PET-NECK: a multicentre randomised Phase III non-inferiority trial comparing a positron emission tomography computerised tomography-guided watch-and-wait policy with planned neck dissection in the management of locally advanced (N2/N3) nodal metastases in patients with squamous cell head and neck cancer

Evidence in support of planned (routine) neck dissection

Until recently the standard care in the UK was to perform a ND with CRT. In other countries, a significant proportion of patients are still treated with planned ND. 5 Proponents of this management policy maintain that CRT does not eradicate large-volume nodal disease in a large proportion of patients (up to 50%), putting them at risk of recurrence. Some believe that for most of these patients, salvage by surgery will not be possible,9 resulting in devastating consequences.

The only randomised controlled trial available in the literature on the subject compared conservative follow-up with planned ND before radical CRT. 10 It found that there was significantly improved disease-specific survival following ND. However, it was small (50 patients) and had several major limitations, casting strong doubts on the validity of its results and findings. Other single-arm and retrospective studies have reported that planned ND demonstrated excellent locoregional control, although some did not detect improvement in overall or disease-specific survival in their series. 11–14

In addition, it has been found that in up to 40% of patients who show a complete clinical response to CRT but who undergo ND 8–10 weeks later tumour deposits can still be detected histologically in the ND specimen. 11,14,15 It has therefore been suggested that CRT does not completely eradicate the tumour in up to 40% of patients. Furthermore, proponents of ND maintain that selecting patients at a high risk of persistent disease is not possible using CT and/or magnetic resonance imaging (MRI), as studies have found that clinical and imaging evidence of complete response (CR) of nodal disease to CRT does not predict a complete pathological response (i.e. it does not correlate with an absence of pathological evidence of the disease in the ND specimen). 12,14

Evidence supporting a watch-and-wait policy in patients with complete response to chemoradiotherapy

Many clinicians now advocate a conservative surveillance policy, performing ND only if there is clinical evidence of persistent nodal disease after CRT. The rationale is that, among patients who exhibit a clinical CR in the neck following CRT, the recurrence rate in the neck is low (< 10%), and similar to that (8%) in patients with pathologically negative neck following ND. 8,16 Other level 2/3 studies9,17,18 have also found that ND in patients who show a CR to CRT does not confer any benefit in terms of improved survival or reduction in nodal recurrence compared with a watch-and-wait policy. Importantly, a large retrospective cohort study found that 43% of patients showed CR on CT, with a control rate of 92% at 5 years. Among those who experienced less than a CR, the 5-year control rate among those who underwent ND was similar, at 90%, but among those who did not undergo ND it was significantly lower (76%). 19

There is also evidence to suggest that, in many cases, the cells in the residual nodal deposits found on histology in ND specimens following CRT are not viable. 7,18 An experimental study, examining a proliferation marker, Ki-67, appears to confirm this hypothesis. 20

Accuracy of PET–CT scanning in the assessment of response to chemoradiotherapy and detection of residual nodal disease

Crucial to a watch-and-wait policy is the ability to detect residual disease in the neck post CRT, so that these patients can be targeted to have a ND. Detection of residual disease in the neck is not accurately assessed by examination in the clinic or by CT and/or MRI. Evidence suggests that PET is able to accurately identify those patients who do not have residual tumour in their neck following CRT (i.e. PET has a high negative predictive value).

Positron emission tomography has been reported in several small single-institution prospective and retrospective series and two meta-analyses21,22 to have a negative predictive value of 90–100% for the detection of persistent nodal metastasis following radiotherapy or CRT, with a variable positive predictive value of approximately 30–60%. 23–31 It has also been shown in several studies to have higher predictive values than CT and/or MRI,21,27,32,33 and better than combined clinical examination and ultrasound. 31 Finally, studies have found that co-registering PET and CT (PET–CT) is even more accurate than PET alone33,34 (Table 1).

There is also evidence from same studies that PET–CT is highly sensitive to the detection of residual tumour at the primary site following CRT and radiotherapy, with a high negative predictive value of 90–100%. 35–37 Indeed, it has been suggested that PET–CT may have similar a predictive value in the exclusion of residual disease at the primary site to examination under anaesthesia (EUA), the current gold standard, and may be capable of replacing it as a result of it being less invasive. 38,39 However, there is no existing level 1 evidence to corroborate this theory.

The study had not intended to measure human papillomavirus (HPV) status at the time of inception. However, over the past 5 years, there have been reports on the significant increasing incidence of HPV-associated OPC, with the proportion of OPCs that are HPV associated increasing from 40.5% to 72% over a period of 20 years. 40 HPV association appears to have a remarkable effect on the prognosis and outcomes of treatment, with a 58% reduction in the risk of death [hazard ratio (HR) 0.42, 95% confidence interval (CI) 0.27 to 0.66]. 41 This has necessitated examining outcomes by HPV status. 42

Amendments to the protocol

Substantial amendments to the trial protocol were approved since conception, which were significant to evaluate the economic costs, to help increase recruitment and to clarify eligibility and which included the following:

• The introduction of health economics evaluation at local sites. A health economic cost analysis was required because the avoidance of a ND with its costs and possible complications and morbidity is expected to be one of the most significant benefits of the watch-and-wait policy. Furthermore, the economic effect of replacing EUA and CT with a PET–CT scan would need to be evaluated to assess its potential for the future. The aim of the economic evaluation was to identify the within-trial (WT) and long-term incremental cost-effectiveness of PET–CT-guided watch-and-wait compared with planned ND in HNSCC patients.

• To extend the inclusion criteria to allow patients with occult primary tumours to enter the PET-NECK trial. Patients occasionally have pathologically occult tumours but, if all other diagnostic procedures point towards a H&N primary, and all other primary sites are excluded, they are routinely diagnosed and treated as having H&N cancer. There was no reason why these patients should not also have been given the opportunity to enter the PET-NECK trial if they wished and if they met all other criteria.

• To allow ND to be performed before or after CRT in patients randomised to receive standard ND (control arm). There had been a change in practice in the previous 2–3 years regarding the timing of planned ND. In the USA and Europe, planned ND is usually carried out after CRT, rather than before, and there has been a gradual shift towards this practice in the UK over the previous 2–3 years. Therefore, amending the control arm to allow this was intended to increase recruitment and ‘future-proof’ the study if this trend continues in the UK.

• To exclude patients with tumours that were not histologically squamous cell carcinoma and patients with primary nasopharyngeal carcinoma. Both non-squamous cell carcinomas and nasopharyngeal carcinomas have different biological behaviours and natural histories from squamous cell carcinoma of the H&N, and they should not be treated in a similar manner. Nasopharyngeal carcinoma is highly sensitive to radiotherapy and should not be treated by a ND. Non-squamous cell carcinoma is much less radiosensitive and, therefore, a ND is indicated in these cases. The literature available pertains to only squamous cell carcinoma in sites of the H&N other than nasopharyngeal.

• To clarify that patients with equivocal PET–CT scans should receive a ND. The management policy in the UK for patients with equivocal PET–CT scans at the time of conducting the study was that they should receive a ND.

Inclusion criteria

Patients who met all of the criteria listed below were eligible to participate in the study:

• had a histological diagnosis of oropharyngeal, laryngeal, oral, hypopharyngeal or occult HNSCC

• had a clinical and CT/MRI imaging evidence of nodal metastases, stage N2 (a, b or c) or N3

• had a multidisciplinary team (MDT) decision to receive curative radical concurrent CRT for primary

• had an indication to receive one of the CRT regimens approved by the study

• were fit for ND surgery

• ND was technically feasible to perform to remove nodal disease (e.g. no carotid encasement, no direct extension between tumour and nodal disease)

• were aged ≥ 18 years

• were able to provide written informed consent.

Patients with N2 or N3 histologically and/or cytologically proven squamous cell carcinoma and an occult primary (after EUA and PET–CT scan) were eligible for the trial if they were going to be treated with CRT.

Exclusion criteria

Patients who met any of the criteria listed below were ineligible:

• had tumours that were not squamous cell carcinomas histologically

• were undergoing resection for their primary tumour, for example resection of the tonsil or base of tongue with flap reconstruction (diagnostic tonsillectomy was not considered an exclusion criteria)

• had N1 stage nodal metastasis

• were receiving neoadjuvant CRT with no concomitant chemotherapy

• were receiving adjuvant chemotherapy

• were undergoing chemotherapy with or without radiotherapy for palliative purposes

• were undergoing radiotherapy alone (this is not an optimal treatment for neck node disease)

• had distant metastases to the chest, liver, bones or other sites

• were unfit for surgery or CRT

• had received previous treatment for HNSCC

• had primary nasopharyngeal carcinoma

• were pregnant

• had had another cancer diagnosis in the past 5 years (with the exception of basal cell carcinoma or carcinoma of the cervix in situ).

Patients undergoing neoadjuvant chemotherapy followed by concomitant CRT were eligible to enter the trial.

The patients in the study were similar to those in the studies that established the efficacy of CRT of the larynx42 and oropharynx,44 and were similar to the patients included in a wide range of studies examining PET–CT in the assessment of persistent nodal disease. 21,22

Administration of Radioactive Substances Advisory Committee approval

Administration of Radioactive Substances Advisory Committee licences were study specific and related only to the PET–CT centre, which was stated on the application.

Therefore, when a licence was held for the PET-NECK trial at a particular PET–CT centre, the site could send their patients to be scanned there as long as the ARSAC licence holder agreed to oversee the safety of the patients. The trials office had a list of ARSAC licence holders and arranged this for the centre. However, when PET–CT centres were not close by or when the site had its own PET–CT centre but did not hold a licence, its radiologist had to make an application under the Ionising Radiation (Medical Exposure) 2000 Regulations47 to the appropriate ARSAC. Approval of applications took around 1 month and patients could not be scanned without the licence being in place. ARSAC licences were issued for 2 years and had to be renewed after that time. The trials office kept a log of all expiry dates for ARSAC certificates and reminded the radiologist to renew them and send a copy of the renewal. A list of ARSAC licence holders is given in the Acknowledgements section of this report.

Quality assurance testing for PET–CT scanners

Before sites could scan patients in the surveillance arm of the trial, the PET–CT scanner to be used had to receive a successful QA review by the study’s central physicist.

All scanners had to be quality assured by the physicist. Each site had to arrange with the PET–CT centre to send a test case to the PET-NECK trial core laboratory for standardisation and quality control for each scanner that would be used in the trial.

The test case had to be an ¹⁸F or ⁶⁸Ge Dicom phantom scan. All Dicom images had to be anonymised with patient trial number and initials and the compact disc or digital versatile disc had to be correctly labelled with ‘PET-NECK Trial’, scan date, site name and coded identifier and be sent along with an anonymised copy of the local report and a completed study Dicom patient scan transfer form.

Providers of PET–CT scanning units used in the study were InHealth (InHealth Group, High Wycombe, UK), Alliance (Alliance Healthcare, Hinckley, UK) and Cobalt Healthcare (Cobalt Healthcare, Cheltenham, UK). Scanners used for the study were either ‘static’, that is, fixed at a specific hospital site, or ‘mobile’, that is, able to travel various hospitals around the country.

Static scanners had to be quality assured every 2 years; however, mobile scanning units had to be tested by the core laboratory half yearly. The trial team kept a log of the pass dates for scanning units and organised with the PET–CT providers to renew the QA review.

A set-up visit to each centre by the trial co-ordinator also took place, during which information about the trial protocol was presented and site staff were given the opportunity to clarify information. Each site was provided with the TNM staging manual and written procedural requirements for PET–CT scanning and reporting. Once all approvals were in place, an activation letter was sent to the principal investigator and trust research and development department for each site to activate the start of recruitment.

Assessments for diagnosis and tumour staging

Each of the following assessments had to be done within the 4 weeks prior to randomisation, or at least CT/MRI or EUA had to have been carried out within the 4 weeks before randomisation:

1. CT/MRI of the primary site and neck (which had to be carried out before biopsy or not less than 2 weeks after biopsy).

2. Examination of the primary site with biopsy. In most cases, this necessitated examination under general anaesthesia. However, it was not mandatory if adequate clinical examination and biopsy could be performed without general anaesthetic.

3. CT of the chest.

4. Assessment for fitness for anaesthetic by treating clinician or anaesthetist.

Patient treatment was discussed by the MDT and each patient was considered for the PET-NECK trial. Eligible patients were invited to the clinic to discuss treatment and were told about the PET-NECK study. The study was discussed in detail with the patient and carer and all questions were answered. Once a patient decided to enter the study, informed consent was taken.

Informed consent

Adequate time was allowed for patients to consider their participation in the trial. There was no pre-agreed specified time by which to consent. Consent was required to be informed and voluntary, with time for questions and reflection. However, the patient also had the right to make an immediate decision to consent.

Consent to participate in the PET-NECK study was sought by the clinician involved in the patient’s care, with the involvement of the research nurse in the consent discussion. Patients were asked to confirm their consent to participate in the PET-NECK trial by initialling the appropriate boxes on the consent form and signing the form in the presence of the person taking consent. A copy was given to the patient, another copy was kept in the patient notes and the original was kept in the local site file.

Patient randomisation

In total, 564 patients were recruited from 37 centres (43 NHS hospitals). For all patients recruited to the study, written informed consent was obtained.

Randomisation occurred centrally through the Warwick Clinical Trials Unit, where the study was being managed. Treatment allocation was performed using a minimisation algorithm and was stratified by the following: centre, timing of planned ND (before or after CRT), chemotherapy schedule {concomitant platinum, concomitant cetuximab [Erbitux^® (Merck Biopharma, Darmstadt, Germany)], neoadjuvant and concomitant platinum, neoadjuvant docetaxel [Taxotere^® (Sanofi-Aventis, Gentilly, France)] and platinum and 5-fluorouracil (TPF) with concomitant platinum} disease site (oropharyngeal, laryngeal, oral, hypopharyngeal or occult), tumour (T) stage (T1–T2 vs. T3–T4) and nodal (N) stage (N2a–N2b vs. N2c–N3) (Figure 1).

FIGURE 1.

Trial schema. a, EUA; b, if the patient’s primary tumour showed no response or an incomplete response to CRT or the patient developed a recurrence in primary site or neck, the patient was immediately referred back to the MDT for consideration for salvage surgery.

At randomisation, the eligibility of the patient was checked and the trial number and treatment arm allocated. Before being informed of the random allocation, each patient was asked to complete a quality-of-life questionnaire. Following randomisation, the patient’s general practitioner (GP) was notified of study participation by letter.

Treatment

Timing of neck dissection

If the ND was to be performed before CRT, it had to be performed within 4 weeks of randomisation.

If the ND was to be performed after CRT, it had to be performed 4–8 weeks after completion of CRT. It was recommended that patients undergo assessment by CT when possible to assess the primary site (but this was not mandatory) and clinical examination with or without EUA before ND.

Type of neck dissection

The recommended surgical procedure was a modified radical ND. This involves the removal of lymphatic structures in levels I–V, with preservation of one or more of the following: spinal accessory nerve, internal jugular vein and sternocleidomastoid muscle. 54

Selective NDs (including lymphatic tissues in fewer than the five levels I–V) were also acceptable provided the following conditions were met:

1. All clinically evident involved nodal groups were included in the ND.

2. The lymphatic tissues of levels I–V that had not been removed in the ND were included in the radiotherapy fields to a minimum of 50 Gy.

The type of access incision was left to the surgeon to decide.

Three-month post-chemoradiotherapy assessment

All patients were assessed for response after CRT. In the case of patients randomised to the PET–CT (surveillance) arm, this assessment took place 9–13 weeks after completion of CRT. In patients who were randomised to ND (control) arm and who received ND before CRT, assessment was performed 9–13 weeks after CRT. In patients in the control arm who received ND after CRT, assessment for response to CRT was done 4 weeks prior to the ND.

The timing of the assessment was chosen to reflect literature findings that PET–CT is most accurate when done at least 8 weeks post CRT, and preferably > 12 weeks after CRT. However, this was balanced by the need to perform the ND as soon as possible to avoid regrowth of the residual tumour post CRT, and the increased technical difficulty in performing ND later than 12 weeks post CRT because of ensuing fibrosis and scarring. As this was a pragmatic trial, some flexibility was afforded.

At any time, if a patient was found to have residual primary disease, he or she was immediately referred back to the treating clinician and MDT for consideration for salvage treatment.

Recommended clinical guidelines for PET–CT scanning

1. Patient scheduling:

   • PET–CT should have ideally been performed between 10 and 12 weeks after completion of the last dose of CRT. However, it could be performed between 9 and 13 weeks after completion of CRT and before any biopsies.

2. Preparation:

   • Patients had to fast for 6 hours prior to the scan.

   • Patients had to be weighed without shoes and coats (ensuring that the weighing device was calibrated).

   • Blood glucose had to be recorded using Boehringer Mannheim’s Glucometer (Boehringer Mannheim, Mannheim, Germany) (ensuring that the device was calibrated).

   • Patients had to drink 2–3 glasses of water prior to the test to ensure hydration.

   • Metal denture fixtures had to be removed whenever possible (to reduce CT artefacts and improve semiquantitative accuracy).

3. Injection:

   • Fludeoxyglucose had to be injected via a butterfly cannula under quiet conditions.

   • The patient had to be asked to remain silent.

   • The injected dose had to be 4.5 MBq of fludeoxyglucose per kilogram of body weight up to a maximum of 400 MBq.

4. Fludeoxyglucose uptake:

   • The patient had to remain inactive in a comfortable and quiet environment during the uptake.

   • The patient had to empty their bladder just prior to positioning on scanner bed.

   • The emission scan had to start at 90 minutes post injection.

5. Positioning:

   • The patient had to be scanned on a regular couch top.

   • Scanning had to begin at the skull vertex and end at the groin.

   • The scan had to be done with arms down if a single whole-body scan was performed. If the body was scanned separately from the H&N, then the body had to be scanned with the arms up.

6. Acquisition parameters:

   • Each PET–CT centre was advised to use routine local protocols.

7. Reconstruction parameters:

   • The PET–CT centre was advised to use ordered subset expectation maximisation, with CT for attenuation correction using local routine parameters.

8. Archive:

   • The reconstructed CT, PET attenuation-corrected and non-attenuation-corrected data had to be archived locally.

9. QA and quality control:

   • This was done under an agreed QA and quality control protocol.

10. Reporting PET–CT findings:

    • A standardised criterion was used for reporting PET–CT findings.

    • The PET–CT findings had to be reported by the local hospital imaging team.

    • The PET–CT findings had to be reviewed at the central laboratory by an expert in the PET core laboratory, who was independent of the local report.

    • Differences in reporting between the local hospital and the central laboratory were resolved by consensus, otherwise a third expert at the core laboratory reported on the scan and the majority view was taken.

11. PET–CT data transfer to the central laboratory:

    This was under an agreed protocol transfer following reconstructed and anonymised files:

    • CT

    • PET attenuation corrected

    • PET non-attenuation corrected

    • PET–CT report from local imaging team.

12. Adverse events and serious adverse events (SAEs) during PET CT:

    • Adverse events and SAEs were required to be recorded and documented by the local investigational team in accordance with agreed protocols. However, none of these was reported.

Quality assurance review of PET–CT scans

For patients in the PET–CT arm of the trial, the PET–CT scan was reviewed by both the local reporter and the trial core laboratory radiologist. When there were significant differences between the reports of the local PET–CT clinician and the central PET–CT reviewer that might have affected patient management, the central PET–CT reviewer (from the core laboratory) and the participating centre radiologist conferred and a final report was agreed. If agreement could not be reached, a second independent reviewer from the core laboratory was asked to review the scans and issue a second opinion report. If there was still a significant difference between the local PET–CT report and the core laboratory reports, the final decision on which report to use rested with the local PET–CT clinician, local treating clinicians and MDT.

Any discordance between the reports of the local PET–CT clinician and central PET–CT reviewer was documented in the patient’s assessment form including which report was used to make the final clinical decision.

Patient assessments

The schedule of assessments is shown in Table 2. All biobank samples collected at baseline and follow-up will be retained for future work.

Patient follow-up

Patient follow-up was monthly for the first year up to the 12th month after randomisation, and 2-monthly for the second year up to the 24th month after randomisation. It was recommended, but not mandatory, that patients should have chest radiography annually.

At each follow-up visit, the patient received a full clinical examination with palpation of the neck and visualisation of primary site when possible, with appropriate trial forms and questionnaires completed at the completion of CRT, and at 6, 12 and 24 months after randomisation.

Adverse effects of PET–CT scanning

If, after administration of the radiopharmaceutical required for PET–CT, the biodistribution within the patient was not as would be expected, the ARSAC certificate holder had to be informed and a check made to determine that the correct radiopharmaceutical was administered. If the correct radiopharmaceutical was administered, the supplier of the radiopharmaceutical had to be informed.

In the case of an incorrect radiopharmaceutical being administered to a patient, the ARSAC licence certificate holder and a senior member of staff had to be informed immediately who was then required to follow the guidance in Health and Safety Guidance Number 95. 55

1. The ARSAC certificate holder was required to inform the patient of the error.

2. The referring clinician had to be informed.

3. The GP of the patient had to be informed.

4. A risk assessment was required to be requested from the radiation protection advisor.

5. The staff member was required to inform the chief investigator and the sponsor.

A report of any incident, including action taken in order to prevent a recurrence, was required to be compiled by the radiation protection supervisor and forwarded to the radiation protection advisor. Any such incident was required to be recorded as an unexpected SAE and also reported through the individual trust critical incident policies.

Overall survival (primary outcome)

Information on death and survival was obtained from centres via death and follow-up forms.

Cost-effectiveness (primary outcome)

Data for the health economic analysis were collected from the trial EuroQoL-5 Dimensions (EQ-5D) and resource use questionnaires, as well as the case report forms. Full details of the health economic analysis are given in Chapter 4.

Complication rates

Complications following ND surgery were reported.

Disease-specific survival

Causes of death were reported on the death form. Deaths were regarded as caused by H&N cancer if they were reported as such, if there was persistent, recurrent or metastatic disease present at death or if the patient died as a result of complications of treatment for H&N cancer.

Recurrence in neck nodes

Details of recurrences were reported at follow-up visits. Any notification of recurrence within 3 months of radiotherapy was regarded as persistent disease and notifications after that date were regarded as recurrences. Recurrences in the neck nodes were reported as ipsilateral or contralateral in the notification of recurrence.

Quality of life

The questionnaires included the following instruments:

• EQ-5D: five 3-point scales and one summary 100-point scale. This was used for the health economics evaluation

• European Organisation for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire for Cancer (with 30 questions) (QLQ-C30): five functional, three symptom and a global scale and six single items for assessment of general quality of life

• EORTC Quality of Life Questionnaire for Cancer head and neck module with 35 questions (QLQ-H&N35): seven scales and 11 single items for H&N cancer-related quality of life

• MD Anderson Dysphagia Inventory (MDADI): an overall function scale and four subscales.

Patients were requested to complete these questionnaires at five time points. The first was completed in clinic prior to randomisation. Later questionnaires were received either in clinic or through the post before the assessment dates were due at these time points: within 2 weeks of CRT completion, 6 months after randomisation, and 12 and 24 months after randomisation.

Accuracy of PET–CT scanning

Positron emission tomography–computerised tomography scans were assessed both by the local radiologist and by the trial core radiologist.

Recruitment

A total of 564 patients were recruited between 2 October 2007 and 23 August 2012: 282 to each trial arm (Figure 2).

FIGURE 2.

Recruitment.

Two patients in the surveillance arm were subsequently found to be ineligible: one patient’s disease was found to be T1N1 and another patient had metastatic disease at the time of randomisation. Follow-up data were collected and the patients are included in analyses.

The participant flow diagram outlines the passage of patients through the study; more detail is given in later sections (Figure 3).

FIGURE 3.

Participant flow diagram. (a) ND arm; and (b) surveillance arm. Adapted from the New England Journal of Medicine, Mehanna H, Wong W-L, McConkey CC, Rahman JK, Robinson M, Hartley AGJ, PET-CT Surveillance versus Neck Dissection in Advanced Head and Neck Cancer, 374, 1444–54. Copyright © (2016) Massachusetts Medical Society. Reprinted with permission. 62

Screening

All patients newly diagnosed with HNSCC were considered potentially eligible and had to be screened by the MDT prior to their clinic appointment.

Both screened patients and those approached for study participation had to be recorded anonymously on the screening log. The log was updated monthly and passed to the co-ordinating centre.

If a screened patient was not eligible for the trial, an anonymous record of the case was recorded on the screening log. Recorded details included the name of the centre, the patient’s initials and the reason for non-randomisation, if not randomised. Patients randomised to the trial were also recorded.

In total, 1792 patients were screened. The number of patients screened by centre is given in Table 3.

Of the 1792 patients screened, 564 were randomised, 1032 did not fulfil the eligibility criteria, and 196 patients refused consent. Table 4 shows reasons for patients not meeting the standard eligibility criteria and Table 5 shows reasons for unwillingness of eligible patients to enter the study.

The majority of patients who were ineligible either had the wrong type or severity of disease or a different type of treatment was indicated. There were 47 patients who were not approached about the study for reasons such as cognitive impairment, psychiatric issues or short-term memory problems that were likely to affect compliance. Other examples include language barriers, too much information to give to a patient as a result of having to tell them about their disease at the same time as giving study information, lack of clinicians available to discuss trial/consent patients, patient travelling elsewhere for treatment, a patient refusing to be treated or a patient being entered into another incompatible study. For 30 patients, the reason for ineligibility was not known.

Of the 196 patients who refused to participate, 74 did not give a reason for non-participation. The reasons for non-participation of the other 122 patients are given below.

One patient was anxious about being randomised in the trial, 11 patients were not interested in participating in a clinical trial, two patients did not want to travel for PET–CT if randomised to the surveillance arm of the trial, one patient felt that more imaging would be too much, four patients were overwhelmed with the amount of information given (trial information as well as diagnosis and treatment information), seven patients were too distressed after receiving diagnosis information to make a decision about the trial, one patient requested treatment at another clinic, one patient wanted to remain private, one patient wanted to retain control over his/her treatment and 93 patients had specific treatment preferences. Of these, 42 patients preferred ND and 36 patients did not want ND. There were two patients whose treatment preference was not specified. One patient wanted radiotherapy only, one wanted photodynamic therapy, four wanted CRT only, one wanted CRT followed by PET and one patient wanted to proceed immediately to adjuvant chemotherapy. Finally, five patients wanted standard CRT with ND, before or after CRT.

Baseline characteristics of participants by trial arm

Treatment allocation by minimisation was balanced by the six characteristics in Table 6. A total of 38 centres took part, randomising mainly oropharyngeal patients (84.4%). Neoadjuvant chemotherapy was intended in 35.8% of patients.

Other baseline characteristics of participants by trial arm

The patient characteristics in Table 7 were collected either at randomisation or via baseline information forms. The mean age was 57.9 years and 83.3% were male. Three-quarters of participants were either current or past smokers and, of those tested for p16 status, 335 out of 446 (75%) were positive.

Further details of the tumour and nodal stages of randomised patients are given in Tables 8 and 9.

Chemotherapy schedules planned at randomisation

Planned chemotherapy schedules are counted separately for patients in whom ND was planned to take place before CRT (Table 12) and for those in whom ND was planned to take place after CRT (Tables 13 and 14). Neoadjuvant schedules were more frequent when the planned timing was ND after CRT. A slightly higher proportion of patients in the ND arm whose ND was planned to be before CRT had a planned concomitant platin schedule.

Type of chemotherapy delivered

Chemotherapy details were received for 272 patients in the ND arm and 277 patients in the surveillance arm. Overall, 10 ND patients and five surveillance patients did not have chemotherapy for the reasons given in Table 15.

One patient in the ND arm, who was planned (at randomisation) to receive neoadjuvant and concomitant chemotherapy, received only concomitant chemotherapy. At the same time, two patients randomised to concomitant platinum received only neoadjuvant and concomitant therapy.

Individual drug details were available for 269 patients in the ND arm and 274 patients in the surveillance arm (Table 16). The type of chemotherapy delivered was very similar in the trial arms.

Chemotherapy dose information

Table 17 gives median doses received as reported on the CRT chemotherapy forms.

Patients in neck dissection arm who did not undergo a neck dissection

Of the 282 patients in the ND arm, 221 underwent a planned ND. In total, 25 refused to have surgery. A summary of reasons for the 61 who did not receive a planned ND is given in Table 21.

Summary of surveillance arm patients who had a neck dissection

According to the trial protocol, patients in the surveillance arm should undergo a ND if the PET–CT findings are positive or equivocal. Fifty-four patients underwent ND, with 48 (36 incomplete responders in nodes and 12 incomplete responders in primary and nodes) of them conforming to protocol in that PET–CT was positive or equivocal. Of the remaining six, two patients did not undergo PET–CT, in two patients PET–CT findings were equivocal or node positive in the central reviewer’s PET–CT read and in two cases the reason is unknown.

Summary of surveillance arm patients who did not have a neck dissection

There were 228 patients in the surveillance arm who had no ND (Table 22).

Surgical details

Surgery forms were received for 275 patients who received ND surgery (Table 23).

From the above, it is important to note that, overall, the number of participants who experienced nerve sacrifice (which is morbid) was considerably lower in the surveillance arm (4 out of 220) than in the ND arm (22 out of 220); similar findings were obtained for sacrifice of the internal jugular vein and sternocleidomastoid muscle.

If the numbers of structures removed are summed (counting potentially one for left and one for right), the mean number of structures removed per ND is 0.59 in the ND arm and 0.56 in the surveillance arm. The mean number of structures removed in the ND before group is 0.78, whereas in the ND after group it is 0.49. Combining the five structures given, fewer patients whose planned ND was after CRT had one or more of the structures removed (see Table 23, p = 0.02). On examination of these data, this does not correlate with more advanced disease.

Summary

In the study, 282 SAEs were reported and are summarised in Table 25. There were 169 in the ND arm and 113 in the PET–CT arm; none was deemed unexpected by the chief investigator.

Tables 26 and 27 show that the most common SAE was hospitalisation or prolongation of hospitalisation due to CRT, which was very high for ND post CRT. The rates of other SAEs are similar in the ND before and after CRT trial arms and in the surveillance ND arm.

Of the SAEs with hospitalisation as a result of to CRT in the ND after CRT group, 67 out of 71 patients (94.4%) had their SAE before surgery (Table 28).

The highest symptom grades of the events are given comparing trial arms in Table 29 and comparing ND before CRT with ND after CRT in Table 30. No difference is seen between groups in either case (p = 0.99 and p = 0.42 for trend).

Primary outcome overall survival

The prespecified analysis takes account of timing of planned ND by stratification (planned ND before CRT and planned ND after CRT). These are plotted separately in Figures 5 and 6, respectively.

FIGURE 5.

Overall survival by treatment arm for planned ND before CRT. Reproduced from the New England Journal of Medicine, Mehanna H, Wong W-L, McConkey CC, Rahman JK, Robinson M, Hartley AGJ, PET-CT Surveillance versus Neck Dissection in Advanced Head and Neck Cancer, 374, 1444–54. Copyright © (2016) Massachusetts Medical Society. Reprinted with permission. 62

FIGURE 6.

Overall survival by treatment arm for planned ND after CRT. Reproduced from the New England Journal of Medicine, Mehanna H, Wong W-L, McConkey CC, Rahman JK, Robinson M, Hartley AGJ, PET-CT Surveillance versus Neck Dissection in Advanced Head and Neck Cancer, 374, 1444–54. Copyright © (2016) Massachusetts Medical Society. Reprinted with permission. 62

Analysing by proportional hazards model (Table 31) stratified by intended timing of ND before or after CRT, the HR is 0.924 (95% CI 0.648 to 1.318) in favour of the surveillance arm. The limit for non-inferiority in the trial design was set at a HR of 1.50, which lies at the 99.63th percentile of the estimated CI for the HR, which can be interpreted as a one-sided p-value of 0.0037, that is, indicating non-inferiority (HR < 1.5). The conventional two-sided p-value for difference between treatments is 0.6621.

Further analyses of overall survival

The early drop in the Kaplan–Meier curve of the group that had a planned ND before CRT suggests some non-proportionality of hazards between treatment groups. This is confirmed by testing an interaction between treatment and log-(time) (p-value for the time interaction is 0.03). No evidence of non-proportionality is seen in the planned ND after CRT stratum (p = 0.38). An alternative analysis to the Cox’s model is stratifying analysis into two risk periods, namely the first year and rest of the period at risk. The result of this additional unplanned analysis gives a very similar HR of 0.922 (instead of 0.924 in Table 31).

Adjusting for centre, T stage, N stage, site of primary, chemotherapy schedule, sex and age gives a very similar result, as does further adjustment for p16 status on a reduced sample (Table 32).

p16 status was highly prognostic, but there was no difference between p16-positive and p16-negative status in terms of the trial result, as shown in Table 32 and Figures 7 and 8.

FIGURE 7.

Overall survival: p16 positive.

FIGURE 8.

Overall survival: p16 negative.

The HRs in Figure 9 are stratified by planned ND timing (before or after CRT). Some of the groups are small and differences between them should be interpreted cautiously. There is some evidence that the ND arm did better in the T stage 1/2 group than in the T3–T4 group. The small group of females in the trial did particularly well in the surveillance arm (treatment interaction p = 0.01), although with a total of only 18 deaths this is not a reliable result. There was no difference in the trial result by p16 status, which was highly prognostic. In unplanned analyses, the ND arm did worse in the first year at risk, and no difference was seen in the trial result by radiotherapy dose, IMRT versus conformal radiotherapy or high- versus low-recruiting centres.

FIGURE 9.

Overall survival of subgroups. Method as described in Early Breast Cancer Trialists’ Collaborative Group. Treatment of Early Breast Cancer. Volume 1. Worldwide Evidence 1985–1990. Oxford: Oxford University Press; 1990. 57 (a) Overall survival of subgroups 1; and (b) overall survival of subgroups 2. Adapted from the New England Journal of Medicine, Mehanna H, Wong W-L, McConkey CC, Rahman JK, Robinson M, Hartley AGJ, PET-CT Surveillance versus Neck Dissection in Advanced Head and Neck Cancer, 374, 1444–54. Copyright © (2016) Massachusetts Medical Society. Reprinted with permission. 62

Deaths attributable to head and neck cancer

The causes of death are summarised in Table 33. These include complications attributable to surgery or CRT and persistent, recurrent or metastatic H&N cancer at death. There were 92 deaths due to H&N cancer and 29 deaths from other causes (detailed in Table 34).

Head and neck cancer mortality: cumulative incidence

There is little difference between the trial arms in the incidence of H&N cancer-related deaths (Figure 10).

FIGURE 10.

Mortality attributable to H&N cancer.

The 2-year cumulative incidence rates for H&N mortality are 13.66% (95% CI 9.88% to 18.04%) in the ND arm and 12.25% (95% CI 8.72% to 16.42%) in the surveillance arm. A Gray’s test for difference between cumulative incidence functions gives a p-value of 0.7992.

Other cause mortality: cumulative incidence

The difference in non-cancer deaths is also small (Figure 11).

FIGURE 11.

Mortality attributable to causes other than H&N cancer.

The 2-year cumulative incidence rates for other causes of mortality are 4.80% (95% CI 2.68% to 7.82%) in the ND arm and 2.88% (95% CI 1.36% to 5.37%) in the surveillance arm. A Gray’s test for difference between cumulative incidence functions gives a p-value of 0.4124.

Completion of quality-of-life questionnaires

The total return rate of quality-of-life questionnaires, based on the number that could have been received if all surviving patients completed all of them, was 76.4% in the ND arm and 80.6% in the surveillance arm. Generally, the return rates (Table 39) were at least 70% at each time point and they were returned at the required times.

The patients who responded to the questionnaire at 24 months were broadly similar in terms of baseline characteristics to those living patients from whom there was no response, but the non-responders had slightly worse survival beyond 24 months.

Questionnaires in both trial arms were completed at the designated times (Table 40). The questions were well answered, allowing most scales to be computable in the majority of cases (Table 41). The questionnaires were well completed, with most scales computable in the majority of cases.

Baseline quality-of-life questionnaire scores

The quality-of-life, H&N function and dysphagia-related scores were well balanced between the treatment arms (Table 42).

The European Organisation for Research and Treatment of Cancer’s Quality of Life Questionnaire for Cancer (with 30 questions) quality-of-life outcomes

In Tables 43–47, ‘mean treatment difference’ is the mean change from baseline in the surveillance arm minus the mean change from baseline in the ND arm. For these preplanned analyses of quality-of-life scores, no p-value was specified for determining a significant difference. In Table 43, global and functioning scales are low = bad, high = good. So, a positive mean treatment difference is good for the PET–CT arm. There are no clear differences as, for example, in global health status depicted in Figure 13, but results are slightly more favourable for the surveillance arm. It suggests that the symptoms and quality-of-life effects of ND are less impactful that those of the CRT.

FIGURE 13.

The EORTC’s QLQ-C30 global health status. Reproduced from the New England Journal of Medicine, Mehanna H, Wong W-L, McConkey CC, Rahman JK, Robinson M, Hartley AGJ, PET-CT Surveillance versus Neck Dissection in Advanced Head and Neck Cancer, 374, 1444–54. Copyright © (2016) Massachusetts Medical Society. Reprinted with permission. 62

The European Organisation for Research and Treatment of Cancer’s Quality of Life Questionnaire for Cancer (with 30 questions) symptom scales

For simplicity of interpretation the symptom scales in Table 44 have been transformed so that, here also, a positive mean treatment difference is good for the PET–CT arm. There are no clear differences, but results are slightly more favourable for the surveillance arm.

The European Organisation for Research and Treatment of Cancer’s Quality of Life Questionnaire for Cancer head and neck module with 35 questions quality-of-life outcomes

All scales in Table 45 have been set to low as bad and high as good, so a positive difference in mean treatment difference is good for the PET–CT arm. There is little to suggest differences between trial treatments. No difference is seen in pain in H&N cancer and in problems with swallowing (Figures 14 and 15). The difference in problems with teeth is small (Figure 16). Reported use of painkillers at 12 months was higher in the ND arm (Figure 17).

FIGURE 14.

The EORTC’s H&N35 pain in H&N cancer: mean scores unadjusted.

FIGURE 15.

The EORTC’s H&N35 problems with swallowing: mean scores unadjusted.

FIGURE 16.

The EORTC’s H&N35 problems with teeth: mean scores unadjusted.

FIGURE 17.

The EORTC’s H&N35 used painkillers: mean scores unadjusted (higher score = less pain).

The MD Anderson Dysphagia Inventory dysphagia scales

All scales in Table 46 have been set to low as bad and high as good, so a positive difference in mean treatment difference is good for the PET–CT arm. There is little to suggest differences between trial treatments (see, e.g., Figure 18).

FIGURE 18.

The MDADI dysphagia total: mean scores unadjusted.

European Quality of Life-5 Dimensions

The EQ-5D is used for the health resource study. In Table 47, positive values for mean treatment difference indicate a better response in the PET–CT arm.

Differences of at least 10% in quality-of-life scales

Another way of presenting these data is to display differences from baseline of at least 10%. Figures 19–25 compare the randomised groups on that basis. Tables 48–54 give similar information, comparing patients who underwent ND before CRT with those who did not. The swallowing measures in Figures 21, 24 and 25 show no difference between trial treatments, with both having a low proportion of improvers.

FIGURE 19.

The EORTC’s QLQ-C30 global health status, percentages changed by ≥ 10%.

FIGURE 20.

The EORTC’s QLQ H&N35 pain scale, percentages changed by ≥ 10%.

FIGURE 21.

The EORTC’s QLQ H&N35 swallowing scale, percentages changed by ≥ 10%.

FIGURE 22.

The EORTC’s QLQ H&N35 problems with teeth, percentages changed by ≥ 10%.

FIGURE 23.

The EORTC’s QLQ H&N35 use of painkillers, percentages changed by ≥ 10%.

FIGURE 24.

The MDADI dysphagia global scale, percentages changed by ≥ 10%.

FIGURE 25.

The MDADI dysphagia total scale, percentages changed by ≥ 10%.

A slightly greater percentage of patients who had an early ND had a decreased global health status. This is not generally the case for the more specific symptom scores in Figures 20–25.

p16 subgroups

The baseline scores are better in the p16-positive group than in the negative group. No difference is seen between treatment groups, but the plots in Figures 26 and 27 suggest that the p16-positive group is affected more by CRT than the p16-negative group.

FIGURE 26.

The EORTC’s QLQ-C30 global health status: p16-positive group.

FIGURE 27.

The EORTC’s QLQ-C30 global health status: p16-negative group.

Concordance rate for nodal disease

The concordance rate, on the basis of the readings reported by the local sites, is 247 out of 270, or 91.5%, but on review and a correct reading of the local report, the last group in the table (3 + 4 + 6 + 2 + 1) should be interpreted as concordant, giving a rate of 263 out of 270, or 97.4%. In total, 243 patients are concordant in primary tumour and neck nodes, 20 are concordant in neck nodes only and seven are discordant in neck nodes. Therefore, the concordance rate for the sites and the central laboratory is 97.4%, with a kappa value of 0.78 (95% CI 0.69 to 0.87). The summary assessments are related to NDs performed in Table 56.

Concordance in primary tumour

Similarly, for primary tumour the concordance rate for the sites and the central laboratory is 243 out of 266 or 91.4%, with a kappa value of 0.63 (95% CI 0.49 to 0.77).

True-negative rate

The number of patients in whom no neck cancer was seen on PET–CT on review and for whom no ND was carried out was 191. Among this group, 26 experienced recurrence, giving a true-negative rate of 86.4%. Only one recurrence was in the neck alone. Similarly, the true-negative rate for primary tumour was 29 out of 192 (84.9%).

Within-trial analysis

Individual patient data collected in the trial were used to determine the costs and quality-adjusted life-years (QALYs) associated with each treatment arm. QALYs were derived using patient EQ-5D questionnaire responses, and costs were calculated using information collected in case report forms and patient- and carer-reported questionnaires. Cost-effectiveness was assessed as the incremental cost-effectiveness ratio (ICER) and incremental net health benefit (INB). Future costs and health outcomes (beyond 1 year) were discounted at an annual rate of 3.5%, as per the National Institute for Health and Care Excellence (NICE)’s Guide to the Methods of Technology Appraisal 2013. 63

Lifetime decision model analysis

In order to estimate the lifetime cost-effectiveness of PET–CT-guided management, a de novo decision-analytical model was constructed. In line with the WT analysis, the base-case model adopts a UK secondary care perspective and future costs and QALYs were discounted at an annual rate of 3.5% in line with NICE guidance. 63 A sensitivity analysis was conducted in which a broader NHS and PSS perspective was adopted. The model was built and analysed using R software.

Model structure

The model structure (Figure 28) is split into two phases: an initial 6-month treatment phase, in which patients receive CRT and (potential) ND; and a follow-up phase, in which patients may go on to recover, develop local recurrence (LR) or distant recurrence (DR), or die. Within the treatment period, cost and QALYs for each of the treatment arms are taken directly from the results of the first 6 months of the WT analysis. For the follow-up period, longer-term cost and QALYs for each of the treatment arms are estimated using a modified Markov model. The model begins at this point in order to enable the analysis to capture the full cost and utility decrements associated with LRs and DRs, as these data were not fully captured in the trial.

FIGURE 28.

Lifetime DM structure.

Markov models describe patient progression over time through a pathway of health states, with movement between the health states being triggered by events such as disease progression or death. Resource use and costs are associated with each health state, and patients accumulate costs and health benefits in each state over monthly cycles. The analysis adopts a lifetime horizon, truncated at 100 years.

During the treatment phase of the model, patients in the standard care arm (arm A) receive planned ND either before or after CRT. Patients in the intervention arm (arm B) receive CRT followed by PET–CT at 9–13 weeks post CRT, which dictates whether or not patients go on to receive ND. In the subsequent follow-up phase of the model, patients enter one of four health states: disease free (DF), LR, DR or death. During any cycle in the follow-up period patients in the DF state may remain DF, develop a LR or DR, or die. Similarly, patients in the LR state may recover, develop a DR or die, and patients in the DR state may remain in DR or die. Entry into the LR state is associated with a one-off treatment cost and utility decrement, after which patients either recover (with costs and utility equal to those in the DF state) or advance to the DR state. Patients who enter the DR state similarly incur a one-off treatment cost and further utility decrement and, subsequently, either remain in the DR state (with an ongoing supportive care cost and utility decrement) or die. Once in the DR state, it is assumed that patients cannot recover and either remain in the DR state or die.

Uncertainty and sensitivity analyses

Probabilistic sensitivity analyses were conducted to assess the impact of joint parameter uncertainty on the results. Probabilistic analysis accounts for joint parameter uncertainty in non-linear models by assigning probability distributions to each of the input parameters and randomly drawing from these probabilities over 10,000 Monte Carlo model simulations to produce different cost and QALY estimates in each simulation of the model. As for the bootstrap WT analysis, the results are presented on the cost-effectiveness plane as a scatterplot, using cost-effectiveness acceptability curves to show the probability that the two strategies are cost-effective across different willingness-to-pay per QALY thresholds.

One-way sensitivity analyses were conducted to test the impact of changes in individual parameter estimates on the results. Individual parameters were altered by ± 25% from their baseline value, with the impact on the expected INB reported in a tornado diagram.

For the lifetime DM analysis, three additional sensitivity analyses were conducted to assess the impact of key assumptions used in the model.

Decision model sensitivity analysis 1: imputation of additional patient-reported cost data (NHS and personal social services perspective)

As for the WT cost-effectiveness analysis, the base-case DM analysis was based on utilisation of data from the trial relating to patients’ consumption of NHS secondary care resources, using the trial case report forms, which were collected for the total trial population. In line with the sensitivity analyses conducted in the WT sensitivity analysis 1.1 and sensitivity analysis 1.2 (see Within-trial analysis, Uncertainty and sensitivity analyses), two corresponding sensitivity analyses were conducted in the model analysis to assess the impact of including additional patient-reported cost data:

1. DM sensitivity analysis 1.1: in this analysis the treatment-phase costs for the model were taken from the treatment-phase costs generated from the WT sensitivity analysis that included the additional patient-reported cost data, assuming that patients correctly reported their use of additional secondary care resources (i.e. corresponding to analysis WT sensitivity analysis 1.1). Similarly, the monthly cost of the DF state was recalculated using the results of WT sensitivity analysis 1.1. In addition, the average monthly patient-reported cost associated with the use of primary, community and additional secondary care (i.e. outside the enrolled oncology department) was applied to the local and DR health states. As this is likely to underestimate the primary and community care costs for these states (because the added cost data are derived from primarily DF patients), no other data were available to determine the exact cost of primary and community care for recurrent patients.

2. DM sensitivity analysis 1.2: this analysis involved the same steps as outlined in DM sensitivity analysis 1.1 above, but instead used data from the WT sensitivity analysis, including additional patient-reported data and excluding secondary care resource use where patients identified the visited hospital as one of the trial centres (i.e. WT sensitivity analysis 1.2; see Within-trial analysis, Uncertainty and sensitivity analyses for further details).

Decision model sensitivity analysis 2: visual analogue scale utility values for the local recurrence and distant recurrence states

In the base-case analysis, utility in the recurrence health states was derived using data reported in de Almeida et al. ,84 based on an elicitation of utility values using the standard gamble method. The authors also reported utilities based on using the VAS tool. A sensitivity analysis was conducted using these alternative values in order to assess the impact of the method of utility derivation on the results.

Decision model sensitivity analysis 3: recurrences beyond 5 years post diagnosis estimated using a parametric survival curve

In the base-case analysis, no recurrence events were assumed to occur beyond 5 years in the model. A sensitivity analysis was conducted assuming that recurrences could occur beyond 5 years: long-term recurrence probabilities for the planned ND arm were estimated using a Gompertz parametric survival curve fitted to the trial baseline Kaplan–Meier data on recurrence-free survival; a HR was then applied to this curve in order to derive DF survival within the PET–CT surveillance arm (using the HR observed across the trial follow-up period), as outlined in Briggs et al. 89 The Gompertz distribution was identified as the best-fitting curve to estimate long-term recurrence events (compared with the exponential, Weibull, gamma and log-normal distributions), based on an analysis of the Akaike information criterion (AIC) and Bayesian information criterion (BIC) (Table 60; better-fitting curves are indicated by lower AIC and BIC values) and a visual inspection of the curve fits (Figure 29).

TABLE 60 - Akaike information criterion and BIC to estimate the goodness of fit of alternative parametric survival curve model specifications for the estimation of long-term recurrences

Trial arm	Model specification
	AIC	BIC
Planned ND
Exponential	679.50	683.14
Gamma	674.96	682.96
Weibull	673.83	681.11
Log-normal	665.86	673.14
Gompertz	659.03	666.31
PET–CT surveillance
Exponential	734.08	737.72
Gamma	735.58	742.86
Weibull	735.16	742.44
Gompertz	727.89	735.17
Log-normal	726.19	733.48

FIGURE 29.

Plot of parametric survival curves (using Gompertz and log-normal model specifications) against trial data on patient recurrence-free survival. (a) Planned ND arm; and (b) PET–CT surveillance arm. Log-normal and Gompertz specifications are shown here, as these were the two models identified as having the best fit via an analysis of AIC and BIC. The Gompertz curve was used in the sensitivity analysis, as this had the best overall fit when taking into account AIC, BIC and a visual inspection of the goodness of fit.

Within-trial analysis