Adjuvant bevacizumab in patients with melanoma at high risk of recurrence (AVAST-M): preplanned interim results from a multicentre, open-label, randomised controlled phase 3 study

Background Bevacizumab, a monoclonal antibody that targets VEGF, has shown restricted activity in patients with advanced melanoma. We aimed to assess the role of bevacizumab as adjuvant treatment for patients with resected melanoma at high risk of recurrence. We report results from the preplanned interim analysis. Methods We did a multicentre, open-label, randomised controlled phase 3 trial at 48 centres in the UK between July 18, 2007, and March 29, 2012. Patients aged 16 years or older with American Joint Committee on Cancer stage (AJCC) stage IIB, IIC, and III cutaneous melanoma were randomly allocated (1:1), via a central, computer-based minimisation procedure, to receive intravenous bevacizumab 7·5 mg/kg, every 3 weeks for 1 year, or to observation. Randomisation was strati ﬁ ed by Breslow thickness of the primary tumour, N stage according to AJCC staging criteria, ulceration of the primary tumour, and patient sex. The primary endpoint was overall survival; secondary endpoints included disease-free interval, distant-metastases interval and quality of life. Analysis was by intention-to-treat. This trial is registered as an International Standardised Randomised Controlled Trial, number ISRCTN81261306. Findings 1343 patients were randomised to either the bevacizumab group (n=671) or the observation group (n=672). Median follow-up was 25 months (IQR 16–37) in


Introduction
Melanoma is a chemoresistant cancer. Fewer than 50% of patients with resected locoregional melanoma survive to 5 years. 1 Adjuvant trials assessing interferon alfa have shown that the drug delays melanoma recurrence, but has only a small overall survival benefi t. 2 The low survival benefi t and high treatment-related toxic eff ects associated with interferon alfa mean that no internationally agreed standard adjuvant treatment is available for patients with high-risk melanoma.
Targeting the proangiogenic VEGF with the monoclonal antibody bevacizumab provides modest survival gains for patients with some types of advanced cancer. 3 Proangiogenic factors play a key part in neovascularisation of invading and metastasising cancer; therefore, VEGF blockade after locoregional surgery might prevent spread of disease. 4,5 VEGF concentration measured in either the tissue or serum of patients with melanoma correlate with disease stage and tumour burden, and might be prognostic. [6][7][8][9][10] Findings from clinical trials of small-molecule inhibitors with selectivity for VEGFR in patients with advanced melanoma have been disappointing; [11][12][13] however, bevacizumab activity was more promising. 14, 15 We therefore did this study of adjuvant bevacizumab versus observation after resection of melanoma to establish whether angiogenesis inhibition would off er clinical benefi t in patients at high risk of recurrence.

Study design and patients
We did this open-label, randomised controlled phase 3 trial between July 18, 2007, andMarch 29, 2012, at 48 centres in the UK (appendix). Eligible patients were at least 16 years old, with histological confi rmation of completely resected American Joint Committee on Cancer stage IIB (T3bN0M0 and T4aN0M0), IIC (T4bN0M0), or III (TxN1-3M0) cutaneous melanoma. Sentinel lymph-node biopsy was recommended but not mandatory; complete lymphadenectomy was done if the sentinel node was positive. Inclusion criteria were an Eastern Cooperative Oncology Group (ECOG) performance status of 0-1; a life expectancy of 6 months or more; and adequate haematological, liver, and renal function. Exclusion criteria were evidence of distant or non-regional lymph-node metastases (established by CT or MRI scanning of the body and head within 8 weeks of randomisation), incomplete resection of melanoma, or adjuvant radiotherapy that was ongoing at randomisation. Treatment assignment could not take place within 4 weeks of surgery or in the presence of unhealed wounds, but had to be within 12 weeks of a patient's latest surgery for melanoma. Previous chemotherapy, immunotherapy, or hormonal therapy for melanoma was not allowed within 12 weeks of randomisation. Patients had to be free from drug administration for 28 days either side of any procedure. Patients were also excluded if they had uncontrolled hypertension, or a history or evidence of any disorder that might increase risk of bleeding.
All patients gave written informed consent. We obtained regulatory approval, and ethics approval from a multicentre research ethics committee. Because this interim analysis was preplanned, and in view of the absence of an international standard adjuvant therapy for melanoma, the independent data monitoring committee supported publication of these results in advance of the fi nal analysis anticipated in 2017.

Randomisation and masking
Eligible patients were randomly assigned centrally in a 1:1 ratio using a computer-based minimisation algo rithm, to receive either adjuvant bevacizumab or standard observation. Randomisation was stratifi ed by Breslow thickness of the primary tumour, N stage according to AJCC staging criteria, 1 ulceration of the primary tumour (absent, present, or unknown), and patient sex. This was an openlabelled trial and therefore participants, investigators, and trial staff were not masked to group allocation.

Procedures
For patients randomly assigned to bevacizumab, 7·5 mg/kg was given via 30 min intravenous infusion once every 3 weeks for 1 calendar year (maximum 17 infusions), or until disease recurred. No dose reductions were permitted, but patients could have a maximum of two drug holidays (at their request), each of no more than 6 weeks. If a patient required any further dose interruptions, treatment was discontinued.   Patients in both arms were assessed every 3 months for 2 years; then every 6 months for 3 years; and annually thereafter until 10 years from randomisation, death, or withdrawal for any other reason. Research nurses contacted patients at 6 month intervals between annual clinic visits to enable survival information to be updated. Medical history, physical examination, full blood count, clotting, clinical chemistry, and protein urinalysis were assessed at baseline and every study visit. Chest radiography was done at 3 and 6 months after randomisation; then every 6 months up to 3 years, thereafter annually up to 5 years. Patients completed the European Organisation for Research and Treatment of Cancer Quality-of-Life Questionnaire (EORTC QLQ-C30, version 3.0) every 3 months for 2 years, then at 2·5, 3, 4, and 5 years. Adverse events reported during the fi rst year by patients in the observation group and for 28 days after the last bevacizumab infusion for those in the bevacizumab group were recorded according to the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE), version 3.0. Locoregional and distant recurrence were managed as per local hospital practice. Treatment upon disease progression was determined by local investigator decision for both patient groups. Treatment options included surgery; systemic therapy; radiotherapy, dependent on type of recurrence; and patient choice.
We ascertained BRAF mutation status in archived tumour tissue with a cobas4800 (Roche Diagnostics, West Sussex, UK) test or by pyrosequencing. NRAS mutation status was identifi ed in BRAF wild-type tumours by pyrosequencing. NRAS status was not established for BRAF mutant tumours, because NRAS and BRAF mutations are generally mutually exclusive.

Outcomes
The primary endpoint of the trial was overall survival, which we defi ned as the time from date of randomisation until date of death from any cause, or censored at the last known date alive. Secondary endpoints were disease-free interval, distant-metastasis-free interval, safety and toxic eff ects, and health-related quality of life. Tertiary endpoints were assessment of biological predictive and prognostic markers. Disease-free interval was defi ned as the time from the date of randomisation until the date of fi rst tumour recurrence (including distant and locoregional recurrence), or date of death due to melanoma. Distant-metastasis-free interval was defi ned as the time from the date of randomisation until the date of fi rst distant recurrent disease, or date of death due to melanoma.

Statistical analysis
Patients who withdrew consent for further follow-up were included in the analysis, but censored at the time of withdrawal. The expected rates of 1 year and 5 year overall survival in the standard observation group were roughly 85% and 40%, respectively. A sample size of 1320 patients (660 patients per group) was needed to detect an absolute increase in 5 year overall survival from 40% to 48%, with 85% power and a 5% signifi cance level, which equates to a hazard ratio (HR) of 0·80. An interim analysis was planned after all patients had fi nished the treatment and  Survival curves were constructed with the Kaplan-Meier method and compared with a log-rank χ² test. A Cox proportional hazard model was used to obtain HRs and associated 95% CIs. HRs were calculated for prognostic subgroups and a HR plot constructed. 16 Multivariable Cox-regression models were used to adjust the treatment eff ect for stratifi cation variables, to assess the independent predictors of overall survival and disease-free interval, and to assess treatment interaction with the tumour mutation status and whether hypertension, fi tted as a time-dependent covariate, aff ected disease-free interval. We did a sensitivity analysis excluding ineligible patients.
We scored quality-of-life data on a scale from 0 to 100 according to the algorithm described in the EORTC QLQ-C30 scoring manual. 17 High scores imply an improvement in function or quality of life, but more severe symptoms. Quality-of-life data were analysed by standardised area-under-the-curve analysis 18 and compared across trial groups with Wilcoxon rank-sum tests.
Reported p values are two-sided. All analyses were done by intention to treat with the SAS statistical package (version 9.3). This trial is registered as an International Standardised Randomised Controlled Trial, number ISRCTN 81261306.

Role of the funding source
The sponsor and funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of the report. AM had full access to all the data in the study and the corresponding author had fi nal responsibility for the decision to submit for publication.
We recorded an improvement in disease-free interval for patients in the bevacizumab group compared with those in the observation group (fi gure 3). At 1 year, 77% (95% CI 73-80) of patients given bevacizumab were disease free, as were 70% (66-73) for those who underwent observation. The treatment eff ect for diseasefree interval remained irrespective of the stratifi cation variables (fi gure 4). Multivariable analysis identifi ed disease stage, Breslow thickness of the primary melanoma, trial group, and ECOG performance status as independent predictors of disease-free interval (table 2). There was a suggestion of an improved distantmetastasis-free interval with bevacizumab, but it was not signifi cant (fi gure 2). Types of recurrence and the treatment methods used for recurrence were similar between treatment groups (table 3). 184 (65%) of the 283 local recurrences and 68 (16%) of the 430 distant recurrences were treated with surgery alone. Results from the analysis of disease-free interval were unchanged in a sensitivity analysis excluding the ineligible patients.
BRAF status was available for 645 (48%) patients (table 1). A BRAF V600 mutation was detected in 291 (45%) of the 645 assessed patients (table 1). Suffi cient material remained to undertake NRAS testing in 239 (68%) of 354 patients with BRAF wild-type; 106 (44%) of these patients had a mutation at codon 61. No codon 12 or 13 mutations were detected. The interaction between treatment and BRAF status was not signifi cant (p=0·10); however, in patients with BRAF mutant tumours, we noted an improvement in the disease-free interval in those given bevacizumab (fi gure 3). Diseasefree interval did not diff er signifi cantly between treatment groups in the wild-type BRAF population (fi gure 3).
Three adverse drug reactions were regarded as both serious and unexpected. One patient developed optic neuritis after the fi rst bevacizumab infusion, which resulted in 90% loss of vision in one eye. A second patient had persistent erectile dysfunction, but continued to receive treatment. A third patient died of a haemopericardium after receiving two bevacizumab infusions, and was later identifi ed to have had signifi cant predisposing cardiovascular risk factors.

Discussion
Our fi ndings show an improvement in disease-free interval with bevacizumab in patients with resected melanoma at high risk of recurrence as compared with observation alone. Median disease-free interval and overall survival have not yet been reached. The curves for disease-free interval remain separate at this interim analysis. Follow-up of patients and timing of clinic visits were similar between treatment groups. Clearly, it will be important to ascertain whether this treatment eff ect ultimately translates into a benefi t in overall survival at fi nal analysis, which is follow-up driven and planned for when all patients have been on study for a minimum of 5 years, at which time 736 events are anticipated. The conditional power for futility of the primary outcome of overall survival at this interim analysis was 35%. To our knowledge, AVAST-M is one of the largest adjuvant melanoma trials and the largest bevacizumab monotherapy study ever undertaken (panel). By contrast with most published adjuvant melanoma trials, we chose overall survival as the primary endpoint in AVAST-M. Patients and regulatory agencies increasingly value relapse-free survival as a primary endpoint in this setting, which is justifi ed now that systemic therapies have been shown to improve overall survival in advanced disease. [19][20][21] However, when we started AVAST-M bevacizumab was a new drug with signifi cant toxic eff ects (albeit reported mainly in combination with cytotoxic chemotherapy), and so we felt that a benefi t in overall survival would need to be shown to change clinical practice. We chose disease-free interval and distant-metastasis-free interval as secondary endpoints to identify the eff ect of bevacizumab on disease recurrence. However, diseasefree interval in this trial could be indiciative of relapsefree survival, because only seven additional deaths from other causes without recurrence would have been included in an analysis of relapse-free survival; these inclusions would not have changed the results from those reported for disease-free interval.
The most frequently used adjuvant melanoma treatment worldwide is interferon alfa. The role of adjuvant interferon is controversial and, despite being used in both the USA and Europe, it is not an international standard of care. Moreover, in many countries adjuvant therapy for resected melanoma is not routinely off ered at all. The controversy regarding use of interferon surrounds the interpretation of its clinical   signifi cance attributable to marginal survival benefi ts alongside potential for harm due to drug side-eff ects. Trials assessing diff erent doses and formulations of interferon alfa have not identifi ed the optimum adjuvant regimen for patients with melanoma, 2 while this treatment is recognised to be associated with substantial toxic eff ects, including liver dysfunction, fatigue, and depression. 22,23 Other immunotherapeutic strategies, including several vaccines, have been rigorously tested in patients with melanoma, with disappointing results overall. 24 This outcome is exemplifi ed by the results of the DERMA randomised trial, in which a MAGE A3 vaccine did not improve relapse-free survival (its fi rst coprimary endpoint) in patients with resected stage III melanoma expressing the MAGE A3 antigen. 25 Our results suggest that the extent of patient benefi t off ered with bevacizumab might be similar to that with interferon alfa for the endpoint of relapse-free survival (HR for interferon alfa 0·82, 95% CI 0·77-0·87), but the overall survival benefi t (0·89, 0·82-0·96) on ifterferon alfa was not noted with bevacizumab. 2 At this interim analysis, few patients had received immunotherapy or targeted therapy at relapse, and the most common intervention was surgery. This fi nding is consistent with the high rate of locoregional recurrence as a result of the fairly low use of sentinel node biopsy and the inclusion of high risk primary tumours, together with the scarcity of active systemic therapy options   available until more recently. The eff ect of subsequent treatments at relapse on overall survival is likely to be small at this time point, but will be of greater importance in the future. Even so, the delay in melanoma recurrence noted with adjuvant bevacizumab in this trial might be clinically relevant.
Bevacizumab monotherapy seems to be well tolerated: grade 3 or 4 adverse events associated with bevacizumab were fewer than those reported with interferon, depending on the dose used, although cross-trial comparisons should be made with caution. 22,23 There was no obvious safety signal associated with surgical interventions or haemorrhage to account for patients' early discontinuation. The protocol specifi ed dose interruptions around surgical interventions, requiring 28 days free from drug administration either side of any procedure, and preventing drug administration in the presence of any unhealed wound. Drugassociated complications in wound healing seem to be negligible, and only a few patients failed to complete planned treatment due to unacceptable delays in wound healing, suggesting that, in routine clinical practice, withholding treatment in such circumstances need not be so stringent. Fairly high patient withdrawal rates have been reported in both low-dose and high-dose adjuvant interferon alpha trials: 15% in the UK AIM HIGH trial 23 and 37% in the EORTC 18991 trial. 22 However, it is worth noting that, in the EORTC 18991 trial, grade 3 and 4 adverse events were reported in 12% of patients receiving placebo. Withdrawal rates will be aff ected by the duration of intended treatment, but low-level chronic side-eff ects in an otherwise fi t population might account for patients electing to stop prematurely and this should be considered in future adjuvant trial designs. A placebocontrolled trial might have improved patient willingness to remain on treatment for longer, but the additional cost of a placebo would have been prohibitive in this charityfunded trial.
Multivariate analyses of disease-free interval and overall survival were undertaken to explore whether some subsets of patients were more likely to benefi t from therapy than others. Disease stage and Breslow thickness were the most signifi cant predictors of disease-free interval; disease stage and ECOG performance status were the most signifi cant predictors of overall survival. A tenth of patients had an ECOG performance status score of 1 on trial entry; the reason for their poor outcome is not apparent, but will be assessed fully in the fi nal analysis.
AVAST-M includes a prespecifi ed translational study, aimed at identifying predictive and prognostic biomarkers. We retrospectively obtained tumour blocks with which to test BRAF mutation status; therefore, the possibility of ascertainment bias cannot be excluded. The 45% BRAF V600 mutation rate within the assessed population is consistent with data derived from other advanced melanoma populations. Firm conclusions about whether bevacizumab might result in a longer disease-free interval for the subgroup of patients with BRAF mutant tumours cannot yet be made, but further analysis including multivariate analyses will be undertaken at the fi nal analysis. The subgroup of patients assessed for NRAS mutation was too small to provide reliable results for disease-free interval and overall survival at this stage. Firm conclusions from the BRAF subgroup analysis cannot yet be made, but we will do multivariate analyses at the fi nal analysis. Investigation into the biological basis for this initially unexpected fi nding is ongoing, but it is consistent with emerging evidence of the MAP kinase pathway playing a role in the control of VEGF expression. Studies in a mouse model of Data are n/N (%) or n (%), unless otherwise indicated. *An additional three patients (one in the bevacizumab group and two in the observation group) who died of melanoma before details of their recurrence being reported are included in the analyses of disease-free interval. †Patients can have more than one site of locoregional recurrence and treatment for recurrence recorded. ‡20 (16%) of the patients with regional lymph-node recurrence had previously had a sentinel lymph-node biopsy.  26 In BRAF mutant mouse xenografts, BRAF inhibition resulted in increased T-cell infi ltration of tumours, an eff ect that was mediated by downregulation of VEGF. If this mechanism applies to melanomas in human beings, it would provide a basis for the selection of patients for bevacizumab therapy, and a rationale for future potential combination regimens. Three trials testing adjuvant bevacizumab in combination with chemotherapy have been reported in patients with colon 27,28 and breast cancer. 29 All three trials had disease-free survival as the primary endpoint, and none identifi ed any improvement. These negative trials have called into question whether inhibition of micrometastatic disease by bevacizumab is a realistic goal. Angiogenesis inhibitors have immunomodulatory eff ects 30,31 and (by contrast with epithelial cancers), melanoma is a highly immunogenic tumour. It is possible that any benefi t attributable to bevacizumab is due to immunological eff ects, which can persist beyond the treatment period. Both immunological and angiogenic biomarkers are now being explored with tumour and blood samples collected from most patients recruited to the AVAST-M trial.
Appetite loss 0 (0-4·2) 0 (0-5·6) 0·38 Constipation 0 (0-6·7) 0 (0-4·2) 0·04 Diarrhoea 0 (0-5·6) 0 (0-5·6) 0·29 Financial diffi culties 0 (0-8·3) 0 (0-4·8) 0·06  Data are n (%). Adverse events are recorded for all patients who had an event of grade 3 or 4 severity and include any other adverse events that were recorded in 10% or more patients. Patients can have more than one type of adverse event. We classifi ed events are classifi ed with the National Cancer Institute Common Terminology Criteria for Adverse Events (version 3.0). Eff ective treatment to prevent melanoma relapse remains an unmet need. Bevacizumab is not yet established as an eff ective treatment for melanoma and other drugs are being tested in the adjuvant setting. The results of these trials, alongside the fi nal results of the AVAST-M trial, will have a key role in identifying future adjuvant strategies for this disease.

Contributors
PC was the chief investigator, responsible for the trial design, trial management, data interpretation, and preparation of the manuscript. AM was the trial statistician who analysed the data, created the tables and fi gures, and contributed to the trial design, trial management, data interpretation, and preparation of the manuscript. JAD contributed to the trial design, trial management, and data interpretation. MRM was responsible for the tumour genotyping and contributed to the trial design, trial management, data interpretation, and revision of the manuscript. PN, MG, ND, SN, CK, MM, SD, EM, SH, RB, AW, JN, MH, SK recruited the highest numbers of patients and were involved in data collection at their sites. GY was involved in data collection and coordination of the trial. PL contributed to the trial design, trial management, data interpretation, and revision of the manuscript. All authors participated in the revision and fi nalisation of the manuscript.

Declaration of interests
PC has received non-fi nancial support from F Hoff man La Roche, grants from Cancer Research UK during the study, and personal fees and non-fi nancial support from F Hoff man La Roche outside the submitted work. AM has received grants from Cancer Research UK during the study. JAD has received grants from Cancer Research UK during the study. MRM has received grants from Cancer Research UK during the study, personal fees from Amgen, grants and personal fees from F Hoff man La Roche, grants from Astrazeneca, grants and personal fees from GlaxoSmithKline, institution study fees from Novartis, institution study fees from Astellas, institution study fees from Millenium, institution study fees from Immunocore, personal fees and institution study fees from Bristol Myers Squibb, institution study fees from Vertex, personal fees and institution study fees from Eisai, institution study fees from Abbott, institution study fees from Clovis, institution study fees from Pfi zer, institution study fees from Merck, outside the submitted work. PDN has received grants and personal fees from F Hoff man La Roche outside the submitted work. MG has received personal fees from F Hoff man La Roche, and advisory board and speakers' bureau fees outside the submitted work. SH reports paid advisory board role for F Hoff man La Roche. REB has received travel grants and honoraria outside the submitted work. AW reports travel and accommodation for sponsored colorectal conference from Roche, outside the submitted work. PL reports being a remunerated consultant to F Hoff man La Roche for unrelated product and support for travel from F Hoff man La Roche, outside the submitted work. All other authors declare that they have no competing interests.