Findings from the systematic review of clinical effectiveness literature

Quantity and quality of research available

Included studies

The process of study selection is shown in the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) flow diagram in Figure 4. The electronic searches yielded 2358 papers, and six additional references5–8,57,58 were identified through other sources. In total, the AG included 93 papers5–8,33,51,52,57–142 reporting on 24 separate studies and reviews: two unique RCTs,51,52 13 unique systematic reviews5–8,33,57,61,93,97,104,127,138,141 and nine unique prospective observational studies.59,77,78,81,88,101,103,126,135

FIGURE 4 The PRISMA flow diagram: studies included in AG’s systematic review. Reproduced with permission from Fleeman et al.54 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Excluded studies

A full list of studies excluded at stage 2 with reasons for exclusion is presented in Appendix 2, Table 32.

Evidence from randomised controlled trials

Only two RCTs were identified as relevant for inclusion in the AG’s systematic review: SELECT and DECISION. Except when stated otherwise, all information about these two trials has been extracted from the two key trial publications.51,52

Trial characteristics

A summary of the characteristics of the two included trials is provided in Table 4. Both trials were Phase III, multicentre, double-blind RCTs designed to compare the intervention of interest (lenvatinib or sorafenib) with placebo. Subjects were randomised in a ratio of 2 : 1 to the intervention and comparator arms of SELECT, whereas they were randomised in a ratio of 1 : 1 in DECISION. In both trials, the primary outcome was progression-free survival (PFS) assessed by blinded independent review. Both trials also reported investigator-assessed PFS. Unless otherwise specified, in the remainder of this AG report on clinical effectiveness, PFS refers to PFS assessed by blinded independent review.

Analysis of clinical efficacy

All efficacy outcomes from both trials, including tumour response evaluations in SELECT, were undertaken using data from the intention-to-treat (ITT) population. Tumour response evaluations in DECISION were undertaken using data from the per-protocol population (i.e. randomised patients who were evaluable for tumour response with imaging data, had received the intervention or placebo as allocated and had no major protocol deviations).

Analysis of safety

Safety analyses for both trials were undertaken using data from the population of patients who were randomised and received at least one dose of study drug and had at least one post-baseline safety evaluation. In SELECT, the numbers of patients included in the ITT and safety populations were identical.

Patients eligible for inclusion

A summary of the criteria describing patient eligibility for entry into SELECT and DECISION is presented in Appendix 3, Table 33. Both trials only included patients with RR-DTC and who had an Eastern Cooperative Oncology Group (ECOG) PS of 0–2. As highlighted in Chapter 1, Radioactive iodine-refractory differentiated thyroid cancer, there is no universally agreed definition of RR-DTC. The definitions used to define RR-DTC in the two trials were broadly similar (see Appendix 3, Table 34, for definitions employed by the trials for RR-DTC).

The main difference in trial eligibility was that SELECT permitted the enrolment of patients who had been previously treated with a vascular endothelial growth factor receptor (VEGFR)-targeted therapy (including sorafenib) and DECISION did not. Age, region and VEGFR-targeted therapy were stratification factors in SELECT, whereas age and region were stratification factors in DECISION.

Dose modifications/interruptions and concomitant therapy

In both trials, the starting dose for treatment with lenvatinib or sorafenib was the licensed dose (24 mg and 800 mg, respectively). Both trials permitted dose modifications or interruptions. The criteria were not stated in the protocol for SELECT but the summary of product characteristics (SmPC)45 includes a dose/toxicity management plan for lenvatinib. For DECISION, Brose et al.72 stated that dose modifications or interruptions were allowed, based on specific criteria, for grade 2 to grade 3 hand–foot syndrome and other adverse events (AEs).

A summary of the concomitant therapies permitted and prohibited in each trial is presented in Appendix 3, Table 35. Although neither trial describes BSC for patients in either arm, permitted concomitant therapies could be considered to be BSC and were available to patients in both arms of both trials. The main difference between the two trials is that palliative radiotherapy, which is commonly available as part of BSC in UK NHS clinical practice, was not permitted in either arm of SELECT.

Subgroup analyses

In SELECT, subgroup analyses were prespecified for patients previously treated with a VEGFR-targeted therapy and for those who were not. Both trials also included prespecified subgroup analyses for age, region, sex and histology. Subgroup analyses were prespecified for PFS, OS and objective tumour response rate (ORR) in SELECT but only for PFS in DECISION. Other prespecified subgroup analyses in SELECT were for race and for patients whose TSH level was highest prior to progression. Other prespecified subgroup analyses in DECISION included site of metastasis, FDG take-up, prior radioactive iodine cumulative dosing, tumour burden as measured by number of target or non-target lesions and as measured by the sum of target diameters. Many other post hoc subgroup analyses were also conducted for both trials (see Appendix 3, Table 37).

Follow-up, dose intensity and treatment crossover and other subsequent therapy received

At the time of the primary data cut-off points for both trials, OS data were immature. Therefore, for both trials, OS was updated at two subsequent data cut-off points. The median duration of follow-up at each data cut-off point was approximately 17 months at the first data cut-off point in both trials and there were approximately 20 months of additional follow-up in both trials by the final data cut-off point (see Appendix 3, Table 36).

Patients were eligible to receive treatment (intervention or placebo) in both trials until disease progression. An important feature of both trials is that, on disease progression, patients were unblinded and permitted to cross over from the placebo arm to the active treatment arm. In both trials, patients who crossed over were entered into an open-label extension phase of the same trial. In DECISION, patients who had progressed on sorafenib were also eligible to enter the open-label extension phase of the trial and receive further sorafenib until further disease progression. However, patients who progressed on lenvatinib in SELECT were not permitted to receive additional lenvatinib in the open-label extension phase. Information on treatment crossover and subsequent treatment received is reported in Table 5; it is evident that the majority of patients in both placebo arms, and particularly in the placebo arm of SELECT, crossed over to receive lenvatinib or sorafenib.

In addition, some patients received subsequent anti-cancer treatments, not part of the trial protocols, on disease progression (see Appendix 3, Table 39). In SELECT, at the first data cut-off point (November 2013), 15.7% of patients randomised to lenvatinib and 12.2% of patients randomised to placebo received subsequent treatment. In DECISION, at the first data cut-off point (August 2012), 20.3% of patients randomised to sorafenib and 8.6% of patients randomised to placebo received subsequent treatments. For the most part, subsequent treatment in both trials comprised antineoplastic and immunomodulating agents. The specific antineoplastic and immunomodulating agents were reported only for SELECT, as data were not collected on the specific agents used during the trial follow-up for DECISION. Most commonly, patients received pazopanib (Votrient^®, Novartis) (17.1% and 18.8% of patients who received subsequent therapy in the lenvatinib and placebo arms, respectively) and/or sorafenib (14.6% and 12.5% of patients who received subsequent therapy in the lenvatinib and placebo arms, respectively).

Methods used for adjusting for treatment crossover

As patients in both trials were permitted to cross over to receive the intervention drug on disease progression, the OS results are likely to be confounded. The authors of the SELECT publication51 employed the rank-preserving structural failure time model (RPSFTM) to adjust the OS results for patient crossover. The OS results from DECISION have been adjusted using both the RPSFTM and the iterative parameter estimation (IPE). The unadjusted and adjusted OS analyses have been reported in conference abstracts for SELECT,87 DECISION58,68,110 and in the company submissions.7,8

As patients were not censored when they received postprogression treatments, the RPSFTM and IPE methods implicitly included all subsequent therapies as an inherent part of the intervention/control treatment effect. In other words, it is assumed that the subsequent therapy administered to patients in each arm of the trial is reflective of the subsequent therapy that would have been offered to patients receiving the same treatment in clinical practice.

The RPSFTM and IPE methods also both rely critically on the ‘common treatment effect’ assumption, that is, the effect of receiving the experimental treatment is the same when received on diagnosis (i.e. in patients initially randomised to the experimental arm) as it is in treatment switchers (i.e. patients from the control arm who switch to receive the experimental treatment). In practice, it is unlikely that the ‘common treatment effect’ assumption will ever be completely true; however, it is appropriate to use RPSFTM/IPE methods if the assumption is likely to be approximately true.143 Clinical advice to the AG was that for both SELECT and DECISION it is reasonable to assume that patients who switched from the placebo arm to receive the experimental treatment (i.e. lenvatinib/sorafenib) would experience the same treatment effect as patients who were originally randomised to the experimental arm.

In addition to the assumptions that are common to both the RPSFTM and the IPE methods, the IPE method also assumes that survival times follow a parametric distribution. To implement this method, a suitable parametric model must be identified, which can be problematic. The AG has been unable to identify information on how the IPE analysis was carried out using data from DECISION, including details of the parametric model chosen, and so is not able to comment on the suitability of this method.

Generally, the key assumption of a ‘common treatment effect’ that underpins RPSFTM appears to be valid, and because a large number of placebo patients crossed over to active treatment in both trials, the AG is of the opinion that RPSFTM is the most suitable method for adjusting for treatment switching in SELECT and DECISION. However, a caveat to the use of the RPSFTM-adjusted OS results for both trials is that differences in poststudy (postprogression) anti-cancer treatments administered to patients in each treatment arm are not accounted for in this analysis.

Participant characteristics

Overall, the baseline characteristics of patients included in SELECT and in DECISION were balanced between treatment arms (Table 6). Nevertheless, there are a few notable differences between the treatment arms and also across the trials.

In SELECT, there was a lower proportion of males in the lenvatinib arm (47.9%) than in the placebo arm (57.3%). The median time from diagnosis of DTC to randomisation was shorter in the lenvatinib arm than in the placebo arm (66.0 vs. 73.9 months). Compared with the placebo arm, a smaller proportion of patients in the lenvatinib arm had metastases in the lung [86.6% (lenvatinib) vs. 94.7% (placebo)] or liver [16.5% (lenvatinib) vs. 21.4% (placebo)].

In DECISION, a higher proportion of patients in the sorafenib arm had metastases in the lymph node (54.6%) or pleura (19.3%) than in the placebo arm (48.1% and 11.4%, respectively). There was a higher proportion of males in the sorafenib arm (50.2%) than in the placebo arm (45.2%).

As previously highlighted, patients in SELECT could have been previously treated with a VEGFR-targeted therapy (including sorafenib) prior to trial entry whereas patients in DECISION could not. Approximately one-quarter (23.7%) of patients in SELECT had received prior treatment with a VEGFR-targeted therapy. In the lenvatinib arm, of 66 patients previously treated with a VEGFR-targeted therapy, 51 patients (77.2%) were treated with sorafenib. In the placebo arm, of 27 patients previously treated with a VEGFR-targeted therapy, 21 patients (77.8%) were treated with sorafenib. Other VEGFR-targeted therapies used prior to trial entry in SELECT included sunitinib (Sutent^®, Pfizer) and pazopanib. The median duration of any prior therapy was ≈11 months in both arms.

A higher proportion of enrolled patients were from North America in SELECT than in DECISION (29.6% vs. 17.3%, respectively) and a lower proportion of patients were from Europe in SELECT than in DECISION (49.7% vs. 59.7%, respectively). A greater proportion of patients were white in SELECT (79.3%) than in DECISION (60.2%). A higher proportion of patients had bone metastases in SELECT than in DECISION (38.8% vs. 27.1%, respectively).

Comparison of assessments of risk of bias

A summary of the risk-of-bias assessments for both trials is reproduced in Appendix 3, Table 48. Overall, the AG considered the risk of bias to be low in both trials.

Consideration of proportional hazards assumption

Cox proportional hazards (PHs) modelling was used to generate PFS, unadjusted OS and adjusted OS hazard ratios (HRs) from data collected during SELECT and DECISION. The validity of this method relies on the event hazards associated with the intervention and comparator data being proportional over time within each trial. The AG assessed the validity of the PH assumption for all analyses, when possible, provided in the submissions from Eisai Ltd8 and Bayer HealthCare7 that included a HR result (see Appendix 6 for methods and results). The AG concluded that the PH assumption was not valid for PFS, unadjusted OS or RPSFTM-adjusted OS in SELECT or for PFS or RPSFTM-adjusted OS in DECISION. This means that the majority of the survival HRs generated using data from SELECT and DECISION and, consequently, statements about the statistical significance of results should be interpreted with caution.

Overall survival

A summary of the unadjusted and adjusted OS findings from the most recent data cut-off points from both trials is presented in Table 7. The findings for all data cut-off points are summarised in Appendix 3, Table 38.

In both trials, there was no statistically significant difference in unadjusted OS between trial arms. However, when RPSFTM was used, patients in the lenvatinib arm had a statistically significant improvement in OS when compared with patients in the placebo arm in SELECT. The difference in OS between sorafenib and placebo was not reported to be statistically significant when using either the RPSFTM or IPE method in DECISION.

Progression-free survival

In both trials, the primary outcome was PFS by blinded independent review. The findings for PFS reported in SELECT and DECISION are summarised for the first data cut-off points (November 2013 and August 2012, respectively) in Table 8 because this was the only data cut-off point for which PFS results had been published for both trials.

In SELECT, there was a median improvement in PFS (blinded independent review) of 14.7 months with lenvatinib when compared with placebo. In DECISION, there was a 5-month median improvement in PFS (blinded independent review) with sorafenib when compared with placebo. The differences in median PFS assessed by investigators were marginally decreased in SELECT (12.9 months) and marginally increased in DECISION (5.4 months). However, the HRs in both trials were similar to those from the assessments by blinded independent review.

The SELECT trial is the only trial that also reports PFS for another data cut-off point.85,86 This was available for investigator-assessed PFS at the third data cut-off point (August 2015). Compared with the first data cut-off point, median PFS was reported to be slightly higher in the lenvatinib arm at the third data cut-off point (19.4 months), but the median PFS remained the same in the placebo arm (3.7 months), a difference of 15.7 months. However, for both data cut-off points, the HR between arms was identical (0.24) and reported to be statistically significant (p < 0.001).

The findings for all data cut-off points are summarised in Appendix 3, Table 41.

Objective tumour response

The findings for objective tumour response are reported in Appendix 3, Table 42. In both trials, the tumour response assessment was conducted by blinded independent review at the first data cut-off point and favoured patients in the intervention arms compared with patients in the placebo arms. It is noticeable that the difference in ORR between the intervention and placebo arms was much greater for patients treated with lenvatinib in SELECT (63.2%) than for those treated with sorafenib in DECISION (11.7%). This is attributable to the much higher proportion of patients who were treated with lenvatinib and had a partial response in SELECT than the proportion of patients treated with sorafenib in DECISION. Complete responses were only reported for patients treated with lenvatinib, albeit there were very few patients (1.5%). ORR was statistically significantly improved in both trials for patients treated with either lenvatinib or sorafenib when compared with placebo.

The objective tumour response evaluations for SELECT were conducted using an ITT analysis. In DECISION, patients for whom it was not possible to evaluate a tumour response were excluded from the analysis (as per the requirements of a per-protocol analysis). If all patients are included in the evaluations using ORR data from DECISION, the ORR is marginally decreased in both arms: 11.6% for sorafenib and 0.5% for placebo.

Time to response was reported only for SELECT. The median time to response was 2.0 months for patients treated with lenvatinib compared with 5.6 months for patients in the placebo arm. The median duration of response was not estimable for patients in SELECT; however, for those treated with lenvatinib, the restricted mean was 17.34 months. Time to response was not reported in DECISION, but the median duration of response was 10.2 months for patients treated with sorafenib.

Both trials also assessed disease control rates (complete response + partial response + stable disease), and SELECT reported clinical benefit rate (complete response + partial response + durable stable disease). In each trial, the findings were statistically significantly in favour of lenvatinib or sorafenib compared with placebo. However, comparisons between trials cannot be easily made as the definition of disease control rate differed across trials because of differences in the length of stable disease required for control. SELECT required a stable disease of ≥ 7 weeks whereas DECISION required a length of ≥ 4 weeks; however, both trials did report the proportion of patients with stable disease of ≥ 6 months. This was similar in the placebo arms of both trials (SELECT, 29.8%; DECISION, 33.2%), whereas in the intervention arms, it was 15.3% for patients treated with lenvatinib and 41.8% for patients treated with sorafenib. Therefore, a clinical benefit at 6 months was reported by 79.5% of patients treated with lenvatinib compared with 31.3% of patients who received the placebo in SELECT, and 54.0% patients treated with sorafenib compared with 33.7% who received the placebo in DECISION. In the submission from Bayer HealthCare,7 it is noted that most sorafenib-treated patients (77%) experienced target lesion tumour shrinkage (compared with 28% of patients in the placebo arm).

Safety findings

Safety data from SELECT and DECISION were reported for the first data cut-off point (November 2013 and August 2012, respectively). For the individual types of AEs experienced by patients, the published paper for SELECT presented data for treatment-related AEs whereas the published paper for DECISION presented data for any treatment-emergent AEs. Therefore, data for specific types of treatment-emergent AEs were extracted from the pharmaceutical company submission (Eisai Ltd8) for SELECT.

All-grade and grade ≥ 3 adverse events

Nearly all of the patients who received lenvatinib or sorafenib reported an AE, and ≈90% of patients who received placebo reported an AE. AEs that were reported by ≥ 30% of patients and grade ≥ 3 AEs that were reported by ≥ 1.5% of patients in any of the arms are summarised in Appendix 3, Tables 44 and 45. All types of AEs were more common in patients treated with lenvatinib or sorafenib than in patients in the placebo arms of both trials. Hand–foot syndrome was reported by approximately three-quarters of patients in DECISION. Approximately two-thirds of patients reported all-grade hypertension or diarrhoea when treated with lenvatinib in SELECT, similar to the proportion treated with sorafenib reporting all-grade diarrhoea or alopecia in DECISION. Weight loss was reported by approximately half of all patients treated with either lenvatinib or sorafenib. By far the most common grade ≥ 3 AEs for patients treated with lenvatinib and sorafenib were hypertension (> 40%) and hand–foot syndrome (> 20%), respectively.

Serious adverse events (including fatal adverse events)

Serious adverse events (SAEs) reported in SELECT and DECISION are summarised in Appendix 3, Table 45. In SELECT, approximately half of the patients in the lenvatinib arm reported a SAE. Just over one-third of patients reported a SAE in the sorafenib arm of DECISION. Approximately one-quarter of patients in the placebo arms of both trials reported a SAE. The only SAE reported by ≥ 2% of patients in both trials was dyspnoea, which was at least as common for patients who received placebo as for those who received lenvatinib or sorafenib. The most common SAEs (reported by ≥ 3% of patients) reported for patients treated with lenvatinib in SELECT were pneumonia and hypertension. The most common SAEs (reported by ≥ 3% of patients) reported by patients treated with sorafenib in DECISION were secondary malignancy and pleural effusion.

Deaths from AEs were reported for 7.7% of patients treated with lenvatinib and 4.6% of patients in the placebo arm in SELECT. Fatal AEs in DECISION were reported for 5.8% of patients treated with sorafenib and 2.9% of patients in the placebo arm.

Treatment-related adverse events

A summary of treatment-related AEs is presented in Appendix 3, Table 46. A very high proportion of all-grade AEs (≥ 96%) were considered to be related to treatment with lenvatinib or sorafenib. The proportion of all-grade AEs considered to be treatment related was high (> 50%) in the placebo arms of both trials.

In SELECT, the causes of death considered to be treatment related in the lenvatinib arm were one case each of pulmonary embolism, haemorrhagic stroke and general deterioration of physical health; three cases were reported as deaths or sudden deaths (not otherwise specified). DECISION was the only trial in which a patient in the placebo arm was considered to have died because of a treatment-related AE. The cause of death for this patient was subdural haematoma. The cause of death for a patient in the sorafenib arm that was considered to be treatment related was myocardial infarction.

Timing of adverse events

In both trials, there have been subsequent analyses of the timing of AE occurrences in the treatment cycle reported. For SELECT, Haddad et al.91 reported the incidence and timing of five AEs: proteinuria, diarrhoea, fatigue/asthenia/malaise, rash and hand–foot syndrome. Hypertension was a notable AE omitted from the analysis. For DECISION, detailed analysis of the AE occurrence patterns in patients is published in a peer-reviewed paper by Worden et al.139 Findings from the two trials cannot be easily compared because Haddad et al.91 reported their findings as median time to first onset and median time to last resolution, whereas Worden et al.139 reported the proportion of AEs occurring during each cycle. The AEs reported included hand–foot syndrome, rash/desquamation, diarrhoea, fatigue, hypertension, weight loss, increased TSH levels and hypocalcaemia. Increased TSH levels were described as a ‘study-specific’ AE, with a maximum severity of grade 1; this AE was reported by 69 patients (33.3%) treated with sorafenib.139

In SELECT, Haddad et al.91 found that generally AEs for patients treated with lenvatinib occurred early in the treatment process and were resolved. Median time to onset for patients treated with lenvatinib ranged from 3.0 weeks with fatigue/asthenia/malaise to 12.1 weeks with diarrhoea. With regard to resolution, this ranged from a median of 5.9 weeks with rash to a median of 20.0 weeks with hand–foot syndrome.

In DECISION, Worden et al.139 found that in patients treated with sorafenib, the incidence of AEs was usually highest in the first cycle or the first two cycles. Severity tended to diminish with each cycle (over the first nine cycles). The prevalence of AEs (defined as the number of patients with a new or continuing AE during a treatment cycle) tended to remain stable. Diarrhoea and TSH were notable exceptions in that prevalence steadily increased over the first five or six cycles, at which point the prevalence peaked. Only weight loss, which was primarily grade 1 or grade 2 and highest in the first four cycles, tended to increase in severity over time (from grade 1 to grade 2: a greater proportion of patients experienced grade 2 toxicity in cycle 9 compared with cycles 1 and 2). The authors noted that in general, AEs with sorafenib were manageable over time following dose modification and/or concomitant medications, such as antidiarrhoeals, antihypertensives or dermatological preparations.

Dose modifications

Dose modifications as a result of AEs were more common for patients treated with lenvatinib and sorafenib than for those who received placebo (Table 9). It is of note that the incidence of dose interruptions with lenvatinib in SELECT was higher than with sorafenib in DECISION. The incidence of dose interruptions and dose reductions were lower in the placebo arm of SELECT than in the placebo arm of DECISION.

It is reported that, in SELECT, the most common AEs developing during treatment that led to a dose interruption or reduction among patients receiving lenvatinib were diarrhoea (22.6%), hypertension (19.9%), proteinuria (18.8%) and decreased appetite (18.0%). It is also noted that four patients in the lenvatinib arm (1.5%) required dose adjustments owing to hypocalcaemia. In the submission from Eisai Ltd,8 it is further noted that 1.1% of patients discontinued treatment because of hypertension and 1.1% of patients discontinued because of asthenia. In DECISION, it is reported that hand–foot syndrome was the most common reason for sorafenib dose interruptions (26.6%), reductions (33.8%) and withdrawals (5.3%).

Health-related quality-of-life findings

It was reported in the European Public Assessment Report (EPAR)27 that, although HRQoL data were not collected in the randomised part of SELECT,51 HRQoL would be assessed in 30 patients who participated in the open-label extension phase of the trial. The AG is unaware of whether or not these findings have been published.

For DECISION, HRQoL was reported in a conference abstract by Schlumberger et al.120 More detailed HRQoL results were also reported in the submission from Bayer HealthCare.7 Cancer-specific HRQoL was measured using the Functional Assessment of Cancer Therapy – General (FACT-G) questionnaire144 and general health status was measured using the generic EuroQol-5 Dimensions, three-level version (EQ-5D-3L) and the EuroQol-5 Dimensions (EQ-5D) visual analogue scale (VAS).145 The FACT-G questionnaire is a validated 27-item questionnaire designed to assess the following dimensions in cancer patients: physical well-being, social/family well-being, emotional well-being and functional well-being. The FACT-G total score ranges from 0 to 108 points, with higher scores representing a better HRQoL. Similarly, the EQ-5D is a validated instrument in which higher scores represent better health status.

All questionnaires were self-administered at baseline and day 1 of every 28-day cycle. The overall questionnaire completion rate during the trial was reported by the authors to be 96%.120 However, the actual number of patients completing the questionnaires reduces with each cycle because only patients with progression-free disease were asked to complete the questionnaires. Thus, for example, as shown in the submission from Bayer HealthCare7 by the response to one of the physical well-being questions, by cycle 13 the number of patients who responded was 87: 40.1% of all patients enrolled into the trial.

Functional Assessment of Cancer Therapy – General

Minimally important differences in the FACT-G total score (i.e. a difference considered to be clinically meaningful) have been reported to range between 3 and 7 points.144 At baseline, it was reported7,120 that FACT-G scores were comparable to a normative adult cancer population, the mean scores being 81 points [standard deviation (SD) 15 points] in the sorafenib arm and 82 points (SD 14 points) in the placebo arm. However, at the first assessment (cycle 2, day 1), the score for the sorafenib arm had fallen to 76 points (SD 15 points), whereas the score in the placebo arm remained very similar to baseline. The authors of the conference abstract120 reported that the scores in the sorafenib arm thereafter remained similar to the scores at the first assessment, whereas the scores remained similar to the baseline scores in the placebo arm. A mixed linear model estimated that, compared with placebo, the FACT-G score was 3.45 points lower in the sorafenib arm (p = 0.0006), representing a clinically meaningful difference between arms in favour of the placebo arm. The authors attributed the diminished HRQoL score to AEs. Indeed, the submission from Bayer HealthCare7 noted that in response to the FACT-G physical well-being domain question ‘I am bothered by side effects’, the proportion of patients in the sorafenib arm who replied ‘quite a bit’ or ‘very much’ increased from 1.5% at cycle 1 to 29.6% at cycle 2. However, this proportion gradually diminished over time: it was 16.8% by cycle 6 and 8.0% by cycle 13.

EuroQol-5 Dimensions index and visual analogue scale

For UK utility scores, a change of 0.10 points on the EQ-5D index has been reported by Pickard et al.146 to be clinically meaningful for all cancers (using ECOG PS as the anchor). Similarly, the same study reported a change of at ≥ 7 points on the VAS to be clinically meaningful.146 It was reported in DECISION7,120 that the patterns for EQ-5D index and VAS were similar to that of the FACT-G; after the first assessment, the scores in the sorafenib arm were lower than the scores in the placebo arm. Although the between-arm differences were statistically significant (p < 0.0001 for both EQ-5D index and VAS), the treatment effects (–0.07 and –6.75, respectively) were of a small magnitude and did not reach the threshold for a clinically meaningful difference. It is reported in the submission from Bayer HealthCare7 that dimensions in the EQ-5D index that are sensitive to AEs include mobility, usual activities and pain/discomfort.

Subgroup analyses from randomised controlled trials

Only subgroup analyses considered by the AG to be of direct relevance to the decision problem have been reported in the remainder of this report. The AG considered the following subgroup analyses to be relevant (with rationale given):

• patients previously treated and not previously treated with TKIs (prespecified subgroup in the NICE scope53 and AG decision problem)
• patients with and without symptomatic disease at baseline (as highlighted in Chapter 1, Treatment options for patients with radioactive iodine-refractory differentiated thyroid cancer, systemic treatment is recommended for patients who have symptomatic disease)
• analyses of subgroups that were prespecified in the trials and in which there appeared to be differences in baseline characteristics within or across trials (as differences in baseline characteristics may influence results).

As previously highlighted, the AG concluded that the assumption of PH does not hold in any of the analyses that it was able to check other than unadjusted OS in DECISION. This means that the majority of the survival HRs generated using data from SELECT and DECISION and, consequently, statements about the statistical significance of results should be interpreted with caution.

Patients previously treated and patients not previously treated with tyrosine kinase inhibitors

Subgroup analyses have been reported for patients previously treated with a TKI (e.g. VEGFR-targeted therapy) in SELECT but only for PFS and ORR.51,105,106 No patients in DECISION had received prior treatment with a TKI.

Results from subgroup analyses using data from SELECT51,105,106 showed that PFS was statistically significantly longer for patients treated with lenvatinib compared with placebo for patients previously treated with VEGFR-targeted therapy (including sorafenib) (Table 10). For patients who were VEGFR-targeted therapy naive, PFS was also statistically significantly longer for patients treated with lenvatinib compared with placebo.

Compared with patients in the placebo arm, ORR was statistically significantly improved for patients treated with lenvatinib whether or not they had been previously treated with a VEGFR-targeted therapy (see Appendix 3, Table 47).51,105,106 In both subgroups, ORRs were similar to the ORRs observed in the overall trial population (lenvatinib 64.8% and placebo 1.5%).

Newbold et al.105,106 reported that any all-grade and grade ≥ 3 AEs were similar in the two subgroups of patients receiving lenvatinib (prior VEGFR-targeted therapy 100.0% and 87.9%, respectively; no prior VEGFR-targeted therapy 99.5% and 86.7%, respectively). However, SAEs were more common in the lenvatinib arm among patients who had received prior VEGFR-targeted therapy (60.6%) than among those who had not (50.8%). For patients in the placebo arm, the opposite was the case, with SAEs being less common among patients who had received prior VEGFR-targeted therapy (18.5%) than among those who had not (25.0%).

Patients who had not received prior VEGFR-targeted therapy were treated with more cycles of lenvatinib (median 16 cycles) than those who had received prior VEGFR-targeted therapy (median 12.5 cycles). The proportion of patients who had at least one lenvatinib dose reduction was also similar between subgroups (prior VEGFR-targeted therapy 81.8% and no VEGFR-targeted therapy 86.7%). Patients with no prior VEGFR-targeted therapy had an earlier median time to first dose reduction (8.9 weeks) than patients with prior VEGFR-targeted therapy (14.8 weeks). Patients with no prior VEGFR-targeted therapy also had a lower median daily dose of lenvatinib than those with prior VEGFR-targeted therapy (16.1 vs. 20.1 mg, respectively).

Patients with and without symptomatic disease at baseline

Subgroup analyses were not conducted for patients with symptomatic or asymptomatic disease at baseline in SELECT. In DECISION, the median PFS for patients who were retrospectively categorised as being symptomatic at baseline was longer for patients who were asymptomatic than those who were symptomatic in the placebo arm but was similar in the intervention arm (Table 11). Patients were classified as being symptomatic if they had symptoms/findings that were consistent with RR-DTC reported in the medical history or pre-treatment AE data set at trial entry.113,119 It is noted in the EPAR26 for sorafenib that ≈20% of patients had symptoms that were likely to be related to thyroid cancer at baseline.

Bayer HealthCare7 has stated that although tumour shrinkage was not always sufficient to be confirmed as an objective response for some patients, it was often sufficient to alleviate symptoms. Further evidence has not been presented to support this statement.

Safety analyses for patients with symptomatic or asymptomatic disease at baseline have not been reported in SELECT or DECISION.

Other subgroup analyses of interest

Some OS subgroup analyses in SELECT have been reported in conference abstracts.67,73,82,89 No OS subgroup analyses have been reported using data from DECISION. For OS (first data cut-off point in November 2013) in SELECT, it has been reported that:

• There was no statistically significant difference in OS between older and younger lenvatinib-treated patients [HR 0.78, 95% confidence interval (CI) 0.49 to 1.26; p = 0.304] but there was a statistically significant difference in the placebo arm, favouring younger patients (HR 0.48, 95% CI 0.27 to 0.85; p = 0.010).67,73
• Median OS was not reached in either arm in patients treated in North America.89
• A statistically significant OS advantage was observed in patients with FTC treated with lenvatinib compared with placebo (HR 0.41, 95% CI 0.18 to 0.97).82

In addition to the subgroup analyses, Haddad et al.91 found from a post hoc exploratory multivariate analysis of SELECT (first data cut-off point) that ECOG PS and histology (favouring FTC vs. PTC) were statistically significantly associated with OS.

For PFS, all prespecified and some post hoc subgroup analyses (first data cut-off points) have also been reported in the appendix to the primary published paper for SELECT51 and in the published paper for DECISION.52 The results for both trials showed that for all subgroups, PFS favoured lenvatinib or sorafenib versus placebo. In the majority of instances, the differences were statistically significant. Regarding PFS for prespecified subgroup analyses, the following results are noted:

• The effect was statistically significantly in favour of lenvatinib (compared with placebo) for patients aged ≤ 65 or > 65 years in SELECT; the effect was statistically significantly in favour of sorafenib (compared with placebo) for patients aged < 60 or ≥ 60 years in DECISION.
• The effect was statistically significantly in favour of lenvatinib (compared with placebo) and for sorafenib (compared with placebo) for males and females in SELECT and DECISION.
• The effect was statistically significantly in favour of lenvatinib (compared with placebo) for patients with PTC, poorly differentiated carcinoma, FTC or Hürthle cell carcinoma in SELECT; the effect was statistically significantly in favour of sorafenib (compared with placebo) for patients with PTC or Hürthle cell carcinoma but not for those with FTC or poorly differentiated carcinoma in DECISION.
• The effect was statistically significantly in favour of lenvatinib (compared with placebo) for patients classified as white or Asian in SELECT; no subgroup analyses have been presented for race in DECISION.
• The effect was statistically significantly in favour of lenvatinib (compared with placebo) for patients treated in Europe and North America or other regions in SELECT; the effect was statistically significantly in favour of sorafenib (compared with placebo) for patients treated in Europe or Asia but not for patients treated in North America in DECISION.
• The effect was statistically significantly in favour of lenvatinib (compared with placebo) for those with and without lung metastases in SELECT and the effect was statistically significantly in favour of sorafenib (compared with placebo) for those with lung metastases only and for those without lung metastases only in DECISION.
• The effect was statistically significantly in favour of lenvatinib (compared with placebo) and for sorafenib (compared with placebo) for patients with and without bone metastases in SELECT and DECISION.

It is recommended by the EMA26 that before initiating treatment, physicians should carefully evaluate the prognosis in the individual patient, considering the maximum lesion size, symptoms related to the disease and the progression rate.

As reported in the appendices to the submission from Bayer HealthCare,7 a post hoc analysis of investigator-assessed PFS by number of target lesions in DECISION found statistically significant improvements with sorafenib compared with placebo for patients with at least three lesions. For patients with fewer than three lesions, PFS was numerically improved with sorafenib compared with placebo. It is also reported that another post hoc subgroup analysis of investigator-assessed PFS showed a treatment effect in favour of sorafenib compared with placebo for patients with maximum tumour size of ≥ 1.5 cm (HR 0.54, 95% CI 0.41 to 0.71). A numerically lower effect was reported for patients with a maximum tumour size of < 1.5 cm (HR 0.87, 95% CI 0.40 to 1.89).

Aside from the caveat surrounding the use of HRs to determine statistical significance as a result of PH assumption being violated, it is important to note that subgroup analyses are not powered to detect statistical significance. Therefore, when no statistically significant differences are reported, it could be that the numbers of patients in the subgroups were not large enough to detect a difference.

Extended open-label phases of SELECT and DECISION

In the extended open-label phase of SELECT, the starting daily dose of lenvatinib was originally 24 mg. This was later modified to 20 mg and then reverted to 24 mg. It is important to note that this phase of the trial only included 115 patients who crossed over from the placebo arm to lenvatinib and therefore does not present evidence from a randomised or controlled patient population. Furthermore, only placebo-treated patients who had confirmed disease progression (independent blinded review) during the randomisation phase and who met protocol-specified eligibility criteria were treated with lenvatinib. Consequently, it is noted in the EPAR27 for lenvatinib that these patients had very advanced disease, because they had experienced two sequential, confirmed disease progressions: the first before randomisation at the time of study entry and the second during treatment with the study drug in the randomisation phase.

The extended open-label phase of DECISION differed to that of SELECT in that as well as including patients who crossed over from the placebo arm to receive sorafenib, it also included patients who remained on sorafenib. In total, 150 patients in the placebo arm crossed over to receive sorafenib at progression and of these, data from 137 patients were evaluable for efficacy. In addition, 55 patients randomised to the sorafenib arm continued on sorafenib treatment in the open-label extension phase, of which 46 patients were evaluable for efficacy. It is reported by Schlumberger et al.122 and Paschke et al.114 that patients evaluable for efficacy had poorer risk features at enrolment compared with patients who were not evaluable. Like the extended open-label phase of SELECT, evidence from this patient population does not comprise evidence from a randomised or controlled patient population.

Findings from the extended open-label phase of SELECT for only ‘. . . the more mature dataset of patients who started treatment at the 24mg lenvatinib dose’ were reported in a conference abstract118 describing the first data cut-off point (November 2013). Findings reported at the second data cut-off point (June 2014) were presented for patients who started treatment at the 20-mg dose of lenvatinib and for patients who started the 24-mg dose of lenvatinib in the EPAR for lenvatinib.27 In the EPAR,27 it is reported that patient characteristics, previous treatments, geographical allocation, on-study placebo exposure, lenvatinib exposure in the extended open-label phase and median follow-up times vary considerably for these two dose regimens. Thus, patients receiving the different dose regimens are considered by the EMA to represent different populations of patients.

In addition to conference abstracts,114,122 the findings from the extended open-label phase of DECISION have been reported in the EPAR46 for sorafenib. Safety data for the extended open-label phase of DECISION are reported in the submission from Bayer HealthCare.7

The efficacy and safety findings from the open-label phases of both trials are summarised in Appendix 7, Tables 56 and 57. OS data have not been reported. With the exception of median PFS for patients receiving sorafenib for a second time, which was 6.7 months, the efficacy findings for PFS from the extended phase of SELECT and DECISION were similar to the findings reported in the randomised phase of the trials. The incidence of AEs for patients treated with lenvatinib and sorafenib in the open-label phases of the two trials tended to be slightly lower than reported during the double-blind phase.

In addition, Kappeler et al.94 and Fassnacht et al.83 have reported exploratory analyses of tumour growth rate in the randomised double-blind and extended open-label phases of DECISION. The authors found that the tumour growth rate (mean changes per month of sum of target lesion diameters) from baseline to nadir was –3.9% then +2.6% from nadir to progression for patients treated with sorafenib in the randomised phase; for those continuing with additional sorafenib in the open-label phase, from second baseline to progression the tumour growth rate was +1.7%. For patients randomised to the placebo arm, the tumour growth rate was +5.0% for all placebo patients and it was +6.1% for placebo patients subsequent to crossing over to receive sorafenib. Patients in the placebo arm who crossed over to sorafenib in the open-label phase then experienced a tumour growth rate pattern similar to that of patients who started on sorafenib in the randomised phase: –4.4% from second baseline to nadir and then +1.8% from nadir to progression.

Associations between tumour response, progression-free survival, overall survival, safety and health-related quality of life

Gianoukakis et al.86 examined the association between ORR and PFS for patients treated with lenvatinib in SELECT. The analysis is based on the third data cut-off point (August 2015) using investigator-assessed ORR [60.2% (the proportion of patients who achieved an objective tumour response)] and investigator-assessed mean PFS (19.4 months). The authors found that the median PFS in patients who received lenvatinib and who demonstrated a tumour response was 33.1 months (95% CI 27.8 months to not estimable). In lenvatinib-treated patients who did not show tumour response, the median PFS was 7.9 months (95% CI 5.8 to 10.7 months). Robinson et al.117 reported that an exploratory multivariate analysis found that percentage change in tumour size at the first assessment was a marginally statistically significant positive predictor for PFS (p = 0.06).

Using data from the first data cut-off point of SELECT, Newbold et al.108 analysed PFS by patients who had responded to treatment with lenvatinib at the first tumour assessment (median time to response 1.9 months) and by those who responded later (median time to response 3.8 months). The authors found that there was no difference in PFS between patients who achieved objective response at the time of first tumour assessment compared with thereafter.

Haddad et al.91 found from a multivariate analysis (first data cut-off point) that in SELECT, all-grade diarrhoea was statistically significantly associated with OS (median OS for lenvatinib-treated patients with diarrhoea was not reached and median OS for lenvatinib-treated patients without diarrhoea was 17.1 months). Choi et al.79 reported that the results of a post hoc analysis showed that lenvatinib-treated patients with hypertension had higher median PFS than those without hypertension (18.8 vs. 12.9 months; p = 0.009). Haddad et al.91 also reported results from multivariate analyses of associations between five other AEs (proteinuria, diarrhoea, fatigue/asthenia/malaise, rash and hand–foot syndrome) and PFS in SELECT. No statistically significant associations between any of the AEs and PFS were found.

Using data from DECISION, Kappeler et al.95 carried out an exploratory analysis to explore the association between tumour growth rate and PFS and OS. It is reported that the data cut-off points used for PFS and OS were the first data cut-off point (August 2012) and third data cut-off point (July 2015), respectively. Values of early tumour growth rate were split into quartiles [by median times derived from Kaplan–Meier (K–M) curves and from modelling with a Weibull distribution] separately by treatment arm. Better prognosis for PFS and OS with sorafenib was associated with the second and third tumour growth rate quartiles.

No other analyses have been conducted for patients treated with either lenvatinib or sorafenib in SELECT or DECISION examining the relationships between any of the efficacy or safety outcomes and HRQoL. As reported earlier (see Health-related quality-of-life findings), it has been speculated that AEs did affect HRQoL based on data from FACT-G and EQ-5D questionnaires, but no formal analyses have been conducted in an attempt to correlate the findings.

Indirect comparison feasibility assessment

In the absence of direct clinical evidence comparing treatment with lenvatinib versus sorafenib, the AG considered whether or not it was appropriate to carry out an indirect comparison to obtain estimates of the relative efficacy and safety of these two treatments.

The first step was to determine whether or not SELECT and DECISION shared a common comparator. The comparator arm of both trials was placebo. From the limited information on the placebos reported by Eisai Ltd,8 Bayer HealthCare7 and in the published papers,51,52 the AG considered that the comparator arms were likely to be similar and that a network could be constructed (Figure 5).

FIGURE 5 Indirect comparison network. (Figure not reproduced here; see the original article.)

The second step was to check the comparability of the participant and trial characteristics between the two trials. As described in Trial characteristics and Participant characteristics, the AG has noted that there are several trial design and participant differences, both within and between SELECT and DECISION. These differences raised concerns about whether or not data from these trials should be included in the same network of evidence.

The final step undertaken by the AG was to examine the PFS K–M data from the placebo arms of SELECT and DECISION to determine the extent to which the risk profiles of the populations in these arms of the two trials were comparable. The AG concluded that the risks were not sufficiently comparable and that these two trials should, therefore, not be included in the same network of evidence.

The Assessment Group’s detailed commentary on progression-free survival Kaplan–Meier data from the placebo arms

An indirect comparison implicitly assumes that the randomised patients are drawn from similar populations with reference to their risk profile for the time-to-event outcomes (PFS and OS). Because PFS is the primary outcome specified in both clinical trials, it is important that the equivalence of the placebo arms of the two trials can be confirmed by comparison of PFS outcomes: any significant discrepancy in progression risk would invalidate an indirect comparison between lenvatinib and sorafenib.

Figure 6 compares the K–M PFS trial results for the placebo arms of the two trials. After similar trends over the first 2 months, the curves separate markedly for more than a year before crossing over in the long term. Visual examination is sufficient to establish that these data are not amenable to either a simple HR adjustment or a time ratio adjustment.

FIGURE 6 Comparison of PFS in the placebo arms of DECISION and SELECT. Reproduced with permission from Fleeman et al.54 This is an Open Access article distributed in accordance with the terms of the Creative Commons Attribution (CC BY 4.0) license, which permits others to distribute, remix, adapt and build upon this work, for commercial use, provided the original work is properly cited. See: http://creativecommons.org/licenses/by/4.0/. The figure includes minor additions and formatting changes to the original text. (Figure not reproduced here; see the original article.)

Further exploration of these data trends through a plot of cumulative hazards in the two trial arms at common time points reveals a clear divergence from a simple linear (PH) relationship (Figure 7). The trial data indicate a higher initial risk of disease progression in SELECT in the first 10 months, followed by a sharp reversal in which the risk in the SELECT placebo arm reduces by > 50%.

FIGURE 7 Comparison of PFS hazard trends in the placebo arms of DECISION and SELECT. (Figure not reproduced here; see the original article.)

The AG considers that the placebo arms of SELECT and DECISION exhibit unexpectedly inconsistent patterns of temporal change that are not compatible with the assumption that these are similar patient groups. Consequently, patients enrolled in the two trials cannot be considered to be derived from a common population and, therefore, conducting an indirect comparison to obtain estimates of relative efficacy for lenvatinib and sorafenib is not appropriate.

Differences in trial and participant characteristics in the placebo arms of the trials

As reported earlier (see Trial characteristics and Participant characteristics), a number of differences in trial and participant characteristics were observed between arms within trials and across trials. Given the apparent differences in the placebo arms of the two trials, as demonstrated by differing hazard trends, the AG highlights the following differences in characteristics between the two placebo arms:

• The SELECT trial permitted the enrolment of patients who had been previously treated with a VEGFR-targeted therapy (including sorafenib), whereas DECISION did not: 20.6% had received prior therapy in the placebo arm of SELECT compared with no patients in the placebo arm of DECISION.
• Palliative radiotherapy, which is commonly available as part of BSC in UK NHS clinical practice, was not permitted for patients in the placebo arm of SELECT.
• The proportion of patients who crossed over from the placebo arm of SELECT was 87.8% at the third data cut-off point, compared with 75.0% in DECISION.
• There were proportionately more males in the placebo arm of SELECT than in the placebo arm of DECISION (57.3% and 45.2%, respectively).
• A higher proportion of patients in the placebo arm of SELECT were classified as being white than were similarly classified in the placebo arm of DECISION (78.6% and 61.0%, respectively), whereas the opposite was the case for patients classified as Asian (18.1% and 24.8%, respectively).
• Proportionately fewer patients in the placebo arm of SELECT were from Europe (48.9%) and proportionately more were from North America (29.8%) compared with the patients in the placebo arm of DECISION (59.5% and 17.1%, respectively).
• A greater proportion of patients in the placebo arm of SELECT had an ECOG PS of ≥ 1 (48.1%) than in the placebo arm of DECISION (38.6%).
• A greater proportion of placebo patients had FTC and poorly differentiated thyroid cancer in the placebo arm of SELECT (16.8% and 14.5%, respectively) than in the placebo arm of DECISION (9.0% and 7.6%, respectively).
• The time from diagnosis to randomisation was greater in the placebo arm of SELECT (73.9 months) than in the placebo arm of DECISION (66.9 months).
• A greater proportion of patients in the placebo arm of SELECT had lung, bone and liver metastases (94.7%, 36.6% and 21.4%, respectively) than in DECISION (86.2%, 26.7% and 14.3%, respectively).

Proportional hazards assumption

As discussed in Consideration of proportional hazards assumption, the AG concluded that the PH assumption was not valid for PFS, unadjusted OS or adjusted OS in SELECT or for PFS or adjusted OS in DECISION. The violation of the PH assumption, for all but unadjusted OS in DECISION, means that the network of evidence is compromised for all outcomes.

Assessment Group summary statement

The AG considers that is not appropriate to conduct an indirect comparison to obtain HRs for lenvatinib versus sorafenib for the outcomes of PFS, unadjusted OS and adjusted OS. This is because the risk profiles of the patients in the placebo arms of the trials are not comparable and any indirect comparison would produce results that could not be considered to be robust. This also precluded indirect comparison for subgroups of patients according to previous treatment with TKIs.

As described in Chapter 3, Methods of analysis/synthesis, in addition to trial characteristics, participant characteristics and outcome data, the AG stated that it would consider the quality of the included trials when conducting its feasibility assessment. The results of the AG’s risk-of-bias assessment are reported in Comparison of assessments of risk of bias. However, given the issues already highlighted, the quality of the trials was not a factor in the AG’s decision to not conduct an indirect comparison.

Systematic review evidence

The AG included 13 systematic reviews5–8,33,57,61,93,97,104,127,138,141 in its review; these reviews included the evidence submissions reporting systematic reviews and indirect comparisons for this MTA from Eisai Ltd8 and Bayer HealthCare,7 and also the evidence reported in a paper by Tremblay et al.57 Although Tremblay et al.57 did not report the conduct of a systematic review, this paper was included as it did report results from an indirect comparison and a matching-adjusted indirect comparison (MAIC) using data from SELECT and DECISION.

A summary of the characteristics of the included systematic reviews is presented in Appendix 5, Table 49. Most of the evidence was derived from observational studies of treatment with sorafenib. However, four of the reviews,7,8,57,97 including the submissions from Eisai Ltd8 and Bayer HealthCare,7 included evidence from SELECT and DECISION and results from indirect comparisons, including MAICs.

The AG’s assessment of the quality of the included reviews is presented in Appendix 5, Table 50. Overall, the AG considered that the quality of nine5–8,61,97,104,127,138 of the identified systematic reviews was good. However, only 45–8 of the 13 reviews included a quality assessment of the included primary studies. Four33,57,93,141 of the reviews were considered to be of poorer quality than the rest. Of these, only one33 reported the use of an adequate search strategy. In addition, methods of cross-checking during either the study selection process or the data extraction process were not reported by the authors of three reviews.33,57,93 No quality assessment of the primary studies was reported in any of these four reviews.33,57,93,141

The conclusions reached by the authors of the systematic reviews are presented in Appendix 5, Table 51. The earliest of the reviews was carried out by Anderson et al.61 and was published in 2013. The authors concluded that certain treatments, notably TKIs, showed promise in Phase II trials. Gruber and Colevas33 concluded that the most likely outcome of treatment with a TKI was stable disease. McFarland and Misiukiewicz104 concluded that sorafenib slowed the progression of disease in the majority of cases. For treating thyroid cancer, Ye et al.141 reported that the clinical effects of sorafenib and lenvatinib outweigh the toxicities [relative risk (RR) 1.27, 95% CI 1.05 to 1.53] and deaths (RR 15.24, 95% CI 6.99 to 33.21). Ye et al.141 concluded that lenvatinib and sorafenib were more useful for thyroid cancer than for RR-DTC, based on the results of the subgroup analyses that were conducted. However, the AG considers that all of the studies that included patients with DTC also included patients with RR-DTC and so the validity of this subgroup analysis and the conclusions reached based on these subgroup analyses are questionable.

Jean et al.93 found AEs reported for sorafenib for treating RR-DTC to be higher than for AEs reported for treating renal cell carcinoma (RCC) or hepatocellular carcinoma (HCC). In two reviews127,138 ORR data and AE data were pooled for sorafenib from seven observational studies59,78,88,101,126,147,148 (five prospective and two retrospective). In the review by Shen et al.,127 all of the studies59,78,88,101,126,147,148 included patients with RR-DTC, whereas the review by Thomas et al.138 included five prospective observational studies of RR-DTC,59,88,101,126,147 a retrospective study of RR-DTC149 and a Phase II study150 of patients with medullary thyroid cancer. Although the incidences of hand–foot syndrome (≥ 73%), diarrhoea (≥ 68%) and weight loss (≥ 50%) included in both meta-analyses were broadly similar to the incidence of the same AEs in DECISION, it was noticeable that the incidences of rash (≥ 66%) and fatigue (≥ 60%) were higher than reported in DECISION. Similarly, the pooled ORR (20.9% to 22%) from the two reviews127,138 was higher than the ORR reported in DECISION. The pooled median PFS (17.9 months) from the review by Thomas et al.138 was also higher than median PFS reported in DECISION, but the pooled analysis for PFS also included patients with medullary thyroid cancer. The key results from these three reviews93,127,138 are summarised in Appendix 5, Table 52.

In addition, Shen et al.127 noted that rare but severe AEs were observed mainly due to intracranial haemorrhage, cardiac arrest, angiooedema, small cell lung cancer, carcinoma of the tongue, and grade 5 event of sudden death. Because of the limited data, the authors did not pool these high-grade AEs. Thomas et al.138 also reported that bleeding at any site occurred in 13.6% of patients, 3.8% of patients reported acute myocardial infarctions and 2.2% experienced congestive heart failure. Severe hypocalcaemia (grade ≥ 3) occurred in 2.5% of patients and 8.7% of patients developed cutaneous squamous cell carcinoma. However, it should be cautioned that in the meta-analyses conducted by Shen et al.127 and Thomas et al.,138 the authors did not investigate the heterogeneity of the studies included in the meta-analyses.

For RR-DTC, all of the indirect comparison results (including results from MAICs7,57) showed that lenvatinib was statistically significantly superior to sorafenib in terms of PFS but not OS.6–8,57,97 Kawalec et al.97 also reported lenvatinib to result in statistically significantly fewer cases of alopecia but statistically significantly more cases of hypertension and treatment-related SAEs, when compared with sorafenib. Bayer HealthCare7 reported sorafenib to result in fewer grade ≥ 3 AEs, SAEs and withdrawals owing to AEs when compared with lenvatinib. However, caveats about the generalisability of the results of the indirect comparisons have been raised,6 and Kawalec et al.97 stated that indirect comparison results should be interpreted with caution because of differences in trial characteristics. Furthermore, during the current appraisal, Bayer HealthCare confirmed that it considered that the data from its indirect treatment comparison (ITCs)7 did not enable a robust comparison of sorafenib and lenvatinib given important differences between SELECT and DECISION.151 Of the indirect comparisons conducted, only the indirect comparison by Kawalec et al.97 was not sponsored by Eisai Ltd or Bayer HealthCare. A summary of the findings from the indirect comparisons is presented in Appendix 5, Tables 53–55.

Evidence from prospective observational studies

The AG included nine prospective observational studies in its review.59,77,78,81,88,101,103,126,135 Five of these studies59,78,88,101,126 were included in the meta-analyses conducted by Shen et al.127 and by Thomas et al.138 Seven of the studies were included in the EPARs26,27 for lenvatinib77,135 and sorafenib.59,78,88,101,126 The study and participant characteristics and efficacy and safety findings are summarised in Appendix 8, Tables 58–68.

All studies included patients whose disease was described as being radioactive iodine-refractory59,77,78,101,126,135 or resistant to radioactive iodine,81,88 or who may have received multiple treatments of radioactive iodine.103 Two studies77,135 investigated the efficacy and safety of lenvatinib, six studies59,78,88,101,103,126 assessed the efficacy and safety of sorafenib and one study81 considered the efficacy of sorafenib. Some patients included in four of the studies59,88,101,135 had anaplastic or medullary carcinoma. Safety data from these four studies59,88,101,135 are, therefore, not reported for RR-DTC only. However, all nine studies59,77,78,81,88,101,103,126,135 reported efficacy findings for patients with RR-DTC only and all efficacy data reported in this section related to patients with RR-DTC only.

The E7080-G000-201 study (hereafter referred to as Study 20177) of lenvatinib was conducted in the UK, France, Italy, Poland, the USA and Australia. The E7080-J081-208 study of lenvatinib (hereafter referred to as Study 208135) was conducted in Japan. Studies of sorafenib were carried out in the UK,59 the Netherlands,126 Italy,103 Greece,81 the USA101 and China.78 The earliest study was conducted between 2004 and 2005,101 and the most recent study was conducted between 2012 and 2015.135 The length of study follow-up varied from a minimum of 3 months78 to a median of 51.6 months.27

The number of patients included in the studies varied from 978 to 58.77 In total, 109 patients were treated with lenvatinib, of whom 83 had RR-DTC; 213 patients were treated with sorafenib, of whom 186 had RR-DTC. In most studies, the majority of patients with RR-DTC had a histology of PTC,59,77,78,88,101,126 the exception being the study by Marotta et al.103 in which the ratio of patients with FTC to patients with PTC was 2 : 1. The average age of participants ranged from 55 years59 to 64 years.101 Four studies59,77,88,101 included a majority of males and three studies81,103,126 had a majority of females. Two studies78,135 did not report information on sex. The authors of only two studies77,101 reported information on race and these included a majority of white participants. Only two studies that reported ECOG status included patients with an ECOG PS of ≥ 2 (6.9%77 and 35.3%103). The same two studies were the only studies to explicitly state that patients could have received a prior TKI (11.8%103 to 29.3%77). There was scant and inconsistent reporting of the sites of metastases.

Median OS was reported in five studies.77,88,101,126,135 Median OS ranged from 31.8 months135 to 32.3 months77 for lenvatinib and from 23 months101 to 34.5 months126 for sorafenib. Median PFS was reported in six studies77,88,101,103,126,135 and ranged from 12.6 months77 to 25.8 months135 for lenvatinib and from 12 months103 to 22.1 months for sorafenib88 (this last finding was reported in a subsequent conference abstract137). Chen et al.78 (sorafenib) reported mean PFS (9.7 months). The ORRs ranged from 50.0%77 to 68.0%135 for patients treated with lenvatinib, and from 15% (histology of PTC)101 to 38.3%88 for those treated with sorafenib (this latter finding was reported in a subsequent conference abstract137). Median time to response and median duration of response were only reported in two studies.77,126 For lenvatinib,77 median time to response was 3.6 months and, for sorafenib,126 all responses were reported to have happened within 6 months. The median duration of response was 12.7 months for lenvatinib77 and 29.6 months for sorafenib.126

Key AEs are summarised in Appendix 8, Tables 61–66. Two studies88,101 (sorafenib) only reported treatment-related AEs. Two of the sorafenib studies,78,81 presented only as abstracts, reported very little information about AEs.

Incidences of the same types of AEs varied across the studies: for lenvatinib, hypertension and proteinuria were very commonly reported; for sorafenib, hand–foot syndrome, rash and alopecia were common; and diarrhoea and fatigue were common with both drugs. Data on SAEs were available only from Study 20177 (lenvatinib). Information on fatal AEs were reported only in two studies77,135 of lenvatinib and in one study101 of sorafenib. For patients treated with lenvatinib, 48% reported a SAE77 and up to 8%135 died from an AE. Only one death from AEs has been reported in one of the studies of sorafenib;101 it is unclear if the lack of reporting of fatal AEs in the other sorafenib studies59,78,81,88,103,126 means that there were no deaths from AEs in these studies. None of the deaths from AEs in any of the three studies77,101,135 reporting fatal AEs were described as being treatment related.

Ongoing studies and studies for which there are no results

The AG identified four ongoing studies,152–155 as summarised in Appendix 9 (see Table 69). None of the study results has been published or reported as a conference abstract. Only the two studies of lenvatinib154,155 are RCTs: E7080-G000-211 (Study 211)154 is a Phase II postauthorisation study that includes a randomised controlled phase, comparing two different starting doses of lenvatinib (24 vs. 18 mg) with placebo; E7080-C086-308 (Study 308)155 is a Phase III RCT being conducted in China comparing lenvatinib at its licensed dose of 24 mg with placebo. Eisai Ltd sponsors both of these trials. The other two studies are prospective observational Phase II studies of sorafenib:152,153 a pilot study sponsored by the Royal Marsden NHS Foundation Trust152 and a postauthorisation study sponsored by Bayer HealthCare.153

In addition, although not strictly meeting the inclusion criteria for the current MTA, the AG is aware of an ongoing global prospective non-interventional study [Radioactive Iodine reFractory asymptomatic patients (RIFTOS), NCT02303444]156 of asymptomatic patients with RR-DTC treated with any type of MKI. The primary objective is to compare the time to symptomatic progression from study entry. Bayer HealthCare sponsors this study. The planned enrolment is approximately 700 patients and the expected study end date is 1 July 2020.

Discussion of clinical effectiveness: interpretation of results

The AG’s assessment of lenvatinib and sorafenib for the treatment of patients with RR-DTC focused on evidence from two RCTs: SELECT (lenvatinib vs. placebo) and DECISION (sorafenib vs. placebo). Supporting evidence was derived from 13 systematic reviews5–8,33,57,61,93,97,104,127,138,141 (including two systematic reviews described in the submissions from Eisai Ltd8 and Bayer HealthCare7) and nine prospective observational studies.59,77,78,81,88,101,103,126,135

Clinical efficacy

Summary and interpretation of evidence: lenvatinib versus sorafenib

The primary objective of the AG’s systematic review was to compare the clinical effectiveness of lenvatinib with that of sorafenib. Results from the AG’s literature search revealed that there have been no head-to-head trials comparing the effectiveness of treatment with lenvatinib with the effectiveness of treatment with sorafenib. However, five studies6–8,57,97 have reported results from indirect comparisons and/or MAICs. Results from all of these analyses show that, compared with sorafenib, treatment with lenvatinib improves PFS but not OS.

The AG explored whether or not it was appropriate to conduct an indirect comparison. Although it was possible to construct a network, the AG identified issues that raised concerns about whether or not evidence from SELECT and DECISION could be included in the same network. First, there were differences between trial characteristics (prior treatment with TKIs, concurrent use of palliative radiotherapy and differences in subsequent treatment received on disease progression). Second, there were differences in participant characteristics (sex, race, geographic region, ECOG PS, time from diagnosis, histology and site of metastases) both within and between the trials. Third, the analysis of the PFS K–M data from the placebo arms of SELECT and DECISION showed that the risk profiles of the two trial populations were not comparable. The reasons for the differences in risk are currently unknown. Fourth, the AG considered that, for the majority of patient survival hazards assessed in the two trials, PHs were violated, the exception being unadjusted OS in DECISION.

The AG is unable to conclude whether or not treatment with lenvatinib is more effective than treatment with sorafenib for patients with RR-DTC. The AG considers that the results from the four published indirect comparisons7,8,57,97 should be interpreted with caution. This warning also extends to the results from the MAICs.7,57 It is unknown whether or not the MAIC adjustments would fully account for all of the differences in the trial populations because the AG was unable to compare the adjusted risk profiles of patients included in the MAIC.

The AG highlights that Kawalec et al.97 stated that their indirect comparison results should be interpreted with caution because of differences in the characteristics of the included trials. In addition, the EMA,27 Scottish Medicines Consortium (SMC)38 and Canadian Agency for Drugs and Technologies in Health (CADTH)6 all highlighted that differences in populations might have contributed to differences in results observed between the two trials. The SMC38 also highlighted that the validity of the results from the MAIC submitted by Eisai Ltd may be limited by weaknesses including heterogeneity across the studies in inclusion criteria, assessment of disease progression and analysis of PFS. The CADTH6 highlighted that the MAIC approach does not have the ability to control for the potential for unobserved or unrecorded differences between trials such as differences in standards of care or baseline characteristics. Furthermore, during the current appraisal, Bayer HealthCare confirmed that it considered that the data from its ITCs7 did not enable a robust comparison of sorafenib and lenvatinib given important differences between SELECT and DECISION (Bayer HealthCare response to AG report to the NICE Appraisal Committee, 6 September 2017).157

Summary and interpretation of evidence: lenvatinib and sorafenib versus best supportive care

The secondary objective of the AG’s systematic review was to compare treatment with lenvatinib and sorafenib with BSC. The AG has assumed that, in both trials, treatment with lenvatinib plus BSC or sorafenib plus BSC is compared with placebo plus BSC. The unadjusted OS results from SELECT and DECISION demonstrated that there was no statistically significant difference in OS between treatment with lenvatinib and treatment with sorafenib versus placebo. After adjusting OS data for treatment crossover using RPSFTM, there was a statistically significant improvement in OS from treatment with lenvatinib compared with placebo; however, the difference in effect of sorafenib versus placebo was not statistically significant. The AG highlights that the unadjusted median OS estimates for patients treated with lenvatinib and sorafenib in SELECT and DECISION are higher than those reported for patients treated with lenvatinib and sorafenib in prospective observational studies.

For PFS and ORR, the results from SELECT and DECISION demonstrated that treatment with both lenvatinib and sorafenib were statistically significantly better than treatment with placebo for patients with RR-DTC. For all of the prespecified subgroups, the results from SELECT and DECISION favoured treatment with the intervention (lenvatinib or sorafenib) when compared with placebo. Median PFS and ORR for patients treated with lenvatinib in SELECT were higher than the prospective observational results from Study 20177 and lower than the results from Study 208.135 In contrast, median PFS and ORR results reported for patients treated with sorafenib (DECISION) were lower than findings from any of the prospective observational studies or the two meta-analyses.127,138

Patients in DECISION were permitted to receive concomitant palliative radiotherapy, a common component of BSC in NHS clinical practice, whereas patients in SELECT were not; full details of the BSC provided in the two trials are not available. Whether or not patients in the trials received BSC similar to that provided by the NHS is unknown and this raises uncertainty about whether or not the trial results are generalisable to NHS patients. If the BSC delivered in the two trials is not comparable, then using the placebo arms to connect the two trials in an indirect comparison becomes even more challenging. However, as the rates of palliative radiotherapy administered to patients in DECISION are low (10.6% of patients treated with sorafenib and 21.4% of patients treated with placebo), then perhaps this issue is not important.

There are two important issues to consider when interpreting the RCT evidence. First, a caveat to the use of the RPSFTM-adjusted OS results from both trials is that the method requires the assumption that postprogression anti-cancer treatments, other than those permitted by treatment crossover, represent routine clinical practice. For patients with RR-DTC, there is currently no standard of care for patients with progressive disease. Therefore, it is unknown whether or not the poststudy anti-cancer treatments administered to patients in SELECT and DECISION reflect the treatments that would be offered to patients in the NHS. Second, the AG’s examination of the PH assumption for OS (unadjusted and adjusted) and PFS in SELECT and DECISION showed that the PH assumption does not hold for any of these outcomes other than unadjusted OS in DECISION. This means that the majority of the HRs reported in the company submissions should be interpreted with caution. However, clinical advice to the AG is that the PFS results for the overall populations of SELECT and DECISION are clinically meaningful.

Safety

Summary and interpretation of evidence: lenvatinib versus sorafenib

The AG did not conduct its own indirect comparison to facilitate a comparison of the effect of treatment with lenvatinib with the effect of treatment with sorafenib for AEs. However, two other reviews7,97 reported results from indirect comparisons of AEs. Kawalec et al.97 reported that treatment with lenvatinib resulted in statistically significantly less alopecia but statistically significantly more hypertension and treatment-related SAEs than sorafenib. Bayer HealthCare7 reported sorafenib to result in fewer grade ≥ 3 AEs, SAEs and withdrawals owing to AEs when compared with lenvatinib.

Summary and interpretation of evidence: lenvatinib and sorafenib versus best supportive care

When compared with placebo, treatment with both lenvatinib and sorafenib resulted in increased AEs. However, although diarrhoea was experienced by just over two-thirds of patients treated with both drugs in SELECT and DECISION, there were some notable differences in the safety profiles. Hypertension and decreased appetite were reported by over half of patients in SELECT, whereas the most common AEs reported by half or more of patients in DECISION were hand–foot syndrome, alopecia and rash. Grade ≥ 3 hypertension was very common in patients treated with lenvatinib (> 40%), and grade ≥ 3 hand–foot syndrome was very common in patients treated with sorafenib (> 20%). Hypertension was also reported to be one of the most common SAEs in SELECT (3.4%). Data on the median time to onset of AEs91,139 from SELECT and DECISION suggest that AEs typically occur early, with a decrease in incidence, prevalence and severity over time. In DECISION, exceptions were diarrhoea that increased in prevalence over the first six cycles and weight loss that increased in severity (from grade 1 to grade 2) over the first nine cycles.

Overall, the safety findings from the RCTs were consistent with the findings from prospective observational studies of lenvatinib77,135 and sorafenib,59,78,81,88,101,126 although it is noticeable that the incidence of some AEs varied quite widely in observational studies for patients treated with sorafenib. However, meta-analyses127,138 of data from observational studies for hand–foot syndrome and diarrhoea reported incidences of all-grade and grade ≥ 3 AEs to be similar to those reported in DECISION. It has, however, been found in a systematic review by Jean et al.93 that the incidence of common all-grade AEs tends to be higher for patients with RR-DTC than for patients with RCC or HCC and also for some patients with grade ≥ 3 hand–foot syndrome and rash.

After diarrhoea, hypertension was the most common reason for dose modifications, as well as being the most common reason (alongside asthenia) for discontinuations in SELECT. In DECISION, the most common reason for dose modifications and discontinuations was hand–foot syndrome. Dose reductions were frequent (> 60%) for patients treated with both lenvatinib and sorafenib. Life-threatening AEs from treatment with lenvatinib and sorafenib were rare. The AG considers that the AEs associated with treatment with lenvatinib and sorafenib can be managed with usual medical care and dose modifications, including treatment withdrawal. Clear guidance for managing AEs is set out in the SmPCs for lenvatinib45 and sorafenib.46

Health-related quality-of-life findings

The HRQoL data were not collected as part of SELECT, and HRQoL data from the 30 patients who participated in the open-label extension phase of SELECT are not yet available. This is disappointing given that the investigators in the earlier DECISION had measured and reported HRQoL outcomes and highlighted that HRQoL may be negatively impacted by treatment with TKIs.7,120 AE rates were high in SELECT and it would have been informative if HRQoL data had been collected. HRQoL research is much needed as HRQoL is one of the most important outcomes to consider, both from the perspective of patients and for assessing comparative cost-effectiveness.

The HRQoL data collected during DECISION demonstrated that the FACT-G scores were higher for patients in the placebo arm than for patients in the sorafenib arm, indicating a higher HRQoL for patients receiving placebo. The negative impact of treatment with sorafenib on HRQoL may be linked to the high rates of AEs.7,120 Indeed, it has been noted by Bayer HealthCare7 that in response to the question on the FACT-G questionnaire ‘I am bothered by side effects’, the proportion of patients in the sorafenib arm who replied ‘quite a bit’ or ‘very much’ increased from 1.5% in cycle 1 to 29.6% in cycle 2 but then gradually diminished over time.

There are, however, limitations to the results from the HRQoL analyses. Although the overall questionnaire completion rate during DECISION was reported to be 96%,120 the number of patients eligible to complete the questionnaires diminished with every cycle because only those who had not experienced progression were asked to complete the questionnaire. This also means that there are no HRQoL data available from patients whose disease has progressed. It is also unknown whether or not there is a direct correlation between HRQoL and AEs and how the different types of AEs experienced by patients treated with lenvatinib (e.g. hypertension) and sorafenib (e.g. hand–foot syndrome) affect HRQoL. Finally, to what extent a patient’s HRQoL is affected by their symptom status (symptomatic vs. asymptomatic) is unknown.

Generalisability of findings

The AG considers that the generalisability of the findings from SELECT and DECISION to NHS clinical practice is questionable. This concern is driven by the fact that clinical advice to the AG is that in clinical practice there are concerns about the toxicity of TKI therapy in patients and effects on the quality of life of patients with asymptomatic disease and so treatment is more commonly given when symptomatic or clinically significant progressive disease develops. Hence, BSC is a common treatment option for this group. The authors of two of the meta-analyses of sorafenib127,138 concluded that the high incidence of AEs associated with sorafenib may affect the quality of patients’ lives and most patients with metastatic disease do not require systemic therapy. This view is supported by several clinical guidelines4,24,25 as patients experiencing RR-DTC symptoms and/or those with rapidly progressing disease are considered to be in greatest need of systemic treatment.31 In addition, the EMA concluded that maximum lesion size, symptoms related to the disease and progression rate should be carefully considered for each individual patient before initiating treatment.26

Although all of the patients in SELECT and DECISION had RR-DTC, it is unclear how many had symptomatic and/or rapidly progressing disease; however, it is reported in the EPAR26 for sorafenib that results from a post hoc subgroup analysis of data from DECISION suggest that 20% of patients were likely to be symptomatic. Clinical advice to the AG is that this is probably typical of the proportion seen in clinical practice. It is unclear how many patients in SELECT were symptomatic and/or had progressive disease.

The post hoc retrospective analysis of data from patients participating in DECISION113,119 categorised patients as having symptomatic disease if they had symptoms/findings that were consistent with RR-DTC reported in the medical history or pre-treatment AE data set at baseline. Clinical advice to the AG is that there are no generally agreed definitions of ‘symptomatic’ or ‘rapidly progressive disease’ and that, in clinical practice, the definition of a patient’s disease status depends on individual patient characteristics.

Results from the post hoc analysis show that median PFS was similar for all patients treated with sorafenib, irrespective of whether or not they were symptomatic or asymptomatic (10.7 months and 10.8 months, respectively, compared with 10.8 months for all patients in the sorafenib arm of the trial).7 However, for patients in the placebo arm, median PFS was much lower for symptomatic patients (3.6 months) than for asymptomatic patients (7.2 months), and it was also lower than for all patients in the placebo arm of the trial (5.8 months).

No analyses have been undertaken to compare the effectiveness of treatment with lenvatinib for symptomatic patients with that for asymptomatic patients. In the absence of such analyses, no assumptions can be made about relative effectiveness. However, clinical advice to the AG is that, like sorafenib, only patients with symptomatic and/or progressive disease are likely to be treated with lenvatinib in the NHS.

The most recent published guidelines for treating RR-DTC, by the NCCN,25 recommend lenvatinib or sorafenib as the treatment of choice for patients with progressive and/or symptomatic disease. However, the choice between lenvatinib and sorafenib should be based on the individual patient, taking into account the likelihood of response and comorbidities.25

There are further important caveats regarding the generalisability of the findings from SELECT and DECISION to NHS clinical practice.

The first caveat is that, although most patients participating in the trials had a diagnosis of PTC, as would be expected in clinical practice, there were proportionately more patients with other types of DTC than would be expected in NHS clinical practice. Patients with these other types of DTC are reported to have a worse prognosis than patients with PTC.15 However, subgroup and exploratory analyses of SELECT data showed that for unadjusted OS, there was a statistically significant OS gain for patients with FTC treated with lenvatinib compared with those treated with placebo,82 and that histology (favouring FTC vs. PTC) was statistically significantly associated with increased OS.91 These exploratory results warrant further investigation.

The second caveat relates to the age of patients. Thyroid cancer incidence is strongly related to age, with the highest incidence rates being in older males (aged > 60 years) and the highest incidence rates in females being in younger and middle-aged women (aged 40–60 years).1 The median age of patients was 61 years in the lenvatinib arm and 64 years in the placebo arm of SELECT and 63 years in both arms of DECISION, and approximately half of the patients in both trials were male. Given that the median time from diagnosis in the trials varied from between 5.5 and 6 years, it appears that, in general, patients were older than may be seen in clinical practice. Moreover, the prognosis of patients tends to differ for patients aged < 45 years and for those aged ≥ 45 years, as reflected in the staging criteria used for DTC.4 Detailed data on the age range of included patients were not reported for either trial.

Other issues of relevance to clinical practice

The relative importance of ORR also warrants some discussion, particularly given the marked reported differences in effect between treatment with lenvatinib and sorafenib indicated by results from SELECT and DECISION and the prospective observational studies.59,77,78,81,88,101,103,126,135 Although studies of lenvatinib51,77,135 suggest that at least half of all patients achieve a response, meta-analyses of data from observational studies of sorafenib127,138 suggest that no more than 22% of patients receiving this treatment respond. This finding reflects the finding from a systematic review of TKIs33 that shows that the most likely outcome of treatment with a TKI is stable disease. Indeed, in DECISION, 42% of patients in the sorafenib arm had stable disease for ≥ 6 months (and 12.2% had an objective tumour response) compared with 33% in the placebo arm (and 0.5% had an objective tumour response). However, given that lenvatinib and sorafenib are likely to be preferred treatment options for patients with clinically significant progressive disease, reducing the rate of disease progression may be a more relevant outcome. The AG notes that in the submission from Bayer HealthCare,7 it is reported that most patients (77%) in the sorafenib arm of DECISION experienced target lesion tumour shrinkage, compared with 28% of patients in the placebo arm. The authors of a systematic review of sorafenib104 for treating RR-DTC concluded that, although the data in the review came primarily from non-randomised Phase II trials (but also included DECISION), the results suggest that treatment with sorafenib slows the progression of disease in the majority of cases.

The findings from the extended open-label phases of SELECT and DECISION should also be considered. These findings show that the median PFS and ORR outcome results for patients previously randomised to the placebo arms but who crossed over to receive lenvatinib or sorafenib at the licensed doses were similar to the median PFS and ORR reported for patients treated with lenvatinib and sorafenib in the double-blind phases of the trials. Given that patients in the placebo arm received no active systemic therapies during the double-blinded phase, these results appear to support the view that patients with progressive disease do not need to be treated immediately and can be treated when showing symptoms and/or rapidly progressing. However, the AG cautions that data on symptoms and/or whether or not patients were rapidly progressing are lacking, although patients were progressing to the extent that, on the basis of RECIST (Response Evaluation Criteria in Solid Tumours) criteria, they were considered to have progressive disease. The AG also cautions that no OS data were available for these specific cohorts of patients.

The results from the open-label phase of SELECT, which included patients who crossed over from placebo to receive treatment with two different doses of lenvatinib, suggest that PFS may be improved for those starting at the 20-mg dose (median PFS not reached) as opposed to the licensed dose of 24 mg (median PFS of 17.5 months). However, the numbers of patients in each group, particularly in the 20-mg dose cohort, were small, and definitive conclusions could not be reached. Study 211,154 an ongoing Phase II RCT, compares two different starting doses of lenvatinib (24 vs. 18 mg) with placebo. This trial is expected to end in October 2020.

Although patients treated with lenvatinib in SELECT were not permitted to receive additional lenvatinib in the extended open-label phase of that trial, around one-quarter of patients had received treatment with a VEGFR-targeted therapy, including sorafenib, prior to enrolment. SELECT subgroup PFS and ORR findings suggest that patients benefited from treatment with lenvatinib, regardless of whether or not they had received prior treatment with a VEGFR-targeted therapy. This result suggests that lenvatinib could be used as a first- or second-line treatment for patients with RR-DTC. Further research is required to identify the effect on OS of treating patients with lenvatinib followed by sorafenib. Furthermore, it has also been reported that SAEs were more common in the lenvatinib arm among patients who had received a prior VEGFR-targeted treatment (60.6%) than among those who had not (50.8%).105,106

Some patients in DECISION who had experienced disease progression while receiving sorafenib were also eligible to receive sorafenib for a second time in the extended open-label phase of DECISION. Clinical advice to the AG was that, in NHS practice, patients could be prescribed sorafenib post progression as there is a view that continued treatment with sorafenib will slow the progression of disease. This expectation is supported, to some extent, by exploratory post hoc findings.83,94 These findings suggest that despite evidence of tumour growth or prior RECIST progression, treatment with sorafenib continued to slow tumour growth for patients who had also been treated with sorafenib during the randomised phase, when compared with tumour growth for patients treated with placebo during the randomised phase.83,94 However, as concluded by authors of other abstracts114,122 reporting results from the open-label extension phase of DECISION, the effect of continued treatment with sorafenib after progression needs to be explored further.

Finally, there are no data for patients treated with sorafenib followed by lenvatinib. Further research is needed to identify the effect on OS and other efficacy and safety outcomes of treating patients with lenvatinib followed by sorafenib, and sorafenib followed by lenvatinib.