Impact of video-assisted thoracoscopic lobectomy versus open lobectomy for lung cancer on recovery assessed using self-reported physical function: VIOLET RCT

Criteria for progression from Phase I to Phase II

During the first phase, processes for trial conduct, including recruitment and consent, were established. Progression from the internal pilot (Phase I) to the full trial (Phase II) was dependent on the following criteria being met when assessed 18 months after the start of recruitment:

• At least 60% of patients undergoing lobectomy are considered eligible for the trial (if necessary, by revising the eligibility criteria).

• At least 50% of patients consent to randomisation after 6 months of recruitment.

• Less than 5% of patients fail to receive their allocated treatment.

• Less than 5% of patients are lost to follow-up (excluding deaths).

In Phase II, the number of study sites was increased and all sites used the optimum methods of recruitment established in Phase I.

Changes to trial design after commencement of the trial

There were several substantial amendments made to the study protocol throughout the course of the trial. The changes are summarised below. The protocol version in use when the trial started was version 2.0. The current full trial protocol can be found in the National Institute for Health and Care Research (NIHR) Journals Library [URL: www.journalslibrary.nihr.ac.uk/programmes/hta/130403/#/ (accessed 27 October 2021)].

Patient population

Adults referred for lung resection for known or suspected lung cancer to one of the participating centres.

Patient eligibility criteria

Patients were eligible to enter the study if all the following applied:

• Aged ≥ 16 years.

• Undergoing lobectomy or bi-lobectomy for treatment of known or suspected primary lung cancer beyond a lobar orifice, or undergoing frozen section biopsy with the intention to proceed with lobectomy or bi-lobectomy if primary lung cancer beyond a lobar orifice is confirmed. (Note that for bi-lobectomy, the distance for the lobar orifice is in reference to the bronchus intermedius.)

• Cancer staging using TMN8:

  • clinical tumour stage 1–3 (cT1–3) [by size criteria, equivalent to TNM Classification of Malignant Tumours, Seventh Edition14 (TNM7) stage cT1a-2b] or cT3 (by virtue of two nodules in the same lobe)

  • node stage 0–1 (N0–1)

  • metastasis stage 0 (M0).

• Multidisciplinary team (MDT) consider the disease is suitable for both VATS lobectomy and lobectomy via open surgery.

• Ability to give written informed consent.

Patients were not eligible to enter the study if any of the following applied:

• Previous malignancy that influences life expectancy.

• Planned pneumonectomy, segmentectomy or non-anatomic resection (e.g. wedge resection).

• Serious concomitant disorder that would compromise patient safety during surgery.

• Planned robotic surgery.

Changes to trial eligibility criteria after commencement of the trial

In July 2015, planned segmentectomy was added as an exclusion.

In June 2017, the inclusion criteria were revised to allow for the inclusion of patients undergoing bi-lobectomy and to update the cancer staging from TNM7 to TMN8. The Trial Management Group (TMG) recommended widening the inclusion criteria to include patients scheduled for a bi-lobectomy (i.e. resection of two lobes rather than one), as this can be performed via VATS or open surgery and, therefore, the research question is equally applicable to patients having this procedure. The revisions to the TNM staging system necessitated a change to the eligibility criteria, which were widened to include patients with cT1–3 tumours (by size criteria, equivalent to TNM7 stage cT1a-2b) or cT3 by virtue of two nodules in the same lobe. The entry criteria for nodal and metastatic involvement remained unchanged at N0–1 and M0, respectively.

VATS lobectomy (experimental)

Surgeons were permitted to undertake the VATS lobectomy using between one and four keyhole incisions. The use of rib spreading was prohibited, as this intraoperative manoeuvre disrupts the intercostal nerves and is thought to be an important cause of pain (and is a key feature of open surgery). The procedure was to be performed with videoscopic visualisation without direct vision. The hilar structures (i.e. vein, artery and bronchus) were dissected, stapled and divided. Endoscopic ligation of pulmonary arterial branches was optional. The fissure was completed and the lobe of lung resected. The incisions were closed in layers and may have involved muscle, fat and skin layers. This definition of VATS lobectomy is a modification of CALGB 39802. 15

Open lobectomy (control)

Conventional open surgery was undertaken through a single incision. Rib spreading was mandated, but rib resection was optional. The operation was performed under direct vision, with isolation of the hilar structures (i.e. vein, artery and bronchus), which were dissected, ligated and divided in sequence, and the lobe of lung resected. Ligatures, over sewing or staplers could be used. The thoracotomy was closed in layers, starting from pericostal sutures over the ribs, muscle, fat and skin layers.

Aspects common to both groups

Primary outcome

The primary end point was self-reported physical function assessed using the European Organisation for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire Core 30 (QLQ-C30) at 5 weeks post randomisation. Physical function was chosen because it is a patient-centred outcome that would reflect the anticipated earlier recovery with VATS. It had also been used in other minimal access surgery trials. The 5-week primary end point (approximately 1-month post surgery) was chosen to capture the early benefits of minimal access surgery on recovery.

The EORTC QLQ-C30 is used to assess quality of life in cancer patients. The EORTC QLQ-C30 comprises 30 questions, which are used to derive an overall measure of global health, and a number of subscales, of which physical function is one. For all scales, higher scores indicate a higher level of functioning, symptoms or problems. The EORTC QLQ-C30 has been validated for use in European cohorts. Version 3 of the questionnaire was used.

Secondary outcomes

The following secondary outcomes were selected to assess the efficacy of the two approaches:

• Time from surgery to hospital discharge.

• Pain scores in the first 2 days post surgery.

• Adverse health events in the period from randomisation to 1 year.

• Uptake of adjuvant treatment (i.e. frequency and time from surgery).

• Frequency of upstaging to pathologic node stage 2 (pN2) disease after the procedure.

• Overall and disease-free survival to 1 year.

• Frequency of complete resection during the procedure.

• Frequency of prolonged incision pain (defined as the need of analgesia for > 5 weeks post randomisation).

• Generic and disease-specific patient-reported health-related quality of life (HRQoL) measured using the EORTC QLQ-C30, Quality of Life Questionnaire Lung Cancer 13 (QLQ-LC13) and EQ-5D-5L questionnaires completed at 2 weeks, 5 weeks, 3 months, 6 months and 1-year post randomisation.

• Resource use during the hospital stay and post discharge to 1 year after randomisation.

Changes to trial outcomes after commencement of the trial

In 2015, minor amendments were made to clarify some secondary outcomes prior to the study opening to recruitment (see Changes to trial design after commencement of the trial for details). In a further amendment in the same year, the collection of patient-reported pain scores at baseline and at 1 day and 2 days postoperatively was added to the table of assessments, but the list of secondary outcomes was inadvertently not updated in line with this addition. This error was corrected in 2019 when pain scores were added to the list of secondary outcomes in the protocol. Pain scores were added to allow further comparison between the two surgical techniques during the early postoperative period. Pain assessments are routinely undertaken by the clinical team to determine whether or not the analgesia provision is sufficient and so a patient verbal assessment of pain represented minimal additional burden to the patient.

Research team

The surgical team, anaesthetist and other staff caring for the participant during the operation were not blinded to the patients’ treatment allocation. However, to minimise the risk of bias in the assessment of outcomes, the randomisation was performed by a member of the research team who was not responsible for the collection of outcome data.

Participants

To ensure that study participants remained blinded during the postoperative period to discharge home, participants were asked to turn their head away from the wound site(s) while wounds were being cleaned and dressed. Participants were advised of how best to care for their wounds when they were considered ‘fit for discharge’. Those participants who asked to know which treatment they had received were informed at this point.

Assessment of blinding

The success of blinding was assessed using Bang et al. ’s Blinding Index. 17 Participants were asked to complete the assessment 2 days postoperatively and at discharge, but before the treatment allocation was revealed. The RNs responsible for data collection and follow-up of participants completed the Blinding Index when the patient was ready for discharge, and after the 5-week and 1-year follow-up appointments.

Overview

Data collection for the trial participants included the following elements:

• A log of patients screened by the MDT for suitability for the trial and the date when patients were given or sent the patient information leaflet (PIL).

• A log of patients assessed against the eligibility criteria and reason(s) if ineligible.

• Audio-recording and transcription of consultations between surgeons and potential participants (see QuinteT Recruitment Intervention for further details).

• Semistructured interviews with a sample of eligible patients, including patients who accept or decline to join the trial.

• Approach and consent details, including reason(s) for non-approach or decline.

• Baseline data, including the participant’s medical history, disease status and HRQoL prior to randomisation.

• Operative details.

• Histopathology of any samples (e.g. biopsies) taken intraoperatively.

• Postoperative care, including analgesia and pain scores.

• AEs and resource use in the period from randomisation to 1 year.

• HRQoL during follow-up.

• Results of scans taken to assess disease status.

An overview of the schedule of data collection is given in Table 1.

Collection of health-related quality-of-life data

Health-related quality-of-life data were collected on paper or online, according to participant preference. If data were not returned, the participant may have been telephoned by the local RN and the data collected over the telephone.

Collection of adverse event data

Serious adverse events and other AEs were recorded and reported in accordance with Good Clinical Practice guidelines. Data were collected from the time of consent until 1-year post randomisation. Events were graded in severity using the CTCAE, which is a standardised classification system used in cancer studies.

As lung resection surgery is a major surgical intervention, events related to the surgery were considered ‘expected’. Many participants would go on to receive adjuvant chemotherapy or radiotherapy after their lung resection surgery. Such treatments have a range of common serious side effects and toxicities, which were also considered ‘expected’ for participants undergoing adjuvant chemotherapy and/or radiotherapy. These expected events were listed in the study protocol. Events that occurred that were not listed in the protocol were considered unexpected.

Safety data were reviewed regularly by the study team and at least annually by the Data Monitoring and Safety Committee (DMSC). Reporting to the sponsor was required only if an AE was considered serious (i.e. resulted in a hospital admission, prolonged a hospital admission, was life-threatening, resulted in significant disability or death) and unexpected or expected and fatal. Reporting to the Research Ethics Committee and DMSC was required if an unexpected SAE was found to be causally related to the intervention.

Overview and aims

A QRI19,20 was integrated throughout the recruitment period of the VIOLET trial because of anticipated recruitment challenges1 arising from the nature of the trial interventions [i.e. different approaches to undertaking lung resection (lobectomy) via VATS or open surgery]. The QRI methods, first developed in the ProtecT (Prostate testing for cancer and Treatment) study,21,22 have been refined and applied in nearly 70 RCTs, including other surgical RCTs. 23–25

The aim of the QRI in the VIOLET trial was to optimise and sustain recruitment and informed consent by preventing recruitment difficulties from arising, identifying new challenges as they arose and addressing those that did arise rapidly. The VIOLET trial QRI began with QRI-informed recruitment training workshops aimed at helping recruiters prepare for impending recruitment activities and preventing the development of recruitment barriers. Next, when recruitment commenced, we employed established QRI methods, which comprised two iterative components19,20 aimed at (1) understanding the recruitment issues in real time and identifying the clear obstacles and hidden challenges to recruitment,26,27 and (2) developing and implementing a plan of action comprising strategies28–31 to overcome the challenges, in collaboration with the chief investigator, TMG, Bristol Trials Centre and the recruiting sites. Evaluation of the QRI was carried out throughout the recruitment period by regularly monitoring recruitment figures [using the screened, eligible, approached, randomised (SEAR) framework]32 and recruitment practice. Each of these methods is described in detail below (for further context regarding the evolution of the QRI, see Appendix 1).

Preventing recruitment difficulties: training and guidance prior to recruitment

We aimed to prevent recruitment difficulties in the VIOLET trial by disseminating strategies to optimise recruitment and informed consent to recruiting surgeons, drawing from the QRI evidence base and using multiple avenues. These activities occurred prior to recruitment and involved study-wide, as well as site-specific, activities.

At the trial set-up phase, PIs and other recruiting surgeons from the five sites in the internal pilot phase were invited to attend a 1-day QRI-informed recruitment training workshop. 29 Surgeons from the new centres were invited to subsequent workshops. The evidence-based training was aimed at raising awareness of, and providing practical tips to manage, the clear obstacles27 (e.g. logistical issues) and hidden challenges to recruitment (e.g. conveying equipoise, addressing patient preferences). 26–28,30,33 The evidence that informed the above workshops was also presented to each site during site initiation visits (SIVs), summarised in a brief tips document circulated to all internal pilot sites and used to identify aspects of patient- and recruiter-facing study documentation (e.g. information leaflets, consent forms and trial protocol) that were potentially unclear, imbalanced or open to misinterpretation.

Understanding recruitment issues

We employed a range of methods, primarily qualitative, but also drawing on descriptive quantitative data obtained from the SEAR logs, to understand the recruitment processes in the VIOLET trial and to identify the challenges to optimal recruitment.

Plan of action: strategies to optimise recruitment and informed consent

Initial anonymised QRI findings that identified factors appearing to hinder recruitment were presented to the chief investigator/TMG (in February 2016, 8 months after recruitment to the VIOLET trial had commenced) so that they could agree on a plan of action to address factors impeding recruitment. This plan was implemented through the first set of group and individual feedback sessions covering all the internal pilot phase sites (May–June 2016), with the aim of optimising recruitment and informed consent. As recruitment progressed and the trial moved into the main phase, there was further QRI data collection, analysis and presentation of findings at TMG and investigator meetings and at VIOLET trial sites, which aimed to sustain the recruitment momentum gained early in the trial. The key findings were also summarised in succinct ‘tips’ documents circulated to all centres, especially during periods of low recruitment activity. Towards the end of the recruitment period, the QRI team, in collaboration with the chief investigator/TMG, developed a rapid communication strategy, which aimed to reiterate the most important QRI findings and strategies and disseminate them through monthly newsletters developed by the trials centre and e-mail communications with infographics on recruitment figures.

Evaluating the impact of the QRI

We evaluated the impact of the QRI in two ways through (1) monthly monitoring of recruitment figures (i.e. SEAR data) and (2) recruitment practice. First, SEAR data were monitored from the onset of recruitment until the achievement of the final recruitment target to check if recruitment commenced well following the training/guidance provided prior to recruitment, if any recruitment momentum gained was sustained thereafter and if there were periods of low recruitment activity that improved after rapid intervention. Second, we monitored recruitment practice by listening to the audio-recordings (and interviews where relevant) to document changes in practice following the provision of trial- and site-specific feedback and instances where recruitment challenges appeared to have been averted following the training/guidance prior to recruitment. A conventional pre- and post-intervention evaluation was not feasible in the VIOLET trial QRI, as it did not have a precise pre-QRI period of recruitment because of the preventative activities undertaken in advance of recruitment.

Summary statistics and analysis population

Data were described using summary statistics, using mean and SD for continuous variables [or median and interquartile range (IQR) if distributions were skewed] and number and percentage for categorical variables. HRQoL questionnaires were scored in accordance with the developer’s scoring instructions and summary scales derived from the questionnaires are reported in summary tables.

Participants were grouped according to the randomised allocation (intention to treat). The analysis population consisted of all randomised participants, excluding those who withdrew and were unwilling for data already collected to be used. Data from any participant who withdrew and was unwilling for their data to be used were included in the study flow chart, but not in any subsequent data tables or figures.

Models used to compare primary and secondary outcomes

The models used to compare longitudinal HRQoL outcomes, including the primary outcome, are presented in Table 2. The adequacy of a model fit was assessed graphically.

The strategy for modelling longitudinal HRQoL outcomes, listed in order of preference, is as follows:

• Joint longitudinal survival model.

• Linear mixed-effects model (chosen if the joint longitudinal survival model did not provide an adequate fit to the data).

• Mixed-effects ordinal logistic regression (chosen if the HRQoL score could take only four possible values, the proportional odds assumptions held and the previous models were not appropriate) or partial proportional odds model (chosen if proportional odds assumption did not hold for all variables to be included in the model).

• Mixed-effects logistic regression (chosen if the ordinal logistic regression model did not converge or the proportional odds assumption did not hold for time or treatment variables).

If the distribution of the HRQoL score was non-monotonic, with a high proportion of participants scoring perfect functioning/health, a two-part model was used. Scores were dichotomised into perfect functioning/health (i.e. a QLQ-C30 score of 100 and an EQ-5D-5L utility score of 1) and less than perfect functioning/health. The first part of the model was an occurrence model, that is a mixed-effects logistic regression model comparing perfect functioning/health with less than perfect functioning/health. The second part was an intensity model, that is a log-linear mixed-effects model for the score, conditional on a less than perfect functioning/health score.

Time-by-treatment interactions were added to all longitudinal models. Overall treatment effects are presented unless the interaction reached 10% statistical significance, in which case treatment effects for each time point are reported. For the primary outcome, the treatment effect at 5 weeks is reported, estimated from the longitudinal model with a time-by-treatment interaction included.

Time-to-event outcomes were compared using Cox proportional hazards models and treatment estimates are presented as hazard ratios (HRs). Time to uptake of adjuvant treatment was analysed using competing-risks regression, with death modelled as a competing risk. Overall survival, progression-free survival (with progression defined as progression of lung cancer or new primary lung cancer or death from any cause, whichever occurred first) and uptake of adjuvant treatment were censored at last follow-up for those who had not experienced the event (or competing risk). For duration of hospital stay, in-hospital deaths were censored at the maximum observed time to discharge for survivors. The exact partial-likelihood method was used to account for tied times. Model assumptions were assessed graphically.

In-hospital pain scores were analysed using a linear mixed-effects model. Binary outcomes [i.e. complete R(0) resection and prolonged incision pain] were analysed using generalised linear models and multinomial outcomes (i.e. upstaging to pathologic node stage 1 (pN1) disease and upstaging to pN2 disease) using generalised structural equation models. Effect estimates are presented as relative risk (RR).

The mean daily dose of each analgesic, averaged over the hospital stay, was derived for each participant. Analgesic agents were combined into groups, where appropriate. The mean ratios were derived for each analgesic group and 95% confidence intervals (CIs) were estimated using bootstrapping.

Open surgery was the reference group in all analyses. Treatment estimates are reported with 95% CIs.

Adjustment in models

The plan was to adjust all models for centre and operating surgeon fitted as random effects (or as stratification variables in time-to-event outcomes). For binary outcomes, a clustered sandwich estimator was used to account for the clustering within surgeon, as random effects were not estimable. All longitudinal analyses were adjusted for baseline preoperative score as a fixed effect.

Subgroup analysis

A prespecified subgroup analysis, comparing in-hospital pain scores by type of analgesia (e.g. paravertebral block, intercostal block, both, neither) was performed. This was implemented by adding a treatment-by-analgesia interaction term into the model, comparing pain scores between groups.

Sensitivity analyses

A sensitivity analysis of the primary outcome, excluding participants with benign disease, was prespecified in the protocol. Sensitivity analyses of overall survival and progression-free survival were also performed, adjusting the model for participant’s disease stage based on pathological findings. These analyses were not in the protocol, but were added to the SAP on recommendation from the DMSC.

Exploratory analyses

Two exploratory analyses of pain scores were undertaken. The first was stated in the protocol and the second was requested by the DMSC:

1. An exploratory analysis comparing in-hospital pain scores by number of incisions (i.e. VATS with a single port site, VATS with multiple port sites and open surgery).

2. An exploratory analysis comparing in-hospital pain scores and QLQ-C30 pain scores by type of thoracotomy (i.e. anterior thoracotomy vs. posterolateral thoracotomy, muscle sparing vs. no muscle sparing, and rib resection vs. no rib resection).

An exploratory analysis comparing length of stay by incisions (i.e. single-port VATS vs. multiport VATS vs. open surgery) was not prespecified in the protocol, but was requested by the chief investigator before any comparative analyses had been undertaken.

Missing data

Missing data are described in footnotes to all tables. Rules for imputing missing data outlined in the SAP were dependent on the level of missing data. All HRQoL analyses and the in-hospital visual analogue scale (VAS) pain score analysis met the threshold for multiple imputation. For other outcomes, participants with missing data were excluded. For HRQoL analyses, each subscale was imputed separately. Imputation by chained equations was used to generate multiple complete data sets [using the Stata^® version 16.1 (StataCorp LP, College Station, TX, USA) -ice- command] and results were combined using Rubin’s rules. Factors included in the imputation models used were baseline HRQoL score, treatment allocation, centre and operating surgeon.

Significance levels and adjustment for multiplicity

For hypothesis tests, two-tailed p-values of < 0.05 were considered statistically significant. Likelihood ratio tests were used in preference to Wald tests. For HRQoL outcomes derived from the QLQ-C30 and QLQ-LC13 questionnaires, significance levels were adjusted for multiplicity using the false discovery rate method proposed by Benjamini and Hochberg. 37 The adjustment was applied within each instrument (e.g. for QLQ-C30 functional scale scores, QLQ-C30 symptom scale scores and QLQ-LC13 scores). No formal adjustment for multiplicity was made for other outcomes. Formal statistical comparisons were not made for outcomes with low-event rates and only prespecified subgroup analyses were performed. The number of statistical tests performed should be considered when interpreting results.

All statistical analyses were performed with the use of Stata software.

Economic evaluation aim

The economic evaluation aimed to estimate the incremental cost-effectiveness of VATS lobectomy compared with open lobectomy for the treatment of lung cancer, in line with the VIOLET trial.

Economic evaluation overview

The perspective of the evaluation was that of the UK NHS and Personal Social Services, as recommended by the National Institute for Health and Care Excellence (NICE). 38 The perspective for outcomes comprised the patients undergoing treatment. The primary outcome measure for the cost-effectiveness analysis was quality-adjusted life-years (QALYs), estimated using the EQ-5D-5L. 39,40 Established guidelines on the conduct of economic evaluations set out by NICE were followed. 38 Table 3 summarises the key aspects of the economic evaluation, and further details are provided below.

Form of analysis, primary outcome and cost-effectiveness decision rules

As advocated by NICE, a cost-effectiveness analysis (specifically a cost–utility analysis) was conducted, using QALYs as the primary outcome measure. 38 QALYs combine both quantity and quality of life into a single measure. Incremental costs (i.e. the difference in mean costs between the VATS and open lobectomy groups) were divided by incremental QALYs (i.e. the difference in mean QALYs between the groups) and presented as the incremental cost-effectiveness ratio (ICER), which quantifies the incremental cost per QALY gained by switching from using open surgery to VATS lobectomy. The economic evaluation analyses were performed on an intention-to-treat basis.

Video-assisted thoracoscopic surgery lobectomy was considered cost-effective if the ICER fell below £20,000, which is generally considered as the threshold that NICE adopts for considering an intervention to be cost-effective. 41

Time horizon

A within-trial analysis, taking a 1-year time horizon, was conducted. It was anticipated that all major resource use (i.e. surgery, complications relating to surgery and adjuvant therapy) would occur within this time frame and would, therefore, be captured.

The starting point for our analysis was from the point of surgery, rather than the point of randomisation, as was the case with the trial effectiveness analysis. Randomisation was performed within 1 week of the planned operation date. The time point for baseline costs and outcomes was not quite the same. The EQ-5D-5L data were collected preoperatively, whereas detailed resource use data collection began on the day of surgery. However, as no difference in resource use was expected between the groups until the time of surgery, we did not anticipate this being an issue. Our time horizon continued until 1-year postoperatively.

Collection of resource use and costs

Resource use data were collected on all significant health service resource inputs for the trial participants to the end of the 1-year follow-up period. Collection of detailed resource use data was integrated into the trial case report forms (CRFs) for the index admission, and captured from telephone calls with participants at 5 weeks, 3 months, 6 months and 1-year post randomisation. The main resource use categories that were captured and costed were initial thoracic surgery, hospital stay post surgery (by ward type), complications, adjuvant therapy, imaging, recurrence/progression of cancer, hospital readmissions, outpatient and emergency department (ED) attendances, and community health and social care contacts. Appendix 2, Table 25, details the sources of unit cost information for each of these categories. Costing decisions (e.g. resource use assumed for complications) were made without knowledge of the allocation of participants to trial groups.

Attaching unit costs to resource use

Unit costs for hospital and community health-care resource use were largely obtained from national sources, for example the National Schedule of Reference Costs42,43 for ward costs, scans and many complications, and Unit Costs of Health and Social Care44 for community costs. Resources were valued in 2018/19 GBP and any unit costs not in 2018/19 prices have been adjusted to 2018/19 prices using the NHS cost inflation index. 45 Where available, costs of drugs given in hospital were taken from the electronic marketing information tool (eMIT),46 which provides the reduced prices paid for generic drugs in hospital. Otherwise, costs were obtained from the British National Formulary. 47 For a summary of the sources of unit cost information, see Appendix 2, Table 25. For further details on all unit costs and their source, see Appendix 3, Tables 28–39.

Measurement of health-related quality of life and quality-adjusted life-years

Missing data

We first summarised the number of missing data for resource use and outcomes (EQ-5D scores) descriptively. Exploratory analyses were conducted to explore the possible mechanisms and patterns of missing data. 51 Logistic regressions were used to explore associations between missingness and baseline variables, and missingness and previously observed outcomes. If the number of missing data was small (< 1% of cases), then unconditional or conditional mean imputation was considered to be sufficient. However, we anticipated that it would be necessary to use multiple imputation to impute missing values. Multiple imputation is a flexible approach, which is valid if data are assumed to be missing at random (i.e. the probability that data are missing does not depend on the unobserved values; it is conditional on the observed data). 51,52 This assumption was assessed.

Multiple imputation uses regression to predict m values for each missing data cell, and enables all key variables used in the economic evaluation and demographic data (i.e. both complete and incomplete) to be used to predict the values of missing data cells. In accordance with guidelines,51,53 multiple imputation using chained equations was conducted and the number of imputations set to be at least equal to the percentage of incomplete cases. 53 Multiple imputation was performed separately for each treatment group.

Multiple imputation can be conducted at an aggregated level of total costs, for example, or at a disaggregated level of individual resource use items or EQ-5D domains. Given that imputing large numbers of variables may make the model difficult to estimate, a balance between the two is likely to be required. The patterns of missing data for resource use/costs and outcomes were used to determine the approach to multiple imputation. For example, data collected on a patient follow-up questionnaire may have similar patterns of missing data, in which case the total costs for that follow-up can be imputed rather than individual resource use items. For each variable with missing data, individual regressions were specified and tailored to the type of data being predicted. Linear regression with prediction mean matching was used, as it is particularly flexible.

Once multiple imputation had been conducted, tabulations and summaries of the observed and imputed data were compared to check the validity of the imputations. Rubin’s rule was then used to summarise data across the m data sets. 54 This approach accounts for the variability both within and between imputed data sets and takes uncertainty in the estimated mean into account.

Adjustment for baseline utility

Given that baseline utility directly contributes to QALY calculations, it is important to control for any potential imbalances in baseline utility in the estimation of the mean difference (MD) in QALYs between treatment groups to avoid introducing bias. 55 Regression adjustment also allows for regression to the mean and increases precision. Therefore, if there is an imbalance at baseline, we planned to adjust QALYs for baseline EQ-5D.

Within-trial statistical analysis of cost-effectiveness results

Analyses were conducted in Stata version 15 and Microsoft Excel^® 2016 (Microsoft Corporation, Redmond, WA, USA).

Initially, descriptive summaries of resource use, costs and HRQoL were performed using means, SDs and standard errors around the means using the central limit theorem. Cost data are typically positively skewed; however, regardless of this, costs were summarised using the arithmetic mean, as it is this combined with the total number of patients that relates to the total budget impact of an intervention.

The ICER was derived from the average costs and QALYs gained in each trial group, producing an incremental cost per QALY gained of VATS lobectomy compared with open surgery. Non-parametric bootstrapping of costs and QALYs was used to quantify the degree of uncertainty around the ICER. Results are expressed in terms of a cost-effectiveness acceptability curve (CEAC), which indicates the likelihood that VATS lobectomy is cost-effective for different levels of willingness to pay for health gain. Although VATS lobectomy is considered cost-effective if the ICER falls below £20,000, the ICERs and CEACs presented allow decision-makers to assess cost-effectiveness at a willingness-to-pay threshold of their choice.

Discounting

Costs and effects were not discounted, as our time horizon was 12 months.

Sensitivity analysis

Univariate sensitivity analyses were used to investigate the impact on costs and cost-effectiveness results of variation in key parameters and major cost drivers, and to investigate the impact of uncertainty on the cost-effectiveness results.

Factors examined in the sensitivity analyses for costing were varying the unit costs for key cost drivers, including surgery and ward stays. The impact of any high-cost participants was also investigated. For outcomes, we examined not adjusting for baseline utility and the impact of any missing survival status.

For details of all sensitivity analyses, see Appendix 4.

Subgroup analysis

No subgroup analyses were pre-planned for the cost-effectiveness analyses; comparing pain scores by type of analgesia, as per the clinical analyses, would not be meaningful here.

Phase I: study sites and dates opened to recruitment

• Royal Brompton Hospital and Royal Brompton and Harefield NHS Foundation Trust (29 July 2015).

• Liverpool Heart and Chest Hospital NHS Foundation Trust (6 October 2015).

• Bristol Royal Infirmary and University Hospitals Bristol NHS Foundation Trust (14 October 2015).

• The James Cook University Hospital and South Tees Hospitals NHS Foundation Trust (29 October 2015).

• Harefield Hospital and Royal Brompton and Harefield NHS Foundation Trust (18 December 2015).

Phase II: study sites and dates opened to recruitment

• John Radcliffe Hospital and Oxford University Hospitals NHS Foundation Trust (25 September 2017).

• Castle Hill Hospital and Hull University Teaching Hospitals NHS Trust (12 October 2017).

• Birmingham Heartlands Hospital and University Hospitals Birmingham NHS Foundation Trust (28 September 2017).

• Royal Infirmary of Edinburgh, NHS Lothian (20 September 2018).

Recruitment rate

When the study was designed, the estimated recruitment rate was expressed in terms of the proportion of eligible patients recruited, rather than as a recruitment per site per month. The proposed study sites were asked to estimate the number of lobectomies performed for early-stage lung cancer each year and from this the anticipated recruitment rate was derived, allowing for a staggered opening of study sites. It was estimated that 60% of patients would be eligible for the trial and that the participating surgical teams would initially recruit 30% of eligible patients, but that with training and feedback provided by the QRI team this might increase to 50% after 6 months. The anticipated recruitment rate for each of the participating centres is documented in Appendix 5, Table 45.

The actual recruitment rate is illustrated in Figure 3. The trial was assessed for progression 18 months after the start of recruitment. Recruitment was ahead of target at that time and remained so throughout the trial, completing 2 months ahead of schedule. The trial over-recruited by five participants. The number of patients recruited by site and the rate per month at each site is given in Appendix 5, Table 46.

FIGURE 3.

Predicted and actual recruitment.

Progression from Phase I to Phase II

Progress against the predefined progression criteria was assessed in December 2016. Performance against the prespecified criteria is shown in Appendix 5, Table 47. The Trial Steering Committee (TSC) recommended progression to Phase II. Following review of the screening data and reasons for ineligibility, the eligibility criteria were widened to include bi-lobectomies to increase the generalisability of the trial results.

Eligibility and surgery

Overall, the number of protocol deviations were low at 13% (66/502) (Table 5). Forty-nine patients did not undergo a lobectomy (31 patients were found to have benign disease on frozen section, 11 patients underwent a wedge resection, three patients underwent a segmentectomy, two patients were found to have extensive malignancy and so no resection was performed, and two patients underwent a pneumonectomy). A further 17 participants who underwent a lobectomy received the other trial intervention to that they were allocated [15 participants randomised to VATS received open surgery (one participant decided preoperatively to have open surgery and the remaining 14 participants were intraoperative conversions necessitated by the surgeon) and two participants randomised to open surgery had a VATS procedure (both participants decided preoperatively to have VATS)]. Reasons for conversions from VATS to open surgery can be found in Table 6.

Adherence to the mandated and prohibited aspects of the surgical procedure (i.e. number of ports used for a VATS procedure and use of rib spreading) and to the criteria for fitness for discharge are presented in Baseline data and operative characteristics and Chapter 5, Exploratory analysis: length of stay, respectively.

Success of blinding

Bang et al. ’s Blinding Index13 asks for individuals to guess which treatment (i.e. method of surgical access) the participant received. Results of the assessment of blinding of participants and RNs responsible for outcome data collection are presented in Table 7. At day 2, 159 of 442 (36.0%) participants correctly identified the surgery they had received. At discharge, 202 of 417 (48.4%) participants correctly identified the surgery they had received. A greater proportion of RNs correctly identified the surgical approach at discharge (275/440, 62.5%). This had reduced by 12 months (208/414, 50.2%).

Patient follow-up

The number of patients for whom follow-up data were available is presented in Figure 2. Follow-up data at 1 year was available for 81% of randomised participants. Of the 50 patients not followed to 1 year, 19 had withdrawn and 31 had died.

Data set

Our data set comprised interview data from 15 staff (recruiting surgeons, n = 11; RNs, n = 3; TMG members, n = 1) involved in VIOLET trial oversight and recruitment. A total of 451 individual patients’ audio-recordings were available from six recruitment sites and a purposively selected sample of 304 individual patient recordings were analysed in detail.

Patient pathways through eligibility and recruitment

A description of the standard patient pathway in VIOLET trial sites is presented, followed by pathway-related recruitment issues and recruiter perceptions regarding eligibility criteria in VIOLET (perceived to be clear and unambiguous, with some reservations expressed for specific groups of patients).

Patient pathways

The pathway for potentially eligible patients for the VIOLET trial was simple and consistent across centres. Recruiters felt that the VIOLET trial was easily integrated into standard clinical practice. Although there was some site variation, patients generally underwent similar processes to those illustrated in Figure 5.

FIGURE 5.

Patient pathway from referral for treatment to decision about trial participation.

Patients referred to secondary care centres received diagnostic tests and their results were discussed in MDTs. PIs screened these patients and confirmed eligibility of potential trial participants. Approaches to invite patients to take part in research occurred during patient consultation with surgeons who would perform the surgery. Surgeons were the primary information providers and recruiters in the VIOLET trial. RNs facilitated study-related paperwork, such as written consent for audio-recording of consultations, trial baseline assessments and follow-up questionnaires.

Recruiter equipoise

In the interviews, recruiters expressed strong enthusiasm for the VIOLET trial, yet there were a number of discernible instances of recruiter biases evident in the interviews, usually in favour of VATS. In the consultations, many recruiters were adept at withholding their personal biases and carried out balanced information provision on the two operations, with an emphasis on uncertainty. There were some instances where these biases were unwittingly conveyed to patients. In addition, recruiter bias towards open lobectomy was observed in some consultations as the trial progressed (mainly from one centre). Some of these findings are detailed below.

Patient preferences

Explaining the VIOLET trial and related concepts

In many instances, recruiters placed the study in the context of existing uncertainty and emphasised that the two operations being compared were established techniques. A number of recruiters also avoided using the word ‘trial’ in their consultations, opting for ‘study’ instead, as recommended in QRI-informed training accessed by VIOLET trial recruiters prior to recruitment (occasional use of ‘trial’ was noticed and was limited to specific centres or recruiters). However, recruiters sometimes presented the VIOLET trial in an apologetic manner and appeared reluctant to present the trial information when they assumed that patients would not want to consider participating in VIOLET trial and would have preferences in line with their own (i.e. a preference for VATS). These beliefs led to a tendency to close down, rather than open up, conversations and patients missing opportunities to hear about the VIOLET trial, with recruiters appearing surprised when patients were willing to consider trial participation (see Appendix 6):

If you don’t want to be part of the study, that is the other option, then you tell me which procedure you want and we will do it.

He was thinking he’d quite like to be randomised.

Like you say there’s no, it’s toss a coin really isn’t it, it’s, yeah, whatever, yeah. I’m all right, yeah.

Right, so you’ve agreed to have the lobotomy operation, what do you think about the VIOLET study? Do you want to be part of that?

Yeah.

You sure?

So, coming to the study, have you got any interest in this or you, just to step out and let . . . no one will force you and, that’s absolutely fine with me and then you say, OK, I’m not interested, and I . . .

No, I would like to do it [patient was subsequently randomised].

Explanations of randomisation were sometimes absent or not well explained in consultations, with recruiters emphasising ‘lack of choice’ or implying that treatment is ‘selected’ or ‘decided’ by a computer. Similarly, some recruiters struggled with explanations of blinding, leading to patient concerns about what the process meant for them (see Appendix 6):

Before that you don’t know if you had keyhole surgery or surgery by thoracotomy, that’s the main let’s say parameter for you.

I’m wary of that, sounds a bit like a lottery. I’m surprised that the experts don’t know which is the best, especially for somebody of my age perhaps.

You and the research nurse looking after you whilst you are an inpatient, the study attempts to blind you to what you had.

Summary of recruitment processes and issues

Recruiters expressed high levels of support for the VIOLET trial. The patient pathway and eligibility criteria were considered straightforward, with some concerns regarding treatment recommendations for VATS made early in the pathway, which recruiters had addressed before it became overly problematic (i.e. from the prior training they received). Eligibility criteria for specific groups of patients was felt to be debatable. It appeared that a number of recruiters in the VIOLET trial had a preference for VATS and yet, for the most part, were able to overcome this, assume a position of equipoise and present balanced information on the two operations to patients (likely as a result of prior training). There were instances, however, where this was more challenging for recruiters, as they offered treatment recommendations and conveyed their own biases to patients. Patient preferences were being addressed by recruiters to some extent in the VIOLET trial, but we observed instances where patient preferences were accepted without further discussion or information. In addition, although concepts, such as the uncertainty underpinning the need for the VIOLET trial, were well presented, there were some noticeable instances of reluctance from recruiters in putting the trial forward to patients, likely stemming from their own preferences for VATS. Recruiters struggled with explanations of randomisation and blinding. Recruiters were found to follow recruitment advice from the training received prior to recruitment to use ‘study’ instead of ‘trial’. Although the blinding component of the study had been challenging to implement at first, the process became easier in most centres over time. In summary, although upfront training helped overcome some recruitment challenges, it did not resolve all issues. A number of key recruitment issues identified through the QRI were addressed, as outlined below.

Preventing recruitment difficulties: training and guidance prior to recruitment

Six surgeons, including PIs of all five pilot sites, attended the QRI-informed recruitment training workshops in May 2015, prior to the internal pilot phase in July 2015. A further pilot site surgeon attended the same workshop in May 2016. After the VIOLET trial moved to the main phase (July 2017), PIs of the new centres were invited to the training workshop in January 2018, which was attended by three VIOLET trial surgeons, including two PIs from two of the four new sites. This occurred in the same month that VIOLET trial-specific feedback was provided to the main phase sites.

Individual feedback

Eleven individual feedback sessions were held with nine recruiting surgeons from five centres (two surgeons received feedback on two occasions each), either face to face or over the telephone. The QRI researchers analysed each surgeon’s audio-recordings of recruitment consultations and prepared a two-page report. The confidential feedback highlighted aspects of their communication that worked well in promoting informed consent and trial participation, and provided suggestions for improving aspects of communication that may benefit from using alternative approaches.

Group feedback

The QRI team conducted eight feedback sessions with six sites (internal pilot sites, n = 5; main phase site, n = 1). Note that one site had three feedback sessions, one of which was for RNs only. Three of the new sites in the main phase did not receive a group feedback session. One centre submitted only two audio-recordings, which were insufficient to provide feedback; one centre did not submit any audio-recordings and provided screening log data that were incomplete or delayed; and one centre opened 4 months before recruitment was completed. Group feedback sessions included a presentation of site-specific recruitment figures (i.e. SEAR data) and anonymised QRI findings, drawing from audio-recordings from the site and the wider findings across other centres. Interactive group discussions focused on elements of good practice and suggestions for improving recruitment practice that were developed in collaboration with recruiters at each site to ensure that strategies were tailored to address their particular recruitment issues.

Recruitment and informed consent guidance (tips) documents

The QRI team produced two tips documents. The first version, disseminated early in the internal pilot phase (October 2015), contained suggestions for recruiters on how to discuss the study purpose and procedures, including randomisation, drawing from evidence in previous QRIs and a few audio-recordings collected in the first months of recruitment. A revised version of the guidance was developed and disseminated (June 2017) when new sites came on board in the main phase of the trial. The revised version was based on VIOLET trial audio-recordings and provided more extensive communication suggestions, including tips to balance information provided about the two operations, exploring patient preferences and how best to close a recruitment appointment.

Trial Management Group, investigator and study update meetings

The QRI findings and updates were regularly discussed at all available opportunities in VIOLET trial meetings (as mentioned in Chapter 2). QRI researchers disseminated findings on recruitment challenges at three TMG meetings. These meetings helped shape the recruitment strategies that were then disseminated more widely to the sites. The presentations were based on the challenges faced by recruiters at the given time. For instance, the TMG organised a meeting (Birmingham; February 2018) with PIs and co-applicants mid-way through the main phase of the study, with the purpose of providing an update on the study’s progress and encouraging sites to continue recruiting to the VIOLET trial after declining randomisation rates in the preceding 3 months. The QRI team’s presentation, therefore, focused on strategies to ensure that the momentum gained in recruitment previously was sustained for the remainder of the recruitment period.

Rapid communication strategy

When recruitment dipped towards the target line (March 2019), the TMG and QRI teams made co-ordinated efforts to support and continue engaging with sites until recruitment target was achieved (February 2019). These activities included the following.

EORTC QLQ-C30 physical function over time

The EORTC QLQ-C30 physical functioning scores over time from randomisation to 1 year are shown in Figure 6 and Appendix 5, Table 50. The improvement in physical function was more marked in the early discharge period and less pronounced after 6 months (see Figure 6). On average, the score was 4.22 points higher with VATS than with open surgery (95% CI 1.48 to 6.97 points; p = 0.009).

FIGURE 6.

QLQ-C30 physical function over time. Higher scores indicate better physical function. Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Complete resection

The total number of lymph node stations harvested (median, IQR 4–6) was very similar in both groups, as was the number of mediastinal nodes harvested (median 3, IQR 3–4) (Table 12). Complete R(0) resection was achieved in 429 of 439 (97.7%) participants and there was no difference between the groups (RR 0.999, 95% CI 0.97 to 1.26; p = 0.94) (see Appendix 5, Figure 27). All participants with residual disease had R1 disease.

Lymph node upstaging

Lymph node upstaging is summarised in Table 13. Upstaging from clinical node stage 0 (cN0) to pN1 and from clinical node stage 0 or 1 (cN0/1) to pN2 was similar in the groups (see Appendix 5, Figure 28).

Pain in the first 2 days post surgery

Visual analogue scale pain scores

In-hospital VAS pain scores in the first 2 days post surgery are shown in Figure 7 and in Appendix 5, Table 51. Pain scores were similar in the two groups on day 1, with a median score of 4 in both groups (MD –0.02, 95% CI –0.46 to 0.41; p = 0.913), but participants in the VATS group had significantly lower pain scores on day 2 than participants in the open-surgery group (median 3 vs. 4, MD –0.54, 95% CI –0.99 to –0.09; p = 0.018). The complete-case sensitivity analysis gave consistent results (see Appendix 5, Table 52).

FIGURE 7.

Pain scores over time. Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Postoperative analgesia

Analgesia is used to manage pain following surgery. The most commonly prescribed analgesia during the intraoperative and postoperative hospital stay, expressed as the ratio of mean daily dose, is shown in Figure 8. Overall, analgesic consumption was 10% lower in the VATS group (mean ratio 0.9, 95% CI 0.80 to 1.01). Additional analgesia prescribed, which are included in the overall estimate but not depicted in Figure 8, are given in Appendix 5, Table 53.

FIGURE 8.

Analgesia prescribed intraoperatively and postoperatively. Data show the mean ratio (95% CI). PCA, patient-controlled analgesia. Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Subgroup analysis: impact of type of analgesia received during surgery

Most participants received an intercostal block (315/502, 63%), with a similar proportion of participants receiving a paravertebral block (96/502, 19%) or neither (85/502, 17%). There was no evidence to suggest the difference in pain scores between VATS and open surgery differed by the type of analgesia received (test for treatment by analgesia interaction p = 0.19) (Figure 9).

FIGURE 9.

Subgroup analysis of pain scores by analgesia received.

Exploratory analyses: pain scores

Impact of type of thoracotomy, use of muscle sparing and rib resection

Differences in pain scores by type of thoracotomy performed, use of muscle sparing and rib resection are shown in Figure 10. In the first 2 days, pain scores did not vary significantly by type of thoracotomy, the use of muscle sparing or with rib resection.

FIGURE 10.

Exploratory analysis of the impact of type of thoracotomy on pain scores.

Impact of number of port sites

Summaries of pain scores by number of port sites are presented in Table 14. Lower pain scores on day 1 were reported in participants receiving single-port VATS than in participants receiving multiport VATS and open surgery (median score 3 vs. 4 vs. 4, respectively). On day 2, the median pain score report by participants was the same for single-port VATS and multiport VATS (i.e. a median score of 3) and lower than open surgery (i.e. a median score of 4). The exploratory analysis comparing pain scores found no significant differences (see Appendix 5, Table 54).

TABLE 14 - Pain scores by number of port sites

Missing data: 17 patients (single-port VATS, n = 1; multiport VATS, n = 7; open surgery, n = 9).

Missing data: 17 patients (single-port VATS, n = 3; multiport VATS, n = 7; open surgery, n = 7).

Missing data: 33 patients (single-port VATS, n = 6; multiport VATS, n = 13; open surgery, n = 14).

Note

Data are presented as median (IQR).

Time point	Patients receiving VATS (N = 208)		Patients receiving open surgery (N = 245)
	Single-port VATS (n = 42)	Multiport VATS (n = 166)
Baselinea	0 (0.0–1.0)	0 (0.0–2.0)	0 (0.0–1.0)
Day 1b	3 (2.0–5.0)	4 (2.0–6.0)	4 (2.0–6.0)
Day 2c	3 (1.0–5.0)	3 (0.0–5.0)	4 (2.0–5.0)

Prolonged incision pain beyond 5 weeks

The proportion of patients with prolonged incision pain (defined as the need for analgesia after 5 weeks post randomisation) is shown in Table 15. A total of 143 (59.6%) patients in the VATS group experienced prolonged incision pain compared with a total of 175 (72.3%) patients in the open-surgery group. This difference was significant (RR 0.82, 95% CI 0.72 to 0.94; p = 0.003) (see Appendix 5, Figure 29).

Time from surgery to hospital discharge: postoperative length of hospital stay

Time from surgery to hospital discharge is presented in Figure 11 and in Appendix 5, Table 55. Median length of stay was lower for patients in the VATS group than for patients in the open-surgery group (at 4 and 5 days, respectively). This difference was statistically significant (HR 1.34, 95% CI 1.09 to 1.65; p = 0.006).

FIGURE 11.

Time from surgery to hospital discharge. Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Overall survival and progression-free survival to 1 year

There were 31 deaths within 1 year of randomisation, 18 in the open-surgery group and 13 in the VATS group (Table 18). Overall, 94.6% of participants were alive at 1 year in the VATS group compared with 92.6% of participants in the open-surgery group (HR for death 0.67, 95% CI 0.32 to 1.40; p = 0.28). Overall survival by group is shown in Appendix 5, Figure 30.

Sixteen VATS participants and 17 open-surgery participants experienced disease recurrence/progression within 1 year. Overall, 90.4% of participants were alive and disease free at 1 year in the VATS group compared with 88.0% of participants in the open group (HR for disease progression 0.73, 95% CI 0.42 to 1.27; p = 0.26) (Table 19). Progression-free survival by group is shown in Appendix 5, Figure 31.

Sensitivity analyses adjusting for pathological disease stage were carried out for survival and progression-free survival outcomes. See Table 19 for data contrasting the results of the primary analyses for these outcomes and the results of the sensitivity analyses, which were very similar.

The locations of recurrence and new cancers are shown in Table 20. There were 24 cases of locoregional recurrence (VATS, n = 11; open surgery, n = 13), 17 cases of distant recurrence (VATS, n = 7; open surgery, n = 10) and 10 cases of new cancer (VATS, n = 4; open surgery, n = 6).

Uptake of adjuvant treatment

Time to uptake of adjuvant treatment was analysed in the primary intention-to-treat population and, for the subset of participants eligible for adjuvant treatment according to NICE guidelines,57 namely participants with a postoperative disease stage of (1) N1–2 and M0 or (2) T2b to 4, N0 and M0 were deemed eligible. Overall, 73 participants had adjuvant treatment, 56 of whom met the eligibility criteria defined by NICE (see Appendix 5, Table 57). The time to uptake of adjuvant treatment was similar between the two treatment groups, both for the primary intention-to-treat population and the NICE-eligible subset of participants (Figures 12 and 13, see also Appendix 5, Table 57).

FIGURE 12.

Uptake of adjuvant treatment in the primary analysis population.

FIGURE 13.

Uptake of adjuvant treatment in the NICE-eligible subset of participants.

EORTC QLQ-C30 quality-of-life questionnaire

The EORTC QLQ-C30 comprises an overall measure of global health and a number of subscales. For all scales, higher scores indicate a higher level of functioning or symptoms or problems.

Global health status and subscales assessing functioning

In addition to the global health status, the questionnaire measures physical, role, social, cognitive and emotional functioning. The physical function subscale was the primary outcome for the VIOLET trial and is reported in Primary outcome: EORTC QLQ-C30 physical function at 5 weeks and EORTC QLQ-C30 physical function over time. Scores for the other scales over time for participants randomised to VATS or open surgery are shown in Appendix 5, Figures 32–34. Global health status, role and social functioning were all significantly higher in the VATS group (Figure 14) than in the open-surgery group, and where cognitive function was impaired, the impairment was less in the VATS group than in the open-surgery group (Figure 15). The effect of surgery on emotional function varied over time. At 2 weeks, fewer participants in the VATS group than in the open-surgery group reported some impaired emotional function, but thereafter the results were similar in the two groups (Figure 16). Summary data and the estimated treatment effects for each scale are given in Appendix 5, Tables 58 and 59.

FIGURE 14.

QLQ-C30 global health status and role and social functioning: treatment effects.

FIGURE 15.

QLQ-C30 cognitive functioning over time. Higher scores indicate better health.

FIGURE 16.

QLQ-C30 emotional functioning over time. Higher scores indicate better health.

Subscales assessing symptoms and problems

Scores for scales measuring symptoms and problems over time for participants randomised to VATS or open surgery are shown in Appendix 5, Figures 35–43. Participants randomised to VATS experienced less pain and fatigue and had less difficulty sleeping in the first 2 weeks than participants randomised to open surgery. These participants were also less likely to experience appetite loss and nausea, and constipation in the early period post surgery. Other measures were similar in the two groups (Figure 17). Summary data and the estimated treatment effects for each subscale are given in Appendix 5, Tables 60 and 61.

FIGURE 17.

QLQ-C30 symptoms and problems subscales: treatment effects.

QLQ-C30 exploratory analysis: pain scores

Differences in pain scores to 1 year by type of thoracotomy performed, use of muscle sparing and rib resection are shown in Figure 18. The data show the same trend as was observed in the pain scores in the first 2 days after surgery (see Figure 10), with pain scores in those who had received an anterior thoracotomy being lower, on average, than those who received a posterior thoracotomy, and there being less pain with muscle sparing and significantly more pain with rib resection (MD 9.8, 95% CI 2.07 to 17.52).

FIGURE 18.

Impact of type of thoracotomy on QLQ-C30 pain scores.

EORTC QLQ-LC13 quality-of-life questionnaire

The QLQ-LC13 measures a range of cancer-related symptoms and problems. As for the QLQ-C30, higher scores indicate a higher level of symptoms or problems. Scores over time for participants randomised to VATS or open surgery are shown in Appendix 5, Figures 44–53. Participants randomised to VATS experienced significantly less pain in the chest and less pain in the arm at 5 weeks than participants randomised to open surgery, but other outcomes were similar in the two groups (Figure 19). Summary data and the estimated treatment effects for each subscale are given in Appendix 5, Tables 62 and 63.

FIGURE 19.

QLQ-LC13 symptoms and problems subscales: treatment effects.

EQ-5D-5L utility score

The EQ-5D-5L utility scores over time are shown in Figure 20. The median utility scores were higher in the VATS group than in the open-surgery group at all post-baseline time points. Participants in the VATS group were less likely to have less than perfect health (i.e. a score < 1) than participants in the open-surgery group [odds ratio (OR) 0.57, 95% CI 0.38 to 0.86; p = 0.007]. Of those participants with less than perfect health, participants in the VATS group had, on average, a higher score (representing better health) than those in the open-surgery group (geometric mean ratio 0.90, 95% CI 0.84 to 0.96; p = 0.003) (see Appendix 5, Table 64, for summary data).

FIGURE 20.

EQ-5D-5L utility scores over time.

Adverse events

Adverse events in the period from surgery to discharge from hospital following surgery

Eighty-one (32.8%) participants allocated to VATS and 113 (44.3%) participants allocated to open surgery experienced at least one AE in the period from surgery to discharge from hospital (RR 0.74, 95% CI 0.66 to 0.84; p < 0.001), but the number of SAEs was similar in the two groups (8.1% in the VATS group vs. 8.2% in the open group) (Table 21). AE and SAEs summarised by Medical Dictionary for Regulatory Activities (MedDRA) system organ class are presented in Figure 21. Details of the events within each system organ class are given in Appendix 5, Table 65.

TABLE 21 - Relative risk of AEs and SAEs following surgery

Adapted with permission from NEJM Evidence, Lim E, Batchelor TJP, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. , Video-assisted thorascopic or open lobectomy in early-stage lung cancer, Volume 1, Copyright © 2022 Massachusetts Medical Society. 56

Outcome	Participant allocation		RR (95% CI)	p-value
	Randomised to VATS (N = 247), n/N (%)	Randomised to open surgery (N = 255), n/N (%)
In hospital before discharge
Any in-hospital AE	81/247 (32.8)	113/255 (44.3)	0.74 (0.66 to 0.84)	< 0.001
Any in-hospital SAE	20/247 (8.1)	21/255 (8.2)	0.98 (0.59 to 1.63)	0.948
After discharge following surgery (events/patients)
Readmissions	117/70 (29.0)	141/88 (35.9)
SAE	142/75 (30.7)	207/94 (37.8)	0.81 (0.66 to 1.00)	0.053

FIGURE 21.

In-hospital AEs summarised by MedDRA system organ class.

Infective complications were the most common in both groups, the majority of which were pneumonia or lower respiratory tract infections (see Appendix 5, Table 65). Comparing the two groups, participants in the VATS group had fewer infective (RR 0.89, 95% CI 0.84 to 0.94), psychiatric (RR 0.98, 95% CI 0.97 to 1.00) and renal (RR 0.96, 95% CI 0.91 to 1.00) complications than participants in the open-surgery group (see Figure 21). There were seven deaths prior to discharge from hospital (VATS group, n = 2; open-surgery group, n = 5). Causes of death for these seven participants are summarised in Table 18.

Serious adverse events in the period after discharge from hospital to 1 year

Seventy-five (30.7%) participants allocated to VATS and 94 (37.8%) participants allocated to open surgery experienced at least one SAE in the period after discharge from hospital to 1 year (RR 0.81, 95% CI 0.66 to 1.00; p = 0.053), of which 158 (93.5%) resulted in an admission to hospital (VATS group, n = 70; open-surgery group, n = 88) (see Table 21). SAEs summarised by MedDRA system organ class are presented in Figure 22. Details of the events within each system organ class are given in Appendix 5, Table 66.

FIGURE 22.

Post-discharge SAEs summarised by MedDRA system organ class.

As with the early postoperative period, infective complications were the most common in both groups, the majority of which were pneumonia or lower respiratory tract infections, followed by respiratory, thoracic and mediastinal disorders (see Appendix 5, Table 66). The events that resulted in admission to hospital are summarised in Appendix 5, Table 67. There were 24 deaths after discharge from hospital (VATS group, n = 11; open-surgery group, n = 13). Causes of death for these 24 participants are summarised in Table 18. Half of the deaths (12/24) were due to disease progression.

Trial conduct

With the support of the QRI to optimise recruitment, the trial successfully recruited to time and target. Average recruitment rates at study sites ranged from 0.5 participants randomised per month in Hull to 3.4 participants randomised per month in Bristol. The lead study site (Brompton) was the second highest recruiting site, with an average of 2.5 participants randomised per month.

Recruitment was paused in Bristol for 3 months in early 2016 (i.e. 5 January 2016 to 29 March 2016) because of issues in the postoperative recruitment pathway that were unblinding the researcher responsible for data collection, which had the potential to bias the cost-effectiveness analysis. During late 2015, it became apparent that the Bristol research team were experiencing difficulties integrating the VIOLET trial into their local practice. These difficulties stemmed from historical hospital guidelines that specified that patients undergoing thoracotomy should recover in a high-dependency unit, whereas patients undergoing VATS should go to a general ward for their recovery. Recruitment was paused while the Bristol PI and colleagues liaised with the policy-makers to encourage the hospital to adopt a more patient-focused, risk-based approach to patient management. These discussions were protracted, but, on 29 March 2016, the hospital agreed to change the guidelines and adopt a risk-based approach to patient management, allowing recruitment to restart immediately. Following this, recruitment at the Bristol site progressed well. Bristol was the highest recruiting site in the VIOLET trial, with 136 participants.

The study sites were engaged with the trial throughout and data completion rates were good. There was one change that related to the transition to TNM813 of the TNM staging system, which had a significant impact on study sites and the trials centre. Following approval of an amendment to the protocol to reflect this change in June 2017, the screening log and CRFs were revised to collect the clinical staging in accordance with TNM8. 13 Subsequent to this, the trials centre became aware that TMN813 was not mandated for clinical use until January 2018, and that at least one study site was still using TMN7,14 which meant that there was not a clear transition from TMN714 to TNM813 in line with the change to the CRF. In addition, there was not a clear mapping from TNM714 to TNM8. 13 Following discussion at a study investigator meeting, it was agreed that all participants recruited before January 2018 would be restaged using TNM813 to ensure that we had consistent data for the full trial. A ‘restaging’ CRF and associated page in the study database were created to facilitate this data collection exercise.

During a regular review of the study data, it also became apparent that the CRF for collecting data on ‘ward movements’ (i.e. length of stay in the intensive care unit/high-dependency unit/ward) was being misinterpreted by some sites. The CRF was redesigned to maximise the accuracy of the data and sites were asked to reconfirm the data submitted for participants recruited before the change. To minimise the burden for study sites, this was carried out at the same time the participant staging was reviewed.

QRI

We applied a range of QRI methods to optimise recruitment and informed consent in the VIOLET trial, focusing on prevention of recruitment challenges and the identification and resolution of potential recruitment issues. Recruitment to the VIOLET trial commenced and progressed well, completing to target and on time. Although examples of good practice were seen (e.g. expressing uncertainty and avoiding the term ‘trial’ in favour of the term ‘study’), the QRI also identified and addressed a number of recruitment challenges (e.g. recruiters recommending a treatment, using imbalanced/loaded terminology to explain the two operations and explaining the study apologetically). QRI actions included upfront training, followed by trial-specific group and individual feedback sessions, tips documents and a rapid communication strategy. These actions were aimed at helping recruiters present the study to all eligible patients, balance information provision on the two operations, provide clearer explanations of randomisation and blinding, and avoid loaded terminology, treatment recommendations or assumptions regarding patients’ views on the trial.

The recruitment issues reported in the VIOLET trial have been previously identified in other RCTs with QRIs. 25–27,30,33 However, the VIOLET trial recruited above target from the very beginning and throughout the recruitment period. Although it is difficult to prove cause and effect, it is plausible that the QRI training contributed towards the good start to recruitment. It is also plausible that the QRI training sustained the momentum thereafter, including when new centres joined the study, and helped recruitment pick up when it dipped towards the target line in the last year of recruitment. An observational study58 that evaluated five QRIs showed that randomisation improved in three of the five RCTs in the post-intervention period. The two RCTs that showed no difference in recruitment after the QRI were similar to the VIOLET trial in that it had completed recruitment on target following training and support before recruitment began and then throughout the recruitment period. 58

It is important to note, however, that the prior training and support received by recruiters did not resolve all challenges in advance of recruitment, with many being identified and addressed later with the help of the QRI, which contributed towards sustaining the recruitment momentum gained early on. The most crucial aspect of QRI evaluation is the change in recruitment practice observed qualitatively. Despite some healthy recruitment practices, there was scope for improvement, and we observed noticeable changes in how recruiters addressed patient preferences, expressed uncertainty or explained the trial following VIOLET trial-specific training.

Trial results

The results of the VIOLET trial suggest that for patients with early-stage lung cancer a VATS approach to lobectomy was associated with less pain, fewer complications and a shorter length of hospital stay, without any compromise to oncologic outcome. The benefits of a VATS approach extended beyond the in-hospital period. The primary outcome of physical function was significantly improved at the 5-week point and improved recovery was consistent for all secondary measures of quality-of-life outcomes up to 1 year. In addition, there were fewer post-discharge readmissions for care and no difference in the measures of oncologic outcome of recurrence-free and overall survival up to 1 year.

Prior to the conduct of the VIOLET trial, the benefit of VATS lobectomy was widely considered to be ‘better recovery’ (i.e. a readily understandable but nebulous term with multifactorial and time-dependent components). When challenged to consider one single most relevant outcome that encompasses all the potential benefits of better ‘recovery’ at a single most relevant time point, physical function at 5 weeks was chosen by our patient and public involvement (PPI) group and unanimously agreed by the Trial Management Committee. With this global composite measure, our results revealed a striking 13-point improvement between the observed median scores in favour of VATS at 2 weeks, which reduced to a modelled difference of approximately 5-point difference at 5 weeks. It took a further 6 months for participants in the open-surgery group to reach similar levels of physical function as the participants in the VATS group. The results shed light on the duration of functional recovery provided by a VATS approach, as a 9-point difference has been proposed to correspond to a 1-point difference in performance status. 16 On a World Health Organization performance scale, 1 unit of change would correspond to a difference from performance status 0 (i.e. ‘able to carry out all normal activity without restriction’) to performance status 1 (i.e. ‘restricted in strenuous activity but ambulatory and able to carry out light work’). Our earliest assessment was undertaken at 2 weeks and, therefore, we may not have captured the full effects, which we now know to be most prominent within the first 2 weeks.

Pain is a near universal trade-off for surgery and one of the most important considerations for improvement. The open-surgery approach requires rib spreaders (metal retractors) to splay the ribs apart. The result is inevitable compression of the intercostal nerves and is considered the main source of pain after open lobectomy. Results from a Danish trial9 reported a lower proportion of patients randomised to VATS with severe pain (38% vs. 63%), but the trial did not report a direct comparison of pain scores between the two groups. 9 On a linear scale, we noted similar pain scores on day 1 and less pain (by 1 point in the VAS) on day 2. The lack of difference on day 1 can be attributed to the effective local anaesthetic blocks that were administered (often taken down on days 1 or 2 to allow patients to mobilise). Having adjusted for total analgesic use, patients randomised to VATS had approximately 10% lower composite analgesic use, with an adjusted estimate of a half-point lower pain score, surmising that a VATS approach is (modestly) less painful. We analysed the different forms of local anaesthetic block (i.e. paravertebral and intercostal block), subtypes of thoracotomy (i.e. posterior and anterior), use of rib resection and number of port sites, and found no evidence of interaction with the surgical approach (similarly effective). After discharge, however, we noted that prolonged incision pain, defined as the need for analgesia after 5 weeks, was lower in the VATS group than in the open-surgery group (59.6% vs. 72.3%), suggesting better recovery with VATS when measured by the cessation for the need of analgesics (at 5 weeks). Pain scores using QLQ-C30 and QLQ-LC13 were consistent in the direction of the estimates in favour of VATS out to 1 year. In addition, we noted that patients who underwent rib resection as part of a thoracotomy experienced significantly more pain, with a 9.8-point difference out to 1 year (an observation that has not been previously described).

Another important consideration when evaluating a new procedure is a demonstration of safety (i.e. if the new procedure can be performed without increasing harm). When we observed the in-hospital AEs, we noted that the RR of harm was in fact lower with VATS (i.e. a 26% reduction in AEs with no difference in SAEs). One specific complication that was noted was that the proportion of patients experiencing intraoperative bleeding (often attributed to blood vessel injury when using a keyhole approach) was similar in the two groups (VATS group, 6.2%; open-surgery group, 3.9%). The main benefits for VATS during the in-hospital phase was a notable reduction in kidney and infective complications. There are a number of hypotheses to be offered for this, including less analgesic use, which may reduce renal complications and improve mobility for lower chest infections, and smaller incisions, resulting in fewer wound infections.

The benefits of a less painful and safer operation culminated in a shorter length of hospital stay, an additional global outcome for ‘better recovery’. The results were consistent in both fitness for hospital discharge (based on discharge criteria) and actual measured time to discharge in favour of VATS by 1 day. The benefits of a better recovery persisted in the year after discharge, with benefits from global scales (e.g. QLQ-C30 and EQ-5D-5L) and individual scales (e.g. dyspnoea, fatigue, appetite loss) broadly consistent in direction in favour of VATS or no difference between the two groups. There was no single measure that was consistently, or was of a clinically important magnitude, in favour of open surgery. AEs after discharge continued in favour of VATS, with fewer readmissions (29.0% vs. 35.9%) and a 19% reduction in the number of SAEs (p = 0.053) up to 1 year.

Perhaps the most important contribution of the VIOLET trial is the study of oncologic outcomes, which, to date, and to the best of our knowledge, has not been comprehensively reported in any RCT. One concern for keyhole surgery has been the ability to perform a cancer operation without the surgeon’s hands entering the chest through small incisions (without direct tactile feedback), using a video camera and television monitor (without direct visualisation) through fixed positions in the skin (without a full range of movement of the instruments). We assessed the quality of the lymph node harvesting both in terms of the number and position of the nodes assessed, and found no difference in between the two groups (five stations harvested and three mediastinal stations for both arms), indicating that the VATS techniques and instruments used were able to effectively access the same extent of lymph node harvesting. Lymph node upstaging is often considered to be a more discriminating assessment of quality of lymph node dissection, with the premise that a more thorough dissection would yield more positive lymph nodes (upstaging). When we assessed lymph node upstaging, we found 6.2% in the VATS group compared with 5.2% in the open-surgery group (cN0 to pN1) and 6.2% in the VATS group compared with 4.8% in the open-surgery group (cN0/1 to pN2). In addition, there was no difference in the ability to achieve complete pathologic resection (97.7% vs. 97.8%) between the VATS and open-group groups, respectively, confirming that there was no difference in the completeness of the cancer operation. After discharge, the uptake of chemotherapy was similar in both groups, at 50.9% in the VATS group and 45.9% in the open-surgery group, indicating that the improvement in time to recovery with VATS did not translate to an important difference in uptake of adjuvant chemotherapy. After discharge to 1 year, the HR for disease-free survival and overall survival was 0.67 (p = 0.366) and 0.74 (p = 0.312), respectively, in favour of VATS, giving assurance on the longer-term oncologic safety of a VATS approach. Our trial was not powered for long-term survival and this is an important area for further research clarification, as systematic review of (mainly) non-randomised studies indicated the possibility of a VATS approach leading to better overall survival. 4 Given that the main oncologic outcomes are similar in both groups in the VIOLET trial, we hypothesise any difference in survival to be more likely associated with secondary improvement in quality of life, rather than technical oncologic surgery.

Health economic evaluation

In the economic evaluation, differences in costs and QALYs favoured the VATS group, and when combined VATS proved to be a cost-effective option for the NHS. The mean QALYs to 12 months were 0.841 and 0.780 in the VATS and open-surgery groups, respectively, resulting in a statistically significant MD of 0.060. The mean total cost from surgery to 12 months was £10,879 in the VATS group and £13,581 in the open-surgery group (i.e. a MD of –£2702, although this was not statistically significant). This cost difference is largely driven by less time in critical care in the index admission and fewer days readmitted to hospital in the VATS group. Results were robust to all the sensitivity analyses performed.

The cost-effectiveness results clearly indicate that VATS is cost-effective across any willingness-to-pay threshold. Based on the point estimates of the cost and QALY differences and on the point estimate of the ICER (–£44,908), VATS is considered cost-effective. VATS surgery is dominant over open surgery as it is both more effective and less costly. The probability that VATS is cost-effective at a willingness-to-pay threshold of £20,000 per QALY, which is generally considered as the threshold that NICE adopts for considering an intervention to be cost-effective, is 1. Indeed, at any willingness-to-pay threshold, VATS surgery is considered cost-effective, and there is negligible uncertainty around this finding so we can be confident of this result.

Our cost-effectiveness results are consistent with the Danish study,11 in which 103 patients were randomised to VATS and 103 patients were randomised to thoracotomy, which concluded that VATS was cost-effective compared with thoracotomy following lobectomy for stage 1 lung cancer. VATS provided a larger number of QALYs and generated lower costs. Specifically, the mean cost per patient for VATS was 103,108 kr (£11,857) and the mean cost per patient for thoracotomy was 134,945 kr (£15,518), making the costs for VATS 31,837 kr (£3661) lower than thoracotomy (p < 0.001). In terms of HRQoL, the difference between the two surgery types for QALYs gained over 1 year of follow-up was 0.021 (p = 0.048). The CEAC presented clearly showed that VATS was superior to thoracotomy more or less regardless of the willingness-to-pay threshold for a QALY, just as was shown here. Results of a French study by Pagès et al. 12 were not available at the time that our report was written.

Strengths

Limitations

Project management team members

Professor Eric Lim, Chief Investigator; Professor Chris A Rogers, Methodological Lead and Statistician; Timothy Brush, Clinical Trial Co-ordinator; Lucy Dabner, Clinical Trial Co-ordinator; Dawn Phillips, Clinical Trial Co-ordinator; Holly McKeon, Clinical Trial Co-ordinator; Chloe Beard, Assistant Trial Co-ordinator; Surinder Kaur, Assistant Trial Co-ordinator; Rosie A Harris, Medical Statistician; Dr Daisy Elliott, Research Fellow, QRI; Dr Sangeetha Paramasivan, Research Fellow, QRI; Dr Alba Realpe, Senior Research Associate, QRI; Professor Sarah Wordsworth, Lead Health Economist; Dr Elizabeth Stokes, Health Economist; Professor Jane Blazeby, Methodologist and Surgeon; Professor Andrew G Nicholson, Pathologist; and David Hutton, Database Manager.

Participating sites members: Phase I

Participating sites members: Phase II

Acknowledgement of non-author contributions

We would like to acknowledge the teams at the VIOLET trial sites, without whom the trial would not have been possible. We also acknowledge the advice and support provided by the members of the VIOLET TSC and DMSC (see Appendix 9).

Peer-reviewed publications

Lim E, Batchelor T, Shackcloth M, Dunning J, McGonigle N, Brush T, et al. Study protocol for VIdeo assisted thoracoscopic lobectomy versus conventional Open LobEcTomy for lung cancer, a UK multicentre randomised controlled trial with an internal pilot (the VIOLET study). BMJ Open 2019;9:e029507.

Lim E, Batchelor T, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. Video-assisted thoracoscopic or open lobectomy in early-stage lung cancer. NJEM Evid 2022;1.

Conference abstracts

Lim E, Brush T, Rogers C. Video Assisted Thoracoscopic Lobectomy Versus Conventional Open Lobectomy for Lung Cancer, a Multi-Centre Randomised Controlled Trial with an Internal Pilot: the VIOLET Study. 14th Annual British Thoracic Oncology Group Conference, Dublin, Ireland, 27–29 January 2016.

Rogers CA, Paramasivan S, Elliott D, Whybrow P, Kanavou S, Harris RA, et al. Audio-Recording Recruitment Consultations – An Exploratory Study in Two RCTs to Investigate the Impact on Randomisation Rates. 4th International Clinical Trials Methodology Conference and the 38th Annual Meeting of the Society for Clinical Trials, Liverpool, UK, 7–10 May 2017.

Lim E, Batchelor T, Dunning J, Shackcloth M, Anikin V, Naidu B, et al. PL02.06 in hospital clinical efficacy, safety and oncologic outcomes from violet: a UK multi-centre RCT of VATS versus open lobectomy for lung cancer. J Thorac Oncol 2019;14:S6.

Lim E, Begum S, Batchelor T, Krishnadas R, Shackcloth M, Dunning J. S23 optimum diagnostic pathway and pathologic confirmation rate of early stage lung cancer: results from VIOLET. Thorax 2019;74:A15.

Press release

International Association for the Study of Lung Cancer. Video Assisted Lung Surgery Reduces Complications and Hospital Stays Compared to Open Surgery. Press release, 9 September 2019.