curie, Paris, France (B Asselain,
P Broet, S Scholl).
EBCTCG steering committee
J Bergh, K I Pritchard (co-chairs), K Albain, S Anderson, R Arriagada,
W Barlow, J Bartlett, E Bergsten-Nordström, J Bliss, F Boccardo,
R Bradley*, E Brain, J Braybrooke*, D Cameron, M Clarke*, A Coates,
R Coleman, C Correa, J Costantino, J Cuzick, N Davidson, C Davies*,
L Davies*, A Di Leo, D Dodwell*, M Dowsett, F Duane*, M Ewertz,
J Forbes, R Gelber, M Gnant, A Goldhirsch, P Goodwin, R Gray*,
D Hayes, C Hill, J Ingle, R Jagsi, W Janni, Z Liu*, S Loibl,
E MacKinnon*, G Mannu*, M Martín, P McGale*, H Mukai, L Norton,
Y Ohashi, S Paik, H Pan*, R Peto*, M Piccart, L Pierce, P Poortmans,
V Raina, P Ravdin, M Regan, J Robertson, E Rutgers, D Slamon,
J Sparano, S Swain, C Taylor*, A Tutt, G Viale, G von Minckwitz,
X Wang, T Whelan, N Wilcken, E Winer, N Wolmark, W Wood,
M Zambetti. *EBCTCG Secretariat, Population Health Research Unit.
Declaration of interests
The writing committee and EBCTCG secretariat declare no competing
interests.
Acknowledgments
This study is supported by core funding to the Population Health
Research Unit and Clinical Trial Service Unit, Nuffield Department of
Population Health, University of Oxford, from Cancer Research UK,
the British Heart Foundation, and the UK Medical Research Council, as
well as by the UK Department of Health grant RRX108 and Cancer
Research UK grant C8225/A21133. The chief acknowledgement is to the
women in these trials and the trial personnel.
References
1 Rubens RD, Sexton S, Tong D, Winter PJ, Knight RK, Hayward JL.
Combined chemotherapy and radiotherapy for locally advanced
breast cancer. Eur J Cancer 1980; 16: 351–56.
2 Mougalian SS, Soulos PR, Killelea BK, et al. Use of neoadjuvant
chemotherapy for patients with stage I to III breast cancer in the
United States. Cancer 2015; 121: 2544–52.

3 Clough KB, Acosta-Marín V, Nos C, et al. Rates of neoadjuvant
chemotherapy and oncoplastic surgery for breast cancer surgery:
a French national survey. Ann Surg Oncol 2015; 11: 3504–11.
4 Vugts G, Maaskant-Braat AJ, Nieuwenhuijzen GA, Roumen RM,
Luiten EJ, Voogd AC. Patterns of care in the administration of
neoadjuvant chemotherapy for breast cancer. A population-based
study. Breast J 2016; 22: 316–21.
5 Fisher B, Gunduz N, Coyle J, Rudock C, Saffer E. Presence of a
growth-stimulating factor in serum following primary tumor
removal in mice. Cancer Res 1989; 49: 1996–2001.
6 Cortazar P, Zhang L, Untch M, et al. Pathological complete
response and long-term clinical benefit in breast cancer:
the CTNeoBC pooled analysis. Lancet 2014; 384: 164–72.
7 von Minckwitz G, Blohmer JU, Costa SD, et al. Response-guided
neoadjuvant chemotherapy for breast cancer. J Clin Oncol 2013;
31: 3623–30.
8 Taucher S, Steger GG, Jakesz R, et al. The potential risk of
neoadjuvant chemotherapy in breast cancer patients—results from
a prospective randomized trial of the Austrian Breast and Colorectal
Cancer Study Group (ABCSG-07). Breast Cancer Res Treat 2008;
112: 309–16.
9 Wolmark N, Wang J, Mamounas E, Bryant J, Fisher B. Preoperative
chemotherapy in patients with operable breast cancer: nine-year
results from National Surgical Adjuvant Breast and Bowel Project
B-18. J Natl Cancer Inst Monogr 2001; 30: 96–102.
10 van der Hage JA, van de Velde CJ, Julien JP, Tubiana-Hulin M,
Vandervelden C, Duchateau L. Preoperative chemotherapy in
primary operable breast cancer: results from the European
Organization for Research and Treatment of Cancer trial 10902.
J Clin Oncol 2001; 19: 4224–37.
11 Powles TJ, Hickish TF, Makris A, et al. Randomized trial of
chemoendocrine therapy started before or after surgery for
treatment of primary breast cancer. J Clin Oncol 1995; 13: 547–52.
12 Gazet JC, Ford HT, Gray R, et al. Estrogen-receptor-directed
neoadjuvant therapy for breast cancer: results of a randomised trial
using formestane and methotrexate, mitozantrone and mitomycin
C (MMM) chemotherapy. Ann Oncol 2001; 12: 685–91.
13 Gianni L, Baselga J, Eiermann W, et al. A. Phase III trial evaluating
the addition of paclitaxel to doxorubicin followed by
cyclophosphamide, methotrexate, and fluorouracil, as adjuvant or
primary systemic therapy: European Cooperative Trial in Operable
Breast Cancer. J Clin Oncol 2009; 27: 2474–81.
14 Mauriac L, MacGrogan G, Avril A, et al. Neoadjuvant chemotherapy
for operable breast carcinoma larger than 3 cm: a unicentre
randomized trial with a 124-month median follow-up. Ann Oncol
1999; 10: 47–52.
15 Scholl SM, Asselain B, Palangie T, et al. Neoadjuvant chemotherapy
in operable breast cancer. Eur J Cancer 1991; 27: 1668–71.
16 Danforth DN Jr, Cowan K, Altemus R, et al. Preoperative
FLAC/granulocyte-colony-stimulating factor chemotherapy for
stage II breast cancer: a prospective randomized trial.
Ann Surg Oncol 2003; 10: 635–44.
17 Ragaz J, Baird R, Rebbeck P, et al. Preoperative (neoadjuvant)
versus postoperative adjuvant chemotherapy for stage I–II breast
cancer. Long-term analysis of British Columbia randomized trial.
Proc Am Soc Clin Oncol 1997; 16: 142a.
www.thelancet.com/oncology Vol 19 January 2018 39
Articles
18 Mieog JS, van der Hage JA, van De Velde CJ. Neoadjuvant
chemotherapy for operable breast cancer. Br J Surg 2007;
94: 1189–200.
19 Mauri D, Pavlidis N, Ioannidis JP. Neoadjuvant versus adjuvant
systemic treatment in breast cancer: a meta-analysis.
J Natl Cancer Inst 2005; 97: 188–94.
20 Feinstein AR, Fosin DM, Wells CK. The Will Rogers phenomenon
stage migration and new diagnostic techniques as a source of
misleading statistics for survival in cancer. N Engl J Med 1985;
312: 1604–08.
21 Bear HD, Anderson S, Smith RE, et al. Sequential preoperative or
postoperative docetaxel added to preoperative doxorubicin plus
cyclophosphamide for operable breast cancer: National Surgical
Adjuvant Breast and Bowel Project Protocol B-27. J Clin Oncol 2004;
24: 2019–27.
22 Early Breast Cancer Trialists’ Collaborative Group. Treatment of
early breast cancer. Volume 1, worldwide evidence 1985–1990.
Oxford: Oxford University Press, 1990.
23 Early Breast Cancer Trialists’ Collaborative Group. Comparisons
between different polychemotherapy regimens for early breast
cancer: meta-analyses of long-term outcome among 100 000 women
in 123 randomised trials. Lancet 2012; 379: 432–44.
24 Stewart LA, Clarke M, Rovers M, et al. Preferred Reporting Items
for a Systematic Review and Meta-analysis of individual participant
data: the PRISMA-IPD Statement. JAMA 2015; 313: 1657–65.
25 Plummer M. Improved estimates of floating absolute risk. Stat Med
2004; 23: 93–104.
26 Early Breast Cancer Trialists’ Collaborative Group. Effects of
chemotherapy and hormonal therapy for early breast cancer on
recurrence and 15-year survival: an overview of the randomised
trials. Lancet 2005; 365: 1687–717.
27 Gelman R, Gelber R, Henderson IC, Coleman CN, Harris JR.
Improved methodology for analyzing local and distant recurrence.
J Clin Oncol 1990; 8: 548–55.
28 Baron LF, Baron PL, Ackerman SJ, Durden DD, Pope TL Jr.
Sonographically guided clip placement facilitates localization of
breast cancer after neoadjuvant chemotherapy. AJR Am J Roentgenol
2000; 174: 539–40.
29 von Minckwitz G, Untch M, Blohmer JU, et al. Definition and
impact of pathologic complete response on prognosis after
neoadjuvant chemotherapy in various intrinsic breast cancer
subtypes. J Clin Oncol 2012; 30: 1796–804.
30 Early Breast Cancer Trialists’ Collaborative Group. Effects of
radiotherapy and of differences in the extent of surgery for early
breast cancer on local recurrence and 15-year survival: an overview
of the randomised trials. Lancet 2005; 366: 2087–106.
31 Gnant M, Harbeck N, Thomssen C. St. Gallen/Vienna 2017: a brief
summary of the consensus discussion about escalation and
de-escalation of primary breast cancer treatment. Breast Care 2017;
12: 102–07.