Standard chemotherapy with or without bevacizumab in advanced ovarian cancer: quality-of-life outcomes from the International Collaboration on Ovarian Neoplasms (ICON7) phase 3 randomised trial.

BACKGROUND: In the Gynecologic Cancer Intergroup International Collaboration on Ovarian Neoplasms 7 (ICON7) trial, bevacizumab improved progression-free survival in patients with ovarian cancer when used in combination with first-line chemotherapy and as a single-drug continuation treatment for 18 cycles. In a preliminary analysis of a high-risk subset of patients, there was also an improvement in overall survival. This study aims to describe the health-related quality-of-life (QoL) outcomes from ICON7.
METHODS: ICON7 is a randomised, multicentre, open-label phase 3 trial. Between Dec 18, 2006, and Feb 16, 2009, after a surgical procedure aiming to debulk the disease, women with International Federation of Gynecology and Obstetrics (FIGO) high-risk stage I-IV epithelial ovarian cancer were randomly allocated (1:1) by computer program and block randomisation to receive either six cycles of standard chemotherapy (total 18 weeks) with carboplatin (area under the curve 5 or 6) and paclitaxel (175 mg/m(2)) alone or with bevacizumab (7·5 mg/kg) given intravenously with chemotherapy and continued as a single drug thereafter (total 54 weeks). The primary QoL endpoint was global QoL from the European Organisation for Research and Treatment of Cancer quality-of-life questionnaire-core 30 at week 54, analysed by ANOVA and adjusted for baseline score. Analyses were by intention to treat. The ICON7 trial has completed recruitment and remains in follow-up. This study is registered, number ISRCTN91273375.
FINDINGS: 764 women were randomly assigned to the standard chemotherapy group and 764 to the bevacizumab group. At baseline, 684 (90%) of women in the standard chemotherapy group and 691 (90%) of those in the bevacizumab group had completed QoL questionnaires. At week 54, 502 (66%) women in the bevacizumab group and 388 (51%) women in the standard chemotherapy group provided QoL data. Overall, the mean global QoL score improved during chemotherapy by 7·2 points (SD 24·4) when analysed for all women with data at baseline and week 18. The mean global QoL score at 54 weeks was higher in the standard chemotherapy group than in the bevacizumab group (76·1 [SD 18·2] vs 69·7 [19·1] points; difference 6·4 points, 95% CI 3·7-9·0, p<0·0001).
INTERPRETATION: Bevacizumab continuation treatment seems to be associated with a small but clinically significant decrement in QoL compared with standard treatment for women with ovarian cancer. The trade-off between the prolongation of progression-free survival and the quality of that period of time needs to be considered in clinical practice when making treatment decisions. FUNDING: Roche and the National Institute for Health Research through the UK National Cancer Research Network.

RCT Entities:   Population Interventions Outcomes

Introduction

Angiogenesis is central to the process of cancer growth and metastasis and has a role in the progression and prognosis of ovarian cancer. VEGF is an important promoter of angiogenesis produced by normal and neoplastic cells. Bevacizumab is a recombinant humanised version of a murine anti-human VEGF monoclonal antibody and has been studied in the management of many tumours. The International Collaboration on Ovarian Neoplasms 7 (ICON7) trial is a Gynecologic Cancer Intergroup phase 3 trial that assessed the effects of adding bevacizumab, concurrently and as a continuation, to standard chemotherapy with carboplatin and paclitaxel in patients with primary peritoneal carcinoma, fallopian tube carcinoma, and epithelial ovarian carcinoma (ovarian cancer). Patient characteristics, progression-free survival, toxicity, and preliminary overall survival data and a summary of quality-of-life (QoL) data have been reported from ICON7. In the standard chemotherapy group, 696 (91%) of 764 women received 18 weeks of chemotherapy by protocol. In the bevacizumab group, 719 (94%) of 764 women received 18 weeks of chemotherapy and bevacizumab and 472 (62%) continued bevacizumab to protocol completion at 54 weeks. The hazard ratio for progression-free survival with standard chemotherapy and bevacizumab was 0·81 (95% CI 0·70–0·94, p=0·004). In patients at high risk of progression, defined as International Federation of Gynecology and Obstetrics (FIGO) stage IV disease or stage III disease with greater than 1·0 cm of residual disease after debulking surgery, the hazard ratio for death in the bevacizumab group was 0·64 (95% CI 0·48–0·85; p=0·002). We noted consistent differences in QoL between the two groups that were less than a 10-point difference, and referred to the evolving interpretations of that difference as QoL data have been described in more detail.

The evaluation of treatments that achieve such gains in survival needs to include a comprehensive understanding of the other effects of these treatments on patients. Detailed health-related QoL assessment can provide a distinct and broad evaluation of the health, function, and wellbeing of people with cancer and of the functional and symptomatic benefits and losses that might result from the medical interventions. QoL is important in advanced ovarian cancer, where treatments can be effective but are only rarely curative. Even when treatments prolong progression-free or overall survival, an analysis of the effects on QoL using validated patient self-reported measures is needed to understand the life quality of that additional period. There are few detailed QoL analyses of outcomes in large first-line ovarian cancer treatment trials, which tend to publish outline data only. However, detailed descriptions of QoL can guide clinicians and patients to make informed treatment decisions and describe the experience of receiving novel cancer treatments. Moreover, well-designed, detailed, and rigorous QoL analyses are increasingly needed for regulatory approvals. We report findings from a QoL substudy, designed at the outset of ICON7, that examined the effect of adding bevacizumab to standard chemotherapy in patients with ovarian cancer.

Methods

Study design

ICON7 is a randomised, multicentre, open-label phase 3 trial designed to assess the safety and efficacy of adding intravenous bevacizumab (7·5 mg/kg, intravenous over 30–90 min on day 1 of every 3-week cycle; F Hoffmann-La Roche, Basel, Switzerland) to standard intravenous chemotherapy with carboplatin (area under the curve 5 or 6 intravenous over 1 h on day 1 of every 3-week cycle) and paclitaxel (175 mg/m2, intravenous over 3 h on day 1 of every 3-week cycle) in the management of patients with FIGO high-risk stage I–IV epithelial ovarian cancer (figure 1). The contributing hospitals all had existing suppliers and continued to use these suppliers during the study. Each participant gave written informed consent before enrolment and the study was approved by the relevant ethics committees and national and local research governance authorities.

Figure 1

Study schema, with measurement of quality of life over time and timing of a-priori quality-of-life outcomes

Every coloured box in the top panel represents one 3-week chemotherapy cycle. FIGO=International Federation of Gynecology and Obstetrics.

Randomisation and masking

Between Dec 18, 2006, and Feb 16, 2009, after a surgical procedure aiming to debulk the disease, 1528 women were recruited at 263 centres in 11 countries (Australia, Canada, Denmark, Germany, Spain, Finland, France, UK, Norway, New Zealand, and Sweden) and randomly allocated (1:1) by a computer program to standard chemotherapy (standard chemotherapy group) or standard chemotherapy plus bevacizumab (bevacizumab group) in the order they were enrolled by block randomisation, stratifying for the following factors: FIGO stage (stage I–III with residual disease ≤1 cm, stage I–III with residual disease >1 cm, FIGO stage IV), intent to start chemotherapy 4 weeks or fewer after surgery versus intent to start chemotherapy more than 4 weeks after surgery, and Gynecologic Cancer Intergroup group. Inclusion and exclusion criteria are listed on the study website. Treatment in the standard chemotherapy group was for 18 weeks and treatment in the bevacizumab group was for 54 weeks; group allocation was not masked.

Procedures

QoL data for the ICON7 QoL substudy were collected on paper by patient self-reporting. Patients completed the QoL questionnaires on their own without conferring with friends or relatives and always before the administration of treatment or medical consultation (eg, baseline questionnaires were completed on day 1 before randomisation), at timepoints defined by the protocol during outpatient attendances (figure 1). QoL data were not collected after disease progression, but a scheduled QoL data collection for all patients still alive at 3 years will be reported in the future.

The 58 items of the European Organisation for Research and Treatment of Cancer quality-of-life questionnaire–ovarian cancer module (EORTC QLQ–OV28) and the EORTC QLQ–core 30 (C30) were used to measure QoL within the preceding 7 days. The C30 contains 30 items, grouped into a global health status scale, five function scales (physical, role, emotional, cognitive, and social functioning), and nine symptom scales or items (fatigue, nausea and vomiting, pain, dyspnoea, insomnia, appetite loss, constipation, diarrhoea, and financial difficulties). The OV28 contains 28 items that are designed to assess QoL in ovarian carcinoma, including abdominal or gastrointestinal symptoms, peripheral neuropathy, other chemotherapy side-effects, hormonal or menopausal symptoms, body image, attitude to disease or treatment, and sexual functioning. For function scales, higher scores suggest more function (hence better QoL), whereas for symptom scales, higher scores suggest more symptoms (hence poorer QoL). These instruments were chosen because of their extensive psychometric validation, acceptability to patients, and availability in 14 languages. They also include core domains of QoL such as abdominal symptoms and physical and social function scales, which are pertinent to examining our a-priori hypotheses.

The definition we used for minimum clinically important difference for the C30 scales was originally defined (until March 24, 2011) as a change of ten points for any C30 scale. However, methodological advances including findings from a review of 152 articles showed that group differences in mean QoL scores reported in trials can be classified reliably by clinicians as trivial, small, moderate, or large. For the C30, the minimum clinically important difference varies from subscale to subscale; for example, in the global QoL scale, a small between-group mean difference was taken as four to seven points (suggesting a small but clinically relevant difference) and a moderate difference as ten to 15 points. We have adopted these definitions for our analyses. For the OV28 subscales, because no evidence-based clinically relevant difference has been defined, we used a clinically significant effect size of 0·3 standard deviations.

The primary QoL endpoint was the global QoL score at 54 weeks on the C30 (range 0–100, with 100 being best QoL). A priori, we specified early, mid, and late hypotheses as secondary QoL endpoints to detect specific differences predicted by the known effects of bevacizumab administration, corresponding to different periods in a patient's treatment (figure 1).

Our early hypothesis, after the first post-baseline assessment, was that, because angiogenesis is related to ascites formation, more rapid resolution of ascites would occur in the bevacizumab group than in the standard chemotherapy group. We estimated resolution of ascites using the abdominal or gastrointestinal subscale of the OV28 over the chemotherapy phase of treatment, with the daily mean gastrointestinal symptom score derived from the area under the QoL–time curve divided by the number of days from randomisation.

Our mid hypothesis, halfway through the chemotherapy phase, at week 9, was that the administration of bevacizumab would slow wound healing and hence increase symptoms from the scar, as assessed by comparison between groups of social functioning, physical functioning, and pain subscales of C30 and the body image scale of OV28.

Our late hypothesis, from the end of the chemotherapy phase to 54 weeks, was that the continuation of treatment in the bevacizumab group would prolong an impairment of social functioning and increase fatigue. To assess this hypothesis, we examined the individual patient change in social functioning and fatigue scales of the C30 between the completion of cytotoxic chemotherapy and completion of 36 weeks of continuation bevacizumab. We also undertook exploratory analyses of scores from other subscales in the C30 and OV28 at week 54.

We also did a sensitivity analysis to examine the effect of missing data. We collected data regarding compliance with questionnaire responses; when a form was missing, researchers recorded whether this was because the patient refused to complete the form, the centre was unable to complete the form (eg, staff absence), the patient's language needs were not met, the site missed the administration, or the patient was too ill.

Statistical analysis

The sample size for ICON7 was calculated to provide 83% power to detect a 28% change in progression-free survival at 18 months of follow-up and 80% power to identify a 19% improvement in overall survival, each with a two-sided significance level of 5%, allowing for up to 5% non-compliance with protocol.

We compared differences between treatment groups in the primary QoL endpoint using ANOVA adjusting for baseline score against a two-sided statistical significance level of 0·05. Secondary QoL endpoints were analysed against a two-sided statistical significance level of p=0·05. Differences between the treatment groups in all other validated subscales were assessed against a two-sided significance level of p=0·01.

After primary analyses of all available data from both groups, we used imputation to characterise the potential effect of missing data. Where data were missing because of administration error at the site, we imputed average global QoL baseline scores for the group, inferring these data were missing at random. Data missing because of patient refusal or illness or missing without explanation were imputed as a reduction in global QoL of 15 points from the mean for that treatment group. Where data were missing because of death we imputed a global QoL of 0. Because of the higher number of women returning QoL data in the bevacizumab group than in the standard chemotherapy group (because time to disease progression was longer), we modelled the potential effect of data that were not sought because of disease progression by imputatation and comparison of a full range of values, examination of the effect of that imputation, and comparison with data within the Medical Research Council (MRC) ICON4 trial of QoL at first relapse in ovarian cancer, to calculate the range of clinically plausible values.

All analyses were undertaken on the intention-to-treat population using Stata version 10 or later.

This study is registered as an International Standard Randomised Controlled Trial, number ISRCTN91273375, and with EudraCT, number 2005-003929-22.

Role of the funding source

The sponsor and funding sources had a role in the study through membership of a professional academic steering group. Roche had a role in study design, data interpretation, and writing of the report, through representation on the steering group for quality of life. The National Institute for Health Research through the UK National Cancer Research Network had a role in study design via peer review. The MRC as sponsor had a regulatory role, but their staff had a role in study design, data collection, data analysis, data interpretation, and writing of the report. ES was the sole representative of Roche on the steering group for the quality of life and was involved in the study design, data collection, data analysis, data interpretation, writing of the report, and the decision to submit for publication along with DS, MN, LE, MS, FH, JB, AL, GV, WQ, and AMS. DS, MN, WQ, and AMS had access to the raw QoL data. DS had full access to all of the data and the final responsibility for the decision to submit for publication.

Results

All 1528 women (764 in both the standard chemotherapy and the bevacizumab groups) enrolled in ICON7 were asked to provide data on QoL. Table 1 shows concordance with data collection by group. At baseline, 684 (90%) women in the standard chemotherapy group and 691 (90%) in the bevacizumab group provided complete data. By 54 weeks, 388 (51%) women in the standard chemotherapy group and 502 (66%) in the bevacizumab group had provided data; 254 (33%) women who received standard chemotherapy and 147 (19%) in the bevacizumab group had either progressive disease or had died (so no QoL data were sought), with an imbalance between the two groups (table 1), and QoL data were missing for 122 (16%) women in the standard group and 115 (15%) of those in the bevacizumab group. Specific reasons for these missing data forms were given for 101 women (6·6%; 50 [6·5%] in the standard chemotherapy group and 51 [6·7%] in the bevacizumab group). For 85 of 101 forms (44 in the standard chemotherapy group and 41 in the bevacizumab group) the reason was centre administrative oversight, but for 15 (six in the standard chemotherapy group and nine in the bevacizumab group) it was because of either a patient declining to complete the form or illness. One patient in the bevacizumab group did not complete a data form because they were on holiday.

Table 1

Attrition in returns of health-related quality-of-life data

	Standard chemotherapy group (n=764)				Bevacizumab group (n=764)
	n (%)	Died (%)	Progression but alive (%)	No data (%)	n (%)	Died (%)	Progression but alive (%)	No data (%)
Screening	684 (90%)	0 (0%)	0 (0%)	80 (10%)	691 (90%)	0 (0%)	0 (0%)	73 (10%)
Week 18	662 (87%)	11 (1%)	15 (2%)	76 (10%)	693 (91%)	7 (1%)	5 (1%)	59 (8%)
Week 54	388 (51%)	71 (9%)	183 (24%)	122 (16%)	502 (66%)	43 (6%)	104 (14%)	115 (15%)

No imbalance of attrition was observed, comparing study groups by age, centre, country, or stage (data not shown).

During the 18-week period of chemotherapy treatment in both groups, mean global QoL for women with data from both baseline and week 18 (333 women in the standard chemotherapy group and 444 women in the bevacizumab group) improved by 7·2 points (SD 24·4). A difference between the group mean global QoL scores of 5·1 points (95% CI 2·9–7·4) was noted at the end of chemotherapy when analysed for all women with available data (table 2). The primary outcome, mean global QoL at 54 weeks, was better for women in the standard chemotherapy group (76·1 points [SD 18·2]) than for those in the bevacizumab group (69·7 points [19·1]). This difference, 6·4 points (95% CI 3·7–9·0), is clinically small by modern criteria but statistically significant (p<0·0001 by ANOVA). The number of women whose global QoL score improved by at least 10 points between baseline and 54 weeks was 221 of 333 (66%) with standard chemotherapy and 250 of 444 (56%) with bevacizumab (odds ratio 0·58, 95% CI 0·42–0·80; p=0·001). This finding was also significant when all available data during the 54-week period were included and mean individual area under the global QoL–time curves were compared between groups, controlled for baseline score (p<0·0001, ANOVA).

Table 2

Mean global quality-of-life score between groups at key timepoints for all patients with available data

	Standard chemotherapy group	Bevacizumab group	Difference	p value
Baseline (before randomisation)	58·6 (20·6)	55·1 (20·8)
Week 18 (end of chemotherapy)	64·4 (20·3)	59·2 (19·4)	−5·1 (−7·4 to −2·9)	<0·0001
Week 54 (end of continuation bevacizumab)	76·1 (18·2)	69·7 (19·1)	−6·4 (−9·0 to −3·7)	<0·0001

Data are mean (SD) or difference (95% CI). Numbers of patients per group are shown in table 1. p values calculated by ANOVA controlled for baseline value.

Analysis of our a-priori hypotheses did not support a difference between groups during the chemotherapy course in gastrointestinal problems or problems related to the surgical scar (table 3). In the continuation bevacizumab phase, we did not note a difference between groups in the trajectory of social functioning or fatigue (table 3).

Table 3

Analyses of three quality-of-life hypotheses

	Mean score		Difference (95% CI)	p value
	Standard chemotherapy group	Bevacizumab group
Early: resolution of ascites
Gastrointestinal symptoms	76·4 (n=566)	74·7 (n=606)	−1·6 (−3·5 to 0·2)	0·43
Mid: scar symptoms
Pain	22·0 (n=596)	22·7 (n=628)	0·7 (−2·0 to 3·4)	0·65
Physical functioning	80·5 (n=594)	78·8 (n=624)	−1·6 (−3·7 to 0·5)	0·18
Social functioning	70·6 (n=584)	70·2 (n=621)	−0·4 (−3·4 to 2·5)	0·61
Body image	36·2 (n=586)	35·4 (n=609)	−0·9 (−4·2 to 2·4)	0·62
Late: social functioning and fatigue
Social functioning	13·7 (n=319)	11·9 (n=430)	−1·8 (−5·1 to 1·6)	0·30
Fatigue	−15·8 (n=324)	−14·9 (n=435)	0·9 (−2·3 to 4·1)	0·58

p values calculated by ANOVA controlled for baseline value.

Table 4 shows data from the exploratory analysis of other subscales in the C30 and OV28 instruments. All differences were in favour of the standard chemotherapy group. Bevacizumab was associated with clinically small but statistically significant decrements of role functioning, financial worries, attitudes to disease or treatment, hormonal symptoms, and rash (all p<0·01).

Table 4

Exploratory analyses of quality-of-life subscales

	Mean score at week 54			p value	Combined SD	Effect size
	Standard chemotherapy group	Bevacizumab group	Difference (95% CI)
EORTC QLQ-C30
Role functioning	84·5 (n=332)	75·6 (n=438)	−8·9 (−12·3 to 5·5)	<0·0001	24·4	Small*
Cognitive functioning	84·9 (n=335)	81·4 (n=445)	−3·5 (−6·5 to 0·4)	0·033	21·6	Small*
Emotional functioning	77·5 (n=335)	75·4 (n=446)	−2·1 (−5·4 to 1·2)	0·21	23·2	Trivial*
Nausea and vomiting	2·6 (n=334)	3·7 (n=444)	1·0 (−0·3 to 2·4)	0·12	9·5	Trivial*
Dyspnoea	11·8 (n=339)	14·8 (n=436)	3·0 (−0·1 to 6·1)	0·10	21·7	Trivial*
Appetite loss	4·2 (n=333)	7·6 (n=439)	3·4 (1·1 to 5·7)	0·0035	16·2	Trivial*
Finance	12·8 (n=330)	18·9 (n=440)	6·1 (2·2 to 9·8)	0·0072	26·7	Small*
Diarrhoea	5·0 (n=332)	7·3 (n=440)	2·3 (0·1 to 4·5)	0·037	15·5	Trivial*
Sleep	24·4 (n=333)	28·6 (n=443)	4·2 (0·1 to 8·3)	0·078	28·8	Small*
Constipation	13·0 (n=335)	15·8 (n=442)	2·7 (−0·8 to 6·2)	0·14	24·6	Trivial*
EORTC QLQ-OV28
Attitude	27·6 (n=329)	33·3 (n=443)	5·7 (2·0 to 9·3)	0·0029	25·7	<small†
Peripheral neuropathy	6·3 (n=320)	9·3 (n=423)	3·0 (0·3 to 5·8)	0·052	19·0	<small†
Hormonal	13·0 (n=326)	17·3 (n=436)	4·3 (1·6 to 7·0)	0·0042	18·8	<small†
Rash	12·4 (n=326)	18·0 (n=433)	5·6 (2·3 to 8·9)	0·0010	23·1	<small†

p values calculated by ANOVA controlled for baseline value. n=number of women who responded to the individual questionnaire items in question at that point in time. EORTC QLQ–C30=European Organisation for Research and Treatment of Cancer quality-of-life questionnaire–core 30. EORTC QLQ–OV28=EORTC QLQ–ovarian cancer module.

According to Cocks and colleagues.

According to Cohen.

After imputation of global QoL of 0 for patients who had died, we imputed a range of mean global QoL decrements (as estimates of the effect of progressive disease upon global QoL). Imputing a decrement of at least a 20-point reduction in global QoL for the effect of relapse of the cancer did not alter our findings of a statistically significant, clinically small, deficit in QoL for the bevacizumab group (figure 2). If the decrement in global QoL from disease progression is imputed as a 30–50-point reduction then there is no statistically significant difference between the groups. If the decrement is imputed as a 60–70-point reduction, then the analyses favour the bevacizumab group. A clinically plausible estimate of global QoL in relapsed ovarian cancer was made from the ICON4 trial (unpublished, Medical Research Council Clinical Trials Unit, London, UK). The mean global QoL (using the C30) for women being treated for ovarian cancer at first relapse in ICON4 was 61 points compared with a baseline global QoL in ICON7 of 57 points.

Figure 2

Sensitivity analyses

The figure shows point estimates with 95% CIs for the difference between the treatment groups (bevacizumab group mean minus standard chemotherapy group mean) after imputing a range of post-progression global quality-of-life values to estimate the potential effect of missing data. Values above the dashed line favour bevacizumab; those below the dashed line favour standard chemotherapy. *No imputation.

Discussion

We present findings from detailed QoL analyses from the ICON7 study of the role of bevacizumab in the first-line treatment of epithelial ovarian cancer. Global QoL at 54 weeks was better for women receiving standard chemotherapy than for those receiving standard chemotherapy plus bevacizumab. The difference between groups was clinically small by modern criteria but statistically significant. Global QoL improved over the period after surgery in both groups, when cytotoxic chemotherapy was administered. From that point forward, there was a clinically small decrement in global QoL in the bevacizumab group compared with the standard chemotherapy group, including during the period of continuation of bevacizumab. This decrement was robust to clinically plausible assumptions made in modelling missing data due to patient or centre actions and data missing by design. Clinicians, women receiving ovarian cancer treatment, and funding authorities need to bear this decrement in global QoL in mind when considering bevacizumab as first-line treatment for ovarian cancer (panel).

What the mechanism of a small decrement in QoL associated with bevacizumab might be remains unclear. We planned at study inception to examine whether bevacizumab seemed to improve the resolution of ascites, to cause a decrement to healing of abdominal wall wounds, or to prolong the disruption in day-to-day life associated with continuation of bevacizumab. None of those effects were noted. The decrease in QoL might be because of toxicity of bevacizumab, such as rash, hormonal symptoms, or the bleeding, thromboembolic or gastrointestinal perforations reported in toxicity data. Findings from exploratory analyses suggested that financial effects and impairment of women's return to their life roles might also be affected by continuation of treatment with bevacizumab compared with standard chemotherapy. There was no evidence of an effect on these QoL differences due to the amount of surgical cancer debulking or of residual ascites (data not shown). There was no evidence of an interaction between the difference reported in global QoL score and clinical risk subgroups identified in the biomedical outcomes (data not shown).

Alongside our results, the findings of the related Gynecologic Oncology Group (GOG) 218 study should be considered. In this study, a higher dose of bevacizumab was used (15 mg/kg rather than 7·5 mg/kg in ICON7), treatment allocation was double blind, and placebo was administered to women in the standard chemotherapy group during the concurrent and continuation phases of the protocol. The Functional Assessment of Cancer Therapy Ovarian Trial Outcome Index was used to assess QoL. A difference in QoL favouring standard chemotherapy with placebo occurred at two timepoints but, by contrast with ICON7, these differences were below the level regarded as important to patients or that would lead a clinician to consider a change in management. The investigators concluded that bevacizumab concurrently and as continuation did not seem to improve or impair QoL. This difference between groups reported in ICON7 but not GOG 218 might be because of differences in measurement or might suggest less difference in QoL between groups in GOG 218 than in ICON7. The use of placebo in GOG 218 led to an equal need for treatment delivery between the two groups, whereas the reported toxicities were not markedly different, raising the possibility that this difference between the two trial results is because of disruption to women's lives from delivery of continuation therapy.

Our study has some strengths. We studied a large sample, representative of the population of women making choices about postsurgical systemic treatment for ovarian cancer, enrolled at 263 international centres, with 10% having high-risk FIGO early stage disease but 70% stage IIIC or IV disease and one quarter having more than 1·0 cm of residual disease after surgical debulking. We present a large amount of QoL data with data collected according to a uniform schedule over multiple continents, languages, and centres. We believe that the collection of the reasons for missing data adds to the confidence in our findings. We have also undertaken a sensitivity analysis, using mean imputation of historical data from a previous trial of relapsed ovarian cancer to model the effect of the different proportions of patients with missing data within the two groups.

However, some study weaknesses should be considered. Our assessment did not include direct data on the women's experience beyond cancer progression. We decided to measure QoL in this way because of issues of patient burden, incomplete data on treatment after relapse, and study cost. The amount of missing data beyond disease progression was not balanced between the groups. Sensitivity analyses were undertaken to examine this issue, using mean imputation for missing data (rather than multivariate imputation taking into account potentially influential factors such as age and stage) at baseline and later, and imputation of clinically plausible QoL after progressive disease from other related trials do not alter our conclusions. ICON7 is an open-label study, which could affect QoL, although it also represents a realistic comparison of treatments with and without the addition of bevacizumab to standard chemotherapy. Based on animal models, there has been debate over the effect of antiangiogenic drugs on cancer invasiveness after resistance has developed. Clinical data so far have not supported the initial animal data. Because we have not captured QoL data beyond progression, we cannot assess whether the clinical benefits of giving bevacizumab add or detract from QoL after progression. Further post-progression data will be available at the next planned analysis when overall survival is formally reported, and we plan to present QoL data 3 years from randomisation for all patients still alive.

The addition of bevacizumab to standard combination chemotherapy in ovarian cancer prolonged progression-free survival. However, QoL seems to be impaired to a clinically small extent during bevacizumab continuation treatment, with no direct evidence that this was because of the specific hypotheses we studied. In future, as further drugs are developed with continuation schedules, the effect on QoL outcomes that are directly reported by patients will be an important consideration alongside improvements in the duration of cancer control.