Correspondence therapy-specific side effects of aromatase inhibi- of interest in patients treated with an aromatase tors are not discussed in their article. inhibitor.1,2 In addition, the method used to assess toxic effects in the trial (National Cancer Clemens Tempfer, M.D. Ruhr University Bochum Institute Common Toxicity Criteria, version 2.0) Bochum, Germany did not include these events as specific adverse clemens.­tempfer@­rub.­de events. However, our data showed no significant No potential conflict of interest relevant to this letter was redifferences in grade 3 or 4 arthritis, muscle weakported. ness, or myositis between the letrozole group 1. Spagnolo F, Sestak I, Howell A, Forbes JF, Cuzick J. Anastroand the placebo group. Similarly, no significant zole-induced carpal tunnel syndrome: results from the International Breast Cancer Intervention Study II Prevention Trial. J Clin between-group differences were detected with Oncol 2016;34:139-43. respect to the adverse events of memory loss or 2. Lombard JM, Zdenkowski N, Wells K, et al. Aromatase incognitive disturbance. As noted in our article, hibitor induced musculoskeletal syndrome: a significant problem with limited treatment options. Support Care Cancer 2016; the low incidence of adverse events with letro24:2139-46. zole and the acceptable side-effect profile of the DOI: 10.1056/NEJMc1610719 drug, as compared with placebo, were probably due to self-selection of study participants who The authors reply: The MA.17R trial showed had had no unacceptable side effects during the that the rates of recurrent and contralateral previous 5 years during which they had received breast cancer were significantly lower among aromatase-inhibitor therapy and who were thus women receiving extended aromatase-inhibitor willing to undergo another 5 years of treatment. therapy than among those receiving placebo, al- Therefore, we believe that the MA.17R trial cohort though the absolute differences were small. As is not comparable to the International Breast stated in our article, this benefit is likely to vary Cancer Intervention Study II cohort, which comaccording to the risk of recurrence of breast can- prised healthy, postmenopausal women with no cer. In our article, we conclude that the decision prior exposure to aromatase inhibitors. to receive prolonged therapy with an aromatase James N. Ingle, M.D. inhibitor (i.e., on the basis of the individual riskMayo Clinic to-benefit ratio, or what Sanz and del Valle call “a Rochester, MN real benefit”) will depend on an individualized assessment of quality of life during receipt of Wendy R. Parulekar, M.D. therapy, the side-effect profile (particularly re- Queen’s University Kingston, ON, Canada lated to bone health), and the estimated risk of wparulekar@­c tg.­queensu.­ca disease recurrence. We believe that for patients Since publication of their article, the authors report no furwho have a high risk of recurrence, have had no ther potential conflict of interest. unacceptable side effects with previous aroma Spagnolo F, Sestak I, Howell A, Forbes JF, Cuzick J. Anastrotase-inhibitor therapy, and have sufficient bone 1. zole-induced carpal tunnel syndrome: results from the Internahealth, prolonged therapy with an aromatase in- tional Breast Cancer Intervention Study II Prevention Trial. J Clin hibitor should be discussed as a treatment option. Oncol 2016;34:139-43. Lombard JM, Zdenkowski N, Wells K, et al. Aromatase inAt the time of the initiation of the trial, aroma- 2. hibitor induced musculoskeletal syndrome: a significant probtase-inhibitor–associated carpal tunnel syndrome lem with limited treatment options. Support Care Cancer 2016; and aromatase-inhibitor–induced musculoskele- 24:2139-46. tal syndrome were not identified as adverse events DOI: 10.1056/NEJMc1610719 Adaptive Randomization of Neratinib in Early Breast Cancer To the Editor: The I-SPY 2 TRIAL (Investigation of Serial Studies to Predict Your Therapeutic Response with Imaging and Molecular Analysis 2) uses adaptive randomization, putting more patients in the treatment group with the current higher rate of pathological complete response. In their Perspective article, Harrington and Parmigiani (July 7 issue)1 state that adaptive randomization makes more sense than fixed randomi­ zation. Why? Adaptive randomization is neither n engl j med 375;16 nejm.org October 20, 2016 The New England Journal of Medicine Downloaded from nejm.org by andre Tartar on October 20, 2016. For personal use only. No other uses without permission. Copyright © 2016 Massachusetts Medical Society. All rights reserved. 1591 The n e w e ng l a n d j o u r na l m e dic i n e ethically nor statistically preferable. Randomization requires equipoise.2 As long as equipoise exists, there is no logical or ethical reason to favor one treatment over the other, and any assignment ratio is acceptable, whether balanced (1:1) or unbalanced (e.g., 2:1). The latter is used merely to expose more patients to an experimental treatment than to a well-known control. Ethically, adaptive randomization is questionable if it is used to lure patients into accepting trial participation on account of a higher probability of receiving a treatment of no proven benefit.3 Statistically, adaptive randomization may bias the treatment comparison, undermining the purpose of randomization. The I-SPY 2 platform has created a collaborative culture and proposed many innovative ideas that may make future trials more efficient.4 However, adaptive randomization is not one of these ideas. Marc Buyse, Sc.D. Everardo D. Saad, M.D. Tomasz Burzykowski, Ph.D. hormone-receptor–negative cancer, the rate of pathological complete response was approximately similar with neratinib plus weekly paclitaxel (46%) but substantially higher with trastuzumab plus weekly paclitaxel (57%), the latter value being above the upper boundary of the 95% probability interval in the I-SPY 2 trial (11 to 54%).2 Few patients were treated with trastuzumab plus weekly paclitaxel in each trial in this subgroup and still fewer had a pathological complete response (how many in the I-SPY 2 trial?). Besides the slight differences between the trials, chance might also in part explain these seemingly quite different rates of pathological complete response in the control (trastuzumab plus weekly paclitaxel) groups. Heikki Joensuu, M.D. International Drug Development Institute Louvain-la-Neuve, Belgium marc.­buyse@­iddi.­com Drs. Buyse and Burzykowski report being employed by and holding stock in the International Drug Development Institute, and Dr. Saad reports being employed by the International Drug Development Institute. No other potential conflict of interest relevant to this letter was reported. 1. Park JW, Liu MC, Yee D, et al. Adaptive randomization of neratinib in early breast cancer. N Engl J Med 2016;375:11-22. 2. Jacobs SA, Robidoux A, Garcia JMP, et al. NSABP FB-7: a phase II randomized trial evaluating neoadjuvant therapy with weekly paclitaxel (P) plus neratinib (N) or trastuzumab (T) or neratinib and trastuzumab (N+T) followed by doxorubicin and cyclophosphamide (AC) with postoperative T in women with locally advanced HER2-positive breast cancer. Cancer Res 2016; 76 Suppl 4:PD5-04. abstract. 1. Harrington D, Parmigiani G. I-SPY 2 — a glimpse of the future of phase 2 drug development? N Engl J Med 2016;375:7-9. 2. Freedman B. Equipoise and the ethics of clinical research. N Engl J Med 1987;317:141-5. 3. Hey SP, Kimmelman J. Are outcome-adaptive allocation trials ethical? Clin Trials 2015;12:102-6. 4. Carey LA, Winer EP. I-SPY 2 — toward more rapid progress in breast cancer treatment. N Engl J Med 2016;375:83-4. DOI: 10.1056/NEJMc1609993 DOI: 10.1056/NEJMc1609993 To the Editor: Park et al. (July 7 issue)1 report that neoadjuvant neratinib plus weekly paclitaxel resulted in a higher estimated rate of pathological complete response than trastuzumab plus weekly paclitaxel (56% vs. 33%), each followed by four cycles of doxorubicin plus cyclophosphamide, in patients with human epidermal growth factor receptor 2 (HER2)–positive, hormone-receptor–negative breast cancer in the I-SPY 2 trial. Curiously, when the same neoadjuvant chemotherapy regimens were investigated in another randomized trial (the National Surgical Adjuvant Breast and Bowel Project [NSABP] FB-7 trial) in roughly similar patients with HER2-positive, 1592 of Helsinki University Hospital Helsinki, Finland heikki.­joensuu@­hus.­f i No potential conflict of interest relevant to this letter was reported. Drs. Berry and Esserman reply: Buyse et al. ask, Why adaptive randomization? The answer is that it makes great sense ethically, statistically, economically, scientifically, and logistically, especially in a platform trial. The I-SPY 2 trial considers many patient subtypes and therapies. Adaptive randomization slows down and even stops assignment to therapies that perform poorly depending on subtype. Consider neratinib: in midtrial, neratinib was performing poorly for HER2negative tumors with a status of high-risk category 1 on a 70-gene profile (independent of hormone-receptor status). The members of the data and safety monitoring board considered stopping neratinib in these two subtypes — until they learned that the adaptive-randomization algorithm had already done so. The trial continued without an amendment because the patients had several good alternatives within the trial. Neratinib did well in other subtypes and eventually n engl j med 375;16 nejm.org October 20, 2016 The New England Journal of Medicine Downloaded from nejm.org by andre Tartar on October 20, 2016. For personal use only. No other uses without permission. Copyright © 2016 Massachusetts Medical Society. All rights reserved. Correspondence reached prespecified thresholds for efficacy in a specific biomarker signature (“graduated”). The I-SPY 2 trial is focused on patients. Because of adaptive randomization, 57% of 115 patients assigned to neratinib were HER2-positive, as compared with only 28% of 78 patients who were concurrently assigned to receive control therapy. This focus means that trial participants are more likely to receive better therapy than with equal randomization. It is also beneficial for future patients because better-performing therapies move through the trial faster and estimates of the efficacy are more precise in subtypes that have a response. The larger sample sizes within subtypes that have a response also mitigate the effects of early false positive results, and the Bayesian covariate analysis reduces such biases still further. Buyse et al. also raise the possibility that adaptive randomization may cause patients to be inappropriately attracted to participate in the trial. Yet participants know that they cannot get a better-performing therapy unless there is one in the trial. Joensuu asks about the differences in the estimated rates of pathological complete response between the I-SPY 2 and FB-7 trials in the signature of neratinib in I-SPY 2 (Table 1 accompanying this reply). The most obvious explanation is variability in such estimates when sample sizes are small, as he suggests. Indeed, the 57% rate of pathological complete response in the FB-7 control group is based on eight such responses in 14 patients, and the 46% rate in the neratinib group is based on six such responses in 13 patients. The confidence intervals are wide. As indicated in Table 1, these estimates are fragile Table 1. Efficacy among Patients with HER2-Positive, Hormone-Receptor–Negative Breast Cancer in Two Trials.* Rate of Pathological Complete Response (95% CI or PI) Trial Control Neratinib percent NSABP FB-7 57 (31–83) 46 (19–73) I-SPY 2 33 (11–54) 56 (37–73) *CI denotes confidence interval, and PI probability interval. indeed. The probability intervals in the I-SPY 2 trial are wide as well. Donald Berry, Ph.D. M.D. Anderson Cancer Center Houston, TX Laura Esserman, M.D., M.B.A. University of California, San Francisco San Francisco, CA laura.­esserman@­ucsf.­edu Since publication of their article, the authors report no further potential conflict of interest. DOI: 10.1056/NEJMc1609993 Drs. Harrington and Parmigiani reply: Buyse and coauthors raise important questions about adaptive randomization in the I-SPY 2 trial and in clinical trials in general. We attempted to give a balanced perspective on the strengths and limitations of both adaptive and constant probability designs, especially in Table 2 of our article (available at NEJM.org). We did not mean to imply that adaptive randomization is always preferable; we do believe that it can be an important tool for trialists in settings in which many potential treatments are emerging and there is identifiable patient heterogeneity. Potentially useful treatments must be identified as soon as possible, and those without promise should be quickly discarded. Ethical standards underlying the enrollment of patients into trials and the obtaining of informed consent are very important and should evolve with emerging paradigms of scientific investigation. From a patient perspective, it seems legitimate to consider the type of randomization in deciding whether to enroll in a trial. For example, a patient may prefer a trial with unbalanced randomization, in which more patients are assigned to the experimental treatment. The crux of the ethical issue is whether equipoise should dictate balance. If one accepts unbalanced randomization as ethical, then it follows that presenting randomization properties to patients is also ethically sound. Undoubtedly, presenting this information effectively and impartially can be a challenge. For example, in an unbalanced design, it is also possible that more patients will be exposed to serious side effects of an experimental treatment that has no benefit. Although adaptivity is ancillary to this ethical question, in practice the complex nature of adaptation will probably mean that the adaptation mechanism will remain a black box to many patients, amplifying the challenges of effective communication. n engl j med 375;16 nejm.org October 20, 2016 The New England Journal of Medicine Downloaded from nejm.org by andre Tartar on October 20, 2016. For personal use only. No other uses without permission. Copyright © 2016 Massachusetts Medical Society. All rights reserved. 1593 The n e w e ng l a n d j o u r na l In a well-conducted multicenter trial with adaptive randomization, the threat to clinical equipoise is no different than in a standard frequentist design with interim monitoring; the accumulating results are not shared with investigators or patients who might participate. In a single-center trial of two treatments, an adaptive randomization might send a signal to participating investigators through the changing randomization fractions, but that is a situation in which the use of adaptive randomization poses greater challenges. Standard summary statistics that are not adjusted for the adaptive randomization and the of m e dic i n e changing mix of patients receiving a treatment would, as Buyse et al. suggest, provide biased estimates of treatment effects, and model-based methods must be used. We raised that point as well. The first question that we raise in the fifth paragraph of our article points to the need to learn more about the robustness of those methods. David Harrington, Ph.D. Giovanni Parmigiani, Ph.D. Dana–Farber Cancer Institute Boston, MA Since publication of their article, the authors report no further potential conflict of interest. DOI: 10.1056/NEJMc1609993 Meningococcal B Vaccine during a University Outbreak To the Editor: Basta et al. (July 21 issue)1 measured protective serum antibody responses in college students immunized with a meningococcal serogroup B vaccine (4CMenB) in response to an outbreak. Although the vaccine elicited in vitro protection against two reference strains, a third of the students did not have a response with protective serum titers against the outbreak strain. This result was unexpected on the basis of test results from the Meningococcal Antigen Typing System,2 which predicted susceptibility of the outbreak strain to antibodies elicited by 4CMenB. My colleagues and I previously investigated the strain responsible for the outbreak.3 Despite high expression of two vaccine antigens, factor H– binding protein (fHbp) and neisserial heparinbinding antigen, the strain was relatively resistant to antibodies elicited in mice, nonhuman primates, and a human immunized with 4CMenB. In other studies, meningococcal strains with similar expression of identical fHbp amino acid sequence variants had discordant susceptibility to anti-fHbp bactericidal activity, which was related to the greater ability of the resistant strain to bind complement factor H, which down-regulates complement.4 Collectively, the data underscore our incomplete understanding of strain susceptibility to vaccine-induced bactericidal activity, which limits accurate prediction of vaccine coverage. Dan M. Granoff, M.D. UCSF Benioff Children’s Hospital Oakland Oakland, CA dgranoff@­chori.­org 1594 No potential conflict of interest relevant to this letter was reported. 1. Basta NE, Mahmoud AAF, Wolfson J, et al. Immunogenicity of a meningococcal B vaccine during a university outbreak. N Engl J Med 2016;375:220-8. 2. Donnelly J, Medini D, Boccadifuoco G, et al. Qualitative and quantitative assessment of meningococcal antigens to evaluate the potential strain coverage of protein-based vaccines. Proc Natl Acad Sci USA 2010;107:19490-5. 3. Rossi R, Beernink PT, Giuntini S, Granoff DM. Susceptibility of meningococcal strains responsible for two serogroup B outbreaks on U.S. university campuses to serum bactericidal activity elicited by the MenB-4C vaccine. Clin Vaccine Immunol 2015;22:1227-34. 4. Giuntini S, Pajon R, Ram S, Granoff DM. Binding of complement factor H to PorB3 and NspA enhances resistance of Neisseria meningitidis to anti-factor H binding protein bactericidal activity. Infect Immun 2015;83:1536-45. DOI: 10.1056/NEJMc1610666 To the Editor: The article on immunogenicity of meningococcal B vaccine during a university outbreak showed that 66.1% of students vaccinated with the 4CMenB vaccine had serum bactericidal antibodies (SBA) against the outbreak strain, in assays that included human complement (hSBA); 33.9% had a negative hSBA result. The media incorrectly reported that one third of students did not acquire immunity against the outbreak strain.1 We want to point out that Goldschneider et al.2 established that although a positive result on hSBA assay indicates protection against meningococcal disease, a negative hSBA result is not an indication of susceptibility, because the assay result (positive vs. negative) underestimates immunity. Indeed, blood from people n engl j med 375;16 nejm.org October 20, 2016 The New England Journal of Medicine Downloaded from nejm.org by andre Tartar on October 20, 2016. For personal use only. No other uses without permission. Copyright © 2016 Massachusetts Medical Society. All rights reserved.