Most patients present with incurable disease. For patients with advanced disease, median survival is 3–6 months,
with the majority requiring intervention for dysphagia.
Brachytherapy might represent an appropriate alternative
but is rarely accessible in UK National Health Service (NHS) settings,
and is unavailable in most low-income countries where the incidence of advanced oesophageal cancer is highest.
Although SEMS is widely implemented for first-line management of dysphagia in the UK, the efficacy of SEMS alone is limited by early problems with pain, a decline in general aspects of quality of life (QoL), and later complications such as haemorrhage and tumour overgrowth.
Median time to recurrent dysphagia in stent comparator
and brachytherapy studies
is 11–12 weeks and it has profound effects on independence, social function, and QoL. Hospitalisation and re-intervention account for most stent-related costs,
imposing a substantial burden on both NHS resources and a vulnerable population with a median overall survival of 3–6 months. In line with Cochrane review research recommendations,
combination of SEMS with other treatments might reduce costs and patient burden by reducing adverse events and re-interventions at a time when patients are approaching the last weeks of life.
The Radiotherapy after Oesophageal Cancer Stenting (ROCS) study was developed in response to a UK National Institute for Health Research (NIHR) call for research proposals into aspects of palliation, and aimed to address uncertainties in the evidence base for interventions combined with SEMS. External beam radiotherapy (EBRT) is rarely used in the UK as a monotherapy for rapid dysphagia relief, but its use in the immediate post-stent period has not been rigorously studied. This study addresses the efficacy of adjuvant EBRT compared with SEMS alone in reducing the risk of dysphagia deterioration, and improving QoL and patterns of service use in patients with advanced oesophageal cancer.
Study design and participants
The ROCS study was designed as a multicentre, parallel-arm, open-label, phase 3 randomised controlled trial with an internal pilot phase examining recruitment. It was done in cancer centres and acute care hospitals across Scotland, England, and Wales.
A summary of changes to the original ROCS protocol are provided in the appendix (p 2). The final trial protocol is available online. Ethics approval for the study was given by the Wales Research Ethics Committee 2 in October, 2012 (reference 12/WA/0230).
unsuitable for radical treatment (oesophagectomy or radical chemoradiotherapy) because of patient choice or medical reasons, and with the ability to provide written informed consent. We excluded patients planned to receive endoscopic treatment of the tumour, other than dilatation, in the peri-stent period (except for required emergency interventions), those with a tumour length of greater than 12 cm (or tumour growth within 2 cm of the upper oesophageal sphincter), those with presence of a tracheo-oesophageal fistula, or pacemaker in the proposed radiotherapy field, those who had previous radiotherapy to the area of the proposed radiotherapy field, and those who were pregnant. Patients in whom brachytherapy or EBRT was already planned after stent insertion were not included as we were concerned that further addition of trial radiotherapy might increase the risk of toxicity. The number of such patients was small (brachytherapy accounts for
). Patients were approached and randomly assigned either before or after stent insertion to allow pragmatic accommodation within the clinical pathway.
Randomisation and masking
Patients were randomly assigned 1:1 to receive EBRT (plus usual care) or usual care alone after stenting by the method of minimisation with a random element (80:20) via a central telephone randomisation system developed by, and based at, the Centre for Trials Research at Cardiff University (Cardiff, UK). Minimisation was stratified to ensure balanced treatment allocation by a number of potential confounding factors: treating centre, stage at diagnosis (I–III vs IV), histology (squamous or non-squamous), and MDT intent to give chemotherapy (yes or no). Participants were enrolled and assigned their trial group by the local principal investigator or research practitioner. The research practitioner was responsible for subsequent follow-up data collection. The study was necessarily open label and neither the patients nor the treating clinicians were blinded to treatment allocation. However, classification of some events was blinded, detailed herein.
SEMS insertion was done in the EBRT group and the usual care group as per standard procedures at each centre. Stent type and length were determined by the treating clinician. When possible, the stent length was chosen to ensure that at least 2 cm of normal oesophagus was covered by the stent above and below the tumour.
Usual care was implemented in both groups according to local MDT practice to include, as needed, post-stent dietetic advice, referral for palliative and supportive care interventions (eg, blood transfusion and supportive oncology), and community-based health-care and social-care follow-up.
In the EBRT group, the study protocol mandated that radiotherapy begin within 4 weeks of stent insertion and preferably 2 weeks. Treatment dose was prespecified at each centre, preferably 20 Gy in five fractions per day over 1 week or, at the treating clinician’s discretion, 30 Gy in ten fractions per day over 2 weeks. Treatment was administered according to each centre’s normal radiotherapy procedures without corrections for inhomogeneity in dose calculation. In the event of severe radiotherapy side-effects or treatment machine unavailability, gaps in treatment of up to 7 calendar days were allowed. If the patient missed more than 7 consecutive calendar days during radiotherapy treatment, then they were withdrawn from the trial and further treatment given at the clinician’s discretion. Radiotherapy quality assurance was monitored by the NIHR Radiotherapy Trial Quality Assurance Group.
and EuroQol 5D questionnaire 3 level (EQ-5D-3L).
Data were also collected on WHO performance status, stent complications, toxicities (as per the National Cancer Institute Common Terminology Criteria for Adverse Events [CTCAE] version 4.03), other treatments, and health-care and social-care resource use. If a home visit was not possible, or if patients preferred, data were collected by phone. Additionally, phone calls every 4 weeks were introduced midway between home visits to maximise capture of the primary outcome data only. Serious adverse events were collected in real time via a designated contact service from time of informed consent until 60 days after stent insertion.
Full results of this qualitative study will be reported separately.
in patient-reported dysphagia score on the EORTC QLQ-OG25, with the first being taken as the event timepoint (consecutive deteriorations were specified because patients undergoing radiotherapy might temporarily show worsening of dysphagia secondary to radiation-induced oesophagitis); one deterioration and no more data possible (patient withdrew completely or died before next visit); one deterioration and patient missing on the next visit, with patient withdrawal or death within 4 weeks of the missed visit; additional dysphagia-related primary events consistent with the relevant change in dysphagia score (additional stent insertion, hospital admission for dysphagia, overgrowth or undergrowth of the stent, grade ≥3 dysphagia [CTCAE v4.03], or additional radiotherapy to the oesophagus or stent region; assessed and confirmed by the chief investigators [DA and AB] as tumour related and reviewed by an independent gastroenterologist, all masked to treatment group, as a dysphagia-related event); or death from any cause. In patients showing deterioration at one assessment but with missing data at a subsequent assessment, deterioration was timed at the previous assessment.
Secondary outcomes were overall survival, QoL (including WHO performance status), morbidity (upper gastrointestinal-related bleeding event or hospital admission for a bleeding event, first dysphagia-related stent complication, or re-intervention), dysphagia deterioration-free survival (DDFS), post-stent chemotherapy or additional radiotherapy, patient experience (to be reported in detail elsewhere), and cost effectiveness. Overall survival was calculated from the date of stent insertion to the date of death from any cause. QoL was measured with the EORTC QLQ-C30, EORTC QLQ-OG25, and EQ-5D-3L. The prespecified main patient-reported outcome items were the global health score from the QLQ-C30 and four scales from the EORTC QLQ-OG25: odynophagia, pain or discomfort, eating restrictions, and eating in front of others. Upper gastrointestinal bleeding events were confirmed by the chief investigators who were masked to the study group and reviewed by a masked independent gastroenterologist. These events could include blood transfusion, haematemesis, other descriptions of upper gastrointestinal haemorrhage or bleeds, or interventions related to bleeding (such as argon plasma coagulation or additional radiotherapy). If there was no clinical evidence that anaemia was due to a bleed then it was not considered. Data on treatment with antiplatelet drugs, anticoagulants, and nonsteroidal anti-inflammatory drugs other than aspirin were collected at each clinical assessment visit. Stent complications were defined as re-stenting, repeat endoscopy, overgrowth or undergrowth of the stent, stent blockage, stent fracture, stent slippage, and stent-related pain (grade ≥2 on the CTCAE v4.03). Dysphagia-related stent events (blindly assessed) only influenced the primary outcome if an event had not already been identified in the patient-reported OG25 assessments. Re-interventions (dyphagia-related or not related) were defined as additional stent insertion, stent removal, endoscopic intervention (including laser therapy and alcohol injection), and other palliative radiotherapy (including brachytherapy and additional EBRT for dysphagia). For monitoring of safety, toxicity was measured throughout follow-up with the CTCAE v4.03.
The changes were approved by the independent trial steering committee and ratified by the funder following further independent review.
All statistical analyses followed a predefined statistical analysis plan agreed with the IDMC. Our modified intention-to-treat (ITT) population was defined as all patients who had a stent inserted (otherwise no benefit from radiotherapy was expected) and returned a baseline EORTC QLQ-OG25 (an eligibility criteria). The per-protocol (PP) population was defined as the subgroup of the modified ITT population that was alive and had not withdrawn from trial treatment at 4 weeks after stent insertion, and, in the EBRT arm, had received at least one fraction of radiotherapy to compare those who could have received radiotherapy in the usual care arm with those who did in the EBRT arm.
Analysis of the primary binary endpoint of deterioration in dysphagia symptoms by 12 weeks was primarily done in the modified ITT population with complete case data. Complete cases were defined as having complete data for the dysphagia subscale of the QLQ-OG25 questionnaire at baseline, week 4, week 8, and week 12, or having died with complete data before week 12. In the absence of a documented dysphagia-related event, missing dysphagia scores between two non-event dysphagia scores were assumed to be no event. Multivariable logistic regression was used to adjust for randomisation stratification factors and obtain odds ratios (ORs) and 95% CIs for any treatment effect in the primary analysis and all sensitivity analyses. We did three sensitivity analyses: using the same complete case population but treating death by 12 weeks without earlier deterioration as no deterioration; imputing missing data using a best-case scenario that assumed no deterioration in a missing QLQ-OG25 form immediately before an QLQ-OG25 form that showed deterioration (or a dysphagia-related primary event), or that assumed no deterioration in a missing QLQ-OG25 form immediately before death; and imputing missing data using a worst-case scenario that assumed deterioration in a missing QLQ-OG25 form immediately before an QLQ-OG25 form that showed deterioration (or a dysphagia-related primary event), or that assumed deterioration in a missing QLQ-OG25 form immediately before death. As further sensitivity analyses, all analyses were repeated in the PP population.
As a secondary endpoint per IDMC guidance, DDFS was calculated in the ITT population from the date of stent insertion to the date of deterioration in dysphagia (as per the primary outcome definition). We analysed overall survival and DDFS using Kaplan-Meier plots and Cox regression (with the usual care group as the reference for the treatment effect measured by hazard ratios [HRs] and 95% CIs), with patients without events being censored at the time of last contact and adjusted for randomisation stratification factors with treating centre included as a shared frailty. We tested the model fit and assumptions using Cox-Snell residuals and Schoenfeld’s global test.
QoL data and WHO performance status scores, prespecified in the statistical analysis plan, were analysed by the same method: the distributions of the variables were tested for normality with the Shapiro-Wilk test, kernel density, normal probability, and normal quantile plots, and either mean scores (or median scores if the was evidence of non-normality) plotted accordingly. Box plots were used to show the median, IQR, upper and lower adjacent values, and any outliers (per STATA 16 procedures) as dots, at each timepoint. Mean values were plotted with 95% CIs against time. Linear mixed models were used to compare differences between trial groups for each subscale or single item on the EORTC QLQ-C30, EORTC QLQ-OG25, and WHO performance status. We included time as a categorical covariate using the week of observation from week 1 to week 16, after which the proportion of missing data became too high (>30% of randomly assigned patients returning questionnaires). If an intermediate value was missing, the corresponding time was skipped. Covariates included trial group, age, time 0 score, and randomisation stratification factors. The mixed model residuals were tested for normality.
Time to first morbidity event was compared between trial groups by competing risks regression (used to calculate subhazard ratios and 95% CIs), with death as a competing risk, adjusted for randomisation stratification factors, and with cumulative incidence functions plotted by trial group and median time to event calculated with the stci command in STATA. Treatment-emergent grade 3–4 toxicity was reported in the modified ITT population. Risk ratios were calculated in a post-hoc analysis to compare rates of toxicities and post-stent chemotherapy or additional radiotherapy between treatment arms.
Costs were applied by calculating number of patients in each stage of the model and multiplying by mean cost for that stage. In the base-case analysis, patient-level data was used to populate the model with the first 12 weeks of data. The health-care costs in primary, secondary and social care for both EBRT and control groups post-randomisation were summated and mean absolute cost difference per patient (including 95% CIs and p values) were calculated with SPSS (version 26). Independent sample t-tests were used for comparisons with a 5% significance level. For the 12-month time horizon, the model structure remained the same. However, costs, utilities, and transition probabilities between the health states of stable or worsening dysphagia were updated to include the data from 13–52 weeks. Where data were missing, mean patient-level interpolation was used.
Sensitivity analyses were undertaken to test the robustness of the cost utility analysis considering the uncertainty in input parameters such as costs and outcomes and in different scenarios. In a deterministic, univariate sensitivity analysis, we changed intervention and health-care costs and outcomes individually within plausible ranges (using 10%, 20%, and 30% of the mean value). Scenario analyses were used to test different assumptions and recalculate the incremental cost per QALY gained (eg, based on different populations: complete cases and all available cases). The time horizon was also extended to 12 months to explore longer term effects of the intervention. In a probabilistic sensitivity analysis, we used non-parametric bootstrapping to address joint parameter uncertainty and assess the effect on the incremental cost during 1000 simulations, undertaken with random sampling from distributions of costs and outcomes (with replacement).
In all analyses, a p value of less than 0·05 was considered to indicate significance. All statistical analyses were done with STATA 16. This study is registered as an International Standard Randomised Controlled Trial, ISRCTN12376468, and with ClinicalTrials.gov, NCT01915693.
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.
Table 1Baseline demographic and clinical characteristics in the modified ITT population
Data are n (%), median (IQR); n, or mean (SD); n. ITT=intention to treat. GOJ=gastric-oesophageal junction. EQ5D=EuroQol-5D-3L.
Table 2Patient status and primary endpoint analyses at 12 weeks after stent insertion in the modified ITT population
Data are n (%) unless otherwise specified. NA=not applicable.
Table 3Upper gastrointestinal-related bleeding events
Data are n (%). NSAID=non-steroidal anti-inflammatory drug.
Table 4Participants with grade 3–4 toxicity at weeks 0–16 after stent insertion
In our qualitative study, participants in the EBRT group described whether potential benefits of radiotherapy were worthwhile against additional burdens of hospital attendance and of pain and fatigue. Participants from both study groups revealed ongoing challenges with eating despite stent placement. They suggested that the technical intervention of stenting does not address physical and social eating concerns and symptoms. Information around diet, symptom control, and general medical management throughout the course of the disease was often scarce. Full results of this quantitative analysis will be reported elsewhere.
Table 5Cost of care resources used per patient in the 12 weeks after randomisation in the modified ITT population
Costs per patient are summarised as mean (SD).
Our results show that palliative radiotherapy does not reduce the proportion of patients with recurrent dysphagia at 12 weeks, nor does it improve overall survival or reduce service use. The addition of radiotherapy to stent insertion in advanced oesophageal cancer is therefore not routinely recommended. Patients in the radiotherapy group did have significantly fewer bleeding events, an effect which persisted and increased with time. From a clinical perspective, these findings suggest that radiotherapy could be considered for patients deemed at increased risk of bleeding rather than for all patients, to minimise treatment burden.
with use of a widely available intervention (EBRT) combined with stenting. The choice to assess palliative stent therapy was intended to reflect real-world UK practice, where SEMS is the predominant option for rapid dysphagia relief in advanced oesophageal cancer, accounting for more than 90% of endoscopic and radiological interventions.
ROCS’ target population therefore excluded patients with non-severe dysphagia being considered for interventions other than a stent, and those too unwell to have a stent. Although intraluminal brachytherapy might be considered an appropriate alternative to stenting particularly in patients with longer term survival prospects,
few services in the UK have access to brachytherapy, and it accounts for less than 2% of dysphagia interventions in the NHS.
Similarly, incorporation of endoluminal radiotherapy with a stent as a single modality has been shown to lower dysphagia scores with time compared with a stent alone,
but again, the equipment and expertise is not widely available and cost is likely to be substantially higher.
By contrast, palliative EBRT is widely available across the UK, and at lower cost. The prespecified dose in this study, 20 Gy in five fractions, reflected the most widely used palliative dose used across the UK at the time of study design,
with the 30 Gy in ten fractions dose available if prespecified by the treating clinician. The timing of the primary endpoint at 12 weeks reflects the mean stent patency reported by Homs and colleagues
and other studies,
and the median overall survival of around 19 weeks in both groups confirms that our participant population accurately reflects the wider clinical population.
The finding that almost twice as many patients in the control group received their preplanned chemotherapy is noteworthy. This might reflect treatment burden in the radiotherapy group discouraging planned chemotherapy uptake, or could reflect participant or clinician assessment that an active treatment had already been given in the form of radiotherapy.
and QoL trade-offs.
The results reflect the challenges of their lived experiences of eating restrictions, concerns over nutrition and diet, and a trial and error approach to combating these. These resonate with other findings
and emphasise the need for more structured, proactive multidisciplinary approaches in patients receiving palliative stent therapy.
Although to our knowledge, this study is the first of its kind and has been completed in a patient group with a poor prognosis, it does have some limitations. Originally the sample size calculation was based on a time-to-event analysis for the primary endpoint requiring 496 participants. Due to recruitment and data capture challenges associated with the poor prognosis of the study population, this approach was revised during the study, on advice of the IDMC, to a sample size calculation based on comparison of patient proportions with an event by week 12. Although the revised primary outcome might have affected the ability of the study to detect a true effect for EBRT, the consistency of the results across sensitivity analyses is robust, including the secondary analysis of DDFS. Whether death was treated as an event or not did not alter the primary outcome. In pragmatically allowing two radiotherapy schedules in the EBRT group, we did not aim to seek a difference between doses and the small number of patients receiving 30 Gy precludes any such analysis (64 [78%] of 97 received 20 Gy in five fractions). We also acknowledge the large number of QoL secondary endpoints assessed in this study and associated issues that arise with multiple comparisons, hence we urge caution in the overinterpretation of significant findings found amongst them. Finally, this study was inevitably open-label and some outcome measures could be prone to assessment bias, particularly the adverse events known to be side-effects of radiation. We attempted to mitigate this bias with the use of a comprehensive panel of secondary outcome measures and use of masked assessors to review bleeding events. The baseline QoL assessments (week 1 post-stent) were done after randomisation (and therefore might have been susceptible to bias), but we noted that most scores were balanced between groups at that timepoint.
In summary, the ROCS study confirms that patients with advanced oesophageal cancer requiring a stent to improve dysphagia will not benefit further from the addition of concurrent palliative radiotherapy. In addition, the study provides detailed data on the poor outcomes in these patients, which are rarely the focus of multicentre prospective research. For patients with a long-term prognosis and considered to have a markedly increased risk of tumour bleeding, concurrent palliative radiotherapy might reduce bleeding risk and the need for associated interventions. Future research should focus on alternative, readily accessible interventions that might be effectively combined with stenting or compared as a monotherapy, and effective, multidisciplinary, supportive interventions that address the multidimensional concerns around eating and nutritional intake.
DA, AB, LN, and JB conceived the idea for this study. DA, AB, AN, TC, JS, JB, JF, and GG were grant applicants (AB and DA were co-chief investigators). CH, CP, AB, and DA wrote the first draft of the manuscript. All authors were involved in acquisition of data and critical revision of the manuscript for important intellectual content. LN provided senior management of the study, overseeing the trial manager (MS) at the CTR. CH provided scientific leadership of the study at the Centre for Trials Research at Cardiff University. CP did the statistical analysis with oversight by CH. AN was responsible for the qualitative research. BS was responsible for the health economic study design and analysis. MJ designed and created the decision analytical model, under supervision of DF. JB was responsible for the quality of life study design. MS was responsible for trial management and contributed at all stages of study reporting. ST and JF were responsible for the patient and public perspective throughout study development and implementation and contributed to study reporting. AM and JS were responsible for radiotherapy quality assurance. CP and CH accessed and verified the data. All authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.
Declaration of interests DA reports grants from the UK National Institute for Health Research (NIHR), during the conduct of the study; grants from Roche and Boehringer Ingelheim, outside the submitted work; and has given advice to Roche on the development of multidisciplinary team software but has received no financial recompense for this. AB reports grants from the NIHR and Marie Curie, during the conduct of the study. DF reports being an active member of the European Organisation for Research and Treatment of Cancer Quality of Life Group. GG reports grants from Janssen-Cilag, Novartis, Astex, Roche, Heartflow, Bristol Myers Squibb, and BioNtech, grants and personal fees from AstraZeneca, and personal fees from Celldex, outside the submitted work. JS reports personal fees and non-financial support from Janssen Oncology, non-financial support from Bayer, and personal fees from Astellas, Novartis, and AstraZeneca, outside the submitted work. ST reports personal fees from the NIHR Evaluation, Trials and Studies Coordinating Centre Prioritisation Panel of the NIHR Health Technology Assessment Programme, outside the submitted work. All other authors declare no competing interests.
This project was funded by the NIHR Health Technology Assessment Programme, project number 10/50/49, and will be published in full in a Health Technology Assessment monograph. AB and AN are supported by a Marie Curie core programme grant (MCCC_FCO_17_C). The Centre for Trials Research at Cardiff University is funded by Cancer Research UK and Health and Care Research Wales. JB is supported by funding from the UK Medical Research Council (MRC) as director of the Collaboration and Innovation in Difficult and Complex Randomised Controlled Trials MRC Methodology Hub, and funding from the NIHR Bristol and Weston Biomedical Research Centre, and is an NIHR senior investigator. We thank all the patients who participated in this trial, and their families and carers. We are indebted to the principal investigators at each site for their dedication to identifying and recruiting patients: Carys Morgan, Andrew Bateman, Olivia Chan, Mathilda Cominos, Serena Hilman, Eleanor James, Danielle Power, Ashraf Rasheed, Angus Robinson, Martin Scott-Brown, Elizabeth Selvaduri, Paul Shaw, David Tsang, Ravi Vohra, Nick Wadd, Jonathan Wadsley, and Julie Walther. We thank current and former staff of Cardiff University for supporting the development and running of this trial, and members of the trial steering and independent monitoring committees.