Bevacizumab, sorafenib tosylate and sunitinib for renal cell carcinoma: a systematic review and economic evaluation

Escudier and colleagues106 report the results of the AVOREN study, an international (Australia, Belgium, Czech Republic, Finland, France, Germany, Hungary, Italy, Israel, Netherlands, Poland, Singapore, Spain, Switzerland and Taiwan) and UK multicentre double-blind and placebo-controlled phase III RCT in which 649 patients with confirmed clear cell metastatic RCC were randomised to receive either bevacizumab and IFN or placebo and IFN. The trial has been reported in one full publication106 and in five abstracts. 86,113–116 The aim of the study was to determine whether first-line bevacizumab plus IFN improves efficacy compared with IFN alone. Primary outcomes were OS and PFS. Overall response rate and safety were secondary outcomes. The study was designed to have 80% power for the log-rank test to detect an improvement in OS with an HR of 0.76, assuming an improvement in median survival from 13 months to 17 months, at a two-sided alpha level of 0.05. One interim analysis was planned, based on 250 deaths, after which the study was unblinded and patients in the IFN arm who had not progressed were offered bevacizumab plus IFN.

To be eligible for entry into the trial participants had to have a diagnosis of predominantly (> 50%) clear cell RCC based on routine assessment of tumour histopathology and were also required to have undergone nephrectomy or partial nephrectomy (if resection margins were clearly negative of disease), to have a Karnofsky performance score of 70% or more, to have normal hepatic, haematopoietic and renal function and to have received no previous systemic therapy for RCC.

Randomisation was performed centrally and patients were stratified according to country and MSKCC risk group. Patients were randomly assigned to receive bevacizumab (10 mg/kg body weight, delivered intravenously once every 2 weeks) (n = 327) or placebo (n = 322) plus IFN-α-2a (9 MIU, delivered subcutaneously three times per week for a maximum of 52 weeks). Treatment was continued until evidence of disease progression, the patient experienced unacceptable toxicity, or withdrawal of consent. No dose reduction of bevacizumab/placebo was allowed. A starting dose of IFN of less than 9 MIU was permitted as long as the full dose was reached within the first 2 weeks of treatment. Dose reduction to 6 MIU or 3 MIU was allowed to manage adverse events of grade 3 or higher that were attributable to IFN.

Median follow-up at data cut-off was 13.3 months (range 0–25.6 months) in the bevacizumab plus IFN group and 12.8 months (range 0–24.2 months) in the control group. Median duration of bevacizumab treatment was 9.7 months (range 0–24.4 months) in the bevacizumab plus IFN group, and median duration of placebo treatment was 5.1 months (range 0–24.0 months) in the control group. Median duration of IFN treatment was 7.8 months (range 0–13.9 months) in the bevacizumab plus IFN group and 4.6 months (range 0.2–12.6 months) in the control group.

Median bevacizumab/placebo dose intensity was 92% (range 24–112%; mean 88%) in the bevacizumab plus IFN arm and 96% (range 39–110%; mean 89%) in the IFN only arm.

No substantial additional clinical effectiveness data were located in the related conference abstracts on this study86,113–116 or in the company submission for bevacizumab. 117

Rini and colleagues101 report the results of the Cancer and Leukaemia Group B (CALGB 90206) phase III, open-label trial of bevacizumab plus IFN versus IFN conducted in 732 patients with previously untreated metastatic clear cell RCC. Patients were randomised to receive either bevacizumab (10 mg/kg intravenously every 2 weeks) plus IFN (9 MIU subcutaneously three times weekly) or IFN alone. Randomisation was stratified by prior nephrectomy and MSKCC risk category. The primary end point was OS. Secondary end points were PFS, response rate (according to RECIST criteria) and safety. The trial was designed with 86% power to detect a difference in the HR of 30% assuming a two-sided significance level of 0.05. Preliminary results were reported at the ASCO Genitourinary Cancers Symposium in February 2008. 68,101

We considered the validity of pooling the data from the two studies of bevacizumab plus IFN; however, as the study by Rini and colleagues is available only in abstract form, several key pieces of information were missing [e.g. the number of patients randomised to each group, the method for assessing progression, whether the analysis was carried out on an intention-to-treat (ITT) basis] and we were unable to fully assess the quality of the study. The authors were contacted to request additional data but were unwilling to comply. We were therefore unable to pool the data.

Motzer and colleagues107 report the results of an international (Australia and USA), multicentre, phase III RCT in which 750 patients with metastatic RCC were randomised to receive either sunitinib or IFN. The trial has been reported in one full publication107 and in five abstracts. 87,88,91–93

The aim was to assess the efficacy of first-line treatment with sunitinib compared with IFN-α in the treatment of metastatic RCC. The primary outcome was PFS, defined as the time from randomisation to the first documentation of objective disease progression or to death from any cause, whichever came first. Secondary end points included the objective response rate, OS, quality of life outcomes and safety. The study was designed to have 90% power for the log-rank test to detect a clinically relevant increase in PFS from 4.7 to 6.2 months in patients treated with sunitinib, at a two-sided alpha level of 0.05.

To be eligible for entry into the trial participants had to have a diagnosis of metastatic RCC with a clear cell histological component confirmed by the participating centres. Patients also had to have measurable disease, an ECOG-PS of 0 or 1 and adequate haematological, hepatic, renal and cardiac function.

Patients were stratified according to baseline levels of lactate dehydrogenase, ECOG-PS and previous nephrectomy and randomly assigned to receive sunitinib (50 mg once daily, orally) in 6-week cycles (4 weeks on, 2 weeks off) or interferon-α-2a (Roferon-A^®, Roche) (9 MIU three times per week, subcutaneously). Treatment was continued until evidence of disease progression, the patient experienced unacceptable toxicity, or withdrawal of consent. Dose reductions (sunitinib to 37.5 mg and then 25 mg per day and IFN to 6 MIU and then 3 MIU three times per week) were permitted to allow management of severe adverse events.

Three scheduled interim analyses were planned. The paper by Motzer and colleagues107 published in 2007 provides the results of the second analysis, after which the study was unblinded. This paper states that, at this time point, patients in the IFN group with progressive disease (PD) were allowed to cross over into the sunitinib group. This analysis therefore provides the most complete results for the randomised population. It is not clear why patients with PD were offered further treatment as according to the protocol all treatment would be stopped on evidence of disease progression.

The median duration of treatment was 6 months (range 1–15 months) in the sunitinib group and 4 months (range 1–13 months) in the IFN group. Reasons for discontinuing treatment were PD (25% and 45% in the sunitinib and IFN groups respectively), adverse events (8% and 13% respectively), withdrawal of consent (1% and 8% respectively) and protocol violation (< 1% in each group). Dose intensity was not reported in the full-text paper. In the company submission, Pfizer report a relative dose intensity (total dose administered/total dose assigned multiplied by 100) of 86.40% for sunitinib and 83.10% for interferon, which is cited as originating from the trial of sunitinib versus IFN. 107 No further details are provided.

The AVOREN trial reported by Escudier and colleagues106 is a good quality, randomised, phase III trial. The evaluation of the trial in relation to study quality is shown in Table 9. Allocation concealment, details of randomisation methods and withdrawals were all adequately reported. The study is described as ‘double blind’, although it is unclear whether all members of the study team were blinded (e.g. patient, pharmacist, doctor and assessor).

The CALGB trial101 has only been reported in abstract form and as such there are not sufficient details to adequately assess the quality of the data.

The study assessing sunitinib versus IFN is a large, good quality, international, multicentre, randomised, phase III study. 107 Although it was not possible to double blind the study because of the differences in route of administration, the assessments of the primary outcome measure and objective response rate were performed by a central and blinded review of radiological images. Further details of the quality assessment can be found in Table 9.

At baseline in the AVOREN study106 the two treatment groups were well matched in terms of demographic characteristics and disease status (Karnofsky performance status, MSKCC risk group and the location of metastases) (Table 10).

As the trial by Rini and colleagues101 has only been reported in abstract format, few details of the population characteristics at baseline are available. Overall, 85% of patients had undergone previous nephrectomy, and 26% were assessed as having favourable prognostic risk, 64% had intermediate risk and 10% had poor risk. No further details are provided.

At baseline in the study by Motzer and colleagues107 the two treatment groups were well matched in terms of demographic characteristics and disease status (ECOG-PS, MSKCC risk factors, the number of patients with a previous nephrectomy and the number and sites of metastases) (Table 10).

Participants in the two main trials106,107 were similar in terms of age, gender distribution, RCC pathology (predominantly clear cell), the proportion that had previously undergone nephrectomy or partial nephrectomy (100% versus 90% for the bevacizumab plus IFN and sunitinib trials respectively), the number with metastatic RCC and the profile of prognosis according to MSKCC criteria (approximately 30% of patients have favourable prognosis, 60% intermediate and 10% poor). Although performance status was evaluated using different instruments, patients appear comparable, with the majority of patients (61%) in the bevacizumab plus IFN trial being assessed as ECOG-PS 0, which equates to ‘fully active, able to carry on all predisease performance without restriction’, and 69% of patients in the sunitinib trial having a Karnofsky performance status of 100 (‘normal, no complaints, no sign of disease’) or 90 (‘capable of normal activity, few symptoms or signs of disease’).

Overall survival, defined as the time between the date of randomisation and death from any cause, was the primary end point in the AVOREN trial. 106 The analysis was performed on an ITT basis with patients without an event being censored on the day of the last follow-up assessment or the last day of study drug administration if no follow-up assessment was carried out. At the time of data cut-off, only 251 (56%) of the 445 deaths required for the final analysis of OS to be powered adequately had occurred. Median OS had not been reached in the bevacizumab plus IFN group and was 19.8 months in the IFN group, with a HR of 0.79 (95% CI 0.62 to 1.02; p = 0.0670). A preplanned exploratory analysis stratified by MSKCC risk group and region produced a similar result [HR 0.75 (95% CI 0.58 to 0.97; p = 0.02670)]. Analysis of OS stratified according to baseline MSKCC risk groups was similar to the unstratified analysis with HRs of 0.69 (95% CI 0.36 to 1.33), 0.74 (95% CI 0.53 to 1.02) and 0.87 (95% CI 0.48 to 1.56) for the favourable, intermediate and poor prognosis groups respectively.

Data on OS from the CALGB trial101 are still pending.

At the time of analysis, median OS had not been reached in either group: 13% of patients in the sunitinib group and 17% in the IFN group had died. There was an improved OS with sunitinib, with a HR for death of 0.65 (95% CI 0.45 to 0.94; p = 0.02); the comparison did not meet the prespecified level of significance for the interim analysis. 107

In the AVOREN study,106 according to ITT analysis there was a statistically significant benefit in terms of median PFS observed for the bevacizumab plus IFN group (10.2 months) compared with the IFN and placebo group (5.4 months) [HR 0.63 (95% CI 0.52 to 0.75; p = 0.0001)]. An analysis stratified by MSKCC risk group and region confirmed these results [HR 0.61 (95% CI 0.51 to 0.73; p < 0·0001)]. A test of interaction indicated that the treatment effect was consistent across the MSKCC risk groups (p = 0.508).

In the CALGB study101 the method of assessing progression was not reported in the abstract. Median time to progression was 8.5 months in patients receiving bevacizumab plus IFN and 5.2 months in the group receiving IFN alone. The stratified estimate of the HR was 0.71 (95% CI 0.61 to 0.83; p < 0.0001). Further details of the analysis are not yet available.

Progression-free survival (primary end point) was assessed by blinded central review of imaging studies. 107 There was a statistically significant difference in PFS in patients receiving sunitinib (11 months; 95% CI 10 to 12 months) compared with those receiving IFN (5 months; 95% CI 4 to 6 months) corresponding to a HR of 0.42 (95% CI 0.32 to 0.54; p < 0.001). Similar results from the investigators’ unblinded assessment of radiological images [11 months versus 4 months; HR 0.42 (95% CI 0.33 to 0.52; p < 0.001)] are also reported.

In the AVOREN trial106 tumour response was assessed by the investigators every 8 weeks up to 32 weeks and every 12 weeks thereafter until disease progression. At the time of analysis the overall number of patients in whom a tumour response was measured was significantly greater (p = 0.0001) in the bevacizumab plus IFN group (n = 96; 31%) than in the IFN group (n = 37; 13%). A small number of patients in both groups were assessed as having a complete response to treatment (four versus six in the bevacizumab plus IFN and IFN groups respectively), and 92 patients (30%) receiving bevacizumab plus IFN and 31 patients (11%) in the IFN group experienced a partial response to treatment (defined as a 30% decrease in the sum of the longest diameters of target lesions).

Few details are provided in the abstract describing the CALGB study. 101 The objective response rate was significantly (p ≤ 0.0001) higher in patients receiving bevacizumab plus IFN [25.5% (95% CI 20.9% to 30.6%)] than in those receiving IFN [13.1% (95% CI 9.5% to 17.3%)].

Tumours were assessed both by independent central review and by the treating physicians at baseline, at day 28 of cycles 1–4 and every 2 weeks thereafter until the end of treatment. Assessments were also made if disease progression was suspected clinically. The objective response rate, assessed by blinded imaging studies, was significantly higher in the sunitinib group (n = 103; 31%) than in the IFN group (n = 20; 6%) (p < 0.001). No patients in either group were assessed as having a complete response. Results obtained from investigator review of images were similar [137 (37%) versus 33 (9%) patients in the sunitinib versus IFN groups, respectively; p < 0.001].

Health-related quality of life was not reported in either of the trials of bevacizumab plus IFN versus IFN. 101,106

Health-related quality of life was assessed using the Functional Assessment of Cancer Therapy – General (FACT-G) and FKSI questionnaires (see Chapter 1, Quality of life), which were administered before randomisation, on days 1 and 28 of each cycle and at the end of treatment. No data are available on the comparability of the groups at baseline on these measures. Using data from all postrandomisation assessments, least-square means were estimated for each treatment group. A higher score indicates a better outcome. Overall differences between the two groups were tested using repeated-measures mixed-effects models controlling for the assessment time, treatment by time interaction and the baseline score. Table 14 shows that the overall results (total score and all subscales of the FACT-G and total score and the FKSI-DRS subscale) were all significantly better for patients in the sunitinib group than for those in the IFN group.

In the AVOREN trial,106 in both groups the most commonly reported ‘any grade’ adverse event was pyrexia (in 45% and 43% of patients treated with bevacizumab plus IFN and IFN alone respectively), followed by anorexia (36% and 30% of patients respectively), fatigue (33% and 27% respectively), asthenia (32% and 28% respectively) and influenza-like illness (24% and 25% respectively). There were 203 grade 3 or worse adverse events reported by patients who received one or more doses of bevacizumab compared with 137 reported by those who did not receive the drug. The frequency of grade 3 and 4 adverse events was low, being between < 1% and 12%, with most grade 3 or 4 adverse events occurring at a frequency of 3% or less. The mean number of grade 3 or worse adverse events per patient was 1.3 in the intervention group and 0.9 in the control group. Details of statistical analyses are not provided. Adverse events that led to treatment discontinuation occurred more frequently in patients who received bevacizumab (n = 95; 28%) than in those who did not (n = 37; 12%). Proteinuria, hypertension and gastrointestinal perforation were the most common reasons for treatment discontinuation (Table 56, Appendix 5). Adverse event-related deaths were reported in eight (2%) patients who received bevacizumab and in seven (2%) patients who did not. Three of the deaths in patients who received bevacizumab (two bleeding events and one gastrointestinal perforation) were believed to be possibly related to bevacizumab.

The abstract of the CALGB study101 states that overall toxicity in the bevacizumab plus IFN group was greater than that in the IFN only group, with significantly more patients reporting grade 3 hypertension (9% versus 0%), anorexia (17% versus 8%), fatigue (35% versus 28%) and proteinuria (13% versus 0%).

The most commonly reported ‘any grade’ adverse events and laboratory abnormalities in the sunitinib group were diarrhoea (53% of patients), fatigue (51% of patients), nausea (44% of patients), leukopenia, neutropenia, anaemia, increased creatinine, thrombocytopenia and lymphopenia (which all occurred in more than 50% of the patients treated with sunitinib). A similar adverse event profile was seen in the IFN group with fatigue (51%), pyrexia (34%), nausea (33%) and chills (29%) being the most frequently reported adverse events and anaemia, lymphopenia and leukopenia the most commonly reported laboratory abnormalities (all occurring in more than 50% of patients treated with IFN). There were statistically significant differences (p < 0.05) between groups in the frequency of reporting of the following adverse events at grade 3 and above: diarrhoea, fatigue, vomiting, hypertension, hand–foot syndrome, leukopenia, neutropenia, thrombocytopenia, lymphopenia, increased lipase and increased amylase, with all but fatigue, anaemia and lymphopenia occurring more often in the sunitinib group than in the IFN group. Approximately 12% of patients in the IFN group experienced grade 3 or 4 adverse events compared with 7% in the sunitinib group; this difference was statistically significant (p < 0.05). Treatment discontinuation as a result of unacceptable adverse events occurred more frequently in the IFN group than in the sunitinib group (13% versus 8%; p = 0.05); no further details are provided. A total of 38% of patients in the sunitinib group and 32% in the IFN group had a dose interruption because of adverse events and in a similar proportion dosage was reduced (32% and 21% in the sunitinib and IFN groups respectively).

It is not clear from the paper whether any deaths occurred during the trial that may have been attributable to the study medication.

From the adverse events reported in these trials the safety profile of both interventions appears to be comparable to that of IFN, with some adverse events particularly associated with bevacizumab plus IFN (proteinuria, hypertension, bleeding events) and sunitinib (hypertension, hand–foot syndrome). However, randomised clinical trials are not designed to detect rare adverse events and we therefore briefly reviewed additional literature, obtained from the results of our initial and updated searches, to identify any further potential safety issues.

Hudes and colleagues108 report the results of the Global Advanced Renal Cell Carcinoma (ARCC) trial. An international (Argentina, Australia, Canada, Czech Republic, Germany, Greece, Hungary, Italy, Latvia, Lithuania, Netherlands, Poland, Russia, Serbia and Montenegro, Slovakia, South Africa, Spain, Sweden, Taiwan and Turkey), multicentre, three-way parallel group, randomised phase III trial in which 626 people with previously untreated metastatic RCC, deemed to have poor prognosis according to criteria based on MSKCC risk score, received either temsirolimus, IFN or a combination of temsirolimus and IFN. The study has been published in one full paper108 and in five abstracts. 89,94–97 The primary outcome was OS. PFS, objective response rate and the ‘clinical benefit rate’ (defined as the proportion of people with stable disease for at least 24 weeks or an objective response) were secondary outcomes. The study (with 200 patients per group) was designed to have 80% power to detect an improvement in OS of 40% for each comparison with the use of a two-sided stratified log-rank test at an overall 2.5% level of significance. Two interim analyses were planned after approximately 164 and 430 deaths and a final analysis, if necessary, after a total of 504 deaths had occurred; the paper by Hudes and colleagues108 provides the results of the second analysis (after 446 patients had died).

Trial eligibility is defined in Table 58 in Appendix 5. Participants were required to have a diagnosis of histologically confirmed RCC, a Karnofsky performance status of 60 or more and measurable disease according to RECIST criteria. All patients had to fulfil prespecified criteria for poor prognosis to be eligible. Although based on the MSKCC classification of prognosis, the criteria used in this trial were slightly different. The MSKCC classification includes five predictors of survival, of which a patient with poor prognosis needs to exhibit three. Participants in this trial were required to exhibit three of six features to be defined as having poor prognosis, the additional feature being ‘metastases in multiple organs’.

Randomisation was performed centrally and patients were stratified according to the geographical location of the centre and whether they had undergone previous nephrectomy. Patients were randomly assigned to receive temsirolimus (25 mg, delivered intravenously, weekly) (n = 209), IFN (18 MIU, delivered subcutaneously three times per week) (n = 207) or a combination of both treatments (n = 210). Treatment was continued until evidence of disease progression, symptomatic deterioration or intolerable adverse events. IFN was started at a dose of 3 MIU for the first week, increased to 9 MIU for the second week and 18 MIU for the third week. Treatments were withheld if grade 3 or 4 adverse events occurred and restarted at a reduced dose after recovery to grade 2 or lower.

The results reported in the full publication108 were obtained from the second interim analysis after 446 deaths. At the time of data analysis, median treatment duration for temsirolimus was 3.92 months (range 0.23–29.08 months) in the temsirolimus alone group and 3.46 months (range 0.23–31.85 months) in the group receiving combination treatment. For IFN the figures were 1.85 months (range 0.23–28.62 months) in the IFN group and 2.77 months (range 0.23–31.85 months) in the combination group.

The mean dose intensity of temsirolimus was 23.1 mg per week or 92% of the planned dose; corresponding figures for IFN are 30.2 MIU per week or 56% of the maximum planned dose in the first 8 weeks of treatment. No further details are provided.

Data from the final analysis were available from a conference abstract89 and were presented in the company submission to NICE. 124 Median treatment duration at this analysis is not reported in either source.

Additional data relating to HRQoL, reported in a conference abstract97 and the company submission, are also included (see section on assessment of clinical effectiveness).

See section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy.

See section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy.

This is a large, international, multicentre randomised clinical trial. Although, on the whole, methods are clearly reported, several aspects are not clear in the paper, making the assessment of quality somewhat difficult. Details of randomisation methods and withdrawals were adequately reported, but details of how the randomisation code was generated were omitted. Site investigators were not blind to treatment allocation, although radiological scans used for assessment of PFS and response rate were assessed both by site investigators and by central blinded review. Only the analysis of the primary end point (OS) was conducted on an ITT basis. Further details can be found in Table 9.

See section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy.

See section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy.

In this assessment we are interested in two of the three patient groups in this trial, temsirolimus alone and IFN alone as the combination of temsirolimus and IFN is not licensed for use in people with advanced and/or metastatic RCC. At baseline these two treatment groups were well matched in terms of demographic characteristics (age, gender, RCC histology) and disease status (Karnofsky performance status, MSKCC risk group and proportion of patients having undergone a previous nephrectomy) (Table 19). Most tumours had clear cell histology (approximately 80%) and most patients had Karnofsky performance scores of < 70 (approximately 80%) and had undergone a previous nephrectomy (approximately 65%). It is interesting to note that, according to MSKCC risk classification, approximately 30% of patients in both treatment groups would have been classified as having intermediate prognosis rather than poor prognosis, and about 5% of patients in both treatment groups did not meet the criteria for entry into the study (i.e. three or more of six factors suggestive of poor prognosis).

In the study by Motzer and colleagues,107 23 (6%) patients receiving sunitinib and 25 (7%) patients receiving IFN had three or more MSKCC risk factors and were therefore classified as having poor prognosis. As described above, this classification is slightly different from that used in the trial of temsirolimus. The baseline population characteristics of the entire trial population are described in the section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy.

In total, 9% (n = 28) of the patients who received bevacizumab plus IFN and 7% (n = 24) of the patients receiving IFN in the trial by Escudier and colleagues106 had three or more MSKCC risk factors for poor prognosis. Again, the definition of poor prognosis differs from that used in the trial of temsirolimus. The baseline population characteristics of the entire population are described in the section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy.

As population baseline characteristics are not presented separately for the poor prognosis subgroups in the trials of sunitinib and bevacizumab, comparison between the studies is problematic. However, assuming that the patients with poor prognosis were characteristic of the trial populations as a whole, the demographics (median age, gender mix) of patients included in all three studies appear similar. There are, however, differences between trials in terms of the proportion having undergone previous nephrectomy (100% versus 90% versus 65% in the trials of bevacizumab plus IFN, sunitinib and temsirolimus respectively) and the proportion of patients with clear cell carcinoma (100% versus 100% versus 80% in the trials of bevacizumab plus IFN versus sunitinib versus temsirolimus respectively).

Overall survival was the primary outcome measure of the Hudes and colleagues trial108 and was analysed on an ITT basis. At the time of the interim analysis, median OS was 7.3 months (95% CI 6.1 to 8.8 months) in the IFN group and 10.9 months (95% CI 8.6 to 12.7 months) in the temsirolimus group, producing a HR of 0.73 (95% CI 0.58 to 0.92; p = 0.008).

In the final analysis, median OS in the IFN group was 7.3 months (95% CI 6.1 to 8.8 months) and in the temsirolimus group was 10.9 months (8.6 to 12.7 months), producing a slightly higher HR of 0.78 (95% CI 0.63 to 0.97; p = 0.0252) indicating that temsirolimus reduced the hazard of death by 22%. 89

These results suggest that temsirolimus may be superior to IFN in this patient group. However, the 95% CIs surrounding the estimates are reasonably wide and approach unity at the upper limit (which would indicate no difference between treatments) highlighting the degree of imprecision of these results.

Data on OS were not presented separately for the poor prognosis subgroups in these trials.

Median PFS (defined as time between randomisation and first documented disease progression or death due to any cause) for patients in the poor prognosis subgroup was 2.2 months for those receiving bevacizumab plus IFN and 2.1 months for those treated with IFN, producing a HR of 0.81 (95% CI 0.46 to 1.42) (Table 21). As the 95% CI crosses unity this result would not be considered statistically significant, but could be interpreted as indicating a possible benefit of treatment with bevacizumab plus IFN compared with IFN in this patient subgroup. The lack of statistical significance could be because bevacizumab plus IFN is not more effective than IFN in patients with a poor prognosis or it may reflect the small number of patients (n = 52) in this subgroup.

The study of sunitinib versus IFN includes results for PFS for subgroups according to baseline factors. For all subgroups the HR favours sunitinib. However, data for the group of patients with three or more MSKCC risk factors are not presented separately. This trial therefore does not provide any additional information about the effectiveness of sunitinib versus IFN in this particular population. A later analysis of the trial (following the decision to allow patients in the IFN group to receive sunitinib) is available as a conference abstract91 and suggests that the benefit of sunitinib over IFN in terms of PFS (by investigator assessment) extends over all MSKCC risk groups.

Progression-free survival (not formally defined in the paper108) was assessed both by the site investigators (who were not blind to treatment allocation) and by independent blinded evaluation of the radiological images (Table 21). In the interim analysis, as determined by the site investigators, median PFS was 1.9 months (95% CI 1.9 to 2.2 months) in the IFN group and 3.8 months in the temsirolimus group (95% CI 3.6 to 5.2 months). 108 Radiological images from 153 patients (74%) in the IFN group and 192 patients (92%) in the temsirolimus group were evaluated in the independent blinded review, the results of which suggest that median PFS was 3.1 months (95% CI 2.2 to 3.8 months) and 5.5 months (95% CI 3.9 to 7.0 months) for the IFN and temsirolimus groups respectively. The authors suggest that the reason for the discrepancy in these results is the inclusion in the evaluation by site investigators of patients with symptomatic deterioration that had begun before scheduled radiological measurements of the tumour. HRs are not provided in the paper, nor is there any indication of the results of statistical testing. However, the abstract of the paper states that patients who received temsirolimus alone had longer PFS than did patients who received IFN alone (p < 0.001).

In the final analysis, median PFS by independent assessment was 5.6 months (95% CI 3.9 to 7.2 months) in the temsirolimus group and 3.2 months (95% CI 2.2 to 4.0 months) in the IFN group, with a HR of 0.74 (95% CI 0.60 to 0.91; p = 0.0042). 124 Again, the investigator evaluation resulted in slightly lower estimates of PFS (3.8 months versus 1.9 months for temsirolimus and IFN respectively). Interestingly, the HR was almost identical (0.74; 95% CI 0.60 to 0.90; p = 0.0028). 89

Tumour response results were not presented separately for the poor prognosis subgroup in these trials. 106,107

Before the start of treatment, the following imaging studies were performed: CT scans of the chest, abdomen and pelvis, a radionuclide bone scan and an MRI or CT scan of the brain. Scanning was repeated at 8-week intervals to evaluate tumour size. Response to treatment was assessed using the RECIST criteria. Objective response rates in the IFN and temsirolimus groups were 4.8% (95% CI 1.9% to 7.8%) and 8.6% (95% CI 4.8% to 12.4%), respectively, and did not differ significantly.

No additional information on the effect of these treatments on HRQoL in patients with poor prognosis was available from these trials. 106,107

No HRQoL outcomes were reported in the full-text paper. 108 In a subsequent conference abstract presented in 2007,97 results for quality-adjusted survival (a predefined end point) are presented. Quality-adjusted survival and toxicity (Quality-adjusted Time Without Symptoms of disease or Toxicity of treatment; Q-TWiST) were estimated by partitioning OS into three distinct health states: time with serious toxicity, time with progression and time without symptoms and toxicity (TWiST). Survival was value weighted when patients completed EuroQoL 5 dimensions (EQ-5D) questionnaires at weeks 12 and 32, when a grade 3 or 4 adverse event was reported, upon relapse or progression, or upon withdrawal from the trial. All 626 randomised patients in the trial were included in the computation of health state durations. This includes patients in all three treatment groups – temsirolimus alone, IFN alone and the combination of temsirolimus and IFN. EQ-5D questionnaires were obtained from 260 of 300 patients upon progression and from 230 of 570 patients after a grade 3 or 4 adverse event. Patients receiving temsirolimus had 38% greater TWiST than those receiving IFN (6.5 months versus 4.7 months for temsirolimus and IFN respectively; p = 0.00048) and 23% greater Q-TWiST than those receiving IFN (7.0 months versus 5.7 months for temsirolimus and IFN respectively; p = 0.0015). Median EQ-5D scores for the total trial population are shown in Table 22.

See section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy. No additional data were provided for those in the poor prognosis subgroup.

See section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy. No additional data were provided for those in the poor prognosis subgroup.

Adverse events were defined and graded according to the CTCAE, version 3.0. No further details were provided. Table 59 in Appendix 5 details all adverse events of any grade reported by at least 20% of patients in any group. The tables include all adverse events, not only those considered to be drug related. Asthenia was the most commonly reported adverse event among patients in all treatment groups. Anaemia, nausea, anorexia, fever and chills were also commonly reported in all treatment groups. Patients treated with temsirolimus experienced more rashes, hyperlipidaemia, infection, peripheral oedema, hyperglycaemia, cough, hypercholesterolaemia and stomatitis than patients receiving IFN, although whether these differences were statistically significant is unclear.

Table 24 shows adverse events classified as grade 3 or 4 based on the adverse events that occurred in more than 20% of patients in any group (shown in Table 59 in Appendix 5). For simplicity, only data for the temsirolimus and IFN groups are presented. More patients in the IFN group than in the temsirolimus group reported grade 3 or 4 adverse events (78% versus 67%; p = 0.02). The most commonly occurring grade 3 or 4 adverse event in the temsirolimus group was anaemia (in 20% of patients). Events that occurred more frequently in the temsirolimus group than in the IFN group include dyspnoea (in 9% and 6% of patients respectively) and rash (in 4% and 0% of patients respectively), although the number of patients affected is relatively small and whether these differences were considered statistically significant is unclear. Treatment was discontinued as a result of adverse events in twice as many people receiving IFN as temsirolimus, although the number of people involved was again small [29 (14%) and 15 (7%) in the IFN and temsirolimus groups respectively)]. The number of deaths as a result of adverse events was not reported.

Based on the data reported in these trials the frequency of treatment-related toxic events associated with bevacizumab plus IFN, sunitinib and temsirolimus appears to be comparable to or slightly better than the frequency of treatment-related toxic events associated with IFN. There are some particular adverse events associated with each of the three interventions: bevacizumab plus IFN – proteinuria, hypertension, bleeding events; sunitinib – hypertension, hand–foot syndrome; and temsirolimus – for example hyperglycaemia, hyperlipidaemia, hypercholesterolaemia, peripheral oedema, rash. However, randomised clinical trials are not designed to detect rare adverse events and we therefore briefly reviewed additional data sources to identify any further potential safety concerns. The results of this review are detailed on pp.28–29 for bevacizumab plus IFN and sunitinib. A systematic review of toxicities associated with the administration of sorafenib, sunitinib and temsirolimus in phase I, II and III clinical trials found that between 1% and 20% of patients experience grade 3 or 4 adverse events with temsirolimus treatment. The most commonly experienced grade 3 and 4 adverse events across all included trials of temsirolimus were anaemia (20%), fatigue/asthenia (11%), hyperglycaemia (11%) and dyspnoea (9%). 119

There is some evidence to suggest that, in the poor prognosis subgroup, the combination of bevacizumab plus IFN is more effective in terms of prolonging PFS than IFN alone (2.2 months versus 2.1 months; HR 0.81, 95% CI 0.46 to 1.42); this is consistent with the results obtained from the entire trial population. No additional safety data were available for this subgroup, but there is also nothing in the trial report to suggest that the adverse event profile would be any different than that seen in the whole trial population.

Although some of the patients included in the trial of sunitinib versus IFN were characterised as having poor prognosis, the results of the trial were not reported according to prognosis and so this trial is also not able to offer any substantial evidence.

From the limited clinical data available, treatment with temsirolimus appears to have clinically relevant and statistically significant advantages over treatment with IFN in people with poor prognosis, in terms of OS, PFS and tumour response. Median PFS was approximately doubled from 1.9 months with IFN to 3.8 months with temsirolimus (HR 0.74; 95% CI 0.60 to 0.90). Data on adverse events suggest that temsirolimus may be associated with a lower frequency of grade 3 or 4 adverse events than IFN, although the overall frequency of adverse events is still relatively high.

Data on patients with and without clear cell carcinoma and previous nephrectomy suggest that temsirolimus is more effective than IFN in all of these subgroups. Whether the results are sufficiently distinct from each other to suggest that people in these subgroups respond differently to temsirolimus is not clear.

Escudier and colleagues report the results of the TARGET (Treatment Approaches in Renal Cancer Global Evaluation Trial) study, an international (Argentina, Australia, Belgium, Brazil, Canada, Chile, France, Germany, Hungary, Israel, Italy, Netherlands, Russia, South Africa, Ukraine, UK and USA), multicentre, double-blind and placebo-controlled phase III RCT in which 903 patients with histologically confirmed metastatic clear cell RCC were randomised to receive either sorafenib (400 mg orally twice daily; n = 451) or matched placebo (n = 452). Results of this trial have been reported in two full publications109,114 and five abstracts. 98,99,102,103,125 The primary outcome was OS. PFS and overall response rate were amongst the secondary outcome measures. Data on safety and HRQoL were also collected. The study was designed to have 90% power to detect a 33.3% difference in survival between the two groups at a two-sided alpha level of 0.04 after 540 patients had died. Patients were stratified according to country and MSKCC prognostic score (low or intermediate).

Eligibility criteria included the presence of histologically confirmed metastatic clear cell RCC that had progressed after one systemic treatment within the previous 8 months, an ECOG-PS of 0 or 1, an intermediate or low risk according to the MSKCC prognostic score and a life expectancy of at least 12 weeks.

Treatment was continued until evidence of disease progression or withdrawal from the study because of adverse events occurred. Dose reductions (to 400 mg once daily and then to 400 mg every other day) were permitted to manage adverse events.

Enrolment of patients took place between 23 November 2003 and 3 March 2005. From November 2003 until April 2005 the sponsor and investigators were unaware of the study group assignments in the evaluation of data. In January 2005 a protocol-defined independent review of the status of 769 patients (384 in the sorafenib group and 385 in the placebo group) was conducted. In April 2005 a decision was made by the independent data and safety monitoring committee that study group assignments should be revealed and that sorafenib should be offered to patients receiving placebo. The initial analysis of OS, which is presented in the main publication,109 is based on data obtained before treatment crossover. A further analysis of OS was performed 6 months later.

The median duration of treatment (at the time of the interim analysis) was 23 weeks in the sorafenib group and 12 weeks in the placebo group. Dose intensity was not reported.

No supplementary additional data were identified in conference abstracts. (Commercial-in-confidence data have been removed.)109

In 2006, Ratain and colleagues110 reported the results of an RDT of sorafenib versus placebo in a total of 202 patients with metastatic clear cell RCC. In an RDT (a study design that was developed in an attempt to assess the clinical activity of a drug whilst minimising exposure to placebo) all patients receive the study drug for an initial run-in period followed by random assignment of potential responders to either the active drug or placebo. The design creates a controlled trial without upfront randomisation and decreases the heterogeneity of randomised patients, resulting in increased statistical power with smaller patient numbers. The study initially permitted enrolment of patients (n = 502) with a variety of tumour types including metastatic RCC and metastatic colorectal cancer. Early indications of activity in patients with RCC caused a refocus on this patient population and resulted in 40% of patients in the overall trial having a diagnosis of metastatic RCC. The paper by Ratain and colleagues110 describes only the RCC population. The primary outcome measure was the percentage of randomly assigned patients who remained progression free at 12 weeks following random assignment. Other end points included PFS after random assignment (randomised subset only), overall PFS (from start of treatment), tumour response rate and safety. The study was designed to have 81% power to detect a drug effect that corresponded to a reduction in the progression rate from 90% to 70% 12 weeks after randomisation.

Sorafenib (400 mg twice a day) was administered to all patients in a 12-week open-label run-in period after which disease status was assessed based on changes in bidimensional tumour measurements from baseline. Patients with ≥ 25% tumour shrinkage continued to receive sorafenib until disease progression or toxicity. Patients with PD (≥ 25% tumour growth or other evidence of progression) discontinued treatment. Patients who had a change in tumour size of < 25% were randomly assigned to either sorafenib (at the same dose) or matched placebo using centrally allocated allocation via a telephone randomisation system. Treatment was stopped on disease progression.

No additional supplementary data were identified either in abstract form or as part of the company submission for sorafenib.

Motzer and colleagues report the results of two similar open-label, single-arm trials of sunitinib as second-line therapy in patients with metastatic clear cell RCC. In both trials, conducted in multiple centres in the USA and reported in 2006, patients received treatment with sunitinib [50 mg per day, self-administered orally, in repeated 6-week cycles (4 weeks on treatment followed by 2 weeks off)] until evidence of disease progression, unacceptable toxicity or withdrawal of consent.

In the earlier trial (n = 63),112 eligible patients had a diagnosis of histologically confirmed metastatic RCC (of any subtype), evidence of failure of one cytokine-based therapy because of disease progression or unacceptable toxicity, and an ECOG-PS of 0 or 1. Entry criteria for the larger trial (n = 105)111 were similar, but entry was restricted to patients with histologically confirmed clear cell typology who had undergone previous nephrectomy. The primary outcome measure in both trials was objective response rate according to the RECIST criteria. 111 A later publication providing OS data is also available. 85

No additional supplementary data were identified within the relevant conference abstracts or the company submission for sunitinib.

The quality assessment of these trials is summarised in Table 9. Both are well-conducted and well-reported large, multicentre trials. In the report of the RCT of sorafenib versus placebo109 the authors state that the final planned analysis of OS (which was undertaken after treatment crossover) was conducted on an ITT basis. It is not clear whether the unplanned analysis of OS (before treatment crossover) was also performed under these conditions. Methods for censoring in these analyses are also not provided.

The company submission to NICE from Bayer includes commercial-in-confidence subgroup analyses from this trial. (Commercial-in-confidence data have been removed.) For several reasons we have not considered the results of this analysis further. The clinical basis underlying an expected difference in response to treatment in these two groups of people is not immediately evident. (Commercial-in-confidence data have been removed.) To be considered eligible for the study, patients were required to have disease that had progressed after one systemic treatment within the previous 8 months; in 17% of patients the nature of this systemic therapy is not reported in the paper. (Commercial-in-confidence data have been removed.)

It appears from the details of the sample size calculation provided in the RDT that the investigators were aiming to recruit 50 randomly assigned patients to each group. In practice, a total of 65 patients was randomly assigned in the study.

We have applied a similar list of quality assessment criteria to the two sunitinib trials as used in other critical appraisals in this assessment (Table 9), with obvious exceptions (e.g. methods of randomisation and concealment, etc.); they appear to be well designed and reported.

In the study by Escudier and colleagues,109 population characteristics at baseline were well balanced between the groups in terms of demographic factors (age and gender distribution) and disease status (ECOG-PS and MSKCC prognostic risk score, the proportion of patients with multiple metastatic sites, the location of metastases, previous systemic therapy, the proportion of patients with previous nephrectomy and the median duration of disease) (Table 29). Approximately half of the people in the trial had an ECOG-PS of 0, most (83%) had had previous cytokine-based treatment and the majority (94%) had undergone previous nephrectomy. To be considered eligible for the study, patients were required to have disease that had progressed after one systemic treatment within the previous 8 months; in 17% of patients the nature of this systemic therapy is not reported in the paper.

A similar group of patients was entered into the RDT110 and again the groups were well balanced at baseline. There were slightly more females in the placebo group, but this difference was not statistically significant.

As already described the two trials of sunitinib85,111,112 included patients with similar baseline characteristics, the main differences between trials being the proportion of patients with clear cell RCC and the proportion of patients with previous nephrectomy (Table 29).

Participants in all four trials were similar in terms of age, gender distribution and disease status. Approximately 50% of people in all four trials had an ECOG-PS of 0 and a favourable prognostic score according to MSKCC criteria. Cytokine-based therapies had failed to halt disease progression in the majority of patients and most had undergone a previous nephrectomy. Almost all patients had two or more sites of metastatic disease with the lung being the most common site for metastases in all trials.

Overall survival (defined as the time between the date of randomisation until the date of death) was the primary end point in the RCT of sorafenib versus placebo. 109 In the analysis performed before treatment crossover, 220 of the 540 deaths required for the comparison to be adequately powered had occurred; 97 deaths in the sorafenib group and 123 deaths in the placebo group. Median actuarial OS had not been reached in the sorafenib group and was 14.7 months in the placebo group with a HR of 0.72 (95% CI 0.54 to 0.94; p = 0.02). This result was not considered statistically significant as it did not reach the O’Brien–Fleming threshold of 0.0005.

Overall survival was not an outcome measure in the RDT. 110

Overall survival was 23.9 months (95% CI 14.1 to 30.7 months) in the larger trial of 105 patients85,111 and 16.4 months (95% CI 10.8 to not yet attained) in the smaller trial (n = 63). 112 Interpretation of these results is difficult because of the lack of a comparator group.

Escudier and colleagues109 determined disease progression on the basis of CT or MRI, clinical progression or death. Imaging studies were performed every 8 weeks and assessed according to the RECIST criteria. Investigators and independent radiologists who were unaware of treatment assignments assessed PFS. No information on the method of censoring of values is provided. Median PFS (defined as the time from the date of randomisation to the date of progression) based on 769 patients at the first preplanned interim analysis was 5.5 months in the sorafenib group and 2.8 months in the placebo group; it is unclear from the paper but we assume that this analysis was based on assessment by independent radiologists. Investigator-assessed PFS at the same time point was 5.9 months in the sorafenib group and 2.8 months in the placebo group, with a HR of 0.44 (95% CI 0.35 to 0.55; p < 0.001).

A similar result was obtained at treatment crossover when investigator-assessed PFS in 903 patients was found to be 5.5 months in the sorafenib group and 2.8 months in the placebo group (HR 0.51; 95% CI 0.43 to 0.60; p < 0.001). It is unclear why the authors have chosen to present results based on investigator assessment rather than on assessment by independent radiologists or if there were any differences in the results obtained by the two methods of assessment.

In the RDT of sorafenib versus placebo,110 at 12 weeks post randomisation (24 weeks from study entry) there was a statistically significant (p = 0.0077) difference between groups in the proportion of patients in whom disease progression was evident (50% of patients treated with sorafenib versus 82% treated with placebo). Median PFS from the date of randomisation was also significantly longer in the sorafenib group than in the placebo group (24 weeks versus 6 weeks; p = 0.0087).

The two trials of sunitinib produced similar results for PFS. In the smaller trial,112 median PFS was 8.7 months (95% CI 5.5 to 10.7 months). Based on independent third-party assessment of response, median PFS in the larger trial85,111 was 8.8 months (95% CI 7.8 to 13.5 months). Interpretation of these results is difficult because of the lack of a comparator group.

In the RCT of sorafenib and placebo,109 at the initial planned interim analysis, tumour response was assessed (by independent reviewers according to RECIST criteria) in 672 patients, although data were missing for 87 (approximately 13%). Data were available for 297 patients in the sorafenib group and 288 in the placebo group. In the sorafenib group seven patients (2%) had a partial response, 261 (78%) patients had stable disease and 29 patients (9%) had PD. In the placebo group no patients were assessed as having a partial response, 186 (55%) had stable disease and 102 (30%) had PD. At the unplanned analysis before treatment crossover, according to blinded investigator assessment, one patient in the sorafenib group exhibited a complete response, 43 had a partial response and 333 had stable disease. In the placebo group the corresponding figures were none, eight and 239. Significantly (p < 0.001) more patients in the sorafenib group than the placebo group had a complete or partial response.

Tumour response was not an outcome measure in the RDT. 110

In the two trials of sunitinib objective tumour response, defined according to RECIST, was the primary end point. Assessments of tumour response were made using CT or MRI and bone scans (if bone metastases were present at baseline) at least after every two cycles (the assessment intervals were slightly different in the two trials) until the end of treatment. In the smaller trial (n = 63)112 partial responses were achieved in 25 patients (40%; 95% CI 28% to 53%). Best response of stable disease for 3 or more months was observed in a further 17 patients (27%). The remaining patients (n = 21; 33%) had either progressive or stable disease of less than 3 months duration or were not assessable. In the larger trial111 tumour response was assessed both by treating physicians and a third-party imaging laboratory (with two radiologists). According to third-party assessment of images, 33% of patients (n = 35) had a partial response and a further 30% of patients (n = 31) had stable disease for 3 or more months. The remainder (n = 39; 37%) were assessed as having PD or stable disease for less than 3 months. These results are difficult to interpret as there was no comparator group.

In the RCT of sorafenib versus placebo,109 the FACT-G and the FKSI were administered to assess the impact of treatment on HRQoL (see Chapter 1, Quality of life). Assessments were made every 6 weeks for the first 24 weeks and then every 8 weeks. Subjects completed the questionnaires before seeing the physician. No further assessments were made after withdrawal from treatment. There was no significant difference between the placebo and sorafenib groups in mean FACT-G physical well-being score nor any numerical or statistical difference in mean FKSI-10 total score between groups over the first 30 weeks of treatment (p = 0.83 and p = 0.98 respectively).

However, there were statistically significant changes in some of the individual items of the FKSI-15 in patients receiving sorafenib compared with those receiving placebo in the first 30 weeks of treatment. These included less coughing (p < 0.0001), fewer fevers (p = 0.0015), a greater ability to enjoy life (p = 0.0119) and less worry about their disease (p = 0.0004). Fewer patients in the placebo group reported being bothered by the side effects of treatment (p < 0.0001). There were no significant differences between groups in terms of patients’ perceptions of fatigue, quality of sleep, pain, weight change or energy levels.

HRQoL was not assessed in the RDT. 110

The EQ-5D questionnaire and the Functional Assessment of Chronic Illness Therapy – fatigue scale (FACIT-fatigue) were used to assess HRQoL in the smaller trial of sunitinib. 112 EQ-5D questionnaires were administered on days 1 and 28 of each cycle, and the FACIT-fatigue questionnaire was completed on day 1 and then weekly for cycles 1–4. Compliance with questionnaires at baseline and subsequent visits was high (at or above 90% at each visit for each instrument).

Assessable baseline questionnaires for EQ-5D were received from 60 patients and compliance with subsequent assessments was high. Mean and median health state visual analogue scale scores indicated that the study population’s quality of life before treatment was similar to that of an age-matched US general population. Mean and median health state visual analogue scale scores were similar to baseline scores throughout the 24 weeks of treatment.

Valid baseline questionnaires for the FACIT-fatigue scale were received from 62 patients. Mean and median baseline scores for the study population were similar to scores of a population with cancer (but no anaemia) but lower than those of a general US population. Median and mean fatigue scores were similar to baseline scores throughout 24 weeks of treatment, although the authors did notice a mild and reversible effect of treatment on fatigue levels.

These results are not easy to interpret or extrapolate as there was no comparator group.

Although we were able to locate four trials relevant to this comparison, all of which included patients with similar baseline characteristics, because there was no common treatment arm we were unable to consider an indirect comparison of sorafenib, sunitinib and BSC.

In the TARGET trial109 the most common adverse events of any grade were fatigue (in 37% and 28% of patients treated with sorafenib and placebo respectively), diarrhoea (43% and 13% of patients respectively), rash or desquamation (40% and 16% respectively), nausea (23% and 19% respectively), hand–foot skin reaction (30% and 7% respectively) and alopecia (27% and 3% respectively). There was a statistically significant difference between groups in the proportion of patients reporting grade 2 hypertension, weight loss, diarrhoea, hand–foot skin reaction, rash, alopecia and pruritus; these events were all more common in the sorafenib group. The difference remained significant for hypertension and hand–foot skin reaction when grade 3 and grade 4 adverse events were considered. Grade 3 or 4 bone pain was reported significantly more often by patients in the placebo group. In addition to the events described in Table 33, cardiac ischaemia or infarction occurred in 12 patients (3%) in the sorafenib group and two patients in the placebo group (1%); this difference was also statistically significant (p = 0.01). Of these events, 11 (including two deaths in the sorafenib group and one death in the placebo group) were considered to be serious adverse events associated with treatment. Serious adverse events leading to hospitalisation or death were reported in 154 patients (34%) in the sorafenib group (46 deaths;10%) and 110 patients (24%) in the placebo group (25 deaths; 6%) (p < 0.01). The most frequent drug-related serious adverse event was hypertension (in 1% and 0% of sorafenib and placebo patients respectively).

In the RDT of sorafenib versus placebo,110 the most common treatment-emergent adverse events were fatigue (73% of patients), rash or desquamation (66%), hand–foot skin reaction (62%), pain (58%) and diarrhoea (58%). The most common grade 3 or 4 adverse event was hypertension, which was observed in 31% of patients. Nine patients discontinued drug treatment as a result of unacceptable toxicity. There were no adverse event-related deaths in the trial.

A similar adverse event profile is reported in both trials,111,112 although these are described as ‘selected treatment-related adverse events’ and full information on all adverse events experienced within the trials is not available. The most commonly reported adverse events were fatigue (38%), diarrhoea (24%), nausea (19%), dyspepsia (19%) and stomatitis (16%) in one trial112 and fatigue (28%), diarrhoea (20%), dyspepsia (16%), hypertension (16%) and hand–foot syndrome (15%) in the other. 111

Decline in ejection fraction was also observed in both trials [eight patients (4.7%);111 seven patients (11%)112], although it is unclear whether this represents incidental observation or the results of active monitoring. The decline was sufficient to warrant removal of four patients from the study. 112 One trial111 reports a total of 31 deaths, 10 of these within 28 days of the last dose of sunitinib; one of these deaths (myocardial infarction) was considered to be possibly related to the study medication.

From the data reported in these trials, treatment with sorafenib appears to be associated with an increased frequency of hypertension, hand–foot skin reaction and some gastrointestinal events such as diarrhoea. Although some of the events were classed as grade 3 (severe and undesirable) and grade 4 (life-threatening or disabling), events of this severity occurred in a small proportion of patients (e.g. 4% and 6% for hypertension and hand–foot skin reaction, respectively, in the TARGET trial). Grade 3 hypertension is defined as needing more than one drug for treatment or more intensive treatment than used previously; hypertension with life-threatening consequences (e.g. hypertensive crisis) is the definition of grade 4 hypertension.

As randomised clinical trials are not designed to collect data on rare adverse events, we briefly reviewed additional literature obtained from the results of our initial and updated literature searches to identify any further safety concerns.

A systematic review of toxicities associated with sorafenib, sunitinib and temsirolimus in phase I, II and III clinical trials found that between 1% and 16% of patients experienced grade 3 or 4 adverse events. The most commonly reported grade 3 and grade 4 adverse events associated with sorafenib treatment across all trials were lymphopenia (13%), hypophosphataemia (13%), elevated lipase (12%), mucositis (6%) and hand–foot syndrome (6%). 119

In an expanded access trial of sorafenib in the USA and Canada (n = 2488),126 the following adverse events were experienced at a frequency of > 2% in patients receiving sorafenib as first-line treatment (n = 1239): hand–foot skin reaction (7.7%), fatigue (4.7%), hypertension (3.8%), rash or desquamation (5.2%), dehydration (2.9%), diarrhoea (2.6%) and dyspnoea (2.6%). These data suggest an adverse event profile similar to that reported in the phase III trial. 126

We identified a systematic review and meta-analysis of the incidence and risk of hypertension with sorafenib in patients with cancer conducted by Wu and colleagues120 and published in February 2008 in Lancet Oncology. They identified nine studies in which 3567 patients with RCC or other solid tumours had received sorafenib, including the TARGET trial109 and the RDT110 described above. The overall incidence of all-grade hypertension amongst patients receiving sorafenib was 23.4% (95% CI 16.0% to 32.9%) with 5.7% (95% CI 2.5% to 12.6%) of patients experiencing grade 3 or 4 hypertension. The authors estimate the relative risk for all-grade hypertension in patients receiving sorafenib as 6.11 (95% CI 2.44 to 15.32; p < 0.001) using data from two RCTs (n = 1089). As with all meta-analyses this analysis is limited by the quality of the data in the contributing studies. The authors note possible areas of ambiguity in the grading of hypertension and the lack of data on baseline measurement of blood pressure, both of which may have influenced the results. Although a large proportion of the patients included in the analysis were from the expanded access programme where measurement of hypertension may not have been as precise as in laboratory conditions, the relative risk was calculated using only data allowing a comparison between events reported with and without sorafenib treatment.

A similar systematic review and meta-analysis of the incidence and risk of hand–foot skin reaction with sorafenib treatment, also published in 2008,127 found a 33.8% (95% CI 24.5% to 44.7%) incidence of all-grade hand–foot skin reaction in patients treated with sorafenib. The relative risk of developing all-grade hand–foot skin reaction with sorafenib was 6.6 (95% CI 3.7 to 11.7; p < 0.001).

Comparison of the safety profile of sunitinib with that of BSC is not possible from the phase II trials. Sunitinib treatment was most frequently associated with fatigue, diarrhoea, nausea, hypertension and hand–foot skin reaction, although whether these events were as a result of the treatment or of the disease process is unclear. Further discussion of the adverse events associated with sunitinib is provided in the section on bevacizumab plus IFN and sunitinib compared with IFN as first-line therapy.

The industry submission presents two levels of base-case analysis: (1) preplanned interim analysis data and (2) unplanned updated analysis data. We caution that the unplanned updated analysis data include patients who have crossed over from IFN to sunitinib, with potential for confounding in the estimates of treatment effect (HRs). Therefore, this summary refers to findings presented against the preplanned interim analysis. The base-case analysis presents a cost per LYG of £21,116, an estimate of £45,736 per progression-free year gained and an estimate of £28,546 per QALY gained, with results reported indicating that sunitinib increased OS by an additional 0.82 years, increased PFS by 0.38 years and resulted in an additional 0.60 QALYs compared with IFN.

One-way sensitivity analyses are reported against a range of scenarios. The most important factors affecting the incremental cost-effectiveness ratio (ICER) are the health-state utilities (values) assigned to the PFS and PD states, and the shapes of the OS and PFS curves (extrapolation method). The PSA reported that, at a willingness-to-pay threshold of £30,000 per QALY, sunitinib has a 54% probability of being cost-effective compared with IFN.

In the comparison of bevacizumab plus IFN versus IFN the manufacturer’s (Pfizer) submission estimates a cost per LYG and a cost per QALY of £81,754 and £107,357 respectively.

For second-line use of sunitinib compared with BSC the submission estimates (base-case assumptions) a cost per LYG and a cost per QALY of £29,061 and £37,519, respectively, with results reported indicating sunitinib increased OS by 0.77 years and PFS by 0.54 years and resulted in an additional 0.60 QALYs compared with BSC.

Sensitivity analyses reported in the submission indicate that the most important factors affecting the ICER are the health-state utilities (values) assigned to the PFS and PD states, and the shapes of the OS and PFS curves (and data source). The PSA reported that at a willingness-to-pay threshold of £30,000 per QALY, sunitinib has a 36% probability of being cost-effective compared with BSC.

In the comparison of sorafenib versus BSC the manufacturer’s (Pfizer) submission estimates a cost per LYG and a cost per QALY of £54,750 and £73,078 respectively.

Pfizer do not perform an indirect comparison between sunitinib and bevacizumab plus IFN even though they state that the patient populations in the sunitinib versus IFN and bevacizumab plus IFN versus IFN RCTs are similar. Nonetheless, they do present a comparison of the cost-effectiveness of bevacizumab plus IFN versus IFN.

The model considers the cost-effectiveness of bevacizumab plus IFN versus IFN in first-line use, and the submission provides a rationale for not comparing bevacizumab plus IFN versus temsirolimus for poor prognosis patients. 117

Although the model structure is simple, considering PFS, PD and death, this seems appropriate given the decision problem and the data available. The time horizon and model cycle length are appropriate. In the model, PFS is estimated separately for IFN and for bevacizumab plus IFN based on extrapolation of the Kaplan–Meier data from the RCT. 106 The OS data are modelled differently given that they are still immature for bevacizumab plus IFN (RCT-reported data). In the model, the RCT data on OS for IFN alone are used (as OS data in the IFN arm are more mature) to extrapolate and estimate the OS for IFN over time. 106 To model OS for bevacizumab plus IFN the baseline progression (IFN alone) is used in conjunction with the relative measure of effectiveness (HR) reported in the RCT. 106 The submission reports that several mathematical survival curves were fitted to the Kaplan–Meier data and that the Gompertz function is used in the model on the basis that it gave the best fit to both the PFS and OS data.

Drug costs are estimated using list prices and recommended dose data. Roche use the average body weight of 76.5 kg from the RCT by Escudier and colleagues106 to estimate average dose and hence the cost of bevacizumab. Patients in the bevacizumab plus IFN arm received 10 mg/kg of bevacizumab every 2 weeks, and IFN three times per week at a dose of 9 MIU. As noted above, the analysis assumes a European-wide ‘dose cap’ scheme (in which costs for bevacizumab are much reduced, i.e. by a mean of £8900 in base-case analysis). Modelling assumes that IFN is administered by patients, with no additional resource use/cost. The model assumes one outpatient visit for every intravenous administration of bevacizumab (every 2 weeks), at a cost of £233 per visit. 139 For bevacizumab this administration cost is assumed to capture all other monitoring costs. In patients taking IFN alone, one outpatient appointment each month is assumed. The drug-related cost of bevacizumab administration (unit cost) was calculated as a weighted average of chemotherapy administration costs from NHS reference cost data. 139 Costs for adverse events are included in the analyses. The manufacturer analysis assumes that patients in the PD health state will be offered second-line drug treatments, such as sunitinib or sorafenib. They assume a cost of £405.50 per month in the bevacizumab plus IFN arm and £495.95 in the IFN arm. These figures are based on data from the RCT by Escudier and colleagues,106 which details second-line treatments. A larger proportion of patients in the IFN arm received second-line treatment than in the bevacizumab plus IFN arm, with differences attributed to the relative lack of effectiveness of IFN. Specifically, the monthly costs of the second-line drug were estimated based on second-line drug use for 8.3 months, the duration of second-line PFS according to the second-line sunitinib trial for RCC patients. 111 The total expected drug cost in PD was thus calculated and then the monthly cost in PD was estimated by spreading this total cost over the time spent in PD in the model (12.7 months). In addition, Roche assumes that all patients in PD had one outpatient appointment per month for monitoring.

As noted above, the health-state utilities (for PFS and PD health states) are taken from the sunitinib versus IFN RCT,107 which used the EQ-5D measure. As noted in the PenTAG review of the sunitinib model (manufacturer submission) these utility data are not published and we are unable to consider them in much detail.

The submission presents findings from probabilistic modelling to address parameter uncertainty in cost per QALY estimates, but other sensitivity analyses are performed only on model structure, reporting against the choice of mathematical function of the survival curves (see below). PenTAG note that the absence of sensitivity analysis is a weakness in the reporting of the cost-effectiveness analysis and suggest that the submission could have performed/reported additional sensitivity analysis to help assess the uncertainty in results.

We have a number of concerns with the model and analysis presented in the Roche submission to NICE.

First, we highlight a concern over the assumptions and data used to estimate dose intensity, which is used to adjust drug and drug-related costs. The manufacturer analysis multiplies the costs of drug acquisition and drug administration using dose intensity data (unpublished data from the bevacizumab plus IFN RCT106). In the model, dose intensity data for bevacizumab is estimated using the average time taking the drug in the trial divided by the average time patients spend in PFS in the model. Similarly for IFN, the estimation is the average time actually taking the drug in the trial divided by the average time patients spend in PFS up to 1 year in the model. In this way the dose intensities are calculated as 62% for bevacizumab, 80% for IFN when used with bevacizumab and 63% for IFN alone (monotherapy). Although these data are not reported in the written submission they are used in the model. These data, applied to adjust costs, are different to those reported in the RCT106 and different to the data quoted in the Roche submission117 (Table 13 in the submission). Dose intensity data used in the model are generally much lower than the mean dose intensities reported in the RCT (88% for bevacizumab, 83% for IFN in bevacizumab plus IFN arm and 89% for IFN alone arm)106 and lower than the median dose intensities quoted in the manufacturer submission. 117 When PenTAG have used the dose intensity data reported in the published RCT106 in the manufacturer model the base-case ICER increases substantially, from £75,000 per QALY to £117,000 per QALY.

Second, we highlight a concern over the clinical effectiveness data (HRs) used in the manufacturer analysis for OS and PFS. The analysis uses the HR for OS from unpublished data on what is classed a ‘safety population’ (not the RCT data), using an OS HR of 0.709. This differs from the OS HR of 0.75 from the RCT reported by Escudier and colleagues. 106 When PenTAG have used the manufacturer model and applied the RCT HR for OS of 0.75 the ICER increases from £75,000 per QALY to £87,400 per QALY. It is not clear why the manufacturer analysis uses data from the safety population. Again, we note that the model uses a HR of 0.609 (95% CI 0.508 to 0.728) for PFS and that this is from a safety population (stratified by risk group) rather than from the data reported in the RCT. The RCT106 reports a HR of 0.63 (95% CI 0.52 to 0.75) in unstratified analysis. However, in the model, a PFS HR is not explicitly applied, because PFS for both treatment arms is fitted to empirical trial data independently (we assume that this HR is implicit in the Kaplan–Meier data).

Also, in sensitivity analysis the submission reports findings in which cost-effectiveness has been assessed using a log-logistic model (instead of the Gompertz methods in the base-case analysis), and PenTAG would question the appropriateness and prominence of this sensitivity analysis. In this case the ICER falls greatly, from £75,000 per QALY to £40,000 per QALY, and, at a willingness-to-pay threshold of £30,000 per QALY, bevacizumab plus IFN has a 9% probability of being cost-effective compared with IFN. However, Roche acknowledge that this ICER may be unrealistic because the log-logistic model results in an expected lifetime that may be unrealistically long (Figure 7). We do not see the log-logistic method as a credible approach, that is, we agree with Roche that it is unreasonable to use the log-logistic distribution to model PFS and OS in the sensitivity analysis because the tail of the distribution is too long.

FIGURE 7.

Comparison of the fit to overall data for IFN, using Gompertz and log-logistic curves (as manufacturer sensitivity analysis). Source: Escudier et al. 106

The model uses three primary health states (PFS, post progression, and death), which is similar to the other models presented for RCC. The model structure appears appropriate given the decision problem and data available. The time horizon is short at 3 years but appears to capture the main impacts of disease and treatment, although it has not been tested in sensitivity analysis. The cycle length is appropriate. The model is based on a set of time-dependent transit probabilities derived from individual patient-level data (not available to PenTAG) from the RCT by Hudes and colleagues. 108 We are unable to consider the derivation of these probabilities in any detail. Three Weibull functions are modelled: (1) PFS to post progression, (2) PFS to death and (3) post progression to death. These functions are used to derive the transition probabilities. For subgroup analysis, the PFS and OS Weibull curves are unique for each patient subgroup: clear cell, non-clear cell, nephrectomy, non-nephrectomy. A discussion on the effectiveness data available to model subgroups can be found in Chapter 2.

In the model, patients start in a PFS health state. The model assumes that patients starting in a PFS state are treated with IFN or temsirolimus and stop treatment when they enter a post-progression health state. After disease progression (post progression), patients take second-line drugs (sunitinib, sorafenib, bevacizumab) or receive BSC only.

We note/assume that when calculating disease progression (transition probabilities) the model uses effectiveness data from all patients in the RCT reported by Hudes and colleagues108 It is important to remember that the definition of poor prognosis used in this trial differs from the MSKCC prognosis scale. Using this scale, only 75% of patients in this trial would be considered to have poor prognosis. The remaining 25% of patients had intermediate prognosis.

In the analysis undertaken (survival analysis/transition probabilities), the shapes of the PFS and OS curves calculated (from transition probabilities) are noticeably different to the empirical Kaplan–Meier curves reported in the RCT108 (Figures 8 and 9 respectively). For PFS we believe that there are two reasons for this difference. First, the data presented in the RCT are empirical survival data, whereas the manufacturer has modelled PFS using transition probabilities calculated from individual patient data. Second, the manufacturer has used the independent assessment of PFS, whereas in the RCT reported by Hudes and colleagues108 the published Kaplan–Meier curve for PFS was based on the site investigator assessment. In the PenTAG analysis, shown in Figures 8 and 9, PFS and OS are modelled based on the data available from the RCT reported by Hudes and colleagues108 The differences apparent in the OS curves (Figure 9) would appear to be due to the use of individual patient-level data by Wyeth to calculate transition probabilities instead of the use of empirical data reported in the RCT.

FIGURE 8.

Progression-free survival for IFN in Wyeth analysis and from RCT. *Source: Hudes et al. 108 (site investigator assessment of progression-free survival).

FIGURE 9.

Fit to empirical overall survival for IFN by Wyeth and PenTAG. Source: Hudes et al. 108

We have some concerns over a number of the assumptions in the model over resource use and cost. We summarise our main concerns below, covering costs associated with administration of IFN and use of dose intensity data. See Appendix 6 for more detailed comments on data inputs.

The costs associated with the administration of IFN, and the cost differences between IFN administration and temsirolimus administration, are an important component in the cost-effectiveness estimates. The manufacturer model assumes that all IFN is administered in the hospital outpatient setting, costing £127.80 per visit. With IFN administered three times per week this leads to a high cost associated with IFN treatment. Based on information from the expert advisory group we do not believe that this is an accurate reflection of current practice. Based on the clinical opinions received we would expect that in most cases IFN injections would be administered in patients’ homes, either by themselves or by friends, relatives or carers. It may be that in some cases a district nurse or community or practice nurse would give injections (in a patient’s home). When we assume resource use based on the clinical opinion received (i.e. we assume that typically 25% of patients have IFN administered by a district nurse, at a cost of £25 per visit, and the remaining 75% self-inject, at no cost; see Table 40), the base-case ICER (using the manufacturer model) increases substantially from £55,814 per QALY to £102,000 per QALY. In subgroup analyses this pattern is also noted: the cost per QALY for the clear cell subgroup increases from £57,731 to £121,300, the cost per QALY for the non-clear cell subgroup increases from £51,159 to £63,100, the cost per QALY for patients with previous nephrectomy increases from £60,575 to £117,000 and the cost per QALY for patients without previous nephrectomy increases from £49,690 to £84,000.

A further concern is that the Wyeth submission assumes that the drug administration costs for temsirolimus should be adjusted using dose intensity data from the RCT,108 that is, costs are reduced. However, Wyeth do not apply this same assumption to costs associated with IFN.

Drug costs are estimated using list prices (expected list prices) and recommended dose data. Patients receive IFN three times per week at a recommended dose of 18 MIU. In the Wyeth analysis the cost of temsirolimus is based on 25 mg per dose, one dose per week, at £20.60/mg, giving £515 per dose. The analysis acknowledges that because of vial size (30 mg each) there will be waste/overfill of 5 mg. There is some outline discussion on the potential for vial sharing schemes (and sensitivity analysis) but no detail is provided.

The temsirolimus model uses health-state utilities of 0.60 for the baseline entry health state of stable disease (analogous to PFS) and 0.446 for post progression. The model also includes an incremental gain in utility in patients in whom a response (positive) to initial treatment is reported, with a value of 0.658. The PD and response states (utility values) do not play a major part in the cost-effectiveness analyses and so we do not dwell on them here, focusing on the more generic states of stable disease/PFS and post progression. The submission reports that utilities were modelled under the Q-TWiST structure, according to whether patients were in the TOX (toxicity) state (suffering grade 3 or 4 adverse events), PD or TWiST state. The submission states that utility values were derived from EQ-5D data collected during the temsirolimus RCT,108 although limited details are available on this. We have some concerns over the lack of transparency in the data used to derive health-state utilities (see PenTAG cost-effectiveness analysis, Health state utilities).

The submission presents one-way sensitivity analyses. However, we are concerned that the manufacturer has performed no sensitivity analyses on the PFS and OS curves, especially as these are major drivers of the ICER. However, in the PSA the submission does incorporate some variation in these curves.

For the comparison between temsirolimus and BSC the submission uses data from the Medical Research Council Renal Cancer Collaborators (MRCRCC) RCT140 and we have concerns over the use of these data. The data are based on patients with a range of prognoses, not just those with poor prognosis. Therefore, we suggest that the results of the indirect comparison should be treated as suggestive only.

Although the model of disease progression is simple, considering PFS, PD and death, we regard this as appropriate given the decision problem and data available. The time horizon and model cycle length are also regarded as appropriate. As above, the model uses trial data to model disease progression, PFS and OS in the main analysis (combined patient groups). Using PFS and OS data, the time that patients spend in the PD state is calculated from estimated time alive minus time in PFS. As acknowledged by the manufacturer in the submission, data available to model subgroups are not of good quality, and the modelling is undertaken using an adjustment of the baseline disease progression against data on PFS and OS in the subgroups, with a ratio of median PFS in the subgroup to median PFS in all patients used for the adjustment. Although the method is clear there is some uncertainty over the data available on subgroups (PFS and OS), and these data are largely unpublished. Therefore, we are unable to comment further.

Drug costs are estimated using list prices and recommended dose data. In the model patients are on sorafenib treatment whilst in the PFS health state and were assumed to receive 400 mg sorafenib twice daily (costing £2721 per month). Although approximately 6% of patients receiving sorafenib in the RCT by Escudier and colleagues109 had dose reductions, it was conservatively assumed that all patients would receive 400 mg sorafenib.

Resource use within the model was estimated via two internet-based surveys of six and 31 UK clinicians. Four clinicians with experience of sorafenib estimated resource use in the PFS state for sorafenib-treated patients, whereas clinicians who had not used sorafenib estimated resource use in patients receiving BSC in the PFS state. Resource use estimates were weighted by performance status (ECOG score), with an assumption of 35% ECOG-PS 0 and 65% ECOG-PS 1. There are no published data on resource use for RCC and there are limited alternatives to estimate resource use. Although we consider the estimates used to be high in some cases (i.e. higher than the estimated costs in the PenTAG analysis), for example the manufacturer estimate of £673 per month for patients treated with BSC in the PFS health state, it is acknowledged that this is an area where judgements may differ. We urge caution when using data from such surveys in small samples, and such caution also applies to the estimates used in the PenTAG analysis. See Appendix 6 for more detailed comments on cost data inputs.

Health-state utility data were collected from a survey of 31 UK clinicians working in the field of RCC using the EQ-5D questionnaire. EQ-5D values for patients on sorafenib were based on views elicited from only five physicians. We have significant concerns over the methods used here and note that physician valuations (descriptions) are not methodologically robust and are inconsistent with the NICE reference case requirements. Utilities were higher in PFS than in PD and higher for ECOG-PS 0 than for ECOG-PS 1: PFS ECOG-PS 0: 0.903 (95% CI 0.858 to 0.948); PFS ECOG-PS 1: 0.648 (95% CI 0.582 to 0.714); OS ECOG-PS 0: 0.692 (95% CI 0.606 to 0.778); OS ECOG-PS 1: 0.471 (95% CI 0.389 to 0.553). The analysis combining both ECOG values (all patient subgroups combined) used the average utility across both ECOG groups weighted by the proportion of patients in ECOG-PS 0 and ECOG-PS 1. These treatment-independent averages were 0.737 for PFS and 0.548 for PD for both sorafenib and BSC.

The submission presents one-way sensitivity analysis and PSA. There is no statement of model checking for consistency and/or accuracy, although there is a reference to an accurate prediction of median PFS in the TARGET trial. 109

Further detail is provided in Appendix 6; we have no other major concerns with the modelling presented in this submission.