Cabozantinib and vandetanib for unresectable locally advanced or metastatic medullary thyroid cancer: a systematic review and economic model

Incidence and prevalence

Thyroid cancer is the most common malignant endocrine tumour, but represents only ≈1% of all malignancies. 1,2 The disease is more common in women than in men. According to Cancer Research UK, 3404 new diagnoses of thyroid cancer were reported in England in 2014: 966 cases (28%) were in men and 2438 cases (72%) were in women. 1 The age-standardised incidence rate of thyroid cancer is reported to be 7 per 100,000 women and 3 per 100,000 men. 1 The UK incidence rate is the 11th lowest in Europe for men and the 15th lowest in Europe for women. The median age at diagnosis is approximately 50 years. 3,4

There are four main types of thyroid cancer: (1) papillary, (2) follicular, (3) medullary and (4) anaplastic. Papillary and follicular thyroid cancer are the most common types of thyroid cancer and account for > 90% of all cases. 3 Medullary thyroid cancer (MTC), the disease type considered in this report, develops from the parafollicular cells (also known as C cells) and commonly presents as a mass in the neck. 2 MTC is very rare and accounts for ≈5% of all thyroid cancers,2 although a lower frequency has been quoted by the American Thyroid Association guidelines. 5 Anaplastic cancers, thyroid lymphomas and metastases to thyroid from other primary tumours are rarer than MTC; anaplastic thyroid cancer accounts for ≈2% of all thyroid cancers. 3 MTC is reported to account for 3% of all thyroid cancers in adults and 10% of all thyroid cancers in children. 2 Based on 2014 estimates of disease incidence,1 the number of new cases of MTC in England in any year would be ≈170 individuals (5% of 3404).

There are four types of MTC: (1) sporadic, (2) multiple endocrine neoplasia (MEN) 2 (formerly MEN 2A), (3) MEN 3 (formerly MEN 2B) and (4) familial MTC. Incidence rates for each type differ by age and sex. 1 Approximately 75% of MTC cases are sporadic in nature, whereas the remaining 25% are genetically determined (MEN 2, MEN 3 and familial MTC). 2,3 The RE-arranged during Transfection (RET) oncogene is central to the development of sporadic and hereditary MTC. 5 Germline testing of the RET oncogene mutation is recommended for all confirmed cases of MTC to establish the possible hereditary basis for the disease within an individual and to facilitate the identification of family members who might be at risk. 2 Almost all patients with MEN 2, MEN 3 and familial MTC have germline RET mutation, whereas approximately 40–50% of patients with sporadic MTC have somatic RET mutations. 2,5 Only germline RET mutation testing is routinely undertaken in the NHS.

Diagnosis and management

In > 75% of cases, patients with MTC will typically present with a lump in the neck (which may represent a thyroid or lymph node mass) or distant metastases. 2 The lumps are not usually associated with other symptoms but may occasionally cause dysphagia (difficulty or discomfort in swallowing) or dysphonia (difficulty in speaking). 2,6 Symptoms might also relate to the effect of metastases, especially diarrhoea, flushing, dyspnoea and bone pain.

Diagnosis is usually made by either fine-needle aspiration cytology of a thyroid nodule or lymph node or core needle biopsy with ultrasound guidance, alongside biochemical investigations of serum-based biomarkers, especially calcitonin (CTN). 2,3,5,7 CTN is the major product secreted by C cells:5 CTN levels of > 100 pg/ml are considered to have a 100% positive predictive value for the presence of MTC. 2,3

The disease is staged and, if appropriate, surgery is performed (usually total thyroidectomy and central compartment node dissection, with or without lateral neck dissection). 2,8,9 Patients with MTC may be classified into three groups: (1) patients with localised disease without evidence of metastases, in whom surgical cure is possible; (2) patients with metastatic disease limited to the neck, in whom surgical cure might be possible but is not always achieved; and (3) patients with distant metastasis, that is the disease has spread outside the neck and in whom surgery is not curative. 3 The only curative treatment for MTC is complete surgical resection, but lymph node or systemic metastases are present at initial diagnosis in around half of MTC cases5 and resection is sometimes incomplete because of extensive lateral spread. 3,4 Patients with unresectable locally advanced or metastatic MTC were the focus of this review. For these patients, the treatment options are limited because MTC is relatively unresponsive to conventional doses of radiation therapy and to all tested chemotherapeutic regimens (see Impact of health problem and Current service provision). 2,3,5 Therefore, patients with symptomatic and progressive disease, according to the Response Evaluation Criteria in Solid Tumours (RECIST) criteria,10 are the principal candidates for systemic treatment. 6

Prognosis

Compared with other advanced solid tumours, MTC can be relatively indolent, but it can sometimes be aggressive; data indicate that survival is influenced by age and stage at diagnosis. 4,5,11 It has been reported that patients who are < 40 years of age at the time of diagnosis have a significantly higher adjusted survival rate than older patients,4,12 and 10-year survival rates are reported to be up to 100% for stage I disease, that is, if tumours are confined to the thyroid gland. 4,5,9,13 In the absence of progressive and symptomatic disease, health-related quality of life (HRQoL) can be maintained for months or years. 2,6 However, reported 10-year survival rates decrease to ≈75% with regional disease spread3,14 and range from 21 to 40% for subjects with metastatic disease at diagnosis. 2,3,5 Distant metastases, which can affect multiple organs, most commonly the liver, lungs and bones, are reported to be present in between 7% and 23% of MTC cases at diagnosis. 3,6 Just under half of all patients with sporadic MTC will present with stage III or IV (advanced) disease. 5

Calcitonin and, to a lesser extent, carcinoembryonic antigen (CEA) are used as biological markers of post-operative MTC burden, progression and survival. 15 CEA levels are not specific to MTC and are less sensitive and less reliable than CTN for diagnosis; however, when measured alongside CTN, they are considered to be potentially useful in assessing disease progression. 5,15 Certain levels of CEA might indicate regional spread to draining lymph nodes or more distant spread to non-regional lymph nodes, but are particularly important as an indicator of disease progression. 3,5 Studies16–20 have indicated that patients with CTN and CEA doubling times of ≤ 24 months have more progressive disease and a reduced survival time compared with patients with CTN and CEA doubling times of > 24 months. A 2005 study16 reported 5- and 10-year survival rates of 25% and 8%, respectively, in MTC patients with post-operative CTN doubling times of < 6 months, compared with 92% and 37%, respectively, in patients with doubling times of between 6 and 24 months. In the same study,16 the 10-year survival rate for patients with CTN doubling times of > 24 months was 100%.

Significance for patients

There is little published research concerning the impact of MTC on patients’ HRQoL. As noted in the Ipsen (Paris, France) submission to the National Institute for Health and Care Excellence (NICE),21 most of the available HRQoL evidence is derived from studies of patients with other more common types of thyroid cancer. MTC is associated with a number of symptoms that may impair patients’ HRQoL, including the presence of a thyroid mass (usually a non-tender thyroid nodule or diffuse thyroid enlargement), cervical lymphadenopathy, airway compromise, pain, dysphagia and dysphonia. Diarrhoea is commonly seen in patients with advanced MTC as a result of hormonal excess caused by increased CTN secretion from the parafollicular cells; this may be debilitating and can lead to problems with nutrition. Distant metastases may result in additional symptoms including spinal cord compression, bone fracture, bronchial obstruction and pain. 5 Debilitating symptoms associated with MTC (e.g. severe diarrhoea) may lead to workplace absence and lost productivity.

Significance for the NHS

Medullary thyroid cancer is a very rare disease and, for many patients, surgery can be curative; hence, the population of patients with advanced or metastatic MTC eligible for treatment with vandetanib and cabozantinib is very small. However, given the list prices of the drugs and the lack of effective alternative treatments, the cost per patient treated may be considerable. Both vandetanib (Caprelsa^®; Cambridge, MA, USA) and cabozantinib (Cometriq^®; Ipsen, Paris, France) are also associated with additional monitoring costs. The Summary of Product Characteristics (SmPC) for vandetanib22 states the following:

An ECG [electrocardiography], and levels of serum potassium, calcium and magnesium and thyroid stimulating hormone (TSH) should be obtained at baseline, at 1, 3, 6 and 12 weeks after starting treatment and every 3 months for at least a year thereafter. This schedule should apply to the period after dose reduction due to QTc [corrected QT interval] prolongation and after dose interruption for more than two weeks. ECGs and blood tests should also be obtained as clinically indicated during this period and afterwards. Frequent ECG monitoring of the QTc interval should be continued.

Serum potassium, serum magnesium and serum calcium should be kept within normal range to reduce the risk of ECG QTc prolongation. Additional monitoring of QTc, electrolytes and renal function are required especially in case of diarrhoea, increase in diarrhoea/dehydration, electrolyte imbalance and/or impaired renal function. If QTc increases markedly but stays below 500 msec, cardiologist advice should be sought.

European Medicines Agency, vandetanib SmPC. © EMA [1995–2018]. Reproduced with permission from the European Medicines Agency22

The SmPC for cabozantinib23 also recommends close monitoring during the first 8 weeks of treatment:

As most events can occur early in the course of treatment, the physician should evaluate the patient closely during the first eight weeks of treatment to determine if dose modifications are warranted. Events that generally have early onset include hypocalcaemia, hypokalaemia, thrombocytopenia, hypertension, palmarplantar erythrodysaesthesia syndrome (PPES), and gastrointestinal (GI) events (abdominal or mouth pain, mucosal inflammation, constipation, diarrhoea, vomiting).

European Medicines Agency, cabozantinib SmPC. © EMA [1995–2018]. Reproduced with permission from the European Medicines Agency23

One of the clinical advisors to the assessment group (AG) noted that although cardiac toxicity is less for cabozantinib than vandetanib, electrocardiographic monitoring may also be required.

Clinical guidelines

There are no clinical guidelines for the management of MTC in the UK. A NICE quality standard for head and neck cancer has recently been published;24 however, this does not include the management of MTC.

Current National Institute for Health and Care Excellence technology appraisal guidance

There is currently no NICE technology appraisal guidance for interventions for the treatment of unresectable locally advanced or metastatic MTC.

Current service cost

The current cost of managing MTC is uncertain. However, MTC is a very rare disease, with an estimated annual incidence for England of around 170 new patients. 1 Prescribing data from the Cancer Drugs Fund (CDF) indicate that in 2016 (confidential information has been removed) new patients received vandetanib and (confidential information has been removed) new patients received cabozantinib (Professor Peter Clark, Chairperson of CDF, 2017, personal communication). The data from 2015 indicate very similar prescribing levels, with (confidential information has been removed) new patients starting vandetanib and (confidential information has been removed) patients starting cabozantinib (Professor Peter Clark, personal communication). Based on current prescribing levels, the cost of treating new MTC patients with cabozantinib and vandetanib for 1 year (assuming full dose and excluding any discontinuation) is approximately £1.96M.

Variation in services and uncertainty about best practice

Clinical advisors to the AG noted that although the indications set out in the marketing authorisations for cabozantinib and vandetanib22,23 relate to patients with progressive disease, this may be determined on the basis of radiographic evidence or the presence of symptomatic disease. They also noted that, elsewhere in Europe, clinicians often initiate treatment earlier on the basis of imaging, whereas clinicians in the UK tend to consider symptomatic progression as the more important time point at which to initiate palliative treatment.

The SmPCs for both vandetanib and cabozantinib state that ‘For patients in whom Rearranged during Transfection (RET) mutation status is not known or is negative, a possible lower benefit should be taken into account before individual treatment decision’ (p. 2) (© EMA [1995–2018]. Reproduced with permission from the European Medicines Agency). 22,23 Clinical advisors to the AG noted that all patients should have an assessment of their germline RET status to check if their disease is sporadic or genetic. This is, however, different from checking if the tumour expresses RET (somatic RET mutation testing). In the UK, it is not routine practice to check the tumour (either primary or metastases) for RET mutations. Although clinicians do not currently have routine access to mutation analysis, this may change in the future. The clinical advisors warned that the RET status of the primary thyroid cancer may not reflect the mutation landscape in the metastases and that it would be inadvisable to base recommendations about the use of vandetanib and cabozantinib in the NHS on RET mutation status without a full and accurate picture of the significance of somatic RET status. Furthermore, the clinicians commented that the thyroid primary may have been removed many years before metastases develop; hence, at the time of relapse, the mutation analysis may no longer be accurate. In addition, as cabozantinib and vandetanib have multiple targets, although a patient may be RET mutation negative in the metastases, they may still obtain a treatment response by virtue of other mutations that are targeted by the individual drug received.

Current treatment pathway

A summary of the treatment pathway, as developed by the AG, is presented in Figure 1. For patients who are ineligible to receive cabozantinib or vandetanib, treatment is likely to comprise palliative treatments. Both cabozantinib and vandetanib are currently available on the CDF as first-line treatments for unresectable, locally advanced or metastatic MTC. 25 The CDF indication for each therapy is the same, as shown in Box 1.

FIGURE 1.

Current treatment pathway for adults with symptomatic and progressive MTC. a, Patient may switch to other TKI if intolerant or severe AEs are experienced within 3 months. Note that the vandetanib licence is in aggressive and symptomatic disease, whereas the licence for cabozantinib is for progressive, unresectable locally advanced or metastatic MTC. b, May include palliative surgery, palliative radiotherapy and/or treatments for bone pain. c, Nuclear medicine therapies, such as MIBG/dotatate, may be considered in some patients. AE, adverse event; BSC, best supportive care; GI, gastrointestinal; MIBG, iodine-123 metaiodobenzylguanidine; TKI, tyrosine kinase inhibitor.

Interventions considered in the scope of this report

This assessment includes two interventions: cabozantinib and vandetanib.

Mode of action

Current usage in the NHS

As noted in Current service cost, both cabozantinib and vandetanib are currently available for use through the CDF. Given the rarity of MTC, total prescribing rates of these products are low. In 2016, (confidential information has been removed) new patients were prescribed cabozantinib or vandetanib through the CDF.

Population

• Adults with unresectable locally advanced or metastatic MTC.

  In December 2016, the marketing authorisation for vandetanib was extended to include adolescents and children aged ≥ 5 years;22 this population is beyond the scope of this appraisal. 26 Clinical advisors to the AG note that the incidence of unresectable locally advanced or metastatic MTC in children and adolescents aged ≥ 5 years is expected to be extremely low.

Interventions

• Cabozantinib (oral).

• Vandetanib (oral).

Relevant comparators

Cabozantinib and vandetanib were compared with:

• each other

• best supportive care (BSC).

Outcomes

The following outcomes are included in this assessment:

• overall survival (OS)

• progression-free survival (PFS)

• response rates

• adverse effects of treatment

• HRQoL.

Although response rates were not included in the final NICE scope,26 this outcome has been included in this assessment as it is a clinically relevant end point in the key trials considered in this report. 27,28

Subgroups

The final NICE scope26 states that ‘If the evidence allows subgroups according to RET mutation status will be considered.’ Based on the guidance of the clinical advisors to the AG (see Chapter 1, Variation in services and uncertainty about best practice), RET mutation status has not been considered within the health economic analysis presented in this report.

Inclusion criteria

The inclusion criteria for the reviews are described in Table 1. These criteria are in accordance with the decision problem set out in the final NICE scope. 26

Searches

A comprehensive literature search was undertaken to systematically identify randomised controlled trials (RCTs) and systematic reviews (for the identification of additional trials) of the clinical effectiveness of cabozantinib and vandetanib for the treatment of unresectable locally advanced or metastatic MTC.

The following electronic databases were searched from inception to November 2016:

• MEDLINE – via Ovid, 1946 to present.

• MEDLINE In-Process & Other Non-Indexed Citations – via Ovid, 1946 to present.

• MEDLINE Epub Ahead of Print – via Ovid, 1946 to present.

• Cumulative Index to Nursing and Allied Health Literature (CINAHL) – via EBSCOhost, 1982 to present.

• EMBASE – via Ovid, 1980 to present.

• Cochrane Database of Systematic Reviews – via Wiley Online Library, 1996 to present.

• Cochrane Central Register of Controlled Trials (CENTRAL) – via Wiley Online Library, 1995 to present.

• Database of Abstracts of Reviews of Effects – via Wiley Online Library, 1995 to 2015.

• Health Technology Assessment Database (HTA) – via Wiley Online Library, 1995 to present.

• Web of Science [Science Citation Index (SCI)] – via Clarivate Analytics (formerly Thomson Reuters), 1900 to present.

• Conference Proceedings Citation Index (CPCI) – via Clarivate Analytics (formerly Thomson Reuters), 1990 to present.

To identify ongoing or recently completed studies, trial registers were searched using the International Clinical Trials Registry Portal of the World Health Organization (WHO),32 which regularly compiles and updates data from > 15 clinical trial registers.

Searches were not limited by language or publication date and were not restricted to published research only. Search terms included medical subject heading (MeSH) terms and free-text synonyms for MTC combined with a RCT or systematic review study design filter. The search strategy was designed to be deliberately broad to capture all intervention studies within the MTC population, that is, studies of cabozantinib and vandetanib, as well as additional evidence for possible comparators, including BSC and radiotherapy, as such studies may be used to inform indirect comparisons. The MEDLINE search strategy is presented in Appendix 1.

To identify additional studies, reference lists of relevant studies, systematic reviews, clinical guidelines and submissions to regulatory authorities and advisory bodies [All Wales Medicines Strategy Group (AWMSG), Scottish Medicines Consortium (SMC), European Medicines Agency (EMA) and the US Food and Drug Administration] were examined. In addition, company submissions (CSs) to NICE related to the interventions within the scope of this review were examined. Citation searches of key included studies using the Web of Science database were also conducted. Clinical advisors to the AG provided advice on whether or not any relevant studies were missing from the search results.

A comprehensive database of relevant published and unpublished articles was constructed using EndNote version 8 [Clarivate Analytics (formerly Thomson Reuters), Philadelphia, PA, USA] software.

Study selection and data extraction

Following standard systematic review processes, two reviewers (CC and EK) independently screened all titles and abstracts using the eligibility criteria outlined in Table 1; full papers were retrieved for any publication that was deemed by a reviewer to be potentially includable. The two reviewers independently screened all full texts to identify studies that satisfied the inclusion criteria. Any discrepancies between reviewers were resolved through discussion. Results were reported in text, tables and a PRISMA flow chart. Data extraction was performed by one reviewer (CC) and was independently checked for errors against the original and published trial reports by the second reviewer (EK). Any discrepancies were resolved through discussion. Results were reported in text and tables.

Quality assessment

For the RCT evidence, critical appraisal of included trials was conducted by one reviewer (CC) using the Cochrane Risk of Bias tool;33 this was checked by a second reviewer (EK) and any discrepancies were resolved through discussion.

Evidence synthesis

Details of the included RCTs, including population characteristics, interventions, comparators and outcomes, were tabulated and discussed in a narrative review. On account of the small number of included studies, with just one study contributing evidence for each of the interventions, pairwise meta-analysis was not appropriate. In the absence of direct evidence comparing cabozantinib with vandetanib, a network meta-analysis (NMA) was performed using the ZETA trial EU-label and Efficacy of XL184 (Cabozantinib) in Advanced Medullary Thyroid Cancer (EXAM) trial intention-to-treat (ITT) populations (see Network meta-analysis).

Quantity and quality of research available

The details of the study selection process are outlined in the PRISMA flow chart (Figure 2). The search identified 1581 references after deduplication, of which 1516 were excluded because they did not satisfy the eligibility criteria. The full texts of 65 studies were retrieved to assess eligibility; 38 of these studies were excluded for the following reasons: absence of a control arm (n = 17), review (n = 6), letter/commentary (n = 6), wrong population (n = 5), wrong intervention (n = 2), animal study (n = 1) and a duplicate (n = 1). A list of excluded full papers, with reasons, is provided in Appendix 2. The excluded studies included two single-arm studies of vandetanib in children and adolescents with unresectable locally advanced or metastatic MTC as a result of MEN 2 (one published study34 and one ongoing study35). These studies may be relevant to the extension to the marketing authorisation for vandetanib;22 however, this population is beyond the scope of this appraisal.

FIGURE 2.

PRISMA flow chart.

There were five potentially relevant controlled trials of comparator interventions, principally other tyrosine kinase inhibitors (TKIs), one of which ended prematurely because of recruitment issues;36 the remaining four studies37–40 are ongoing. There is also one published retrospective study41 comparing MTC patients who received radioactive iodine therapy with those whose did not. As a result, there was no appropriate additional controlled trial evidence of other potential comparators to cabozantinib or vandetanib (e.g. radiotherapy) that could may have been used to inform a NMA.

The final result was 27 publications and protocols relating to five RCTs. For cabozantinib, this included 13 publications28,42–53 relating to the Phase III EXAM trial,28,42 which compared 140 mg per day of cabozantinib with placebo, and two publications54,55 relating to the ongoing EXAMINER trial,54 which compares 140 mg per day with 60 mg per day of cabozantinib, and seeks to recruit 188 participants (expected completion date: March 2019). 55 For vandetanib, this included 10 publications18,27,56–63 relating to the Phase III ZETA trial,27,56 which compares 300 mg per day of vandetanib with placebo, and two publications64,65 relating to two ongoing vandetanib trials: one trial64 comparing 300 mg per day with 150 mg per day of vandetanib, and one65 comparing vandetanib with vandetanib plus bortezomib (Velcade^®, Takeda, Osaka, Japan).

No additional relevant papers or studies were identified from the reference lists of included studies or reviews, or from citation searching of the key publications for the EXAM or ZETA trials. The clinical advisors to the AG were satisfied that no other relevant studies were missing.

The two pivotal Phase III trials, EXAM and ZETA, were international, multicentre, placebo-controlled trials. The characteristics of the EXAM and ZETA trials are presented in Table 2.

The clinical evidence submitted to NICE by the manufacturers of cabozantinib21 and vandetanib66 included data from six studies. All of these studies were identified by the search for this review, but only four studies satisfied the review eligibility criteria: for cabozantinib, the EXAM trial and ongoing EXAMINER trial,54 and for vandetanib, the ZETA trial and the ongoing NCT01496313 trial. 64 The submissions also included data from a Phase I, non-controlled, single-arm cabozantinib, dose-escalation trial, which included a subset of relevant MTC patients;67,68 a controlled study to assess the addition of an outreach programme to vandetanib treatment;69 and two ‘real-world’, non-controlled, single-arm vandetanib studies. 70–72 All of these studies were identified by the search but were excluded from this review because they did not satisfy the eligibility criteria; either they were single-arm cohort studies without a control group or the intervention evaluated in the trial did not relate to either cabozantinib or vandetanib (see Appendix 2).

The inclusion and exclusion criteria of the two trials were virtually identical, with the exception that the cabozantinib EXAM trial participants were required to have radiographic evidence of progressive disease (PD) at baseline. This was not an eligibility criterion for the vandetanib ZETA trial as the percentage of participants with ‘aggressive and symptomatic disease’ at baseline is reported to be 56% (186/331). 57 The cabozantinib trial had a median follow-up of 13.9 months, compared with 24 months for the vandetanib trial. The two trials had common primary (PFS) and secondary [OS, objective response rate (ORR), RET mutation status, CTN and CEA levels] end points. The cabozantinib trial assessed quality of life using the MD Anderson Symptom Inventory for thyroid cancer (MDASI-Thy), whereas the vandetanib trial also assessed disease control rate, and measured quality of life using the Functional Assessment of Cancer Therapy – General (FACT-G) tool and time to worsening of pain (TWP). It is noteworthy that the MDASI-Thy and TWP were both listed in the protocols but were not reported in the publications of the EXAM trial [only in the clinical study reports (CSRs)], whereas the FACT-G was not listed in any publication of the ZETA trial, but its results were reported in the Sanofi CS. 66

The definitions of PFS used in the trials were similar (i.e. the time from random assignment to the date of disease progression or death) and both trials employed a central committee to confirm investigator assessments. However, the EXAM trial used the modified RECIST (mRECIST) criteria10 and employed a blinded independent review committee, whereas the ZETA trial used the standard RECIST criteria, and it is unclear whether or not the central review was blinded.

The EXAM and ZETA trials had 330 and 331 participants, respectively (Table 3). Both trials randomised patients 2 : 1 to receive the active drug or placebo, respectively. In terms of baseline characteristics, the two arms of the cabozantinib EXAM trial are generally well balanced with the possible exceptions of Eastern Cooperative Oncology Group (ECOG) performance status of 0 (56.2% in the cabozantinib arm vs. 50.5% in the placebo arm), the proportion who had received prior systemic therapy for MTC (37% in the cabozantinib arm vs. 42% in the placebo arm) and positive RET mutation status (46.1% in the cabozantinib arm vs. 52.3% in the placebo arm), indicating that the control group might have had more severe disease. RET mutation status was unknown in 39% of participants as a result of missing sequence data or the presence of a mutation of unknown significance. 28 The two arms of the vandetanib ZETA trial were also generally well balanced, albeit with higher proportions of participants in the control arm than the treatment arm also potentially having more severe disease on account of a WHO performance status of 1–2 (42% for the placebo arm vs. 33% for the vandetanib arm) and having involvement of two or more organs (92% for the placebo arm vs. 87% for the vandetanib arm).

Comparing the two trials, the vandetanib ZETA trial included substantially greater proportions of participants with hereditary disease (12% in the vandetanib arm vs. 6% in the cabozantinib intervention arm) and participants with a performance status of 0 (67% in the vandetanib arm vs. 56% in the cabozantinib arm). However, the principal difference between the EXAM and ZETA trial populations concerns the presence of PD: participants in the EXAM trial were required to have evidence of PD, whereas participants in the ZETA trial were not. The two ITT populations are therefore sufficiently different to invalidate a standard indirect comparison.

In both trials, participants discontinued study treatment if there was evidence of disease progression or toxicity. The ZETA trial, however, also permitted treatment continuation or treatment switching post progression. 27 During the randomised phase, if there was disease progression based on investigator assessment, participants discontinued study treatment, but were offered the opportunity to receive vandetanib post progression as unblinded open-label treatment until normal discontinuation criteria applied (e.g. toxicity or progression). 27 In the vandetanib arm during the randomised stage of the trial, 120 out of 231 (52%) participants discontinued treatment because of progression or toxicity (compared with 55% in the cabozantinib trial28), but 44 of these 120 (37%) participants continued to receive vandetanib in the open-label phase. In the placebo arm of the ZETA trial, 71 out of 99 (72%) discontinued ‘treatment’ because of progression or toxicity (compared with 86% in the cabozantinib trial), and 58 of these 71 (82%) participants then switched to receive vandetanib in the open-label phase. All efficacy and safety data reported subsequently are subject to bias because of treatment switching, unless otherwise stated. This raises issues of confounding for some of the outcome data from the ZETA trial.

The marketing authorisation for vandetanib states that it is indicated ‘for the treatment of aggressive and symptomatic medullary thyroid cancer (MTC) in patients with unresectable locally advanced or metastatic disease’. 22 The terms ‘aggressive’ and ‘symptomatic’ are not defined in the licence, but were defined post hoc. The Sanofi CS for vandetanib66 presents PFS and OS outcome data from post hoc analyses on two preplanned subpopulations within the ZETA trial (and, as such, are more restrictive than the overall population recruited to this trial):

• Patients with unresectable, locally advanced or metastatic MTC and whose disease is ‘progressive and symptomatic’ (defined as having ‘documented progression 12 months prior to enrolment and at least one of the following symptoms at baseline: pain score > 4, ≥ 10 mg/day opioid use, diarrhoea, flushing, fatigue, pain, nausea, dysphagia, dysphonia, respiratory symptoms, and weight loss.’57 This corresponds to the ‘EU-label’ or ‘progressive and symptomatic’ population (n = 186) referred to in the Sanofi CS. 66 In the post hoc analyses conducted by the company, the data reported by Kreissl et al. 57 could not be replicated exactly, and the number reported is n = 190 for PFS and n = 189 for OS data in the Sanofi CS (see the Sanofi CS,66 appendix 6, tables 5 and 7, respectively). Numbers from the published Kreissl et al. 57 analyses are used throughout the clinical-effectiveness section, whereas the cost-effectiveness section is based on the slightly larger subgroup defined for the purposes of the NICE submission.

• Patients with unresectable, locally advanced or metastatic MTC whose disease is ‘progressive and symptomatic’ (as above) and is ‘aggressive’, that is, with CTN and CEA doubling times of < 24 months from screening. This is the so-called ‘restricted EU-label population’ [n = (confidential information has been removed)] presented in the Sanofi CS. The Sanofi CS claims that ‘This population closely reflects UK clinical practice for TKI treatment’ (Sanofi CS,66 pp. 11 and 54). However, clinical advice received by the AG suggests that CTN and CEA monitoring would not usually inform decisions about whether or not to commence TKI therapy, as this is principally determined by radiographic evidence of progression and symptoms.

The data presented for these groups are partly unpublished (only the PFS and ORR data for the EU-label population are published)57 and are reported here because they are used to inform the health economic model developed by the AG. The baseline characteristics of these subgroups are presented in Table 4, together with the comparable baseline data for the EXAM trial ITT population. Despite the EXAM trial ITT population being ‘progressive’ and the EU-label ZETA trial population being ‘progressive and symptomatic’, clinical advice received by the AG confirmed that these two populations were comparable.

It should also be noted that, among the EU-label population, (confidential information has been removed) of patients in the intervention group continued to receive vandetanib in the open-label phase, whereas (confidential information has been removed) of patients in the placebo arm ‘crossed over’ to receive open-label vandetanib (see Sanofi clarification response,73 question 3). In the restricted EU-label population, (confidential information has been removed) of patients in the intervention group continued to receive vandetanib in the open-label phase, whereas (confidential information has been removed) of patients in the placebo arm ‘crossed over’ to receive open-label vandetanib (Sanofi CS,66 pp. 17 and 63). All efficacy and safety data reported subsequently for this group are subject to bias because of treatment switching, unless otherwise stated. This raises issues of confounding for some of the trial data, including for the restricted EU-label population.

The risk of bias in the EXAM and ZETA trials was assessed using the Cochrane Risk of Bias Tool (Table 5). These assessments made use of the protocols (published and unpublished), the trial publications and unpublished CSRs for each trial.

The AG considers the EXAM trial to be of generally good quality, being assessed as having a low risk of performance, detection and attrition bias on account of measures to ensure blinding and the management of dropouts. The EXAM trial is at unclear risk of selection bias because full details of the randomisation and allocation concealment processes were absent from the documents identified from the searches, or from those made available during this appraisal. It was at moderate risk of reporting bias on account of the failure to report the results of some outcomes in published documents, and at moderate risk of other bias owing to potential conflicts of interest and the failure to control for the possible treatment effect modifier of CTN and CEA doubling times.

Overall, the AG considers that the ZETA trial was at moderate to high risk of bias across most domains. As with the EXAM trial, the likelihood of attrition bias was considered to be low and the risk of selection bias was unclear. However, there was a moderate risk of reporting and other bias because of the presence of selective reporting and some potential conflicts of interest, although post hoc analyses were conducted on the potential treatment effect modifier of CTN and CEA doubling times. In contrast to the EXAM trial, performance bias and detection bias were assessed as being of moderate to high risk because there was a lack of detail on blinding procedures and certain outcomes, and their results were potentially confounded by the inclusion of patients switching to open-label treatment within the analysis.

Assessment of effectiveness

In the EXAM trial, at the data cut-off point (15 June 2011), the median duration of follow-up was 13.9 months. At this time point, 98 out of 219 (45%) participants in the cabozantinib arm were still receiving blinded study treatment, whereas only 15 out of 111 (14%) participants in the placebo arm were still receiving blinded study treatment. 28 In the ZETA trial, at the data cut-off point (July 2009), the median duration of follow-up was 24 months. At this time point, 111 out of 231 (48%) participants in the vandetanib arm were still receiving blinded study treatment, whereas only 28 out of 100 (28%) participants in the placebo arm were still receiving blinded study treatment. 27

Progression-free survival

Both pivotal trials reported PFS as their primary outcome using similar definitions and based on tumour measurements performed at screening and every 12 weeks. Both treatments resulted in a significantly reduced risk of progression. For cabozantinib, the hazard ratio (HR) for PFS was reported to be 0.28 [95% confidence interval (CI) 0.19 to 0.40; p < 0.001] by central review and 0.29 (95% CI 0.21 to 0.42; p < 0.001) by investigator read28,43 (Table 6). For vandetanib, the HR for PFS was reported to be 0.46 (95% CI 0.31 to 0.69; p < 0.001) by central review of all patients (ITT population), 0.28 (95% CI 0.18 to 0.42; p < 0.001) by central review excluding open-label patients, and (confidential information has been removed).

In the post hoc analysis, PFS was also calculated for the EU-label (n = 186) and restricted EU-label (confidential information has been removed) populations. For the vandetanib EU-label population, the HR for PFS was reported to be 0.47 (95% CI 0.29 to 0.77; p = 0.0024) by central review66 and 0.33 (95% CI 0.20 to 0.53; p = 0.0226) by investigator read. 57 The HR by central review but excluding open-label patients57 was reported to be 0.32 (95% CI 0.19 to 0.54; p < 0.001). According to the Sanofi CS (p. 55),66 the median PFS for the restricted EU-label group was (confidential information has been removed) in the placebo arm compared with (confidential information has been removed) in the vandetanib arm (confidential information has been removed).

The investigator-read risk of progression, compared with placebo, for the comparable EXAM trial (n = 331) and ZETA trial EU-label (n = 186) populations was HR 0.29 (95% CI 0.21 to 0.42; p < 0.001) for cabozantinib and HR 0.33 (95% CI 0.2 to 0.53; p = 0.0226) for vandetanib, respectively.

The proportion of randomised patients progressing was similar in the treatment and placebo groups across the two trials. The EXAM trial publication (i.e. Elisei et al. )28 states that 57 out of 219 (26%) participants randomised to cabozantinib had progressed at follow-up compared with 67 out of 111 (60%) participants in the placebo group. The ZETA trial publication27 reported data on 124 participants who progressed: 73 out of 231 (32%) participants randomised to vandetanib had progressed (previously reported as 37% at 24 months58) and 51 out of 100 (51%) participants randomised to placebo had progressed.

In the EXAM trial, the Kaplan–Meier estimates for the proportion of participants alive and progression free at 1 year were reported to be 47.3% for cabozantinib compared with 7.2% for placebo. 28 In the ZETA trial, the proportion of participants in the ITT population alive and progression free at 6 months was reported to be 91% for vandetanib compared with 74% for placebo. 59

Subgroup analyses according to prespecified subgroups were conducted for PFS for both cabozantinib and vandetanib. For both interventions, all subgroups demonstrated a beneficial effect with treatment (HR < 1.0), although 95% CIs indicated non-statistically significant treatment effects for some small subgroups, as may be expected. Subgroups considered included sex, performance status, and number of previous anticancer regimens or other TKIs received and response to those therapies. 27,28,43,44,77 The Ipsen CS21 for cabozantinib reported that PFS was also prolonged in a subgroup of cabozantinib patients (n = 34) who had received prior vandetanib (median PFS was 12.8 months for cabozantinib and 2.8 months for placebo, and ORR was 28%, when prior vandetanib use reported). PFS for cabozantinib was also consistent across subgroups according to age and the presence of bone metastases28 and PFS for vandetanib was not sensitive to ethnicity. 27

Subgroup analyses based on RET mutation status (as specified in the final NICE scope26) were also conducted for the EXAM trial. Details of the number of participants in each of these groups within the EXAM trial are presented in Tables 7 and 8. The data demonstrate that cabozantinib was associated with a beneficial effect compared with placebo for all subgroups tested (see Tables 7 and 8) although the treatment effect was not statistically significant at the 95% level (p = 0.21) for the RET-negative subgroup, and PFS improvement was least pronounced in the small subset of RET mutation-negative participants who were also RAt Sarcoma (RAS) mutation negative. 45,46

With respect to vandetanib, the Sanofi CS states that, ‘subgroups relating to two different definitions for “aggressive disease” were included in a pre-specified subgroup analysis: calcitonin (CTN) doubling time (DT) ≤ 24 months and CEA DT ≤ 24 months’ (Sanofi CS,66 section 4.3, p. 45). Subgroup analyses by these criteria were reported in this CS66 and the unpublished CSR. 74 These found that all subgroups demonstrated a beneficial effect on PFS (HR < 1.0), with a statistically significant treatment effect observed between patients with a CTN doubling time of ≤ 24 months and patients with a CEA doubling time of ≤ 24 months (Figure 3).

FIGURE 3.

Progression-free survival according to subgroups in the ZETA trial (reproduced from Sanofi CS66, figure 4, p. 51 and AstraZeneca’s unpublished CSR dated July 201174).

Safety outcomes

In order to be considered for safety outcomes, participants had to receive at least one dose of the study drug. 27,28

Justification for conducting a network meta-analysis

In the absence of head-to-head evidence comparing cabozantinib with vandetanib, an indirect comparison using a NMA was considered. An indirect comparison has previously been published as an abstract93 and is presented in the Ipsen CS;21 however, owing to the differences between the ITT population of the EXAM and ZETA trials, this analysis was not deemed appropriate for formal consideration within this assessment. The validity of the NMA depends on the assumption that there is no difference in the distribution of trial-level treatment effect modifiers between the populations in the two trials. This is unlikely to be the case for the ITT populations of the ZETA and EXAM trials in particular, because participants in the EXAM trial had confirmed disease progression, whereas the ZETA trial recruited a broader population of participants with no requirement for established disease progression. HRs for the effectiveness of vandetanib compared with placebo for investigator-assessed PFS in the ZETA trial were reported for the symptomatic and progressive subgroup (n = 186, HR 0.33, 95% CI 0.20 to 0.53) and the full analysis set excluding symptomatic and progressive participants (n = 139, HR 0.49, 95% CI 0.27 to 0.58) within the Sanofi CS. 66 This suggests that progression may be a treatment-effect modifier, with a greater treatment effect observed for the subgroup with confirmed progression (though a statistically significant difference between the two groups cannot be inferred).

Despite differences in the ITT populations, the AG considered a NMA based on the EU-label subgroup of the ZETA population to be appropriate. There was a marked difference in the median PFS in the control groups of the two studies [EXAM trial: 4.0 months, ZETA trial EU-label population: 16.4 months (by central review)]; however, differences in baseline characteristics of the included studies as a result of differences in study protocols are to be expected and do not invalidate an indirect comparison. For a NMA to be valid, it is important that there is not an imbalance in treatment-effect modifiers. Clinical advisors to the AG identified severity of disease as an important potential treatment-effect modifier. Information on ECOG/WHO performance status at baseline was not available for the ZETA trial EU-label population, so balance across the two studies could not be assessed, and there is no clear evidence to demonstrate the balance of this potential treatment-effect modifier. However, subgroup analyses indicated consistent treatment effects according to performance status at baseline for both interventions,27,28 and clinical advice received by the AG suggested that the ZETA trial EU-label and EXAM trial ITT populations could be considered to be broadly comparable. Therefore, on the basis of clinical advice, and since there was no evidence to invalidate indirect comparison, a NMA was considered to be justified.

Methods for the network meta-analysis

The NMA was conducted by the AG to provide an indirect comparison between cabozantinib and vandetanib for central-read PFS and investigator-read PFS. For OS, the HRs for both treatment groups are confounded by treatment switching; therefore, a NMA was not conducted for this outcome, as it would not provide a meaningful comparison.

The network diagram is presented in Figure 4 and data contributing to the NMA are presented in Table 14. Analyses were conducted using a Bayesian random-effects model, as described by Dias et al. 94 Given that there is potential heterogeneity between the trials, a random effects model was considered to be the most appropriate so that this uncertainty is appropriately reflected in the estimated treatment effects. There was insufficient information to estimate the between-study variance from the data alone, hence a weakly informative prior was used for this parameter (log-normal –2.56, 1.742 based on the recommendation in Turner et al. 95), which has a median of 0.08 and 95% range of 0.003 to 2.34 on the untransformed scale. This prior was also truncated such that the ratio of the upper and lower 95% CI of the prior does not exceed 10, based on evidence from Speigelhalter et al. 96 and Smith et al. 97 that the between-study treatment effects are unlikely to vary by more than an order of magnitude.

FIGURE 4.

Network diagram for the NMA.

Analyses were conducted in the freely available software packages WinBUGS98 (MRC Biostatistics Unit, Cambridge, UK) and R (The R Foundation for Statistical Computing, Vienna, Austria) using the R2WinBUGS interface package. Convergence to the target posterior distributions was assessed using the Gelman–Rubin statistic, as modified by Brooks and Gelman,99 for two chains with different initial values. A burn-in of 50,000 iterations of the Markov chain was used with a further 20,000 iterations retained to estimate parameters. There was no evidence of high autocorrelation between successive iterations of the Markov chain.

It should be noted that the results from the NMA are not used to inform the health economic model developed by the AG (see Chapter 4, Independent assessment group model). The NMA utilises HRs, which are averaged estimates of treatment effect, and their use in the health economic model would be appropriate only if the hazards are proportional over the entire extrapolation period. However, the AG’s health economic model considers a broader range of parametric functions, not all of which conform to the proportional hazards assumption; hence the use of HRs from the NMA would not be appropriate. Instead, estimation of the treatment effects and baseline model is conducted using the same parametric model type (see Time to event analysis using individual patient data), conforming to the recommendation in Guyot et al. 100

Results of the network meta-analysis

The results of the NMA are shown in Figure 5 for investigator-read PFS and Figure 6 for central-read PFS. Based on investigator-read PFS, the results of the two treatments are broadly similar [vandetanib vs. cabozantinib, HR 1.14, 95% credible interval (CrI) 0.41 to 3.09]. The magnitude of the treatment effect is more favourable towards cabozantinib when the comparison is based on central-read PFS (HR 1.68, 95% CrI 0.61 to 4.62); however, the difference between the two interventions is not statistically significant.

FIGURE 5.

Results of the NMA for investigator-read PFS. PrI, prediction interval.

FIGURE 6.

Results of the NMA for central-read PFS. PrI, prediction interval.

Review of existing economic evaluations: methods

A comprehensive search was undertaken to systematically identify economic evaluations of treatments for locally advanced or metastatic MTC and studies reporting on the HRQoL of patients with locally advanced or metastatic thyroid cancer (including MTC as well as other commoner forms of thyroid cancer). In anticipation of the likely lack of relevant evidence, the AG’s search strategy was designed to be intentionally broad.

The following electronic databases were searched from inception to 3 November 2016:

• MEDLINE – via Ovid, 1946 to present

• MEDLINE In-Process & Other Non-Indexed Citations – via Ovid, 1946 to present

• MEDLINE Epub Ahead of Print – via Ovid, 1946 to present

• CINAHL – via EBSCOhost, 1982 to present

• EMBASE – via Ovid, 1980 to present

• HTA database, 1995 to present

• NHS Economic Evaluation Database (NHS EED), 1995 to 2015

• Web of Science (SCI) – via Clarivate Analytics (formerly Thomson Reuters) 1899 to present

• CPCI – via Clarivate Analytics (formerly Thomson Reuters) 1990 to present.

The search strategy comprised MeSH or Emtree Thesaurus terms and free-text synonyms for ‘thyroid cancer’. Searches were translated across databases and were not limited by either language or publication date. The search strategies are presented in Appendix 1. Search filters designed to identify economic evaluations and HRQoL studies were applied in MEDLINE and other databases, when appropriate. Reference and citation searching of included papers was also undertaken.

Potentially includable studies were sifted by title and abstract by one reviewer (PT). In keeping with the breadth of the search strategy, the inclusion criteria were also defined broadly and the sifting process followed an inclusive approach in order to maximise sensitivity. Given that the cost-effectiveness search also identified studies relating to health utilities (e.g. those used within models), and the HRQoL search also identified health economic evaluation studies, the results of both searches were sifted together using a common set of inclusion criteria (Box 2). Although the inclusion criteria for the review of existing economic evaluation studies were specific to MTC, HRQoL studies were considered to be potentially includable if they were undertaken in patients with MTC or other types of thyroid cancer (papillary, follicular, Hürthle cell carcinoma).

Review of existing economic evaluations: results

Figure 7 presents the study selection results. Before deduplication, the searches yielded 3161 citations (HRQoL search, n = 1282 studies; economic evaluation search, n = 1879 citations). Following the initial sift, 3057 of these studies were excluded. Full texts of the remaining 104 potentially includable studies were retrieved for further examination. However, none of these studies contained an economic evaluation of treatments for MTC; hence all studies were excluded from the review. In addition, none of these studies reported health utilities for patients with locally advanced or metastatic MTC. One study102 reported health utilities for patients with radioactive iodine-refractory differentiated thyroid cancer; this study is discussed in further detail in Independent assessment group model and Health-related quality of life.

FIGURE 7.

Study selection results.

Review of models submitted by the companies

The Sanofi submission66 includes a health economic evaluation of vandetanib for the treatment of locally advanced or metastatic MTC, together with a fully executable health economic model. The Ipsen submission21 does not include any economic evidence for this appraisal.

Sanofi model structure

The economic analysis presented by Sanofi takes the form of a cohort-level partitioned survival model implemented using the discretely integrated condition event (DICE) simulation methodology103 (Figure 8). The model includes three health states: (1) progression free, (2) post progression and (3) dead. The model operates as follows. At any time t, the probability that a participant allocated to treatment group k is alive is given by S(t)_{OS_}k, whereas the probability that a participant allocated to treatment group k is alive and progression free is given by S(t)_{PFS_}k. The probability that a participant is alive following disease progression is calculated as the difference between the two survivor functions: S(t)_{OS_}k – S(t)_{PFS_}k for any time t. Given the presence of censoring, parametric survivor functions were fitted to Kaplan–Meier curves for OS and PFS from the ITT/safety populations of the ZETA trial, including adjustment for two covariates: (1) ‘SympProg’ (presence of symptomatic and progressive disease) and (2) ‘BiomarkerChg’ (CEA and CTN doubling times of ≤ 24 months). Weibull functions were selected to model both OS and PFS, assuming independent (non-proportional) hazards between treatment groups. The DICE routine is evaluated using a monthly cycle length over a 20-year lifetime horizon and includes a half-cycle correction to account for the timing of events.

FIGURE 8.

Schematic of the Sanofi DICE model (reproduced from the Sanofi CS66).

The model assumes that health utility is determined by the presence/absence of disease progression, with higher utilities applied to the progression-free state. In addition, a once-only QALY loss is applied to each group to account for the incidence of grade 3/4 AEs.

The model includes the following resource costs: (1) vandetanib drug acquisition costs, (2) monitoring for participants receiving vandetanib, (3) BSC costs, (4) palliative care costs and (5) costs associated with managing AEs.

The model employs the following structural assumptions:

• Health-related quality of life is determined according to the presence/absence of disease progression and the incidence of grade 3/4 AEs.

• Progression-free survival and OS are modelled using Weibull functions assuming independent (non-proportional) hazards.

• Survival models were fitted to the overall ITT population for PFS and the safety population for OS, including covariate adjustments to reflect the characteristics of the restricted EU-label population.

• No adjustment is made to account for logical inconsistencies [i.e. when S(t)_PFS > S(t)_OS].

• The modelling of costs and health outcomes includes the level of open-label vandetanib use (either post progression or in any patient following the January 2010 protocol amendment73) observed in the ZETA trial.

• Adverse events are assumed to affect both costs and HRQoL. According to the Sanofi CS,66 AE impacts on HRQoL apply only during the first month of the time horizon. This aspect of the model is subject to a programming error (see Critical appraisal of the economic analysis presented by Sanofi) and was corrected by the company in its clarification response73 (question A18).

• Palliative care costs are assumed to be incurred only during the final month of life.

Sanofi model results

Sanofi central estimates of cost-effectiveness (excluding Patient Access Scheme, including error corrections)

Table 17 presents the company’s base-case estimates of cost-effectiveness using the list price for vandetanib. Based on the probabilistic version of the company’s model, vandetanib is expected to generate an additional 1.34 QALYs at an additional cost of £42,215 compared with BSC; the ICER for vandetanib versus BSC is expected to be £31,546 per QALY gained. The deterministic version of the model produces a slightly higher ICER of £31,731 per QALY gained.

TABLE 17 - Sanofi’s base-case estimates of cost-effectiveness (excluding PAS)

Includes post-progression vandetanib costs.

Note

Sanofi probabilistic sensitivity analysis results.

Option	Absolute		Incremental
	QALYs	Costs (£)	QALYs	Costs (£)	ICER (£)
Probabilistic model
Vandetaniba	3.53	181,130	1.34	42,215	31,546
BSCa	2.19	138,915	–	–	–
Deterministic model
Vandetaniba	3.49	175,316	1.36	43,024	31,731
BSCa	2.13	132,292	–	–	–

Figure 9 presents the CEACs for vandetanib and BSC, generated by the AG using the corrected version of the Sanofi model. The CEAC indicates that, assuming willingness-to-pay (WTP) thresholds of £20,000 and £30,000 per QALY gained, the probability that vandetanib produces more net benefit than BSC is approximately 0.33 and 0.48, respectively.

FIGURE 9.

Cost-effectiveness acceptability curves generated using the Sanofi model: vandetanib vs. BSC (redrawn by the AG).

Sanofi’s deterministic sensitivity analysis results

Figure 10 presents the results of the company’s DSAs. The most influential parameters (of those assessed by the company) relate to the probability of vandetanib continuation beyond progression, the probability of treatment switching in the BSC group and the vandetanib discontinuation parameter applied to the vandetanib group during the progression-free phase. The use of the log-logistic and log-normal functions for PFS and OS (analyses not shown in Figure 10) did not have a substantial impact on the ICER for vandetanib versus BSC (log-normal PFS and OS ICER = £37,227 per QALY gained; log-logistic PFS and OS ICER = £28,879 per QALY gained). It should be noted that a higher proportion of vandetanib participants are alive at 20 years (> 8%) using these functions rather than the Weibull model (< 2%).

FIGURE 10.

The DSA results generated using the Sanofi model (reproduced from the Sanofi model). Note: tornado plot shows absolute change to base-case ICER.

Model scope

The scope of the AG’s analysis is summarised in Table 19. The AG’s analyses are presented across two populations of patients with locally advanced or metastatic MTC: (1) patients with progressive and symptomatic disease (the EU-label population for vandetanib) and (2) the restricted EU-label population for vandetanib. Within the broader symptomatic and progressive population, pairwise economic comparisons are made for cabozantinib versus BSC based on the ITT population of the EXAM trial28 (AG analysis 1) and for vandetanib versus BSC based on the post hoc EU-label (symptomatic and progressive) subgroup of the ZETA trial66,78 (AG analysis 2). It should be noted that these analyses are limited in that they do not include all relevant treatment options. As the AG did not have access to the underlying IPD (including data on relevant covariates) from the ZETA trial, it was not possible to implement statistical adjustments to account for open-label vandetanib use in either treatment group, or to adjust for other potential baseline imbalances in the subgroup. Consequently, the comparison of vandetanib versus BSC is subject to potential confounding. To provide more meaningful estimates of the cost-effectiveness of vandetanib and cabozantinib, two sets of fully incremental analyses of all options are also presented. The first of these (AG analysis 3) uses the EXAM trial data for cabozantinib and BSC and applies the PFS treatment effect for vandetanib versus placebo from the ZETA trial EU-label subgroup to the EXAM trial placebo group baseline; OS is assumed to be the same for both TKIs (equivalent to the cabozantinib arm in the EXAM trial). The second incremental analysis (AG analysis 4) assumes that PFS and OS outcomes for vandetanib are equivalent to those for cabozantinib. Although these incremental analyses necessarily reflect potentially strong assumptions concerning transferable/equivalent treatment effects between vandetanib and cabozantinib, they are not subject to potential confounding caused by post-progression vandetanib use within the clinical data. A further pairwise comparison (AG analysis 5) that evaluates vandetanib versus BSC within the restricted EU-label population (patients with symptomatic and progressive MTC with CEA/CTN doubling times of ≤ 24 months) is also presented as equivalent covariate data were not available from the EXAM trial, cabozantinib could not be included within this analysis. Across all five sets of analyses, cost-effectiveness is evaluated in terms of the incremental cost per QALY gained from the perspective of the NHS and Personal Social Services (PSS) over a 20-year (lifetime) horizon. Costs and health outcomes were discounted at a rate of 3.5% per annum. 109 Costs were valued at 2016/17 prices.

Model structure

The structure of the AG model is presented in Figure 11. As shown in the diagram, the AG model structure is broadly similar to that adopted within the Sanofi model (see Sanofi model structure). The AG model adopts a partitioned survival approach, based on three health states: (1) progression free, (2) post progression and (3) dead. For any time, t, the probability that a patient is alive and progression free is given by the cumulative survival probability for PFS, whereas the probability that a patient is alive is given by the cumulative survival probability for OS. The probability that a patient is in the post-progression state at any time t is given by the difference between the cumulative survival probabilities for PFS and OS. The model includes an adjustment for logical inconsistency, whereby, if the probability of PFS is greater than that of OS, PFS is constrained to reflect the lower OS probability. As with the Sanofi model, HRQoL is defined according to the presence or absence of disease progression and a separate QALY loss is applied to account for the incidence of grade 3/4 AEs during the first model cycle. The model includes costs associated with drug acquisition, health-state costs incurred while receiving cabozantinib and vandetanib [consultant-led outpatient visits, nurse-led outpatient visits, ECG, blood tests and computerised tomography (CT) scans], costs associated with managing grade 3/4 AEs, BSC-related costs [consultant-led outpatient visits, CT scans, magnetic resonance imaging (MRI) scans, specialist palliative care visits, palliative radiotherapy, palliative surgery and bisphosphonates for bone metastases] and end-of-life care costs.

FIGURE 11.

The AG model structure. N/A, not applicable. a, Applies only to patients not receiving vandetanib/cabozantinib. b, Applies only to open-label vandetanib costs within pairwise comparisons of vandetanib vs. BSC.

The model employs the following structural assumptions:

• Health-related quality of life is assumed to be determined according to the presence/absence of disease progression and the incidence of grade 3/4 AEs.

• The model includes an adjustment to account for logical inconsistencies [i.e. when S(t)_PFS>S(t)_OS].

• In the pairwise comparisons of vandetanib versus BSC (see Table 19, AG analyses 2 and 5), the modelling of costs and health outcomes includes the level of treatment switching and continued vandetanib use post progression observed in the ZETA trial subgroups. This was included as the company’s attempts to adjust for treatment switching and treatment continuation post progression were reported to have been unsuccessful (see Critical appraisal of the economic analysis presented by Sanofi).

• Grade 3/4 AEs are assumed to affect both costs and HRQoL. Health losses resulting from AEs are assumed to be transient and resolved quickly: a QALY loss is applied during the first model cycle only (1-month duration).

• As patients receiving BSC, by definition, have progressed disease, the costs associated with BSC are assumed to be the same in both the progression-free and post-progression states.

• Health state resource use (including additional TKI monitoring requirements) incurred during the progression-free period are assumed to differ between the three treatment options.

• Palliative care costs are incurred only during the final month of life.

Evidence used to inform the model’s parameters

Model evaluation methods

Uncertainty was evaluated using PSA and DSA. PSA was undertaken using simple Monte Carlo sampling methods (2000 samples). The choice of distribution assumed for each parameter group is summarised in Table 62 in Appendix 4. The results of the PSA are presented as CEACs. DSAs were undertaken to explore the impact of alternative assumptions regarding discount rates, choices of parametric survivor functions, disutilities associated with AEs, and resource use and cost assumptions.

Model validation

The AG adopted a number of approaches to ensure the credibility of the model. These included scrutiny of the implemented model coding and formulae by two modellers, black-box testing, double-programming of the deterministic base case for all pairwise comparisons, checking the accuracy of all model inputs against the original sources, consultation with clinical experts, peer review of the model assumptions by clinical experts and peer review of the report by two third-party modellers (see Acknowledgements).

Assessment group model results

This section presents the results based on the AG model for each of the five sets of analyses.

Analysis 1: EU-label population (symptomatic and progressive medullary thyroid cancer), cabozantinib versus best supportive care (pairwise)

Table 26 presents the results of the pairwise comparison of cabozantinib versus BSC within the EU-label (symptomatic and progressive) MTC population. Disaggregated life-years gained (LYGs), QALYs and costs are presented in Table 63 in Appendix 5. Based on the probabilistic version of the AG model (assuming the log-logistic function for both PFS and OS), cabozantinib is expected to generate 0.48 additional QALYs at an additional cost of £72,734 compared with BSC; the ICER for cabozantinib versus BSC is expected to be £150,874 per QALY gained. The deterministic version of the model (based on point estimates of parameters) produces similar results (deterministic ICER = £148,169 per QALY gained). The disaggregated results show that a considerable amount of the OS gain in both groups is accrued in the post-progression state.

TABLE 26 - Analysis 1: EU-label population (symptomatic and progressive MTC), cabozantinib vs. BSC (pairwise), central estimates of cost-effectiveness (PFS = log-logistic, OS = log-logistic for both options)

Option	Absolute		Incremental		ICER (£)
	QALYs	Costs (£)	QALYs	Costs (£)
Probabilistic model
Cabozantinib	2.28	88,527	0.48	72,734	150,874
BSC	1.79	15,793	–	–	–
Deterministic model
Cabozantinib	2.27	87,960	0.49	72,287	148,169
BSC	1.79	15,672	–	–	–

Figure 12 presents CEACs for the pairwise comparison of cabozantinib versus BSC within the EU-label (symptomatic and progressive) MTC population. Assuming a WTP threshold (λ) of £30,000 per QALY gained, the probability that cabozantinib produces more net benefit than BSC is zero.

FIGURE 12.

Analysis 1: EU-label population (symptomatic and progressive MTC), cabozantinib vs. BSC (pairwise), CEACs (PFS = log-logistic, OS = log-logistic for both options).

Table 27 presents the results of the DSAs for the pairwise comparison of cabozantinib versus BSC within the EU-label (symptomatic and progressive) MTC population. As shown in the table, the ICER remains in excess of £135,000 per QALY gained across all scenarios. The alternative scenarios regarding health utilities, AE impacts and health state resource use do not have a marked impact on the cost-effectiveness of cabozantinib. The exclusion of dose reductions for cabozantinib increases the ICER to £174,297 per QALY gained. The choice of survivor functions for PFS and OS produces ICERs for cabozantinib versus BSC in the range £138,259 to £239,141 per QALY gained; the curves used in the AG’s base-case analysis (PFS = log-logistic, OS = log-logistic) are close to the most favourable scenario.

TABLE 27 - Analysis 1: EU-label population (symptomatic and progressive MTC), cabozantinib vs. BSC (pairwise), DSA results

The AG’s base-case curve choice.

Scenario	Incremental		ICER (£)
	QALYs	Costs (£)
Base case	0.49	72,287	148,169
Undiscounted health outcomes and costs	0.57	77,243	135,531
Sanofi CS utilities	0.47	72,287	154,582
DECISION study utilities	0.43	72,287	166,890
Cabozantinib SMC utilities	0.44	72,287	165,816
AE disutility doubled	0.48	72,287	150,159
AE disutility halved	0.49	72,287	147,194
AE management costs doubled	0.49	72,498	148,601
AE management costs halved	0.49	72,182	147,954
Health state resource use doubled	0.49	72,959	149,546
Health state resource use halved	0.49	71,951	147,481
No cabozantinib dose reductions	0.49	85,034	174,297
Curve choice
PFS – exponential; OS – exponential	0.45	71,195	158,030
PFS – exponential; OS – Weibull	0.42	71,012	170,550
PFS – exponential; OS – Gompertz	0.31	70,525	227,293
PFS – exponential; OS – log-normal	0.47	71,298	150,146
PFS – exponential; OS – log-logistic	0.46	71,251	153,284
PFS – exponential; OS – gamma	0.43	71,061	166,964
PFS – Weibull; OS – exponential	0.38	55,213	147,111
PFS – Weibull; OS – Weibull	0.34	55,035	161,300
PFS – Weibull; OS – Gompertz	0.24	54,530	232,034
PFS – Weibull; OS – log-normal	0.40	55,345	138,424
PFS – Weibull; OS – log-logistic	0.39	55,297	141,864
PFS – Weibull; OS – gamma	0.35	55,093	157,191
PFS – Gompertz; OS – exponential	0.36	52,776	147,369
PFS – Gompertz; OS – Weibull	0.32	52,593	162,336
PFS – Gompertz; OS – Gompertz	0.22	52,105	239,141
PFS – Gompertz; OS – log-normal	0.38	52,879	138,259
PFS – Gompertz; OS – log-logistic	0.37	52,831	141,855
PFS – Gompertz; OS – gamma	0.33	52,642	157,984
PFS – log-normal; OS – exponential	0.46	70,719	152,833
PFS – log-normal; OS – Weibull	0.43	70,551	164,542
PFS – log-normal; OS – Gompertz	0.32	70,024	217,141
PFS – log-normal; OS – log-normal	0.49	70,909	145,511
PFS – log-normal; OS – log-logistic	0.48	70,834	148,443
PFS – log-normal; OS – gamma	0.44	70,617	161,210
PFS – log-logistic; OS – exponential	0.47	72,176	152,470
PFS – log-logistic; OS – Weibull	0.44	72,008	163,867
PFS – log-logistic; OS – Gompertz	0.33	71,481	214,567
PFS – log-logistic; OS – log-normal	0.50	72,342	145,282
PFS – log-logistic; OS – log-logistica	0.49	72,287	148,169
PFS – log-logistic; OS – gamma	0.45	72,070	160,627
PFS – gamma; OS – exponential	0.39	57,437	147,094
PFS – gamma; OS – Weibull	0.36	57,260	160,678
PFS – gamma; OS – Gompertz	0.25	56,743	226,874
PFS – gamma; OS – log-normal	0.42	57,582	138,733
PFS – gamma; OS – log-logistic	0.41	57,535	142,051
PFS – gamma; OS – gamma	0.37	57,318	156,755

Analysis 2: EU-label population (symptomatic and progressive MTC), vandetanib versus best supportive care (pairwise)

Table 28 presents the results of the pairwise comparison of vandetanib versus BSC within the EU-label (symptomatic and progressive) MTC population. It should be noted that this analysis is subject to potential confounding as a result of the open-label use of vandetanib in the ZETA trial; hence, post-progression vandetanib costs are included for both treatment groups. Disaggregated LYGs, QALYs and costs are presented in Table 64 in Appendix 5. Based on the probabilistic version of the AG model (assuming the log-logistic function for both PFS and OS), vandetanib is expected to generate 0.23 additional QALYs at an additional cost of £79,745 compared with BSC; the ICER for vandetanib versus BSC is expected to be £352,508 per QALY gained. The deterministic version of the model yields a lower ICER of £336,896 per QALY gained. The disaggregated results indicate that, based on the log-logistic model, OS is expected to be higher in the BSC group than in the vandetanib group: this is likely to be a consequence of confounding as a result of open-label vandetanib use in the placebo group. It is also noteworthy that, based on the selected OS functions, a similar proportion of patients in each group (11–12%) are predicted to still be alive at 20 years as a consequence of the flattening of the tails of the modelled curves; additional analyses undertaken by the AG indicate that the ICER for vandetanib versus BSC remains stable over longer time horizons (the ICER using a 30-year time horizon, excluding any general population mortality constraints, is £345,284 per QALY gained).

TABLE 28 - Analysis 2: EU-label population (symptomatic and progressive MTC), vandetanib vs. BSC (pairwise), central estimates of cost-effectiveness (PFS = log-logistic, OS = log-logistic for both options)

Option	Absolute		Incremental		ICER (£)
	QALYs	Costs (£)	QALYs	Costs (£)
Probabilistic model
Vandetanib	4.02	255,677	0.23	79,745	352,508
BSC	3.79	175,932	–	–	–
Deterministic model
Vandetanib	4.02	255,114	0.23	79,044	336,896
BSC	3.78	176,070	–	–	–

Figure 13 presents CEACs for the pairwise comparison of vandetanib versus BSC within the EU-label (symptomatic and progressive) MTC population. Assuming a WTP threshold (λ) of £30,000 per QALY gained, the probability that vandetanib produces more net benefit than BSC is approximately 0.01.

FIGURE 13.

Analysis 2: EU-label population (symptomatic and progressive MTC), vandetanib vs. BSC (pairwise), CEACs (PFS = log-logistic, OS = log-logistic for both options).

Table 29 presents the results of the DSAs for the pairwise comparison of vandetanib versus BSC within the EU-label (symptomatic and progressive) MTC population. Across the range of DSAs considered, the ICERs for vandetanib versus BSC remain above £123,000 per QALY gained. In several scenarios in which the Gompertz function is used to model PFS, vandetanib is expected to be dominated by BSC. The DSAs indicate that the choice of utility values used in the base-case analysis produces a considerably more favourable ICER for vandetanib versus BSC than the alternative sources identified. The scenarios surrounding health state resource use assumptions do not substantially alter the ICER; however, the exclusion of post-progression vandetanib costs in both groups produces a marked increase in the ICER for vandetanib (ICER = £752,136 per QALY gained). In addition, setting the vandetanib discontinuation parameter equal to zero leads to an increase in the ICER for vandetanib (ICER = £378,272 per QALY gained). The choice of survival curves produces ICERs for vandetanib versus BSC ranging from £123,723 per QALY gained to dominated; the parametric survivor functions selected for use in the AG’s base case do not represent the most optimistic case for vandetanib, nor do they represent they least favourable.

TABLE 29 - Analysis 2: EU-label population (symptomatic and progressive MTC), vandetanib vs. BSC (pairwise), DSA results

The AG’s base-case curve choice.

Scenario	Incremental		ICER (£)
	QALYs	Costs (£)
Base case	0.23	79,044	336,896
Undiscounted health outcomes and costs	0.25	81,248	320,133
Sanofi CS utilities	0.10	79,044	822,117
DECISION study utilities	0.05	79,044	1,532,109
Cabozantinib SMC utilities	0.07	79,044	1,161,487
AE disutility doubled	0.23	79,044	340,951
AE disutility halved	0.24	79,044	334,904
AE management costs doubled	0.23	79,134	337,283
AE management costs halved	0.23	78,998	336,702
Post-progression vandetanib costs excluded	0.23	176,468	752,136
Vandetanib discontinuation parameter equal to zero	0.23	88,751	378,272
Health state resource use doubled	0.23	80,593	343,500
Health state resource use halved	0.23	78,269	333,593
No vandetanib dose reductions	0.23	85,802	365,703
Curve choice
PFS – exponential; OS – exponential	0.46	59,484	130,328
PFS – exponential; OS – Weibull	0.46	62,545	137,196
PFS – exponential; OS – Gompertz	0.59	72,938	123,723
PFS – exponential; OS – log-normal	0.39	49,372	128,083
PFS – exponential; OS – log-logistic	0.37	49,310	134,230
PFS – exponential; OS – gamma	0.43	60,268	139,406
PFS – Weibull; OS – exponential	0.22	37,245	165,924
PFS – Weibull; OS – Weibull	0.22	40,327	179,916
PFS – Weibull; OS – Gompertz	0.36	50,707	141,776
PFS – Weibull; OS – log-normal	0.15	27,155	176,631
PFS – Weibull; OS – log-logistic	0.14	27,093	199,768
PFS – Weibull; OS – gamma	0.20	38,051	189,697
PFS – Gompertz; OS – exponential	–0.08	53,486	Dominated
PFS – Gompertz; OS – Weibull	–0.08	56,486	Dominated
PFS – Gompertz; OS – Gompertz	0.07	64,762	969,254
PFS – Gompertz; OS – log-normal	–0.15	43,375	Dominated
PFS – Gompertz; OS – log-logistic	–0.17	43,313	Dominated
PFS – Gompertz; OS – gamma	–0.11	54,271	Dominated
PFS – log-normal; OS – exponential	0.39	97,481	249,691
PFS – log-normal; OS – Weibull	0.39	100,596	257,665
PFS – log-normal; OS – Gompertz	0.53	110,381	209,110
PFS – log-normal; OS – log-normal	0.32	87,433	273,140
PFS – log-normal; OS – log-logistic	0.30	87,371	289,324
PFS – log-normal; OS – gamma	0.37	98,325	267,980
PFS – log-logistic; OS – exponential	0.32	89,180	275,834
PFS – log-logistic; OS – Weibull	0.32	92,278	285,560
PFS – log-logistic; OS – Gompertz	0.46	101,633	218,981
PFS – log-logistic; OS – log-normal	0.25	79,106	312,992
PFS – log-logistic; OS – log-logistica	0.23	79,044	336,896
PFS – log-logistic; OS – gamma	0.30	90,002	300,416
PFS – gamma; OS – exponential	0.28	41,060	147,850
PFS – gamma; OS – Weibull	0.28	44,151	159,114
PFS – gamma; OS – Gompertz	0.41	54,525	132,686
PFS – gamma; OS – log-normal	0.21	30,979	149,603
PFS – gamma; OS – log-logistic	0.19	30,917	163,617
PFS – gamma; OS – gamma	0.25	41,875	164,911

Analysis 3: EU-label population (symptomatic and progressive MTC), fully incremental analysis of all options using vandetanib progression-free survival treatment effect from combined model, central estimates of cost-effectiveness

Table 30 presents the results of the fully incremental analysis of all options within the EU-label (symptomatic and progressive) MTC population based on the EXAM trial baseline, together with the PFS treatment effect derived from the EU-label population of the ZETA trial. It should be noted that this analysis assumes that OS for vandetanib is equal to that of cabozantinib, which, given the increased hazard rate/acceleration factor for PFS, may be seen to be optimistic for vandetanib. Disaggregated LYGs, QALYs and costs are presented in Table 65 in Appendix 5. Based on the probabilistic version of the model (assuming the log-logistic function for both PFS and OS), the ICER for vandetanib versus BSC is expected to be £138,405 per QALY gained, whilst the ICER for cabozantinib versus vandetanib is expected to be £195,593 per QALY gained. The deterministic version of the model produces similar results (vandetanib vs. BSC ICER = £134,817 per QALY gained; cabozantinib vs. vandetanib ICER = £195,053 per QALY gained). The disaggregated results indicate that a considerable amount of the OS gain for all options is accrued in the post-progression state.

TABLE 30 - Analysis 3: EU-label population (symptomatic and progressive MTC), fully incremental analysis of all options using vandetanib PFS treatment effect from combined model, central estimates of cost-effectiveness (PFS = log-logistic, OS = log-logistic for all options)

Option	Absolute		Incremental		ICER (£)
	QALYs	Costs (£)	QALYs	Costs (£)
Probabilistic model
Cabozantinib	2.28	88,527	0.11	20,559	195,593
Vandetanib	2.17	67,968	0.38	52,175	138,405
BSC	1.79	15,793	–	–	–
Deterministic model
Cabozantinib	2.27	87,960	0.11	21,094	195,053
Vandetanib	2.16	66,866	0.38	51,193	134,817
BSC	1.79	15,672	–	–	–

Figure 14 presents CEACs for the pairwise comparison of cabozantinib, vandetanib and BSC within the EU-label (symptomatic and progressive) MTC population, including the PFS treatment effect for vandetanib from the ZETA trial. Assuming a WTP threshold (λ) of £30,000 per QALY gained, the probability that either cabozantinib or vandetanib produces more net benefit than BSC is zero.

FIGURE 14.

Analysis 3: EU-label population (symptomatic and progressive MTC), fully incremental analysis of all options using vandetanib PFS treatment effect from combined model, CEACs (PFS = log-logistic, OS = log-logistic for all options).

Table 31 presents the results of the DSAs for the fully incremental analyses of cabozantinib, vandetanib and BSC within the EU-label (symptomatic and progressive) MTC population, including the PFS treatment effect for vandetanib from the ZETA trial. Across the range of DSAs considered, the ICERs for vandetanib remain above £85,000 per QALY gained, whilst the ICERs for cabozantinib remain above £148,000 per QALY gained. In several scenarios in which the Gompertz function is used to model OS, vandetanib is ruled out of the analysis because of extended dominance. The DSAs indicate that the choice of utility values used in the base-case analysis produces a considerably more favourable ICER for cabozantinib than for the alternative sources identified. The scenarios surrounding alternative health state resource use assumptions do not substantially alter the ICER. Setting the vandetanib discontinuation parameter equal to zero leads to a situation in which vandetanib is ruled out because of extended dominance; the ICER for cabozantinib versus BSC is estimated to be £148,169 per QALY gained. The choice of survival curves produces ICERs for vandetanib in the range £85,217 per QALY gained to extendedly dominated and ICERs for cabozantinib in the range £180,985 to £239,141 per QALY gained. The parametric survivor functions selected for use in the AG’s base case do not represent the most optimistic case for either drug, nor are they the least favourable.

TABLE 31 - Analysis 3: EU-label population (symptomatic and progressive MTC), fully incremental analysis of all options using vandetanib PFS treatment effect from combined model, DSA results

The AG’s base-case curve choice.

Scenario	ICER (£) (vs. next-best comparator)
	Cabozantinib	Vandetanib
Base case	195,053 (vs. vandetanib)	134,817 (vs. BSC)
Undiscounted health outcomes and costs	192,555 (vs. vandetanib)	119,397 (vs. BSC)
Sanofi CS utilities	298,889 (vs. vandetanib)	128,932 (vs. BSC)
DECISION study utilities	379,753 (vs. vandetanib)	135,577 (vs. BSC)
Cabozantinib SMC utilities	351,244 (vs. vandetanib)	136,191 (vs. BSC)
AE disutility doubled	203,651 (vs. vandetanib)	135,495 (vs. BSC)
AE disutility halved	191,021 (vs. vandetanib)	134,480 (vs. BSC)
AE management costs doubled	196,428 (vs. vandetanib)	134,980 (vs. BSC)
AE management costs halved	194,366 (vs. vandetanib)	134,735 (vs. BSC)
Vandetanib discontinuation parameter equal to zero	148,169 (vs. BSC)	Extended dominance
Health state resource use doubled	173,521 (vs. vandetanib)	142,718 (vs. BSC)
Health state resource use halved	205,819 (vs. vandetanib)	130,866 (vs. BSC)
No vandetanib or cabozantinib dose reductions	273,909 (vs. vandetanib)	145,927 (vs. BSC)
Curve choice
PFS – exponential; OS – exponential	204,220 (vs. vandetanib)	147,531 (vs. BSC)
PFS – exponential; OS – Weibull	204,220 (vs. vandetanib)	162,113 (vs. BSC)
PFS – exponential; OS – Gompertz	227,293 (vs. BSC)	Extended dominance
PFS – exponential; OS – log-normal	204,220 (vs. vandetanib)	138,620 (vs. BSC)
PFS – exponential; OS – log-logistic	204,220 (vs. vandetanib)	142,141 (vs. BSC)
PFS – exponential; OS – gamma	204,220 (vs. vandetanib)	157,880 (vs. BSC)
PFS – Weibull; OS – exponential	197,918 (vs. vandetanib)	133,290 (vs. BSC)
PFS – Weibull; OS – Weibull	197,908 (vs. vandetanib)	150,033 (vs. BSC)
PFS – Weibull; OS – Gompertz	232,034 (vs. BSC)	Extended dominance
PFS – Weibull; OS – log-normal	197,873 (vs. vandetanib)	123,454 (vs. BSC)
PFS – Weibull; OS – log-logistic	197,873 (vs. vandetanib)	127,303 (vs. BSC)
PFS – Weibull; OS – gamma	197,895 (vs. vandetanib)	145,084 (vs. BSC)
PFS – Gompertz; OS – exponential	207,886 (vs. vandetanib)	135,751 (vs. BSC)
PFS – Gompertz; OS – Weibull	207,886 (vs. vandetanib)	152,470 (vs. BSC)
PFS – Gompertz; OS – Gompertz	239,141 (vs. BSC)	Extended dominance
PFS – Gompertz; OS – log-normal	207,886 (vs. vandetanib)	125,894 (vs. BSC)
PFS – Gompertz; OS – log-logistic	207,886 (vs. vandetanib)	129,755 (vs. BSC)
PFS – Gompertz; OS – gamma	207,886 (vs. vandetanib)	147,537 (vs. BSC)
PFS – log-normal; OS – exponential	204,639 (vs. vandetanib)	142,355 (vs. BSC)
PFS – log-normal; OS – Weibull	204,672 (vs. vandetanib)	155,650 (vs. BSC)
PFS – log-normal; OS – Gompertz	217,141 (vs. BSC)	Extended dominance
PFS – log-normal; OS – log-normal	204,981 (vs. vandetanib)	134,340 (vs. BSC)
PFS – log-normal; OS – log-logistic	204,897 (vs. vandetanib)	137,538 (vs. BSC)
PFS – log-normal; OS – gamma	204,722 (vs. vandetanib)	151,833 (vs. BSC)
PFS – log-logistic; OS – exponential	194,919 (vs. vandetanib)	139,808 (vs. BSC)
PFS – log-logistic; OS – Weibull	194,936 (vs. vandetanib)	153,657 (vs. BSC)
PFS – log-logistic; OS – Gompertz	214,567 (vs. BSC)	Extended dominance
PFS – log-logistic; OS – log-normal	195,113 (vs. vandetanib)	131,503 (vs. BSC)
PFS – log-logistic; OS – log-logistica	195,053 (vs. vandetanib)	134,817 (vs. BSC)
PFS – log-logistic; OS – gamma	194,966 (vs. vandetanib)	149,667 (vs. BSC)
PFS – gamma; OS – exponential	180,990 (vs. vandetanib)	97,633 (vs. BSC)
PFS – gamma; OS – Weibull	180,990 (vs. vandetanib)	122,911 (vs. BSC)
PFS – gamma; OS – Gompertz	226,874 (vs. BSC)	Extended dominance
PFS – gamma; OS – log-normal	180,985 (vs. vandetanib)	85,217 (vs. BSC)
PFS – gamma; OS – log-logistic	180,985 (vs. vandetanib)	89,881 (vs. BSC)
PFS – gamma; OS – gamma	180,989 (vs. vandetanib)	114,798 (vs. BSC)

Analysis 4: EU-label population (symptomatic and progressive medullary thyroid cancer), cabozantinib and vandetanib assumed equivalent

Table 32 presents the results of the fully incremental analysis of all options within the EU-label (symptomatic and progressive) MTC population, assuming equivalent PFS and OS outcomes for cabozantinib and vandetanib, using time-to-event data from the EXAM trial. Disaggregated LYGs, QALYs and costs are presented in Table 66 in Appendix 5. Based on the probabilistic version of the model (assuming the log-logistic function for both PFS and OS), cabozantinib is expected to be dominated; this is a consequence of the more favourable grade 3 or higher AE profile and the slightly lower total relative dose intensity-adjusted drug costs for vandetanib. The probabilistic ICER for vandetanib versus BSC is estimated to be £144,841 per QALY gained. The deterministic version of the model produces a similar result (deterministic ICER = £142,279 per QALY gained). The disaggregated results indicate that a considerable proportion of the total OS gain for all options is accrued in the post-progression state.

TABLE 32 - Analysis 4: EU-label population (symptomatic and progressive MTC), cabozantinib and vandetanib assumed equivalent, central estimates of cost-effectiveness (PFS = log-logistic, OS = log-logistic for all options)

Option	Absolute		Incremental		ICER (£)
	QALYs	Costs (£)	QALYs	Costs (£)
Probabilistic model
Vandetanib	2.28	86,276	0.49	70,482	144,841
Cabozantinib	2.28	88,527	–	–	Dominated
BSC	1.79	15,793	–	–	–
Deterministic model
Vandetanib	2.28	85,736	0.49	70,063	142,279
Cabozantinib	2.27	87,960	–	–	Dominated
BSC	1.79	15,672	–	–	–

Figure 15 presents CEACs for the pairwise comparison of vandetanib versus BSC within the EU-label (symptomatic and progressive) MTC population for the analysis in which PFS and OS outcomes are assumed to be equivalent for both drugs. Assuming a WTP threshold (λ) of £30,000 per QALY gained, the probability that either cabozantinib or vandetanib produces more net benefit than BSC is zero.

FIGURE 15.

Analysis 4: EU-label population (symptomatic and progressive MTC), cabozantinib and vandetanib assumed equivalent, CEACs (PFS = log-logistic, OS = log-logistic for all options).

Table 33 presents the results of the DSAs for the fully incremental analysis of all options based on the assumption of equivalent PFS and OS outcomes for cabozantinib and vandetanib, using time-to-event outcome data from the EXAM trial. Cabozantinib remains dominated across all scenarios, except for the scenario in which the vandetanib discontinuation parameter is set equal to zero. In this scenario, the ICER for cabozantinib versus BSC is estimated to be £148,169 per QALY gained, whilst the ICER for vandetanib versus cabozantinib is estimated to be in excess of £1.35M per QALY gained. Across the remaining scenarios, the ICER for vandetanib versus BSC remains > £130,000 per QALY gained. The DSAs indicate that the choice of utility values and assumptions regarding AE impacts and health state resource use do not have a marked impact on the conclusions of the analysis. The choice of survival curves produces ICERs for vandetanib versus BSC in the range of £132,998 to £227,918 per QALY gained; the parametric survivor functions selected for use in the AG’s base case are close to the most favourable scenario for vandetanib.

TABLE 33 - Analysis 4: EU-label population (symptomatic and progressive MTC), cabozantinib and vandetanib assumed equivalent, DSA results

The AG’s base-case curve choice.

Scenario	ICER (£) (vs. next-best comparator)
	Cabozantinib	Vandetanib
Base case	Dominated	142,279 (vs. BSC)
Undiscounted health outcomes and costs	Dominated	130,280 (vs. BSC)
Sanofi CS utilities	Dominated	148,377 (vs. BSC)
DECISION study utilities	Dominated	160,069 (vs. BSC)
Cabozantinib SMC utilities	Dominated	159,049 (vs. BSC)
AE disutility doubled	Dominated	142,831 (vs. BSC)
AE disutility halved	Dominated	142,005 (vs. BSC)
AE management costs doubled	Dominated	142,405 (vs. BSC)
AE management costs halved	Dominated	142,217 (vs. BSC)
Vandetanib discontinuation parameter equal to zero	148,169 (vs. BSC)	1,354,088 (vs. cabozantinib)
Health state resource use doubled	Extended dominance	148,745 (vs. BSC)
Health state resource use halved	Dominated	139,047 (vs. BSC)
No vandetanib or cabozantinib dose reductions	Dominated	154,164 (vs. BSC)
Curve choice
PFS – exponential; OS – exponential	Dominated	151,561 (vs. BSC)
PFS – exponential; OS – Weibull	Dominated	163,420 (vs. BSC)
PFS – exponential; OS – Gompertz	Dominated	216,938 (vs. BSC)
PFS – exponential; OS – log-normal	Dominated	144,080 (vs. BSC)
PFS – exponential; OS – log-logistic	Dominated	147,058 (vs. BSC)
PFS – exponential; OS – gamma	Dominated	160,026 (vs. BSC)
PFS – Weibull; OS – exponential	Dominated	141,362 (vs. BSC)
PFS – Weibull; OS – Weibull	Dominated	154,796 (vs. BSC)
PFS – Weibull; OS – Gompertz	Dominated	221,301 (vs. BSC)
PFS – Weibull; OS – log-normal	Dominated	133,120 (vs. BSC)
PFS – Weibull; OS – log-logistic	Dominated	136,386 (vs. BSC)
PFS – Weibull; OS – gamma	Dominated	150,910 (vs. BSC)
PFS – Gompertz; OS – exponential	Dominated	141,640 (vs. BSC)
PFS – Gompertz; OS – Weibull	Dominated	155,804 (vs. BSC)
PFS – Gompertz; OS – Gompertz	Dominated	227,918 (vs. BSC)
PFS – Gompertz; OS – log-normal	Dominated	132,998 (vs. BSC)
PFS – Gompertz; OS – log-logistic	Dominated	136,411 (vs. BSC)
PFS – Gompertz; OS – gamma	Dominated	151,689 (vs. BSC)
PFS – log-normal; OS – exponential	Dominated	146,684 (vs. BSC)
PFS – log-normal; OS – Weibull	Dominated	157,787 (vs. BSC)
PFS – log-normal; OS – Gompertz	Dominated	207,458 (vs. BSC)
PFS – log-normal; OS – log-normal	Dominated	139,734 (vs. BSC)
PFS – log-normal; OS – log-logistic	Dominated	142,517 (vs. BSC)
PFS – log-normal; OS – gamma	Dominated	154,630 (vs. BSC)
PFS – log-logistic; OS – exponential	Dominated	146,363 (vs. BSC)
PFS – log-logistic; OS – Weibull	Dominated	157,175 (vs. BSC)
PFS – log-logistic; OS – Gompertz	Dominated	205,085 (vs. BSC)
PFS – log-logistic; OS – log-normal	Dominated	139,536 (vs. BSC)
PFS – log-logistic; OS – log-logistica	Dominated	142,279 (vs. BSC)
PFS – log-logistic; OS – gamma	Dominated	154,103 (vs. BSC)
PFS – gamma; OS – exponential	Dominated	141,316 (vs. BSC)
PFS – gamma; OS – Weibull	Dominated	154,181 (vs. BSC)
PFS – gamma; OS – Gompertz	Dominated	216,482 (vs. BSC)
PFS – gamma; OS – log-normal	Dominated	133,382 (vs. BSC)
PFS – gamma; OS – log-logistic	Dominated	136,532 (vs. BSC)
PFS – gamma; OS – gamma	Dominated	150,469 (vs. BSC)

Analysis 5: restricted EU-label population (symptomatic and progressive medullary thyroid cancer with carcinoembryonic antigen/calcitonin doubling times of ≤ 24 months), vandetanib versus best supportive care (pairwise)

Table 34 presents the results of the pairwise comparison of vandetanib versus BSC for the restricted EU-label population (symptomatic and progressive MTC plus CEA/CTN doubling times of ≤ 24 months). Disaggregated LYGs, QALYs and costs are presented in Table 67 in Appendix 5. This analysis closely reflects the economic analysis presented within the Sanofi CS,66 but includes survival models fitted directly to the observed data for the ZETA trial restricted EU-label subgroup, alternative assumptions regarding the vandetanib discontinuation parameter, different health state costs and different utility values. It should also be noted that this analysis is subject to potential confounding as a result of the open-label use of vandetanib; hence, post-progression vandetanib costs are included in both treatment groups. Based on the probabilistic version of the AG model (assuming the log-normal function for PFS and the Gompertz function for OS), vandetanib is expected to generate 1.61 additional QALYs at an additional cost of £107,780 compared with BSC; the ICER for vandetanib versus BSC is expected to be £66,779 per QALY gained. The deterministic version of the model yields a slightly lower ICER of £65,184 per QALY gained. The disaggregated results indicate that the majority of the incremental OS gain for vandetanib is accrued in the progression-free state. It is also noteworthy that, based on the selected Gompertz OS function, ≈12% of the vandetanib cohort are still alive at 20 years (indicated by the tail of the modelled curve). Additional analyses undertaken by the AG indicate that the ICER for vandetanib versus BSC is similar over longer time horizons (the ICER using a 30-year time horizon, excluding any general population mortality constraints, is £63,357 per QALY gained). However, the AG considers that the level of survival at 20 years may be an overestimate, and that the true ICER for vandetanib may therefore be > £67,000 per QALY gained. The impact of assuming alternative OS functions is explored within the sensitivity analyses (see Table 35).

TABLE 34 - Analysis 5: restricted EU-label population (symptomatic and progressive MTC with CEA/CTN doubling times of ≤ 24 months), vandetanib vs. BSC (pairwise), central estimates of cost-effectiveness (PFS = log-normal, OS = Gompertz for both options)

Option	Absolute		Incremental		ICER (£)
	QALYs	Costs (£)	QALYs	Costs (£)
Probabilistic model
Vandetanib	3.45	204,539	1.61	107,780	66,779
BSC	1.83	96,759	–	–	–
Deterministic model
Vandetanib	3.46	205,457	1.64	106,762	65,184
BSC	1.82	98,695	–	–	–

Figure 16 presents CEACs for the pairwise comparison of vandetanib versus BSC within the restricted EU-label MTC population. Assuming a WTP threshold (λ) of £30,000 per QALY gained, the probability that vandetanib produces more net benefit than BSC is approximately 0.02.

FIGURE 16.

Analysis 5: restricted EU-label population (symptomatic and progressive MTC with CEA/CTN doubling times of ≤ 24 months), vandetanib vs. BSC (pairwise), CEACs (PFS = log-normal, OS = Gompertz for both options).

Table 35 presents the results of the DSAs for the pairwise comparison of vandetanib versus BSC within the restricted EU-label population. As shown in the table, the ICER remains > £51,000 per QALY gained across all scenarios. The DSAs indicate that the choice of utility values used in the base-case analysis produces a slightly less favourable ICER for vandetanib versus BSC within this population compared with the alternative sources identified. The alternative assumptions regarding health state resource use and AEs do not have a marked impact on the cost-effectiveness of vandetanib. In this population, excluding the post-progression vandetanib costs increases the ICER to £84,438 per QALY gained. Setting the vandetanib discontinuation parameter equal to zero increases the ICER to £76,352 per QALY gained. The choice of survival curves produces ICERs for vandetanib versus BSC in the range of £51,194 to £71,128 per QALY gained; the curves used in the AG’s base-case analysis (PFS = log-normal, OS = Gompertz) represent neither the most favourable nor the least favourable scenario for vandetanib within the restricted EU-label population.

TABLE 35 - Analysis 5: restricted EU-label population (symptomatic and progressive MTC with CEA/CTN doubling times of ≤ 24 months), vandetanib vs. BSC (pairwise), DSA results

The AG’s base-case curve choice.

Scenario	Incremental		ICER (£)
	QALYs	Costs (£)
Base case	1.64	106,762	65,184
Undiscounted health outcomes and costs	2.23	137,406	61,584
Sanofi CS utilities	1.76	106,762	60,576
DECISION study utilities	1.69	106,762	63,186
Cabozantinib SMC utilities	1.68	106,762	63,683
AE disutility doubled	1.64	106,762	65,295
AE disutility halved	1.64	106,762	65,128
AE management costs doubled	1.64	106,853	65,239
AE management costs halved	1.64	106,717	65,156
Post-progression vandetanib costs excluded	1.64	138,298	84,438
Vandetanib discontinuation parameter equal to zero	1.64	125,054	76,352
Health state resource use doubled	1.64	115,552	70,551
Health state resource use halved	1.64	102,367	62,500
No vandetanib dose reductions	1.64	116,928	71,390
Curve choice
PFS – exponential; OS – exponential	1.30	81,931	63,007
PFS – exponential; OS – Weibull	1.30	82,041	63,165
PFS – exponential; OS – Gompertz	1.50	90,264	60,296
PFS – exponential; OS – log-normal	1.28	73,914	57,821
PFS – exponential; OS – log-logistic	1.06	56,920	53,857
PFS – exponential; OS – gamma	1.27	80,262	63,172
PFS – Weibull; OS – exponential	1.25	77,205	61,602
PFS – Weibull; OS – Weibull	1.25	77,316	61,765
PFS – Weibull; OS – Gompertz	1.45	85,538	58,993
PFS – Weibull; OS – log-normal	1.23	69,188	56,193
PFS – Weibull; OS – log-logistic	1.01	52,195	51,687
PFS – Weibull; OS – gamma	1.22	75,537	61,739
PFS – Gompertz; OS – exponential	1.40	99,812	71,119
PFS – Gompertz; OS – Weibull	1.41	99,165	70,439
PFS – Gompertz; OS – Gompertz	1.61	106,531	66,060
PFS – Gompertz; OS – log-normal	1.38	91,856	66,516
PFS – Gompertz; OS – log-logistic	1.16	74,863	64,564
PFS – Gompertz; OS – gamma	1.38	97,861	71,128
PFS – log-normal; OS – exponential	1.44	98,830	68,718
PFS – log-normal; OS – Weibull	1.44	98,899	68,821
PFS – log-normal; OS – Gompertza	1.64	106,762	65,184
PFS – log-normal; OS – log-normal	1.42	90,824	64,128
PFS – log-normal; OS – log-logistic	1.19	73,831	61,791
PFS – log-normal; OS – gamma	1.41	97,169	68,989
PFS – log-logistic; OS – exponential	1.44	100,247	69,779
PFS – log-logistic; OS – Weibull	1.44	99,816	69,348
PFS – log-logistic; OS – Gompertz	1.64	107,120	65,132
PFS – log-logistic; OS – log-normal	1.41	92,230	65,198
PFS – log-logistic; OS – log-logistic	1.19	75,237	63,056
PFS – log-logistic; OS – gamma	1.41	98,433	69,923
PFS – gamma; OS – exponential	1.25	76,695	61,206
PFS – gamma; OS – Weibull	1.25	76,806	61,368
PFS – gamma; OS – Gompertz	1.45	85,028	58,651
PFS – gamma; OS – log-normal	1.23	68,678	55,789
PFS – gamma; OS – log-logistic	1.01	51,685	51,194
PFS – gamma; OS – gamma	1.22	75,027	61,334

Assessment group analysis 1: EU-label population (symptomatic and progressive medullary thyroid cancer), pairwise economic evaluation of cabozantinib versus best supportive care

Based on the AG’s probabilistic model (assuming the log-logistic function for both PFS and OS), the ICER for cabozantinib versus BSC is expected to be £150,874 per QALY gained. The DSAs indicate that the AG’s base case is close to the most favourable scenario.

Assessment group analysis 2: EU-label population (symptomatic and progressive medullary thyroid cancer), pairwise economic evaluation of vandetanib versus best supportive care

Based on the probabilistic version of the AG model (assuming the log-logistic function for both PFS and OS), the ICER for vandetanib versus BSC is expected to be £352,508 per QALY gained. The DSAs indicate that the AG’s base case does not represent the most optimistic case for vandetanib, nor does it reflect the most pessimistic scenario.

Assessment group analysis 3: EU-label population (symptomatic and progressive medullary thyroid cancer), fully incremental analysis, vandetanib progression-free survival treatment effect applied to EXAM trial placebo baseline, vandetanib overall survival assumed equivalent to cabozantinib overall survival

Within this analysis, the ICER for vandetanib versus BSC is expected to be £138,405 per QALY gained, whilst the ICER for cabozantinib versus vandetanib is expected to £195,593 per QALY gained. The DSAs indicate that the AG’s base case represents neither the most favourable nor the least favourable scenario for either drug.

Assessment group analysis 4: EU-label population (symptomatic and progressive medullary thyroid cancer), fully incremental analysis, progression-free and overall survival outcomes assumed equivalent for vandetanib and cabozantinib

Based on the probabilistic version of the model (assuming the log-logistic function for both PFS and OS), cabozantinib is expected to be dominated; this is a consequence of the more favourable grade 3 or higher AE profile and the slightly lower total relative dose intensity-adjusted drug costs for vandetanib. The probabilistic ICER for vandetanib versus BSC is expected to be £144,841 per QALY gained. The DSAs indicate that the AG’s base case represents one of the more favourable scenarios for vandetanib.

Assessment group analysis 5: restricted EU-label population (symptomatic and progressive medullary thyroid cancer plus carcinoembryonic antigen/calcitonin doubling times of ≤ 24 months), pairwise economic evaluation of vandetanib versus best supportive care

Based on the probabilistic version of the AG model (assuming the log-normal function for PFS and the Gompertz function for OS), the ICER for vandetanib versus BSC is expected to be £66,779 per QALY gained. The DSAs indicate that the AG’s base case represents neither a highly favourable nor a highly unfavourable scenario for vandetanib.

Table 38 highlights the key differences between the AG model and the Sanofi model. Although the two models are very similar in terms of their structure and definition of parameters, the key differences between the analyses relate to (1) the scope of the economic comparisons, (2) the time-to-event data used to inform the analyses (covariate-adjusted ITT/safety data set vs. actual subgroup data), (3) the source of health utility values, (4) assumptions regarding the costs associated with BSC and (5) assumptions regarding the costs of vandetanib in patients who discontinue therapy prior to disease progression.

Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R)

Date range searched: 1946 to present.

Date searched: 2 November 2016.

Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R)

Date range searched: 1946 to present.

Date searched: 3 November 2016.

EMBASE

Date range searched: 1974 to 1 November 2016.

Date searched: 3 November 2016.

Web of Science Core Collection

Science Citation Index Expanded (1900–).

Conference Proceedings Citation Index – Science (1990–).

Date searched: 3 November 2016.

Cochrane Database of Systematic Reviews: Wiley Online Library

Health Technology Assessment database: Wiley Online Library.

NHS Economic Evaluation Database: Wiley Online Library.

Date range searched: 1995–2015.

Date searched: 3 November 2016.

Cumulative Index to Nursing and Allied Health Literature

Date range searched: 1982 to present.

Date searched: 3 November 2016.

Epub Ahead of Print, In-Process & Other Non-Indexed Citations, Ovid MEDLINE(R) Daily and Ovid MEDLINE(R)

Date range searched: 1946 to present.

Date searched: 3 November 2016.

EMBASE

Date range searched: 1974 to 1 November 2016.

Date searched: 3 November 2016.

Web of Science Core Collection

Science Citation Index Expanded (1900–).

Conference Proceedings Citation Index – Science (1990–).

Date searched: 3 November 2016.

Cochrane Database of Systematic Reviews: Wiley Online Library

Health Technology Assessment database: Wiley Online Library.

NHS Economic Evaluation Database: Wiley Online Library.

Date range searched: 1995–2015.

Date searched: 3 November 2016.

Cumulative Index to Nursing and Allied Health Literature

Date range searched: 1982 to present.

Date searched: 3 November 2016.