Aromatase inhibitors versus tamoxifen in early breast cancer: patient-level meta-analysis of the randomised trials

Articles www.thelancet.com Vol 386 October 3, 2015 1341 Aromatase inhibitors versus tamoxifen in early breast cancer: patient-level meta-analysis of the randomised trials Early Breast Cancer Trialists’ Collaborative Group (EBCTCG)* Summary Background The optimal ways of using aromatase inhibitors or tamoxifen as endocrine treatment for early breast cancer remains uncertain. Methods We undertook meta-analyses of individual data on 31 920 postmenopausal women with oestrogen-receptorpositive early breast cancer in the randomised trials of 5 years of aromatase inhibitor versus 5 years of tamoxifen; of 5 years of aromatase inhibitor versus 2–3 years of tamoxifen then aromatase inhibitor to year 5; and of 2–3 years of tamoxifen then aromatase inhibitor to year 5 versus 5 years of tamoxifen. Primary outcomes were any recurrence of breast cancer, breast cancer mortality, death without recurrence, and all-cause mortality. Intention-to-treat log-rank analyses, stratifi ed by age, nodal status, and trial, yielded aromatase inhibitor versus tamoxifen fi rst-event rate ratios (RRs). Findings In the comparison of 5 years of aromatase inhibitor versus 5 years of tamoxifen, recurrence RRs favoured aromatase inhibitors signifi cantly during years 0–1 (RR 0·64, 95% CI 0·52–0·78) and 2–4 (RR 0·80, 0·68–0·93), and non-signifi cantly thereafter. 10-year breast cancer mortality was lower with aromatase inhibitors than tamoxifen (12·1% vs 14·2%; RR 0·85, 0·75–0·96; 2p=0·009). In the comparison of 5 years of aromatase inhibitor versus 2–3 years of tamoxifen then aromatase inhibitor to year 5, recurrence RRs favoured aromatase inhibitors signifi cantly during years 0–1 (RR 0·74, 0·62–0·89) but not while both groups received aromatase inhibitors during years 2–4, or thereafter; overall in these trials, there were fewer recurrences with 5 years of aromatase inhibitors than with tamoxifen then aromatase inhibitors (RR 0·90, 0·81–0·99; 2p=0·045), though the breast cancer mortality reduction was not signifi cant (RR 0·89, 0·78–1·03; 2p=0·11). In the comparison of 2–3 years of tamoxifen then aromatase inhibitor to year 5 versus 5 years of tamoxifen, recurrence RRs favoured aromatase inhibitors signifi cantly during years 2–4 (RR 0·56, 0·46–0·67) but not subsequently, and 10-year breast cancer mortality was lower with switching to aromatase inhibitors than with remaining on tamoxifen (8·7% vs 10·1%; 2p=0·015). Aggregating all three types of comparison, recurrence RRs favoured aromatase inhibitors during periods when treatments diff ered (RR 0·70, 0·64–0·77), but not signifi cantly thereafter (RR 0·93, 0·86–1·01; 2p=0·08). Breast cancer mortality was reduced both while treatments diff ered (RR 0·79, 0·67–0·92), and subsequently (RR 0·89, 0·81–0·99), and for all periods combined (RR 0·86, 0·80–0·94; 2p=0·0005). All-cause mortality was also reduced (RR 0·88, 0·82–0·94; 2p=0·0003). RRs diff ered little by age, body-mass index, stage, grade, progesterone receptor status, or HER2 status. There were fewer endometrial cancers with aromatase inhibitors than tamoxifen (10-year incidence 0·4% vs 1·2%; RR 0·33, 0·21–0·51) but more bone fractures (5-year risk 8·2% vs 5·5%; RR 1·42, 1·28–1·57); non-breast-cancer mortality was similar. Interpretation Aromatase inhibitors reduce recurrence rates by about 30% (proportionately) compared with tamoxifen while treatments diff er, but not thereafter. 5 years of an aromatase inhibitor reduces 10-year breast cancer mortality rates by about 15% compared with 5 years of tamoxifen, hence by about 40% (proportionately) compared with no endocrine treatment. Funding Cancer Research UK, Medical Research Council. Copyright © Early Breast Cancer Trialists’ Collaborative Group (EBCTCG). Open Access article distributed under the terms of CC BY. Introduction Treatment for 5 years with the selective oestrogen receptor (ER) modulator tamoxifen reduces recurrence rates in ER-positive early breast cancer by about half during treatment and about one-third in the subsequent 5 years, and reduces breast cancer mortality by almost one-third throughout the fi rst 15 years.1 Further reductions in breast cancer mortality during years 10–14 are achieved by extending tamoxifen treatment to 10 years.2,3 In postmenopausal women only, aromatase inhibitors can greatly reduce oestrogen concentrations, hence avoiding stimulation of ER-positive breast cancer cells. Aromatase inhibitors, given either for 5 years or for 2–3 years after 2–3 years of tamoxifen, produce greater reductions in recurrence than 5 years of tamoxifen alone,4 but the eff ect on breast cancer mortality, and the optimal way to schedule aromatase inhibitors and tamoxifen in the treatment of early breast cancer, remain uncertain. Lancet 2015; 386: 1341–52 Published Online July 24, 2015 http://dx.doi.org/10.1016/ S0140-6736(15)61074-1 See Comment page 1317 *Full list of members available at http://www.ctsu.ox.ac.uk/ research/meta-trials/ebctcg/ ebctcg-page Correspondence to: EBCTCG Secretariat, Clinical Trial Service Unit, Nuffi eld Department of Population Health, Richard Doll Building, Oxford OX3 7LF, UK [email protected] Articles 1342 www.thelancet.com Vol 386 October 3, 2015 American Society of Clinical Oncology (ASCO) clinical practice guidelines refl ect this, recom mending that postmenopausal women with early ER-positive breast cancer be off ered either tamoxifen for 10 years, an aromatase inhibitor for 5 years, tamoxifen initially for 5 years followed by an aromatase inhibitor for up to a further 5 years, or tamoxifen for 2–3 years followed by an aromatase inhibitor for up to a further 5 years.5 To help clarify the relative benefi ts of aromatase inhibitors and tamoxifen and the eff ect of diff erent scheduling during 5 years of endocrine therapy, we undertook collaborative meta-analyses of individual patient data from the trials of aromatase inhibitors versus tamoxifen. Methods Identifi cation of studies and collection of data Trial identifi cation, data checking, analysis, and involvement of trialists are as described in previous Early Breast Cancer Trialists’ Collaborative Group (EBCTCG) reports.1,6,7 Eligible trials began by 2005 and randomised postmenopausal women with ER-positive early breast cancer between 5 years of an aromatase inhibitor versus 5 years of tamoxifen (comparison A); 5 years of aromatase inhibitor versus 2–3 years of tamoxifen, then aromatase inhibitor to year 5 (comparison B); 2–3 years of tamoxifen, then aromatase inhibitor to year 5 versus 5 years of tamoxifen (comparison C); 5 years of aromatase inhibitor versus 2 years of aromatase inhibitor, then tamoxifen to year 5 (comparison D); or 2 years of aromatase inhibitor, then tamoxifen to year 5 versus 5 years of tamoxifen (comparison E). Separate analyses are provided for each of these comparisons (A–E), then results from some of them are combined. Information was sought during 2012–14 for each individual patient on randomisation date, allocated treatment, age, menopausal status, body-mass index (BMI), tumour diameter, grade, spread to locoregional lymph nodes, ER, progesterone receptor (PR), and HER2 receptor status, and dates of any locoregional, contralateral, or distant breast cancer recurrence, other second primary cancer, bone fractures, death, and cause of death. Outcomes The primary outcomes were any recurrence of breast cancer (distant, locoregional, or new primary in the contralateral breast); breast cancer mortality; death without recurrence; and all-cause mortality. Secondary outcomes were incidence and site of second primary cancers, and bone fracture. Prespecifi ed primary subgroup investigations were of site of recurrence, age, nodal status, PR status, histological grade, and follow-up period. Statistical analyses Statistical methods (stratifi ed log-rank statistics, KaplanMeier graphs) are described elsewhere.1,6,7 Time-to-event analyses were stratifi ed by age, nodal status, and trial. Within each stratum, they compared all those allocated aromatase inhibitor versus all those allocated tamoxifen, regardless of treatment compliance (yielding intention-totreat analyses). Log-rank statistics were used to assess the eff ects (aromatase inhibitor vs tamoxifen) on various outcomes, and, for each, to estimate fi rst-event-rate ratios (RRs) and their CIs. If a log-rank statistic (o – e) has variance v, then, defi ning z=(o – e)/√v and b=(o – e)/v, b has variance 1/v and the event RR (newer treatment vs control) is estimated as exp(b) with SE=(RR – 1)/z. CIs for RR are derived from those for b (by normal approximations). 2p indicates two-sided signifi cance. The breast cancer mortality rate in each year is the overall mortality rate among all women minus that among women of similar age without recurrence. Breast cancer mortality RRs are estimated from the corresponding log-rank analyses of mortality with recurrence (obtained by subtracting log-rank analyses of mortality without recurrence [ie, censored at recurrence] from those of overall mortality). Analyses used EBCTCG Fortran programs. The policy on data sharing from this study is available online. Role of the funding source The funders of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. The secretariat had full access to all data and the writing committee had fi nal responsibility for the decision to submit for publication. Results Individual patient datasets were provided for nine trials,8–16 including 35 129 (98%) of the 35 718 women randomised between aromatase inhibitor and tamoxifen as part of about 5 years of adjuvant endocrine treatment (appendix). This report is restricted to the 31 920 (91%) with ER-positive tumours of these 35 129 patients. All were randomised evenly between aromatase inhibitor and tamoxifen, though one trial (BIG 1-988) included a four-way randomisation that contributes data to all fi ve comparisons (A–E); the aggregated analyses avoid double counting its results. When reports emerged that patients on tamoxifen had their recurrence risk reduced by switching after 2–3 years to an aromatase inhibitor, crossover to an aromatase inhibitor from the tamoxifen-only group was systematically off ered in two trials (BIG 1-98,8 25% [619/2459] crossover; ABCSG-8,9 18% [341/1949] crossover). In eight trials, compliance was similar in both groups, but in TEAM10 56% (2698/4814) of those allocated tamoxifen then aromatase inhibitor versus 30% (1438/4852) of those allocated only aromatase inhibitor discontinued treatment prematurely. In comparison A (5 years of aromatase inhibitor vs 5 years of tamoxifen: two trials, n=9885), recurrence and mortality were both signifi cantly reduced (fi gure 1). The numbers with recurrence were 827 in the aromatase inhibitor group versus 964 in the tamoxifen group (p<0·00001), with separately signifi cant reductions during years 0–1 after surgery (RR 0·64, 95% CI For the CTSU policy on data sharing see http://www.ctsu.ox. ac.uk/research/data-accesspolicies/data-access-andsharing-policy/view See Online for appendix Articles www.thelancet.com Vol 386 October 3, 2015 1343 0·52–0·78) and during years 2–4 (RR 0·80, 0·68–0·93), but no signifi cant further eff ect after the scheduled treatment period, and little follow-up beyond year 10. The 10-year recurrence risk was 19·1% in the aromatase inhibitor group versus 22·7% in the tamoxifen group (diff erence 3·6%, 95% CI 1·7–5·4). Distant recurrence (RR 0·86, 95% CI 0·77–0·96; 2p=0·007), local recurrence (RR 0·74, 0·58–0·95; 2p=0·020), and contralateral recurrence (RR 0·62, 0·48–0·80; 2p=0·0003) were all reduced (appendix). Breast cancer mortality was also reduced (RR 0·85, 95% CI 0·75–0·96; 2p=0·009), as was all-cause mortality (936 vs 1000 deaths; RR 0·89, 0·81–0·97; 2p=0·010), even though half the deaths were from non-breast cancer causes that are little aff ected by treatment. In comparison B (5 years of aromatase inhibitor vs 2–3 years of tamoxifen then aromatase inhibitor to year 5: three trials, n=12 779), recurrence was signifi cantly reduced only during years 0–1 (RR 0·74, 95% CI 0·62–0·89; 2p=0·002), ie, when the treatments diff ered, and was similar during years 2–4 (RR 0·99, 0·85–1·15), when both groups were receiving an aromatase inhibitor (fi gure 2). There was no signifi cant further eff ect after year 5, but little follow-up beyond year 7. Perhaps because the period during which the treatments diff ered lasted only half as long as in comparison A, the absolute reductions in recurrence and mortality appeared smaller. The total numbers with recurrence were 705 in the aromatase inhibitor group versus 765 in the tamoxifen group (2p=0·045). Although breast cancer mortality appeared somewhat reduced (RR 0·89, 95% CI 0·78–1·03; 2p=0·11), this was not signifi cant, and nor were the eff ects on other mortality or all-cause mortality. In comparison C (2–3 years of tamoxifen then aromatase inhibitor to year 5 vs 5 years of tamoxifen: six trials, n=11 798), recurrence and mortality were both signifi cantly reduced (fi gure 3). Four trials did not randomise until after 2 years of tamoxifen, but two randomised at year 0; for comparability with the other four, only patients who completed 2 years of tamoxifen without recurrence or a second primary are included, but sensitivity analyses (appendix) show this exclusion made little diff erence. Starting from when treatments diverged, the numbers with recurrence were 753 in the aromatase inhibitor group versus 863 in the tamoxifen group (2p=0·0001). Allocation to an aromatase inhibitor reduced the recurrence rate during years 2–4 (RR 0·56, 95% CI 0·46–0·67; p<0·0001), with no signifi cant further eff ect on recurrence after the treatment period, and little follow-up beyond year 10. The 10-year recurrence risk was 17·0% in the aromatase inhibitor group versus 19·0% in the tamoxifen group (diff erence 2·0, 95% CI 0·2–3·8). Distant recurrence (RR 0·86, 95% CI 0·77–0·97; 2p=0·02), and contralateral recurrence (RR 0·67, 0·51–0·87; 2p=0·002) were both reduced (appendix). Breast cancer mortality was also reduced (RR 0·84, 95% CI 0·72–0·96; 2p=0·015), as was all-cause mortality (639 vs 764 deaths; RR 0·82, 0·73–0·91; 2p=0·0002), helped by what might have been a chance reduction in non-breast cancer mortality. The recurrence results already described for comparisons A–C are summarised in the appendix, using black squares for periods when the treatments diff ered (aromatase inhibitor in one group vs tamoxifen in the other) and open squares for periods when they did not. It also gives the comparisons D and E, which both derive from BIG 1-98.8 Comparison D was restricted to the 2558 women who were recurrence free and still on treatment after 2 years of aromatase inhibitor. Although they suggest no apparent gain from continuing to take an aromatase inhibitor rather than switching to tamoxifen after 2 years, the CIs were wide. Comparison E included 3060 women; the proportional recurrence reduction during years 0–1 (when the treatments diff ered) was similar to that in earlier comparisons, and the apparent fl uctuations in the recurrence RR during the period when the treatments no longer diff ered could well be chance. In each of comparisons A–C there was signifi cant benefi t only when treatments diff ered and not when they were the same in both groups. This pattern is even clearer when results from all fi ve comparisons are aggregated by time period (fi gure 4). Recurrence RRs favoured aromatase inhibitors during periods when treatments diff ered (RR 0·70, 95% CI 0·64–0·77), but not signifi cantly thereafter (RR 0·93, 0·86–1·01; 2p=0·08). The recurrence rate was about 30% lower with an aromatase inhibitor than with tamoxifen in years 0–1 (RR 0·70, 95% CI 0·61–0·80; 2p<0·0001), and in years 2–4 (RR 0·71, 0·62–0·80; 2p<0·0001). Combining trials where treatments diff ered only during years 0–1 and not during years 2–4, there was no reduction in recurrence during years 2–4 (RR 1·03, 95% CI 0·87–1·22). There was little further eff ect during years 5–9 when no further treatment was scheduled (RR 0·92, 95% CI 0·83–1·01), and little follow-up beyond year 10. Breast cancer mortality was reduced both while treatments diff ered (RR 0·79, 95% CI 0·67–0·92), and subsequently (RR 0·89, 0·81–0·99), and for all periods combined (RR 0·86, 0·80–0·94; p=0·0005; appendix). All-cause mortality was likewise reduced (RR 0·90, 95% CI 0·84–0·95; 2p=0·0005). To enhance statistical power, the main subgroup analyses of recurrence are restricted to the periods when aromatase inhibitor was directly compared with tamoxifen (fi gure 5). The fi rst such analyses compare the six components from previous fi gures that contribute to this: the recurrence RRs during years 0–1 were, as expected, similar in comparisons A, B, and E, but the recurrence RRs during years 2–4 appeared somewhat more extreme after 2–3 years of previous tamoxifen (RR 0·56, 95% CI 0·46–0·67) than after 2–3 years of aromatase inhibitor versus tamoxifen (RR 0·83, 0·69–1·00), or after 2 years of aromatase inhibitor (RR 1·08, 0·70–1·68). Figure 5 subdivides the aggregated result from the periods when treatments diff ered by aromatase Articles 1344 www.thelancet.com Vol 386 October 3, 2015 inhibitor drug, site of fi rst recurrence, entry age, BMI, and tumour characteristics: PR status, nodal status, tumour diameter, tumour grade, and HER2 status (available for only one-third of patients). The recurrence RRs were similar with diff erent aromatase inhibitors (each p<0·0001), with local recurrence, contralateral 0 5 10 0 10 20 30 40 50 A Recurrence (%) Recurrence rate/year (%), events/woman-years and log-rank statistics Allocation AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 0–1 1·62 (157/9691) 2·41 (230/9542) 0·64 (0·52–0·78) –41·1/92·8 Years 2–4 2·14 (285/13 336) 2·62 (338/12 906) 0·80 (0·68–0·93) –34·1/149·0 Years 5–9 2·33 (365/15 648) 2·48 (372/14 985) 0·92 (0·79−1·06) −15·5/177·2 Year 10+ 3·23 (20/619) 4·54 (24/529) 0·72 (0·39−1·30) −3·6/10·7 9885 women, 1791 events RR=0·80 (95% CI 0·73–0·88) 0 5 10 0 10 20 30 40 50 B Breast cancer mortality (%) Death rates (%/year: total rate minus rate in women without recurrence) and log-rank statistics Allocation AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 0–1 0·52 (0·39−0·66) 0·51 (0·39−0·67) 0·98 (0·66−1·46) −0·4/24·3 Years 2–4 1·23 (1·05−1·41) 1·60 (1·38−1·83) 0·74 (0·60−0·91) −27·2/90·7 Years 5–9 1·66 (1·46−1·86) 1·81 (1·60−2·02) 0·90 (0·76−1·07) −13·6/129·4 Year 10+ 1·93 (0·88−2·99) 1·88 (0·77−2·99) 1·01 (0·45−2·33) 0·1/5·6 9885 women, 1066 deaths RR=0·85 (95% CI 0·75–0·96) 0 5 10 Years 0 10 20 30 40 50 C Death without recurrence (%) Death-without-recurrence rate/year (%), events/woman-years and log-rank statistics Allocation AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 0–1 0·54 (52/9691) 0·60 (57/9542) 0·89 (0·61−1·30) −3·2/27·0 Years 2–4 0·99 (132/13 336) 0·95 (122/12 906) 1·03 (0·81−1·32) 2·0/62·5 Years 5–9 1·46 (229/15 648) 1·63 (244/14 985) 0·88 (0·73−1·05) −14·9/114·4 Year 10+ 3·07 (19/619) 2·84 (15/529) 1·28 (0·64−2·56) 2·0/8·1 9885 women, 870 deaths RR=0·94 (95% CI 0·82–1·07) 0 5 10 Years 0 10 20 30 40 50 D Death from any cause (%) Death rate/year (%), events/woman-years and log-rank statistics Allocation AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 0–1 1·05 (103/9775) 1·10 (106/9661) 0·93 (0·71−1·23) −3·6/51·3 Years 2–4 2·18 (301/13 777) 2·50 (338/13 506) 0·85 (0·72−0·99) −25·2/153·2 Years 5–9 3·06 (500/16 333) 3·35 (530/15 805) 0·89 (0·79−1·01) −28·4/243·9 Year 10+ 4·76 (32/672) 4·44 (26/586) 1·16 (0·69−1·99) 2·1/13·6 9885 women, 1936 deaths RR=0·89 (95% CI 0·8–0·97) 10-year gain 3·6% (95% CI 1·7 to 5·4) Log-rank 2p<0·00001 Tamoxifen 22·7% AI 19·1% 12·1% 9·0% 10-year gain 2·1% (95% CI 0·5 to 3·7) Log-rank 2p=0·009 Tamoxifen 14·2% AI 12·1% 5·8% 4·5% 10-year gain 1·1% (95% CI –0·4 to 2·6) Log-rank 2p=0·34 Tamoxifen 11·9% AI 10·8% 4·0% 4·0% 10-year gain 2·7% (95% CI 0·1 to 4·7) Log-rank 2p=0·01 Tamoxifen 24·0% AI 21·3% 9·4% 8·2% Figure 1: 5 years of aromatase inhibitor versus 5 years of tamoxifen (A) Recurrence, (B) breast cancer mortality, (C) death without recurrence, and (D) death from any cause. RR=rate ratio (with 95% CI). AI=aromatase inhibitor. O–E=observed minus expected. V=variance of O–E. Articles www.thelancet.com Vol 386 October 3, 2015 1345 breast cancer, and distant recur rence all substantially reduced by aromatase inhibitor compared with tamoxifen. In the aggregated data, the RRs while treatments diff ered appeared similar in every subgroup, suggesting that age, BMI, and tumour characteristics cannot usefully predict the RR. Figure 2: 5 years of aromatase inhibitor versus tamoxifen to years 2–3 then aromatase inhibitor to year 5 (A) Recurrence, (B) breast cancer mortality, (C) death without recurrence, and (D) death from any cause. RR=rate ratio. AI=aromatase inhibitor. O–E=observed minus expected. V=variance of O–E. 0 1 2 3 4 5 6 7 1 2 3 4 5 6 7 0 10 20 30 40 50 A Recurrence (%) Recurrence rate/year (%), events/woman-years and log-rank statistics Allocation AI Tamoxifen then AI Rate ratio (95% CI) from (O-E)/V Years 0–1 1·64 (204/12 435) 2·22 (273/12290) 0·74 (0·62−0·89) −34·0/115·0 Years 2–4 2·31 (360/15 589) 2·29 (348/15 183) 0·99 (0·86−1·15) −1·2/170·6 Year 5+ 2·43 (141/5811) 2·52 (144/5715) 0·96 (0·76−1·22) −2·6/69·2 Years 0–1 0·46 (0·35−0·58) 0·55 (0·42−0·68) 0·84 (0·59−1·20) −5·3/30·7 Years 2–4 1·41 (1·23−1·57) 1·49 (1·30−1·69) 0·93 (0·78−1·13) −7·5/110·9 Year 5+ 1·77 (1·44−2·10) 2·08 (1·72−2·44) 0·84 (0·65−1·09) −9·6/55·5 12 799 women, 1470 events RR=0·90 (95% CI 0·81–0·99) 0 0 10 20 30 40 50 B Breast cancer mortality (%) Death rates (%/year: total rate minus rate in women without recurrence) and log-rank statistics Allocation AI Tamoxifen then AI Rate ratio (95% CI) from (O-E)/V Death-without-recurrence rate/year (%), events/woman-years and log-rank statistics Allocation AI Tamoxifen then AI Rate ratio (95% CI) from (O-E)/V Years 0–1 0·52 (65/12 435) 0·50 (61/12290) 1·04 (0·73−1·48) 1·2/31·1 Years 2–4 0·94 (146/15 589) 0·82 (124/15 183) 1·11 (0·88−1·42) 7·1/66·2 Year 5+ 1·14 (66/5811) 1·07 (61/5715) 1·02 (0·72−1·46) 0·8/30·8 Years 0–1 0·98 (123/12 576) 1·04 (130/12 463) 0·94 (0·73−1·20) −4·1/61·8 Years 2–4 2·31 (374/16 200) 2·27 (361/15 885) 1·00 (0·86−1·16) −0·4/177·1 Year 5+ 2·84 (175/6152) 3·08 (187/6064) 0·90 (0·73−1·12) −8·8/86·3 Death rate/year (%), events/woman-years and log-rank statistics Allocation AI Tamoxifen then AI Rate ratio (95% CI) from (O-E)/V 12 799 women, 827 deaths RR=0·89 (95% CI 0·78–1·03) 1 2 3 4 5 6 7 1 2 3 4 5 6 7 0 Years 0 10 20 30 40 50 C Death without recurrence (%) 12 799 women, 523 deaths RR=1·07 (95% CI 0·90–1·28) 0 Years 0 10 20 30 40 50 D Death from any cause (%) 12 799 women, 1350 deaths RR=0·96 (95% CI 0·86–1·07) Tamoxifen then AI 14·5% AI 13·8% 10·7% 9·6% 7-year gain 0·7% (95% CI –0·9 to 2·2) Log-rank 2p=0·045 AI 6·1% Tamoxifen then AI 5·9% 3·8% 3·5% 7-year gain 0·2% (95% CI –1·0 to 1·3) Log-rank 2p=0·42 Tamoxifen then AI 9·3% AI 8·2% 5·5% 5·1% 7-year gain 1·1% (95% CI –0·2 to 2·5) Log-rank 2p=0·11 Tamoxifen then AI 14·5% AI 13·6% 8·7% 8·6% 7-year gain 0·9% (95% CI –0·7 to 2·5) Log-rank 2p=0·46 Articles 1346 www.thelancet.com Vol 386 October 3, 2015 Tumour characteristics were, however, importantly predictive of the absolute risk of recurrence, and hence of the absolute eff ect on breast cancer outcomes of giving an aromatase inhibitor rather than tamoxifen (appendix). For example, in the aggregate of the trials that contribute to the black squares in fi gure 4, the 0 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10 2 3 4 5 6 7 8 9 10 0 10 20 30 40 50 A Recurrence (%) Recurrence rate/year (%), events/woman-years and log-rank statistics Allocation Tamoxifen then AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 2–4 1·48 (170/11 515) 2·64 (300/11 360) 0·56 (0·46–0·67) −65·3/111·5 Years 5–9 2·48 (495/19 920) 2·51 (479/19 101) 0·97 (0·86−1·11) −5·9/234·0 Year 10+ 3·26 (88/2696) 3·35 (84/2505) 0·92 (0·68−1·25) −3·3/40·8 11 798 women, 1616 events RR=0·82 (95% CI 0·75–0·91) 0 0 10 20 30 40 50 B Breast cancer mortality (%) Death rates (%/year: total rate minus rate in women without recurrence) and log-rank statistics Allocation Tamoxifen then AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 2–4 0·37 (0·25–0·48) 0·56 (0·43–0·70) 0·65 (0·44–0·96) −11·0/25·8 Years 5–9 1·28 (1·12–1·44) 1·40 (1·26–1·56) 0·91 (0·77–1·08) −11·9/132·0 Year 10+ 1·68 (1·63–1·72) 2·54 (2·45–2·59) 0·69 (0·48–1·00) −10·6/28·9 11 798 women, 789 deaths RR=0·84 (95% CI 0·72–0·96) 0 Time since allocated treatments differ (years) 0 10 20 30 40 50 C Death without recurrence (%) Death-without-recurrence rate/year (%), events/woman-years and log-rank statistics Allocation Tamoxifen then AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 2–4 0·51 (59/11 515) 0·55 (63/11 360) 0·91 (0·64–1·30) −2·8/30·0 Years 5–9 0·85 (169/19 920) 1·15 (220/19 101) 0·73 (0·60–0·89) −30·3/95·3 Year 10+ 1·85 (50/2696) 2·11 (53/2505) 0·95 (0·63–1·42) −1·3/23·8 11 798 women, 614 deaths RR=0·79 (95% CI 0·67–0·93) 0 Time since allocated treatments differ (years) 0 10 20 30 40 50 D Death from any cause (%) Death rate/year (%), events/woman-years and log-rank statistics Allocation Tamoxifen then AI Tamoxifen Rate ratio (95% CI) from (O-E)/V Years 2–4 0·88 (102/11 644) 1·10 (128/11 618) 0·78 (0·60–1·01) −13·9/55·9 Years 5–9 2·09 (437/20 949) 2·47 (509/20 593) 0·83 (0·73–0·95) −42·3/227·3 Year 10+ 1·86 (50/2696) 2·11 (53/2505) 0·95 (0·63–1·42) −1·3/23·8 11 798 women, 1403 deaths RR=0·82 (95% CI 0·73–0·91) Tamoxifen 19·0% Tamoxifen then AI 17·0% 12·1% 9·5% 10-year gain 2·0% (95% CI 0·2 to 3·8) Log-rank 2p=0·0001 Tamoxifen 8·8% Tamoxifen then AI 7·0% 4·2% 3·2% 10-year gain 1·7% (95% CI 0·3 to 3·2) Log-rank 2p=0·005 Tamoxifen 17·5% Tamoxifen then AI 14·6% 8·8% 7·1% 10-year gain 2·9% (95% CI 1·1 to 4·7) Log-rank 2p=0·0002 Tamoxifen 10·01% Tamoxifen then AI 8·7% 5·0% 4·2% 10-year gain 1·5% (95% CI 0·1 to 2·9) Log-rank 2p=0·01 Figure 3: Tamoxifen to years 2–3 then aromatase inhibitor to year 5 versus 5 years of tamoxifen: events in women alive and free of recurrence when treatments diverged (A) Recurrence, (B) breast cancer mortality, (C) death without recurrence, and (D) death from any cause. RR=rate ratio. AI=aromatase inhibitor. O–E=observed minus expected. V=variance of O–E. Articles www.thelancet.com Vol 386 October 3, 2015 1347 overall Kaplan-Meier estimate of the 5-year recurrence risk was reduced by 2·5% (7·3% vs 9·8%, appendix). But, in this same data set, the 5-year recurrence risks for women with N0, N1–3, and N4+ disease were reduced by 1·2%, 3·7%, and 6·4%, respectively. Similar sets of subgroup analyses for each separate category of comparisons A–E are in the appendix, but with so many subgroup analyses the apparent fi ndings should be interpreted cautiously, as striking false-positive and false-negative results can easily arise just by chance. For example, the hypothesis from ATAC15 of a more extreme recurrence RR in ER-positive PR-negative than in ER-positive PR-positive disease is not supported by evidence from other trials (fi gure 5, appendix). Likewise, the hypothesis from comparison C of equivalent effi cacy of aromatase inhibitors and tamoxifen in node-negative disease is not supported by evidence from the other comparisons. Such patterns might be due mainly to chance. Results for cause-specifi c mortality, second cancer incidence, and bone fracture before any breast cancer recurrence are in the appendix. There was a signifi cant reduction in mortality without recurrence in comparison C (tamoxifen then aromatase inhibitor vs tamoxifen alone) that was not explained by any particular cause and is unlikely to be due to misclassifi ed breast cancer deaths (partly because the non-breast-cancer mortality rates were sharply agerelated whereas breast cancer mortality rates were similar in all age groups). There were fewer uterine cancers and more bone fractures with aromatase inhibitors than with tamoxifen. Aggregating the fi ve comparisons, the 10-year incidence of endometrial cancer (defi ned as any uterine cancer except cervix cancer) was 0·4% in the aromatase inhibitor group versus 1·2% in the tamoxifen group (absolute diff erence 0·8%, 95% CI 0·6–1·0; p<0·0001), including fi ve versus nine deaths. The proportional decrease in endometrial cancer incidence with aromatase inhibitors (RR 0·33, 0·21–0·51) was approximately independent of age and persisted for some years after treatment ended. As endometrial cancer increases with age, the absolute excess with tamoxifen was 0·7% (95% CI 0·5–0·9) at ages 55–69 and 1·4% (95% CI 0·5–2·4) at older ages (appendix). There was no signifi cant eff ect on any other type of cancer (except for contralateral breast cancer). The incidence of bone fractures was increased among aromatase-inhibitor-allocated patients during years 0–4 (RR 1·42, 95% CI 1·28–1·57; p<0·0001), and remained signifi cantly higher through years 5–9 (RR 1·29, 1·09–1·53; 2p=0·003) despite fractures being monitored less reliably after the 5-year treatment period. The 5-year fracture risk was 8·2% in the aromatase inhibitor group versus 5·5% in the tamoxifen group (absolute excess 2·7%, 95% CI 1·7–3·7). Again, the proportional increase appeared approximately independent of age and the absolute incidence increased with age. Hence, among women of age younger than 55, 55–69, and older than 70 years at randomisation, the absolute excess risks (aromatase inhibitor vs tamoxifen) of having a fracture Figure 4: Recurrence reductions by time since surgery, combining data from diff erent comparisons of aromatase inhibitor (AI) versus tamoxifen treatment as part of 5 years of endocrine therapy Black squares show periods when the protocol specifi ed that one group should receive an aromatase inhibitor and the other should receive tamoxifen; open squares show periods when the treatments should have been the same in both groups. *Aggregated totals are adjusted to avoid double counting of events in the four-way randomisation in BIG 1-98. AI=aromatase inhibitor. O–E=observed minus expected. Events/women–years (%/year) Allocated AI Allocated tamoxifen (a) AI vs tamoxifen Years 0–1 AI vs tamoxifen Years 2–4 AI vs tamoxifen Subtotal 361/22 068 (1·6) 425/23 324 (1·8) 786/45 398 (1·7%/year) 502/21 786 (2·3) 581/22 803 (2·5) 1083/44 589 (2·4%/year) 0·70 (0·58–0·84) 0·71 (0·60–0·83) 0·70 (0·64–0·77) 2p<0·00001 –74·3 –83·7 –158·1 207·6 240·2 447·7 (b) Same or no treatment Years 2–4 same treatment Years 5–9 no treatment Years 10+ no treatment Subtotal Total (a+b) 99% or 95% CIs Difference between treatment effects in two subtotals χ2 1=21·1; 2p<0·00001 Heterogeneity within subtotals χ2 3=2·0; 2p=0·6 Heterogeneity between five comparisons χ2 4=23·1; 2p=0·0001 287/11 340 (2·5) 906/38 062 (2·4) 109/3384 (3·2) 1302/52 786 (2·5%/year) 2088/98 184 (2·1%/year) 266/10 954 (2·4) 940/36 499 (2·6) 114/3111 (3·7) 1320/50 564 (2·6%/year) 2403/95 153 (2·5%/year) 1·03 (0·83–1·29) 0·92 (0·81–1·03) 0·85 (0·60–1·21) 0·93 (0·86–1·01) 2p=0·08 0·829 (0·781–0·880) 2p<0·00001 4·2 –39·0 –8·6 –43·5 –201·5 133·1 443·7 53·1 629·9 1077·7 AI events Ratio of annual event rates AI:tamoxifen Rate ratio (CI) Log-rank O–E Variance of O–E AI better Tamoxifen better Treatment effect 2p<0·00001 0 0·5 1·0 1·5 2·0 Articles 1348 www.thelancet.com Vol 386 October 3, 2015 AI better Tamoxifen better Treatment effect 2p<0·00001 0 0·5 1·0 1·5 2·0 Events/women-years (%/year) Allocated AI Allocated tamoxifen (a) Treatment comparison (χ2 5=13·3; p=0·02) Years 0–1 comparison Years 0–1 comparison Years 0–1 comparison Years 2–4 comparison Years 2–4 comparison Years 2–4 comparison (b) AI agent (χ2 2=0·5; 2p=0·8) Anastrozole Exemestane Letrozole (c) Site of recurrence (χ2 2=1·5; 2p=0·5) Distant Isolated local Contralateral (d) Age (trend χ2 1=2·2; 2p=0·14) <45 years 45–54 years 55–69 years 70+ years (e) BMI (trend χ2 1=0·3; 2p=0·6) <20 20–24 25–29 30+ Unknown (f) PR status (χ2 1=5·7; 2p=0·02) ER positive PR negative ER positive PR unknown ER positive PR positive (g) Nodal status (trend χ2 1=0·0; 2p=1·0) N0/N– N1–3 N4+ N other/unknown (h) T stage (trend χ2 1=2·8; 2p=0·09) 1–20 mm (T1) 21–50 mm (T2) >50 mm (T3/T4) Other/unknown (i) Tumour grade (trend χ2 1=0·0; 2p=1·0) Well differentiated Moderately Poorly Grade unknown (j) HER2 status (χ2 2=0·0; 2p=1·0) HER2 positive HER2 negative Unknown Total A 157/9676 (1·6) B 204/12 418 (1·6) E 35/2995 (1·2) A 285/13 313 (2·1) C 170/11 498 (1·5) D 43/2505 (1·7) 291/16 842 (1·7) 233/13 164 (1·8) 262/15 392 (1·7) 617/45 398 (1·4) 101/45 398 (0·2) 68/45 398 (0·1) 4/215 (1·9) 108/6636 (1·6) 472/27 362 (1·7) 202/11 185 (1·8) 21/1506 (1·4) 194/11 001 (1·8) 202/11 397 (1·8) 166/6814 (2·4) 203/14 195 (1·4) 171/6928 (2·5) 49/2259 (2·2) 566/36 243 (1·6) 243/26 174 (0·9) 280/13 746 (2·0) 236/4067 (5·8) 27/1411 (1·9) 324/28 854 (1·1) 401/14 714 (2·7) 42/1116 (3·8) 19/708 (2·7) 60/8857 (0·7) 320/21 782 (1·5) 238/7557 (3·1) 168/7193 (2·3) 79/3264 (2·4) 112/8854 (1·3) 595/33 280 (1·8) 786/45 398 (1·7%/year) 230/9526 (2·4) 273/12 273 (2·2) 45/3008 (1·5) 338/12 886 (2·6) 300/11 333 (2·6) 41/2491 (1·6) 387/16 415 (2·4) 341/12 990 (2·6) 355/15 184 (2·3) 819/44 590 (1·8) 158/44 590 (0·4) 106/44 590 (0·2) 9/183 (4·9) 144/6430 (2·2) 616/27 152 (2·3) 314/10 824 (2·9) 26/1328 (2·0) 272/10 695 (2·5) 264/11 016 (2·4) 203/7477 (2·7) 318/14 073 (2·3) 284/6687 (4·2) 57/2385 (2·4) 742/35 556 (2·1) 338/26 097 (1·3) 390/13 285 (2·9) 308/3850 (8·0) 47/1357 (3·5) 409/28 134 (1·5) 576/14 738 (3·9) 74/1008 (7·3) 24/710 (3·4) 84/8839 (1·0) 452/21 382 (2·1) 318/7287 (4·4) 229/7081 (3·2) 108/3074 (3·5) 166/8733 (1·9) 809/32 782 (2·5) 1083/44 589 (2·4%/year) 0·64 (0·49–0·84) 0·74 (0·59–0·95) 0·78 (0·43–1·40) 0·80 (0·64–0·98) 0·56 (0·44–0·71) 1·08 (0·61–1·92) 0·71 (0·58–0·87) 0·67 (0·54–0·84) 0·73 (0·59–0·91) 0·73 (0·63–0·84) 0·64 (0·46–0·88) 0·64 (0·43–0·94) 0·73 (0·52–1·01) 0·75 (0·64–0·88) 0·60 (0·48–0·76) 0·79 (0·36–1·74) 0·71 (0·55–0·90) 0·72 (0·56–0·92) 0·79 (0·60–1·04) 0·63 (0·51–0·79) 0·57 (0·45–0·73) 0·91 (0·55–1·51) 0·74 (0·64–0·86) 0·72 (0·58–0·89) 0·68 (0·56–0·84) 0·73 (0·58–0·92) 0·59 (0·32–1·09) 0·76 (0·63–0·92) 0·70 (0·59–0·82) 0·51 (0·30–0·85) 0·71 (0·46–1·09) 0·68 (0·57–0·83) 0·70 (0·56–0·87) 0·71 (0·54–0·92) 0·66 (0·45–0·99) 0·67 (0·49–0·92) 0·71 (0·62–0·82) 0·702 (0·640–0·771) 2p<0·00001 −41·1 −34·0 −5·2 34·1 −65·3 1·6 −55·8 −54·4 −46·3 −109·4 −29·1 −19·7 −0·8 −19·2 −75·9 −62·2 −2·5 −37·9 −36·4 −20·8 −61·0 −61·3 −2·4 −94·8 −48·0 −61·6 −39·1 −9·4 −49·6 −84·4 −17·1 −2·6 −12·2 −71·3 −47·2 −33·1 −17·5 −26·1 −114·4 −158·1 92·8 115·0 19·5 149·0 111·5 20·1 163·2 137·9 149·6 344·5 64·3 43·4 1·4 60·4 262·2 123·7 10·7 109·1 111·1 86·8 133·7 109·2 25·8 320·1 143·4 162·4 124·4 17·5 179·2 233·8 25·3 9·7 35·1 188·0 131·3 95·5 43·0 65·4 339·3 447·7 AI events Ratio of annual event rates AI:tamoxifen Rate ratio (CI) Log-rank O–E Variance of O–E 99% or 95% CIs Figure 5: Subgroup analyses of recurrence risk reductions combining data from fi ve comparisons of aromatase inhibitors versus tamoxifen including only data during periods when treatments diff ered Grey squares show unknown status within the subgroup. Results are plotted as black squares with horizontal lines that denote 99% rather than 95% CIs to allow for multiple hypothesis testing. Total is plotted as a white diamond that denotes 95% CI. AI=aromatase inhibitor. O–E=observed minus expected. Comparison A=5 years of aromatase inhibitor versus 5 years of tamoxifen. Comparison B=5 years of aromatase inhibitors versus 2–3 years of tamoxifen, then aromatase inhibitor to year 5. Comparison C=2–3 years of tamoxifen, then aromatase inhibitor to year 5 versus 5 years of tamoxifen. Comparison D=5 years of aromatase inhibitor versus 2 years of aromatase inhibitor, then tamoxifen to year 5. Comparison E=2 years of aromatase inhibitor, then tamoxifen to year 5 versus 5 years of tamoxifen. ER=oestrogen receptor. PR=progesterone receptor. Articles www.thelancet.com Vol 386 October 3, 2015 1349 within 5 years were, respectively, about 1%, 2%, and 4% (appendix). Diff erences in vascular mortality, aromatase inhibitor versus tamoxifen, were not signifi cant: thromboembolic, 14 versus 19 deaths; cerebrovascular, 44 versus 52 deaths; and cardiac, 137 versus 128 deaths. Discussion Individual trials have already shown reduced recurrence rates with aromatase inhibitor compared with tamoxifen but none has shown in intention-to-treat analyses that breast cancer mortality is reduced, nor did previous meta-analyses.4 Now, with longer follow-up, the present meta-analyses establish that breast cancer mortality and all-cause mortality are also reduced, better characterise time-dependent eff ects on recurrence, and allow informative investigation of diff erential effi cacy within subgroups and of uncommon adverse events. There was a fairly consistent pattern of substantial recurrence reductions during periods when one group was receiving an aromatase inhibitor and the other tamoxifen, but little further reduction during subsequent periods when both groups were receiving the same endocrine treatment or after scheduled endocrine treatment had ended in both groups. However, this fi nding should not be interpreted as aromatase inhibitors not having the carry-over benefi ts of tamoxifen,1 rather that 5 years of endocrine therapy that includes an aromatase inhibitor reduces recurrence by about one-third during years 5–9, as does 5 years of tamoxifen. The most extreme recurrence reduction appeared to be in comparison C in which, after 2 years of tamoxifen, an aromatase inhibitor was compared with tamoxifen during years 2–4. This result is not explained by diff erences in effi cacy between diff erent aromatase inhibitors, as indirect comparisons in fi gure 5, and direct randomised comparisons,16 show little diff erence between drugs. It has been hypothesised that the superiority of aromatase inhibitors over tamoxifen is greater after previous exposure to tamoxifen,17 and the larger recurrence reductions reported in years 5–9 in trials of aromatase inhibitor versus no further treatment18–20 after 5 years of tamoxifen than in trials of 10 versus 5 years of tamoxifen2,3 provide some support for this. However, the directly randomised fi ndings in comparison B do not show any eff ect of the type of endocrine therapy during years 0–1 on the effi cacy of treatment during years 2–4, so the apparent heterogeneity of benefi t from indirect comparisons could be largely chance. In comparison E, after an initial 2–3 years of an aromatase inhibitor there appeared to be no benefi t from continuing an aromatase inhibitor to 5 years rather than switching to tamoxifen, but this result was based on one trial with few events. Hence, it remains uncertain whether, after 2–3 years of an aromatase inhibitor, any loss of benefi t occurs from switching to tamoxifen—reassuringly for women who do not tolerate aromatase inhibitors. Results of ongoing trials comparing diff erent durations of aromatase inhibitor treatment will determine whether, as with tamoxifen, longer is better.2,3,21 The reduction in breast cancer mortality with aromatase inhibitor compared with tamoxifen is only slight, as expected in an already relatively good-prognosis population, but persists during years 0–4 and 5–9, signifi cantly reducing 10-year breast cancer mortality. Overall 10-year mortality was also signifi cantly reduced, even though about half the deaths were not due to breast cancer. Non-breast cancer death rates were similar with aromatase inhibitor and tamoxifen except that, after 2–3 years of tamoxifen, there appeared to be fewer such deaths with an aromatase inhibitor than with continuing tamoxifen. This fi nding was unexpected, not explained by any one cause, and not replicated in the other comparisons. Though likely to be a chance fi nding, it is reassuring for the safety of aromatase inhibitors. Bone fractures are a concern with aromatase inhibitors, though the absolute excess of about 0·5% per year might be partly explained by a bone-protective eff ect of 5 years of tamoxifen vs none: EBCTCG previous meta-analysis1 (n=10 645) 5 years of aromatase inhibitor vs 5 years of tamoxifen: present meta-analyses* (n=34 882) 5 years of aromatase inhibitor vs none: estimated eff ects (product of two RRs†) RR (95% CI) p value RR (95% CI) p value RR (95% CI) p value Breast cancer recurrence During years 0–4 0·53 (0·48–0·57) 2p<0·0001 0·70 (0·64–0·77) 2p<0·0001 0·37 (0·33–0·42) 2p<0·0001 During years 5–9 0·68 (0·60–0·78) 2p<0·0001 0·92 (0·83–1·01) 2p=0·082 0·63 (0·53–0·74) 2p<0·0001 Breast cancer mortality During years 0–4 0·71 (0·62–0·80) 2p<0·0001 0·79 (0·67–0·92) 2p=0·002 0·56 (0·46–0·68) 2p<0·0001 During years 5–9 0·66 (0·58–0·75) 2p=0·0001 0·91 (0·80–1·02) 2p=0·12 0·60 (0·50–0·72) 2p<0·0001 EBCTCG=Early Breast Cancer Trialists’ Collaborative Group. RR=rate ratio. *Estimated from the aggregated data (appendix). †Estimated rate ratio for 5 years of aromatase inhibitor vs none (RR3) is obtained by direct multiplication of the rate ratio for 5 years of tamoxifen vs none (RR1) by the rate ratio for 5 years of aromatase inhibitor vs 5 years of tamoxifen (RR2) estimated from the aggregated data; 95% confi dence limits for RR3 are exp[(o – e)1/v1 + (o – e)2/v2) – 1·96 √(1/v1 + 1/v2)] and exp[(o – e)1/v1 + (o – e)2/v2) + 1·96 √(1/ v1 + 1/v2)], respectively, where (o – e) and v are the observed minus expected statistics and their variances for the comparisons of 5 years of tamoxifen vs none and 5 years of aromatase inhibitor vs 5 years of tamoxifen (estimated from aggregated data from trials contributing to subtotal (a) in fi gure 4). Table: Estimation of the eff ect of 5 years of an aromatase inhibitor versus no endocrine treatment