Meta-Analysis of Breast Cancer Outcomes in Adjuvant Trials of Aromatase Inhibitors Versus Tamoxifen

Meta-Analysis of Breast Cancer Outcomes in AdjuvantTrials of Aromatase Inhibitors Versus TamoxifenMitch Dowsett, Jack Cuzick, Jim Ingle, Alan Coates, John Forbes, Judith Bliss, Marc Buyse, Michael Baum,Aman Buzdar, Marco Colleoni, Charles Coombes, Claire Snowdon, Michael Gnant, Raimund Jakesz,Manfred Kaufmann, Francesco Boccardo, Jon Godwin, Christina Davies, and Richard Peto

From the Academic Department ofBiochemistry, Royal Marsden Hospital;Cancer Research UK Centre for Epide-miology, Mathematics and Statistics,Queen Mary University of London; Clin-ical Trials Group, Royal Free and Univer-sity College London Medical School;Department of Cancer Medicine, Char-ing Cross Hospital, London; ClinicalTrials and Statistics Unit, The Instituteof Cancer Research, Surrey; ClinicalTrials Service Unit and EpidemiologicalStudies Unit, University of Oxford,Oxford, United Kingdom; Mayo Clinic,Rochester, MN; International BreastCancer Study Group, Bern, Switzerland;Department of Surgical Oncology,University of Newcastle, Newcastle,Australia; Department of Breast Medi-cal Oncology, M. D. Anderson CancerCenter, Houston, TX; Department ofMedicine, Istituto Europeo di Oncolo-gia, Milan; University and NationalCancer Research Institute, Genoa, Italy;Department of Surgery, Medical Univer-sity of Vienna; Division of GeneralSurgery, Vienna Medical School,Vienna, Austria; OB/GYN/Breast Unit,University Hospital Frankfurt, Frankfurt,Gernany.

Submitted March 17, 2009; acceptedAugust 26, 2009; published onlineahead of print at www.jco.org onNovember 30, 2009.

Written on behalf of the AromataseInhibitor Overview Group.

The Early Breast Cancer Trialists’Collaborative Group secretariat isfunded by the core Clinical Trial ServiceUnit and Epidemiological Studies Unit,support from Cancer Research UK, andby the Medical Research Council.National Health Service funding wasprovided to the Royal Marsden NationalInstitute for Health Research BiomedicalResearch Centre.

Authors’ disclosures of potential con-flicts of interest and author contribu-tions are found at the end of thisarticle.

Corresponding author: Mitch Dowsett,PhD, Royal Marsden Hospital, 237Fulham Rd, London SW3 6JJ, UnitedKingdom; e-mail: [email protected].

© 2009 by American Society of ClinicalOncology

0732-183X/10/2803-509/$20.00

DOI: 10.1200/JCO.2009.23.1274

A B S T R A C T

PurposeTo conduct meta-analyses of randomized trials of aromatase inhibitors (AIs) compared withtamoxifen either as initial monotherapy (cohort 1) or after 2 to 3 years of tamoxifen (cohort 2).

Materials and MethodsData submitted to the Early Breast Cancer Trialists’ Collaborative Group were used in separatemeta-analyses of two cohorts. Primary analyses involve postmenopausal women with tumorsreported to be estrogen receptor positive. Log-rank P values are two-sided.

ResultsCohort 1 comprised 9,856 patients with a mean of 5.8 years of follow-up. At 5 years, AI therapywas associated with an absolute 2.9% (SE � 0.7%) decrease in recurrence (9.6% for AI v 12.6%for tamoxifen; 2P � .00001) and a nonsignificant absolute 1.1% (SE � 0.5%) decrease in breastcancer mortality (4.8% for AI v 5.9% for tamoxifen; 2P � .1). Cohort 2 comprised 9,015 patientswith a mean of 3.9 years of follow-up. At 3 years from treatment divergence (ie, approximately 5years after starting hormonal treatment), AI therapy was associated with an absolute 3.1%(SE � 0.6%) decrease in recurrence (5.0% for AI v 8.1% for tamoxifen since divergence;2P � .00001) and an absolute 0.7% (SE � 0.3%) decrease in breast cancer mortality (1.7% forAI v 2.4% for tamoxifen since divergence; 2P � .02). There was no convincing heterogeneityin the proportional recurrence reduction with respect to age, nodal status, tumor grade, orprogesterone receptor status and no indication of an increase in nonbreast deaths with AIs ineither cohort.

ConclusionAIs produce significantly lower recurrence rates compared with tamoxifen, either as initialmonotherapy or after 2 to 3 years of tamoxifen. Additional follow-up will provide clearerinformation on long-term survival.

J Clin Oncol 28:509-518. © 2009 by American Society of Clinical Oncology

INTRODUCTION

Tamoxifen is a selective estrogen receptor (ER)modulator that acts (along with its metabolites)largely or wholly by competitive binding to the re-ceptor protein.1 In the 75% to 80% of patients withearly breast cancer who have ER-positive disease,treatment with 5 years of tamoxifen immediatelyand substantially reduces local, contralateral, anddistant recurrence rates and reduces 15-year breastcancer mortality, with little effect on the aggregateof all other causes of death.2,3 However, there islittle or no effect on breast cancer outcomes inER-negative disease. The early results from thefirst clinical trials of tamoxifen showed only asmall absolute gain in 5-year survival,3 but theadditional mortality reductions during years 5 to 9

and 10 to 14 were each about as great as the mortalityreduction during years 0 to 4 (the 5 years whiletamoxifen was still being taken).2 Hence, the abso-lute reduction in breast cancer mortality producedby 5 years of tamoxifen was almost three times asgreat 15 years after diagnosis as it had been only 5years after diagnosis.

The development of well-tolerated inhibitorsof aromatase, the enzyme that synthesizes estrogensfrom androgens, has provided an alternative meansof depriving breast tumors of stimulation by endog-enous estrogens in postmenopausal women whoseovaries no longer produce significant amounts ofthe hormone4,5 and in premenopausal women inwhom ovarian function has been suppressed or theovaries have been removed. In recent years, severallarge randomized trials in early breast cancer with 5

JOURNAL OF CLINICAL ONCOLOGY S P E C I A L A R T I C L E

VOLUME 28 � NUMBER 3 � JANUARY 20 2010

© 2009 by American Society of Clinical Oncology 509

195.55.130.44Information downloaded from jco.ascopubs.org and provided by at HOSPITAL ROYO VILLANOVA on May 14, 2011 from

Copyright © 2010 American Society of Clinical Oncology. All rights reserved.

years of tamoxifen as the control arm have reported on the use of oneof three third-generation aromatase inhibitors (AIs)—anastrozole,letrozole, or exemestane; other large third-generation AI trials have yetto be reported.5 Although the trials considered in these meta-analyseshave generally found reduced recurrence rates compared with 5 yearsof tamoxifen, none on its own has yet clearly demonstrated a reduc-tion in breast cancer mortality, and none possesses adequate statisticalpower to do so or to yield reliable subgroup analyses of the effects onrecurrence. Therefore, the trialists have collaborated through the Ox-ford Early Breast Cancer Trialists’ Collaborative Group (EBCTCG)overview process to pool data from all currently available trials of AIversus tamoxifen, using the methods of previous EBCTCG meta-analyses. When results from individual trial publications are amal-gamated in a systematic review, no allowance can be made for thedifferent duration of follow-up in each study or for the differingdefinitions of some end points, and subgroup analyses may be re-ported inconsistently. For example, the primary end point of mosttrials in this meta-analysis was disease-free survival but variably in-cluded or excluded occurrence of a new nonbreast cancer. Individualpatient meta-analysis as carried out here allows each of these issues tobe addressed.

The trials have been grouped into two cohorts, and these cohortshave been analyzed separately. Cohort 1 consists of the trials compar-ing 5 years of an AI with 5 years of tamoxifen, both starting soon aftersurgery,6,7 and cohort 2 consists of the trials in which tamoxifentherapy is switched after 2 to 3 years to 2 to 3 years of an AI comparedwith 2 to 3 more years of tamoxifen (making a total of 5 years ofhormonal therapy in both groups).8-10 Some of the trials included asmall proportion of patients with ER-negative or ER untested tumors.These have both been excluded from the present analyses because ofthe lack of any material effect of 5 years of tamoxifen in ER-negativedisease.2 Trials comparing treatment with an AI after 5 years of tamox-ifen versus no continuation of any hormonal therapy are not coveredby this overview. The database for this analysis was locked beforeupdates from some of the trials were presented at the San AntonioBreast Cancer Symposium in December 2008. Those updates areconsequently not included in this report.

MATERIALS AND METHODS

Trial identification and data handling procedures have been described previ-ously.2 Information was sought from all trials that had started by the year 2000in which two arms differed for some period of time only in one having an AIand the other having tamoxifen. The only two such trials that are known to beunavailable are, first, the Austrian Breast and Colorectal Cancer Study Group(ABCSG) XII trial, which started in 1999 and randomly assigned 1,803 pre-menopausal women, whose ovarian function was being suppressed by gosere-lin, to 3 years of AI (anastrozole) or 3 years of tamoxifen and, second, the twoswitching arms from the Breast International Group (BIG) 01-98/Interna-tional Breast Cancer Study Group (IBCSG) 18-98 trial, which started in 1998.This trial randomly assigned 6,193 women four ways (between 5 years of AI[letrozole], 5 years of tamoxifen, switching from tamoxifen to AI, and switch-ing from AI to tamoxifen); 3,095 of women were randomly assigned to the twosequencing arms.

The statistical methods and the format of presentation of the results areas described in the most recent EBCTCG report,2 with minor amendments.One amendment is to the format of the forest plots in which squares andhorizontal lines are used to describe event rate ratios (AI v tamoxifen) invarious subgroups. As before, a solid vertical line is used to denote no effect (ie,

an event rate ratio of 1.0), and a broken vertical line is used to denote theaverage treatment effect in a meta-analysis of all the available evidence. In thepresent report, the broken vertical line is made particularly bold in the figuresthat present multiple subgroup analyses. This is to help emphasize tests ofheterogeneity between subgroups, thereby reducing (but not eliminating) therelevance of tests of the null hypothesis within particular subgroups.11,12 An-other amendment is to present, among other things, analyses of time to firstdistant recurrence, ignoring any prior locoregional or contralateral recur-rences, as well as analyses of various types of recurrence as a first event.(Recurrence was defined as the first reappearance of breast cancer at any siteand thus included second primary breast cancer or locoregional or distantrecurrence of the original cancer.) As before,2 log-rank analyses are presentedfor any death, death without recurrence, and, by log-rank subtraction, deathwith recurrence. Because the incidence of recurrence was monitored carefullyin these trials, the final amendment is that deaths from an unknown causewithout any recorded recurrence are taken as non–breast cancer deaths.

Analyses are presented entirely separately for the trials of approximately5 years of AI versus the same duration of tamoxifen (cohort 1) and for theswitching trials of some years of tamoxifen and then some years of AI versusthe same duration of tamoxifen (cohort 2). In the switching trials, the analysesare of time since the treatments diverged, which is approximately 2 to 3 yearsafter originally starting adjuvant tamoxifen shortly after diagnosis. All but oneof the switching trials randomly assigned patients after the first few years oftamoxifen (to switch or continue). The remaining trial (ABCSG VIII trial)8

randomly assigned patients at the start of adjuvant tamoxifen between 2 yearsof tamoxifen and then 3 years of AI and, as control, 5 years of tamoxifen. Theanalyses of this one trial are of time since the treatments would have diverged,and any patients who experienced recurrence, death, or loss to follow-upwithin the first 2 years are thus excluded from numerators, denominators, andall analyses.

Keeping to the convention of previous EBCTCG analyses, the first year ofadjuvant treatment is referred to as year 0.2,3 To help allow for the fact thatsome trials are more mature than others, the follow-up in cohort 1 trials issubdivided into years 0 to 1 (first 2 years of treatment), 2 to 4 (last 3 years oftreatment), and 5� (after treatment). For cohort 2, follow-up is subdividedinto years 0 to 2 (the 3 years of randomly assigned treatment) and 3� (aftertreatment) after the treatments were to diverge; these divisions equate to years2 to 4 and 5�, respectively, from the start of adjuvant hormonal treatmentbecause of the initial 2 years of treatment of all patients in this cohort withtamoxifen. For both cohorts, Kaplan-Meier graphs are plotted to 8 years fromthe start of adjuvant hormonal treatment. The lock date for the data analysiswas September 30, 2006 for all trials.

RESULTS

Cohort 1: 5 Years of AI Versus 5 Years of Tamoxifen

Two trials (Arimidex, Tamoxifen, Alone or in Combination[ATAC] and BIG 01-98/IBCSG 18-98; Appendix Fig A1, online only)contribute data, including a total of 9,856 women and more than55,000 woman-years of follow-up (mean, 5.8 years). Mean and me-dian follow-up times were 7.1 and 7.9 years, respectively, for ATACand 4.3 and 4.0 years, respectively, for BIG 01-98. Most of the recur-rences thus far recorded occurred during the first 5 years, while bothgroups were still on active treatment (Fig 1A). Use of an AI rather thantamoxifen was associated with a highly significant 23% (SE � 5%)proportional reduction in these early recurrences (2P � .00001; Ap-pendix Fig A1A). There seemed to be greater proportional decreases inisolated local recurrence (30%; SE � 10%; 2P � .003; Appendix FigA1B) and in contralateral disease (41%; SE � 12%; 2P � .0009;Appendix Fig A1C) than in distant recurrence either as a first event(16%; SE � 6%; 2P � .009; Appendix Fig A1D) or at any time (18%;SE � 6%; 2P � .002; Appendix Fig A1E), although this apparent

Dowsett et al

510 © 2009 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY



heterogeneity of effect between recurrence sites was not statisticallysignificant (�2

2 � 5.0, P � .08).The proportional reduction in the risk of recurrence was not

significantly different during the first 2 years and during the last 3 yearsof treatment (Fig 2A; in both periods, it is separately significant), andthere seemed to be some further risk reduction in subsequent years,although this further reduction was not, on its own, statistically signif-icant. At 8 years after the start of hormonal therapy, the absolutedifference in recurrence risk was 3.9% (SE � 1.0%; 15.3% for AI v19.2% for tamoxifen).

Subgroup analyses of the recurrence results with respect to pro-gesterone receptor (PgR) status, age, nodal status, and tumor grade areshown in Figures 2B to 2E. The global test of heterogeneity between allof these subgroups was not statistically significant (2P � .1), so undueemphasis on any one of them may yield misleading conclusions. Therewas no apparent heterogeneity with respect to age, nodal status, orgrade between the proportional reductions in recurrence, but therewas with respect to PgR status, which was known in 8,745 patients(89%). The proportional reduction in recurrence (AI v tamoxifen)was 40% (SE � 9%) in the 22% of patients categorized as ER positive/

5-year gain, 2.9% (SE, 0.7%)8-year gain, 3.9% (SE, 1.0%)Log-rank 2P < .00001

Recurrence rates (% / year) and log-rank analyses Years 0–1 Years 2–4 Years 5+AI 1.69 (163 / 9,647) 2.31 (261 / 11,297) 2.33 (160 / 6,879)Tamoxifen 2.46 (234 / 9,510) 2.81 (307 / 10,938) 2.78 (180 / 6,478)Rate ratio, from 0.67 (SE, 0.08) 0.81 (SE, 0.08) 0.83 (SE, 0.10)(O−E) / V -38.4 / 96.6 -29.5 / 137.9 -15.7 / 83.0

A

0

Recu

rren

ce (%

, ± S

E)

Time (years)

50

40

30

20

10

5 8

5-year gain, 1.1% (SE, 0.5%)8-year gain, 0.5% (SE, 0.8%)Log-rank 2P = .1

Death rates (%/year: total rate—rate in women without recurrence) & log-rank analyses Years 0–1 Years 2–4 Years 5+AI 0.59 (SE, 0.08) 1.26 (SE, 0.10) 1.78 (SE, 0.16)Tamoxifen 0.57 (SE, 0.08) 1.60 (SE, 0.12) 1.79 (SE, 0.16)Rate ratio, from 1.01 (SE, 0.19) 0.77 (SE, 0.10) 1.01 (SE, 0.13)(O−E) / V 0.2 / 27.5 -20.5 / 80.2 0.4 / 61.5

B

0

Brea

st C

ance

r Mor

talit

y (%

, ± S

E)

Time (years)

50

40

30

20

10

5 8

5-year loss, 0.2% (SE, 0.5%)8-year loss, 0.3% (SE, 0.8%)Log-rank 2P = .9

Death rates (% / year) and log-rank analyses Years 0–1 Years 2–4 Years 5+AI 0.54 (52 / 9,647) 1.10 (124 / 11,297) 1.60 (110 / 6,879)Tamoxifen 0.59 (56 / 9,510) 1.02 (112 / 10,938) 1.62 (105 / 6,478)Rate ratio, from 0.92 (SE, 0.19) 1.07 (SE, 0.14) 0.99 (SE, 0.14)(O−E) / V -2.1 / 26.8 4.1 / 58.3 -0.4 / 52.9

C

0

Deat

h W

ithou

t Rec

urre

nce

(%, ±

SE)

Time (years)

50

40

30

20

10

5 8


Death rates (% / year) and log-rank analyses Years 0–1 Years 2–4 Years 5+AI 1.12 (109 / 9,743) 2.31 (272 / 11,757) 3.29 (240 / 7,286)Tamoxifen 1.15 (111 / 9,656) 2.57 (297 / 11,569) 3.26 (232 / 7,107)Rate ratio, from 0.97 (SE, 0.13) 0.89 (SE, 0.08) 1.00 (SE, 0.09)(O−E) / V -1.9 / 54.3 -16.4 / 138.5 0.0 / 114.4

D

0

Any

Deat

h (%

, ± S

E)

Time (years)

50

40

30

20

10

5 8

10.5%10.0%

5.9%

4.8%

AITamoxifen

19.2%

15.3%12.6%

9.6%

9.1%8.8%4.4%

4.2%

18.0%17.8%

9.6%

8.8%

AITamoxifen

AITamoxifen

AITamoxifen

Fig 1. Life-table curves of (A) recurrence; (B) breast cancer mortality; (C) death without recurrence; and (D) any death, for estrogen receptor–positive patients in trialsof approximately 5 years of aromatase inhibitor (AI) versus tamoxifen. O, observed; E, expected; V, variance.

Meta-Analysis of Aromatase Inhibitors Versus Tamoxifen

www.jco.org © 2009 by American Society of Clinical Oncology 511



PgR poor and only 17% (SE � 6%) in patients categorized as ERpositive/PgR positive. Although the apparent heterogeneity betweenthese two proportional risk reductions is formally statistically signifi-cant (2P � .02), in view of the nonsignificant global test of heteroge-

neity, the total number of subgroups considered with respect to thisand other factors, and the PgR-associated effect being significant inone of the two trials (ATAC) but not the other (BIG 01-98/IBCSG18-98), the apparent heterogeneity with respect to PgR status could be

0 0.5 1.0 1.5 2.0

AI better Tamoxifen betterGlobal heterogeneity (B−E): χ24 = 6.9, P > .1, NS

99% or 95% confidence intervals

Events/Women AI events Allocated Allocated Log-rank Variance Ratio of annual event ratesCategory AI tamoxifen O−E of O−E AI : Tamoxifen

(A) Time period (trend χ21 = 2.0, 2P > .1, NS)

Years 0−1 163/4,954 234/4,902 −38.4 96.6 0.67 (SE 0.08) (3.3%) (4.8%)

Years 2−4 261/4,670 307/4,557 −29.5 137.9 0.81 (SE 0.08) (5.6%) (6.7%)

Years 5+ 160/2,939 180/2,803 −15.7 83.0 0.83 (SE 0.10) (5.4%) (6.4%)

(B) PgR status (χ21 = 5.6, 2P = .02)

PgR poor 118/958 182/937 −36.9 72.5 0.60 (SE 0.09) (12.3%) (19.4%)

PgR+ 381/3,452 443/3,398 −37.3 201.8 0.83 (SE 0.06) (11.0%) (13.0%)

PgR unknown 85/544 96/567 −4.4 43.6 0.90 (SE 0.14) (15.6%) (16.9%)

(C) Entry age (trend χ21 = 0.4, 2P > .1, NS)

Age < 50 25/225 33/202 −5.9 13.9 0.65 (SE 0.22) (11.1%) (16.3%)

50–59 175/1,680 228/1,674 −30.0 98.3 0.74 (SE 0.09) (10.4%) (13.6%)

60–69 222/1,898 266/1,904 −26.4 118.8 0.80 (SE 0.08) (11.7%) (14.0%)

70+ 162/1,151 194/1,122 −21.1 86.1 0.78 (SE 0.10) (14.1%) (17.3%)

(D) Nodal status (N0/N− v all N+ χ21 = 0.6, 2P > .1, NS)

N0/N− 118/1,937 171/2,021 −23.9 71.2 0.72 (SE 0.10) (6.1%) (8.5%)

N1−3 83/734 91/702 −6.9 42.3 0.85 (SE 0.14) (11.3%) (13.0%)

N4+ 324/1,866 372/1,779 −36.8 167.4 0.80 (SE 0.07) (17.4%) (20.9%)

N+ unspec. 45/254 65/244 −12.1 26.5 0.63 (SE 0.16) (17.7%) (26.6%)

N unknown 14/163 22/156 −4.5 8.9 0.60 (SE 0.26) (8.6%) (14.1%)

(E) Tumor grade (trend χ21 = 0.3, 2P > .1, NS)

Well differentiated 63/1,186 93/1,217 −14.6 38.4 0.68 (SE 0.13) (5.3%) (7.6%)

Moderately 265/2,417 331/2,388 −40.4 145.2 0.76 (SE 0.07) (11.0%) (13.9%) Poorly 181/854 209/823 −23.8 92.1 0.77 (SE 0.09) (21.2%) (25.4%) Grade unknown 75/497 88/474 −9.8 39.3 0.78 (SE 0.14) (15.1%) (18.6%)

Total 584/4,954 721/4,902 −83.6 317.5 0.769 (SE 0.049) (11.8%) (14.7%) 2P < .00001

Treatment effect 2P < .00001

Fig 2. Subgroup analyses of recurrencesby (A) time period; (B) progesterone re-ceptor (PgR) status; (C) entry age; (D)nodal status; and (E) tumor grade, forestrogen receptor–positive patients in tri-als of approximately 5 years of aromataseinhibitor (AI) versus tamoxifen. Dashed ver-tical line corresponds to zero heterogeneityof effect. NS, not significant; unspec., un-specified; O, observed; E, expected.

Dowsett et al




largely or wholly a chance finding. If so, the overall proportionalreduction in recurrence rate (AI v tamoxifen) may provide the bestguide to the treatment effect in each of the subgroups in Figures 2Bto 2E.

Thus far, there has been only a small and nonsignificant propor-tional reduction (11%; SE � 7%) in breast cancer mortality in cohort1 (Figs 1B and 3A). The absolute difference between AI and tamoxifenin breast cancer mortality was 1.1% (SE � 0.5%) at 5 years. Althoughit seemed to be smaller (only 0.5%; SE � 0.8%) at 8 years (Fig 1B), themean duration of follow-up is only 5 years, so this could change withthe accumulation of further data. Of the 1,261 deaths from any cause,559 (44%) were of women who had not experienced a recurrence andhad died of an unrelated cause. There was no apparent differencebetween AI and tamoxifen in these non–breast cancer deaths (Figs 1Cand 3B). Hence, the difference in overall mortality was also nonsignif-icant (Figs 1D and 3C).

Cohort 2: 2 to 3 Years of Tamoxifen for All Women

and Then 2 to 3 Years of AI Versus Tamoxifen

Four trials (German Adjuvant Breast Cancer Group/Arimidex-Nolvadex, Intergroup Exemestane Study/BIG 02-97, Italian Tamox-ifen Anastrozole Trial (ITA), and ABCSG VIII) contribute data,including a total of 9,015 women with more than 33,000 woman-yearsof follow-up after the time of switching (mean, 3.9 years); mean andmedian follow-up times were 6.1 and 6.0 years, 4.4 and 4.4 years, 5.9and 6.0 years, and 3.0 and 2.9 years, respectively. The analyses ofrecurrence and death are calculated from the time the allocated treat-ments would differ. (The time from diagnosis is indicated in thegraphs as being approximately 2 years greater than this.)

There was an overall 29% (SE � 6%) proportional reduction inrecurrence, with no significant heterogeneity between the contribut-ing trials (Appendix Fig A2A, online only). The absolute cumulative

risk of recurrence was reduced by 3.1% (SE � 0.6%) 3 years afterswitching (ie, approximately 5 years after diagnosis) and by 3.6%(SE � 1.1%) 6 years after switching (ie, approximately 8 years afterdiagnosis; Fig 4A). There was a significantly greater risk reductionduring the first 3 years after switching (ie, while treatment with AI ortamoxifen was continuing) than during subsequent years (ie, aftercessation of treatment; 2P � .003 for heterogeneity; Fig 5A). Again,there seemed to be greater proportional reductions in isolated localrecurrence (40%; SE � 13%; 2P � .002; Appendix Fig A2B) andcontralateral disease (35%; SE � 17%; 2P � .03; Appendix Fig A2C)than in distant recurrence either as the first event (24%; SE � 7%;2P � .001; Appendix Fig A2D) or at any time (24%; SE � 7%;2P � .0007; Appendix Fig A2E), but this apparent heterogeneity ofeffect between different sites of recurrence was not significant.

Subgroup analyses of the recurrence results with respect toPgR status, age, nodal status, and tumor grade are shown in Figures5B to 5E. The global test of heterogeneity for these four subgroupanalyses was not significant. There was no apparent heterogeneitywith respect to any of the four subgroups in the proportionalreductions in recurrence. PgR status was known in 8,184 patients(91%). The proportional reduction in recurrence was 37%(SE � 12%) in women with ER-positive/PgR-poor disease and21% (SE � 8%) in women with ER-positive/PgR-positive disease,but these proportional risk reductions are not significantly differ-ent from each other.

There was a significant reduction in breast cancer mortality(proportional reduction in mortality with recurrence � 22%;SE � 9%; 2P � .02; Fig 6A). The absolute difference in breastcancer mortality was 0.7% (SE � 0.6%) approximately 5 years afterdiagnosis and 1.7% (SE � 0.8%) approximately 8 years after diag-nosis (Figs 4B and 6A), but the latter result in particular maychange somewhat when more women have been observed to this

0 0.5 1.0 1.5 2.0

AI better Tamoxifen better


Deaths/Women Al deathsYear code Allocated Allocated Log-rank Variance Ratio of annual death ratesand study name Treatment comparison AI tamoxifen O−E of O−E AI : Tamoxifen

(A) Mortality with recurrence (χ21 = 0.2, 2P > .1, NS)

96J CTC ATAC, UK (Ana v Tam) 5 yr 223/2,548 240/2,514 −10.6 111.3 98M IBCSG 18-98/BIG 1-98 (Let v Tam) 5 yr 112/2,406 127/2,388 −9.3 57.8

Subtotal 335/4,954 367/4,902 −19.9 169.1 0.89 (SE 0.07) (6.8%) (7.5%) reduction 2P > .1, NS(B) Mortality without recurrence (χ2

1 = 0.1, 2P > .1, NS)

96J CTC ATAC, UK (Ana v Tam) 5 yr 213/2,548 200/2,514 2.4 101.8 98M IBCSG 18-98/BIG 1-98 (Let v Tam) 5yr 73/2,406 73/2,388 −0.9 36.2 Subtotal 286/4,954 273/4,902 1.5 138.0 1.01 (SE 0.09) (5.8%) (5.6%) increase 2P > .1, NS(C) Any death (χ2

1 = 0.3, 2P > .1, NS)

96J CTC ATAC, UK (Ana v Tam) 5 yr 436/2,548 440/2,514 −8.2 213.1 98M IBCSG 18-98/BIG 1-98 (Let v Tam) 5 yr 185/2,406 200/2,388 −10.2 94.0

Subtotal 621/4,954 640/4,902 −18.4 307.1 0.94 (SE 0.06) (12.5%) (13.1%) reduction 2P > .1, NS

Fig 3. Trial-specific forest plot of (A) mortality with recurrence; (B) mortality without recurrence; and (C) any death, for estrogen receptor–positive patients in trials ofapproximately 5 years of aromatase inhibitor (AI) versus tamoxifen (Tam). O, observed; E, expected; NS, not significant; CTC, Cancer Trials Centre; ATAC, Arimidex, Tamoxifen,Alone or in Combination; UK, United Kingdom; Ana, anastrozole; IBCSG, International Breast Cancer Study Group; BIG, Breast International Group; Let, letrozole.





point. Of the 638 deaths from any cause, 276 (43%) were of womenwho had not experienced a recurrence and had died of an unrelatedcause. There was a nonsignificant reduction in such deaths (Figs 4Cand 6B). The absolute difference in overall mortality was 1.1%(SE � 0.5%) approximately 5 years after diagnosis (Figs 4D and6C). It seemed to grow larger over the next few years and was 2.2%(SE � 1.0%; 2P � .004) approximately 8 years after diagnosis, but

this too may change with further follow-up in these trials and withthe availability of more trials.

DISCUSSION

Each cohort in this overview included between 9,000 and 10,000patients, and in aggregate, the two cohorts included almost 19,000

3-year gain, 3.1% (SE, 0.6%)6-year gain, 3.6% (SE, 1.1%)Log-rank 2P < .00001

Recurrence rates (% / year) and log-rank analyses Years 0–2 (≈2–4) Years 2–4 (≈5–7) Years 5+ (≈8+)AI 1.68 (187 / 11,134) 2.78 (147 / 5,298) 3.21 (23 / 716)Tamoxifen 2.76 (303 / 10,962) 2.99 (149 / 5,007) 3.87 (27 / 697)Rate ratio, from 0.60 (SE, 0.07) 0.92 (SE, 0.11) 0.85 (SE, 0.27)(O−E) / V -61.0 / 118.4 -6.0 / 71.9 -2.0 / 12.1

A

0 0

0 0

Recu

rren

ce (%

, ± S

E)

Time Since Allocated Treatments Differed (years)(and approximate time since diagnosis)

50

40

30

20

10

03 6


Death rates (%/year: total rate—rate in women without recurrence) & log-rank analyses Years 0–2 (≈2–4) Years 2–4 (≈5–7) Years 5+ (≈8+)AI 0.54 (SE, 0.07) 1.60 (SE, 0.17) 1.17 (SE, 0.39)Tamoxifen 0.79 (SE, 0.08) 1.83 (SE, 0.18) 1.80 (SE, 0.48)Rate ratio, from 0.68 (SE, 0.14) 0.88 (SE, 0.14) 0.65 (SE, 0.34)(O−E) / V -13.9 / 36.5 -5.8 / 45.5 -2.5 / 5.7

B

Brea

st C

ance

r Mor

talit

y (%

, ± S

E)


50

40

30

20

10

0

0 0

3 6


Death rates (% / year) and log-rank analyses Years 0–2 (≈2–4) Years 2–4 (≈5–7) Years 5+ (≈8+)AI 0.57 (64 / 11,134) 1.04 (55 / 5,298) 1.12 (8 / 716)Tamoxifen 0.70 (77 / 10,962) 1.22 (61 / 5,007) 1.58 (11 / 697)Rate ratio, from 0.82 (SE, 0.15) 0.85 (SE, 0.17) 0.68 (SE, 0.38)(O−E) / V -7.0 / 35.0 -4.9 / 28.5 -1.8 / 4.7

C

Deat

h W

ithou

t Rec

urre

nce

(%, ±

SE)


50

40

30

20

10

3 6


Death rates (% / year) and log-rank analyses Years 0–2 (≈2–4) Years 2–4 (≈5–7) Years 5+ (≈8+)AI 1.11 (125 / 11,300) 2.59 (144 / 5,566) 2.20 (17 / 771)Tamoxifen 1.47 (166 / 11,301) 2.95 (161 / 5,455) 3.22 (25 / 776)Rate ratio, from 0.75 (SE, 0.10) 0.87 (SE, 0.11) 0.66 (SE, 0.25)(O−E) / V −20.9 / 71.5 −10.8 / 74.2 -4.3 / 10.3

DAn

y De

ath

(%, ±

SE)


50

40

30

20

10

3 6

AITamoxifen

AITamoxifen

AITamoxifen

AITamoxifen

16.0%

12.6%8.1%

5.0%

7.9%6.3%2.4%

1.7%

2.1%

1.7%

13.0%

10.8%

4.4%

3.3%

(≈0) (≈2) (≈5) (≈8) (≈0) (≈2) (≈5) (≈8)

(≈0) (≈2) (≈5) (≈8) (≈0) (≈2) (≈5) (≈8)

5.8%5.0%

Fig 4. Life-table curves of (A) recurrence; (B) breast cancer mortality; (C) death without recurrence; and (D) any death, for estrogen receptor–positive patients in trialsof 2 to 3 years of tamoxifen and then 2 to 3 years of aromatase inhibitor (AI) versus tamoxifen. O, observed; E, expected; V, variance.

Dowsett et al




patients in only six trials. These relatively large numbers per trialreflected the expectation of only a small incremental improvement inrecurrence-free survival given the relatively good prognosis oftamoxifen-treated, ER-positive patients.2 Overview analysis remains

of value for the characterization of the time dependence of the effectson recurrence; for the eventual effects on 5, 10, and 15-year survival;for subgroup analyses; and for identification of any uncommon favor-able or unfavorable adverse effects (not addressed here).

0 0.5 1.0 1.5 2.0

AI better Tamoxifen betterGlobal heterogeneity (B−E): χ24 = 2.7, P > .1, NS


Events/Women AI events Allocated Allocated Log-rank Variance Ratio of annual event ratesCategory AI tamoxifen O−E of O−E AI : Tamoxifen

(A) Time since treatments differed* (trend χ21 = 8.7, 2P = .003)

Years 0−2 (≈2−4) 187/4,508 303/4,507 −61.0 118.4 0.60 (SE 0.07) (4.1%) (6.7%)

Years 3+ (≈5+) 170/2,978 176/2,854 −7.9 84.0 0.91 (SE 0.10) (5.7%) (6.2%)

(B) PgR status (χ21 = 1.6, 2P > .1, NS)

PgR poor 64/704 107/764 −19.1 41.1 0.63 (SE 0.12) (9.1%) (14.0%)

PgR+ 240/3,375 294/3,341 −30.7 130.5 0.79 (SE 0.08) (7.1%) (8.8%)

PgR unknown 53/429 78/402 −16.0 31.5 0.60 (SE 0.14) (12.4%) (19.4%)

(C) Entry age (trend χ21 = 0.0, 2P > .1, NS)

Age < 50 12/119 15/130 −1.2 6.5 (10.1%) (11.5%)

50–59 114/1,423 141/1,393 −17.9 61.8 0.75 (SE 0.11) (8.0%) (10.1%)

60–69 145/1,903 216/1,869 −38.7 87.6 0.64 (SE 0.09) (7.6%) (11.6%)

70+ 86/1,063 107/1,115 −9.9 46.5 0.81 (SE 0.13) (8.1%) (9.6%)

(D) Nodal status (N0/N− v all N+ χ21 = 1.0, 2P > .1, NS)

N0/N− 96/2,399 123/2,414 −13.9 54.0 0.77 (SE 0.12) (4.0%) (5.1%)

N1−3 34/565 55/531 −12.3 21.4 0.56 (SE 0.16) (6.0%) (10.4%)

N4+ 20/124 39/144 −8.0 13.9 0.56 (SE 0.20) (16.1%) (27.1%)

N+ unspec. 156/912 207/897 −30.9 86.4 0.70 (SE 0.09) (17.1%) (23.1%)

N unknown 51/508 55/521 −2.1 26.2 (10.0%) (10.6%)

(E) Tumor grade (trend χ21 = 0.1, 2P > .1, NS)

Well differentiated 46/788 49/786 −1.7 22.8 (5.8%) (6.2%)

Moderately 181/2,676 248/2,703 −37.7 103.6 0.69 (SE 0.08) (6.8%) (9.2%)

Poorly 88/529 88/492 −4.6 41.4 0.89 (SE 0.15) (16.6%) (17.9%)

Grade unknown 42/515 94/526 −25.3 32.5 0.46 (SE 0.12) (8.2%) (17.9%)

Total 357/4,508 479/4,507 −68.9 202.4 0.711 (SE 0.060) (7.9%) (10.6%) 2P < .00001

Treatment effect 2P < .00001

Fig 5. Subgroup analyses of recurrencesby (A) time period; (B) progesterone re-ceptor (PgR) status; (C) entry age; (D)nodal status; and (E) tumor grade, forestrogen receptor–positive patients in tri-als of 2 to 3 years of tamoxifen and then 2to 3 years of aromatase inhibitor (AI) ver-sus tamoxifen. Dashed vertical line corre-sponds to zero heterogeneity of effect. (*)Values in parentheses are approximate timesince diagnosis. O, observed; E, expected;unspec., unspecified; NS, not significant.





Most of the trials recruited patients known to be hormone recep-tor (ER or PgR) positive. Patients recorded as having ER-negative/PgR-positive tumors, who were included in that hormone receptor–positive definition, are commonly treated with hormone treatment,but the overview analyses of 5 years of tamoxifen versus no treatmentshow no significant benefit in this subgroup.2 The analyses presentedhere therefore exclude them and are restricted to the ER-positivesubset of patients. In recent years, progressively smaller proportions ofER-negative tumors have been reported as PgR positive, and increas-ingly, the evidence suggests that most tumors recorded as ER-negative/PgR-positive generally represent either a false-negative ERmeasurement or a false-positive PgR measurement. Therefore, it re-mains reasonable to treat such patients with endocrine therapy in casesome can respond. As a sensitivity analysis, the recurrence reductionswere recalculated for all patients with ER-positive and/or PgR-positivedisease, and there was no material difference in the findings (datanot shown).

Each of the third-generation AIs used in the trials in this meta-analysis has been shown to achieve near-complete inhibition of aro-matase in vivo and near-complete suppression of plasma estrogenlevels.13,14 Minor pharmacologic differences exist between the drugs,and there are ongoing trials to determine whether there are result-ant differences between them in efficacy and tolerability; attemptsto compare different AIs by indirect between-trial comparisonsare inappropriate.

However, the two cohorts of trials were analyzed separatelybecause cohort 1 includes patients who would experience relapseon tamoxifen in the first 2 to 3 years after surgery and cohort 2does not. Direct comparison would be invalid because cohort 2excludes patients with a particularly unfavorable outcomeon tamoxifen.

Cohort 1 comprises data from ATAC at 95 months of medianfollow-up and from BIG 1-98 at 48 months of median follow-up.Accordingly, approximately two thirds of the statistical informationfor the overview results in cohort 1 is provided by ATAC, and thepreponderant influence of ATAC in cohort 1 is particularly marked inlonger follow-up (5 to 8 years). Updated results from BIG 1-9815 haverecently been published elsewhere.

The use of 5 years of an AI immediately after surgery (cohort 1)further reduced recurrence by approximately 23% relative to 5years of tamoxifen. In cohort 2, when the switch to an AI is madeafter 2 to 3 years of tamoxifen, there seems to be a particularlymarked reduction in risk of recurrence (by 40%) during the fol-lowing 3 years. There is evidence that disease relapsing after tamox-ifen treatment can show substantial differences in molecularphenotype from the disease at presentation.16 If this occurs insubclinical disease during adjuvant therapy, it is possible that thedisease may take on an enhanced sensitivity to AI treatment thatcould underpin this effect, but there are no substantive data todemonstrate this.

0 0.5 1.0 1.5 2.0

AI better Tamoxifen better


Deaths/Women Al deathsYear code Allocated Allocated Log-rank Variance Ratio of annual death ratesand study name Treatment comparison AI tamoxifen O−E of O−E AI : Tamoxifen

(A) Mortality with recurrence (χ23 = 2.8, 2P > .1, NS)

94V GABG/ARNO Germany Tam 2 yr, Ana v Tam 3 yr 15/447 22/455 −3.4 9.0 96Y1 Austrian BCSG VIII Tam 2 yr, Ana v Tam 3 yr * 15/1,820 24/1,807 −4.2 9.5 98C IES / BIG 02-97 Tam 2−3 yr, Exe v Tam 3−2 yr 117/2,021 133/2,021 −8.6 60.4 98R ITA Italy Tam 2 yr, Ana v Tam to ≈ 5 yr 12/220 24/224 −6.0 8.8 Subtotal 159/4,508 203/4,507 −22.2 87.7 0.78 (SE 0.09) (3.5%) (4.5%) reduction 2P = .02

(B) Mortality without recurrence (χ23 = 5.0, 2P > .1, NS)

94V GABG/ARNO Germany Tam 2 yr, Ana v Tam 3 yr 5/447 14/455 −4.5 4.7 96Y1 Austrian BCSG VIII Tam 2 yr, Ana v Tam 3 yr * 56/1,820 56/1,807 −1.4 27.6 98C IES / BIG 02-97 Tam 2−3 yr, Exe v Tam 3−2 yr 61/2,021 77/2,021 −9.1 34.2 98R ITA Italy Tam 2 yr, Ana v Tam to ≈ 5 yr 5/220 2/224 1.3 1.7

Subtotal 127/4,508 149/4,507 −13.7 68.2 0.82 (SE 0.11) (2.8%) (3.3%) reduction 2P = .10

(C) Any death (χ23 = 2.6, 2P > .1, NS)

94V GABG/ARNO Germany Tam 2 yr, Ana v Tam 3 yr 20/447 36/455 −7.9 13.8 96Y1 Austrian BCSG VIII Tam 2 yr, Ana v Tam 3 yr * 71/1,820 80/1,807 −5.6 37.1 98C IES / BIG 02-97 Tam 2−3 yr, Exe v Tam 3−2 yr 178/2,021 210/2,021 −17.7 94.6 98R ITA Italy Tam 2 yr, Ana v Tam to ≈ 5 yr 17/220 26/224 −4.8 10.5

Subtotal 286/4,508 352/4,507 −36.0 156.0 0.79 (SE 0.07) (6.3%) (7.8%) reduction 2P = .004

Fig 6. Trial-specific forest plot of (A) mortality with recurrence; (B) mortality without recurrence; and (C) any death, for estrogen receptor–positive patients in trials of2 to 3 years of tamoxifen (Tam) and then 2 to 3 years of aromatase inhibitor (AI) versus tamoxifen. (*) Random allocation at start of tamoxifen. O, observed; E, expected;NS, not significant; GABG, German Adjuvant Breast Cancer Group; ARNO, Arimidex-Nolvadex; Ana, anastrozole; Austrian BCSG, Austrian Breast and Colorectal CancerStudy Group; IES, Intergroup Exemestane Study; BIG, Breast International Group; ITA, Italian Tamoxifen Anastrozole trial; Exe, exemestane.

Dowsett et al




When examined in relation to time on treatment versus timeafter cessation of treatment, there was some evidence for an attenua-tion of the reduced risk of recurrence that was stronger for cohort 2.However, for both cohorts, the CIs of the hazard ratios for the periodafter cessation of therapy were wide, underscoring the importance ofcontinued follow-up and the ongoing clinical trials examining exten-sions to the duration of AI therapy.

The global test for heterogeneity of the subgroups was not signif-icant in either cohort, although in cohort 1, the subset according toPgR status individually was significant. Previous trials have demon-strated that 5 years of tamoxifen is highly, and approximately equally,effective in both ER-positive/PgR-positive and ER-positive/PgR-poordisease.2 The current data seem to suggest that patients reported to bePgR poor derive greater benefit from an AI. However, the two trialscontributing these data differ substantially in their findings accordingto PgR status. In addition, central analyses of PgR (as opposed to thepredominately locally performed receptor analyses in the overview)suggest similar proportional treatment benefit for PgR-positive andPgR-poor populations in both trials.17,18 Thus, the apparent hetero-geneity of effect with regard to PgR status in cohort 1 may well be achance finding. Widely acknowledged difficulties in reproducingPgR status may contribute to these variable findings. Cohort 2showed no suggestion of increased effectiveness of AIs over tamox-ifen in the PgR-poor group. The overview indicates that the enhancedefficacy of AIs over tamoxifen is proportionally similar across allcohorts examined.

In the trials studying approximately 5 years of adjuvant tamox-ifen versus no tamoxifen in ER-positive disease, the main effect oftamoxifen on breast cancer recurrence was seen in the first 5 years(absolute gain of 12.5% at 5 years and roughly constant thereafter),but the main effect on breast cancer mortality was seen later (absolutegain of 3.3% at 5 years and 9.1% at 15 years).2 In the present trials ofAIs versus tamoxifen, the mean follow-up time thus far available is, inboth cohorts, only about 6 years after diagnosis (mean follow-upduration after random assignment was 5.8 years in cohort 1 and 3.9years in cohort 2), and the absolute further recurrence reduction (AI vtamoxifen) 5 years after diagnosis was approximately 3% and highlysignificant in both cohorts (2.9%, SE � 0.7% in cohort 1; and 3.1%,SE � 0.6% in cohort 2). Analogy with the recurrence and mortalitypatterns in the tamoxifen trials suggests that the absolute additionaleffects (AI v tamoxifen) on breast cancer mortality 5 years after diag-nosis would be expected to be approximately 1% in both cohorts,whereas the observed differences were 1.1% (SE � 0.5%) in cohort 1and 0.7% (SE � 0.3%) in cohort 2, but with further follow-up, thebreast cancer mortality differences may increase. Limited follow-upbeyond 5 years after surgery is so far available from either cohort; atthis stage, the data from cohort 2 but not from cohort 1 provideevidence for such a further reduction in breast cancer mortality. Aswith other end points, the mortality results from cohorts 1 and 2should not be compared directly.

There was no indication that the AIs were associated with anincrease in non–breast cancer deaths. This is reassuring regarding theoverall safety of AIs relative to tamoxifen. The overall mortality reduc-tions for cohort 2 but not cohort 1 (AI v tamoxifen) achieved statisticalsignificance. As recently discussed,19 the combination of safety analy-ses (integrated here as non–breast cancer deaths) with efficacy analy-ses (breast cancer–specific deaths) to a summary measure (overallmortality) can lead to substantial loss of information by mixing im-

portant signals with irrelevant noise. Analyses of cause-specific mor-tality have not been conducted at this time but will be an importantcomponent of later overviews.

This overview provides clear evidence that third-generation AIsachieve modest absolute improvements in breast cancer end pointswith significant reductions in recurrence in both cohorts and in breastcancer–specific mortality in cohort 2. Because the proportional bene-fits do not vary significantly with factors affecting prognosis, the abso-lute gain is greater for patients with poorer prognosis. There are alsosignificant differences in adverse effects between tamoxifen and AIsthat have not been reviewed here; AI use is associated with fewerendometrial cancers and thromboembolic events than tamoxifen butwith more arthralgia and fractures.20,21 The decision on whether toinitiate treatment with an AI or tamoxifen or whether to switch to anAI after 2 to 3 years of tamoxifen rather than continuing with tamox-ifen for 5 years depends on a careful evaluation of these factors inindividual patients. This overview has provided greater confidence inthe efficacy aspects of this evaluation.

AUTHORS’ DISCLOSURES OF POTENTIAL CONFLICTSOF INTEREST

Although all authors completed the disclosure declaration, the followingauthor(s) indicated a financial or other interest that is relevant to the subjectmatter under consideration in this article. Certain relationships markedwith a “U” are those for which no compensation was received; thoserelationships marked with a “C” were compensated. For a detaileddescription of the disclosure categories, or for more information aboutASCO’s conflict of interest policy, please refer to the Author DisclosureDeclaration and the Disclosures of Potential Conflicts of Interest section inInformation for Contributors.Employment or Leadership Position: Marc Buyse, IDDI (C); MichaelGnant, Merck (C) Consultant or Advisory Role: Mitch Dowsett,AstraZeneca (C), Novartis (C); Jack Cuzick, AstraZeneca (C); MichaelGnant, Novartis (C), Wyeth (C); Raimund Jakesz, Advisory BoardParticipation (C) Stock Ownership: Marc Buyse, IDDI Honoraria:Mitch Dowsett, AstraZeneca, Novartis; Jack Cuzick, AstraZeneca,Novartis, Eli Lilly; John Forbes, AstraZeneca, Novartis; Judith Bliss,Pfizer; Michael Baum, AstraZeneca; Aman Buzdar, Novartis,AstraZeneca; Marco Colleoni, Novartis; Michael Gnant, AstraZeneca,Novartis, Pfizer, Roche, sanofi-aventis, Amgen; Raimund Jakesz,AstraZeneca, Novartis; Manfred Kaufmann, AstraZeneca, Novartis,Pfizer Research Funding: Mitch Dowsett, AstraZeneca, Novartis; JackCuzick, AstraZeneca; Alan Coates, Novartis; Judith Bliss, Pfizer; AmanBuzdar, AstraZeneca; Marco Colleoni, Novartis; Charles Coombes,Pfizer; Michael Gnant, AstraZeneca, Novartis, sanofi-aventis, Roche,Amgen; Raimund Jakesz, Selected research projects Expert Testimony:None Other Remuneration: Alan Coates, Novartis

AUTHOR CONTRIBUTIONS

Conception and design: Mitch Dowsett, Jack Cuzick, Jim Ingle,Richard PetoAdministrative support: Jon Godwin, Christina DaviesProvision of study materials or patients: Alan Coates, John Forbes,Judith Bliss, Michael Baum, Aman Buzdar, Marco Colleoni, CharlesCoombes, Claire Snowdon, Michael Gnant, Raimund Jakesz, ManfredKaufmann, Francesco BoccardoCollection and assembly of data: Mitch Dowsett, Jack Cuzick, Jim Ingle,Alan Coates, John Forbes, Judith Bliss, Claire Snowdon, Raimund Jakesz,Jon Godwin, Christina Davies, Richard Peto





Data analysis and interpretation: Mitch Dowsett, Jack Cuzick, JimIngle, Alan Coates, John Forbes, Judith Bliss, Marc Buyse, MichaelBaum, Aman Buzdar, Marco Colleoni, Michael Gnant, Jon Godwin,Christina Davies, Richard PetoManuscript writing: Mitch Dowsett, Jack Cuzick, Jim Ingle, AlanCoates, John Forbes, Judith Bliss, Marc Buyse, Michael Baum, AmanBuzdar, Marco Colleoni, Charles Coombes, Michael Gnant, Raimund

Jakesz, Manfred Kaufmann, Francesco Boccardo, Jon Godwin,Richard PetoFinal approval of manuscript: Mitch Dowsett, Jack Cuzick, Jim Ingle,Alan Coates, John Forbes, Judith Bliss, Marc Buyse, Michael Baum,Aman Buzdar, Marco Colleoni, Charles Coombes, Claire Snowdon,Michael Gnant, Raimund Jakesz, Manfred Kaufmann, FrancescoBoccardo, Jon Godwin, Christina Davies, Richard Peto

REFERENCES

1. Jordan VC: The science of selective estrogenreceptor modulators: Concept to clinical practice.Clin Cancer Res 12:5010-5013, 2006

2. Early Breast Cancer Trialists’ CollaborativeGroup: Effects of chemotherapy and hormonal ther-apy for early breast cancer recurrence and 15-yearsurvival: An overview of the randomised trials. Lan-cet 365:1687-1717, 2005

3. Early Breast Cancer Trialists’ CollaborativeGroup: Effects of adjuvant tamoxifen and of cyto-toxic therapy on mortality in early breast cancer: Anoverview of 61 randomised trials among 28,896women. N Engl J Med 319:1681-1692, 1988

4. Dowsett M, Howell A: Breast cancer: Aro-matase inhibitors take on tamoxifen. Nat Med8:1341-1344, 2002

5. Smith IE, Dowsett M: Aromatase inhibitors inbreast cancer. N Engl J Med 348:2431-2442, 2003

6. Forbes JF, Cuzick J, Buzdar A, et al: Effect ofanastrozole and tamoxifen as adjuvant treatment forearly-stage breast cancer: 100-month analysis of theATAC trial—Arimidex, Tamoxifen, Alone or in Com-bination (ATAC) Trialists’ Group. Lancet Oncol 9:45-53, 2008

7. Coates AS, Keshaviah A, Thurlimann B, et al:Five years of letrozole compared with tamoxifen asinitial adjuvant therapy for postmenopausal womenwith endocrine-responsive early breast cancer: Up-date of study BIG 1-98. J Clin Oncol 25:486-492,2007

8. Jakesz R, Jonat W, Gnant M, et al: Switch-ing of postmenopausal women with endocrine-responsive early breast cancer to anastrozole after 2years’ adjuvant tamoxifen: Combined results ofABCSG trial 8 and ARNO 95 trial. Lancet 366:455-462, 2005

9. Boccardo F, Rubagotti A, Guglielmini P, et al:Switching to anastrozole versus continued tamox-ifen treatment of early breast cancer: Updated re-sults of the Italian Tamoxifen Anastrozole (ITA) trial.Ann Oncol 17:vii10-vii14, 2006 (suppl 7)

10. Coombes RC, Kilburn LS, Snowdon CF, et al:Survival and safety of exemestane versus tamoxifenafter 2-3 years’ tamoxifen treatment (IntergroupExemestane Study): A randomised controlled trial.Lancet 369:559-570, 2007

11. Cuzick J: Forest plots and the interpretation ofsubgroups. Lancet 365:1308, 2005

12. Early Breast Cancer Trialists’ CollaborativeGroup: Treatment of Early Breast Cancer. Volume 1:Worldwide Evidence, 1985-1990. Oxford, UnitedKingdom, Oxford University Press, 1990

13. Geisler J, Haynes B, Anker G, et al: Influenceof letrozole and anastrozole on total body aromati-zation and plasma estrogen levels in postmeno-pausal breast cancer patients evaluated in arandomized, cross-over study. J Clin Oncol 20:751-757, 2002

14. Geisler J, King N, Anker G, et al: In vivoinhibition of aromatization by exemestane, a novelirreversible aromatase inhibitor, in postmenopausalbreast cancer patients. Clin Cancer Res 4:2089-2093, 1998

15. Mouridsen HT, Giobbie-Hurder A, GoldhirschA, et al: Letrozole therapy alone or in sequence withtamoxifen in women with breast cancer. N ENglJ Med 36:766-776, 2009. Cancer Res 69:66s, 2009(suppl)

16. Gutierrez MC, Detre S, Johnston S, et al:Molecular changes in tamoxifen-resistant breastcancer: Relationship between estrogen receptor,HER-2, and p38 mitogen-activated protein kinase.J Clin Oncol 23:2469-2476, 2005

17. Viale G, Regan MM, Maiorano E, et al: Prog-nostic and predictive value of centrally reviewedexpression of estrogen and progesterone receptorsin a randomized trial comparing letrozole and tamox-ifen adjuvant therapy for postmenopausal earlybreast cancer: BIG 1-98. J Clin Oncol 25:3846-3852,2007

18. Dowsett M, Allred C, Knox J, et al: Relation-ship between quantitative estrogen and progester-one receptor expression and human epidermalgrowth factor receptor 2 (HER-2) status with recur-rence in the Arimidex, Tamoxifen, Alone or in Com-bination trial. J Clin Oncol 26:1059-1065, 2008

19. Cuzick J: Primary end-points for randomizedtrials of cancer therapy. Lancet 371:2156-2158,2008

20. Lin NU, Winer EP: Advances in adjuvant en-docrine therapy for postmenopausal women. J ClinOncol 26:798-805, 2008

21. Lønning PE, Geisler J: Indications and limita-tions of third-generation aromatase inhibitors. Ex-pert Opin Investig Drugs 17:723-739, 2008

■ ■ ■

Dowsett et al




Meta-Analysis of Breast Cancer Outcomes in Adjuvant Trials of Aromatase Inhibitors Versus Tamoxifen

Documents

Transcript of Meta-Analysis of Breast Cancer Outcomes in Adjuvant Trials of Aromatase Inhibitors Versus Tamoxifen