Sunitinib for the treatment of gastrointestinal stromal tumours

In the submission, the manufacturer presented evidence on the clinical effectiveness of sunitinib for the treatment of unresectable and/or metastatic malignant GIST that was within the marketing authorisation for sunitinib and in line with the appraisal scope. The main clinical-effectiveness evidence came from one randomised controlled trial (RCT). The RCT, A6181004, compared the effect of sunitinib plus best supportive care (n=207) with placebo plus best supportive care (n=105). Best supportive care was defined as symptom control, palliative care without active treatment and monitoring of progression. The trial was conducted in people with a good performance status (Eastern Cooperative Oncology Group [ECOG] status 0 or 1). Approximately 4% of people in the trial had initial imatinib intolerance and approximately 80% had received more than 400 mg of imatinib daily before trial entry. People in the trial were stratified according to:

• best outcome of prior imatinib treatment (progressive disease within 6 months of starting imatinib treatment [primary resistance] or progressive disease after 6 months of starting imatinib treatment [secondary resistance] or imatinib intolerance) and

• baseline McGill pain questionnaire score (0 versus 1 or more).

The primary outcome in the study was time to tumour progression, which was defined as the time from the first dose of study drug to first documentation of progressive disease. The blinded phase of the RCT was stopped early when the first planned interim analysis demonstrated that the time to tumour progression in people randomised to receive sunitinib was statistically significantly longer than in people randomised to receive placebo. A total of 84% of people randomised to receive placebo plus best supportive care crossed over and received sunitinib plus best supportive care. Median time to tumour progression in the intention-to-treat (ITT) population was 27.3 weeks in the sunitinib arm and 6.4 weeks in the placebo arm (hazard ratio [HR] 0.33; 95% confidence interval [CI] 0.23 to 0.47, p<0.0001). The median time to tumour progression for the group that crossed over from placebo to sunitinib was similar to that for the group originally randomised to receive sunitinib.

During the blinded phase of the study, more than half of the people in both study arms of the trial were alive. However, the interim ITT analysis showed that overall survival was significantly longer for those who received sunitinib compared with those who received placebo (HR 0.491; 95% CI 0.290 to 0.831, p=0.007). The ITT analysis of the entire study (that is, blinded plus open-label phase) showed that there was no statistically significant difference in overall survival for people who received sunitinib plus best supportive care (overall survival 73 weeks) compared with people who received placebo plus best supportive care (overall survival 65 weeks) (HR 0.876; 95% CI 0.679 to 1.129, p=0.306).

The manufacturer also presented analyses of overall survival using a rank preserved structural failure time (RPSFT) model. This model was a 'post-hoc' approach taken by the manufacturer to control for the crossover from the placebo arm to the sunitinib arm. The RPSFT method estimated the overall survival of people randomised to receive placebo assuming that they had not crossed over; that is, as if they had remained on placebo for the duration of the trial. This method was therefore based on a comparison of the groups according to the way they were randomised. The RPSFT method proportionally 'shrinks' the estimated amount of additional survival conferred to people who crossed over to receive sunitinib, thereby changing the estimate of the hazard ratio in the ITT analysis used later in the economic analyses.

The initial RPSFT analysis suggested a statistically significantly longer overall survival for people who received sunitinib plus best supportive care (overall survival 73 weeks) compared with those who received placebo plus best supportive care (overall survival 39 weeks) (HR 0.505; 95% CI 0.388 to 0.658, p<0.0001) for the entire study. However, following a recommendation from an independent statistician, the manufacturer revised the 95% confidence interval associated with the hazard ratio derived from the RPSFT approach. The revised RPSFT method did not change the significance, which remained that of the unadjusted ITT analysis (p=0.306). It did, however, provide a lower estimate of the hazard ratio adjusted for crossover and a revised 95% confidence interval for the hazard ratio which, naturally, included one (that is, unity). The revised 95% confidence interval was 0.262 to 1.134.

Quality of life was measured in the RCT using the European quality of life (EuroQoL) health state questionnaire (EQ-5D). More than 75% of people completed the EQ-5D questionnaire at each time point and there were no statistically significant differences reported between the treatment groups. Treatment-related adverse events and serious adverse events were more common in the sunitinib arm than in the placebo arm. A total of 83% of people in the sunitinib arm and 59% of people in the placebo arm experienced treatment-related adverse events of any severity. The manufacturer stated that the adverse events reported were generally of mild to moderate intensity and were easily managed by dose reduction, dose interruption or standard supportive medical treatments. A total of 9% of people randomised to sunitinib and 8% of people randomised to receive placebo discontinued treatment because of adverse events.

The manufacturer also provided details of an ongoing, open-label expanded access programme (EAP). This cohort study (A6181036) was set up to allow people with GIST, who might not have access to the drug because of study inclusion criteria or lack of regulatory approval where they live, to receive sunitinib. As of December 2007, 1,126 people were enrolled in the EAP and the ITT population comprised 1,117 people. People in the EAP were of ECOG performance status 0 to 4 and 68% of people had received dosages of more than 400 mg of imatinib daily before joining the study. In the EAP, sunitinib treatment was given for as long as there was evidence of disease control according to the investigator. The EAP is scheduled to end in December 2009. At the time of data analysis, 50% of the ITT population were alive. The median time to tumour progression was 41 weeks (95% CI 36 to 47) and the median overall survival was 75 weeks (95% CI 68 to 84).

The manufacturer developed a Markov model to assess the cost effectiveness of sunitinib compared with best supportive care in people with unresectable and/or metastatic malignant GIST after failure of imatinib therapy because of resistance or intolerance. The model had 3 distinct health states: progression free, progressive disease (no active therapy) and death. All people entered the progression-free state of the model, assuming that imatinib therapy had failed. The model had a cycle length of 6 weeks and the time horizon was 6 years; the manufacturer stated that this reflected the maximum life expectancy of the population in the model. No subgroup analyses were conducted by the manufacturer.

The model used effectiveness data from the RCT (A6181004) described in section 3.1. For progression-free survival, Weibull curves were fitted to the ITT data from the placebo plus best supportive care and sunitinib plus best supportive care arms independently. For overall survival, Weibull curves were fitted to the ITT data from the sunitinib plus best supportive care arm and to the RPSFT-adjusted data from the placebo plus best supportive care arm independently. In a sensitivity analysis, the manufacturer fitted a Weibull curve to the unadjusted ITT data for overall survival with placebo plus best supportive care.

The utility values used in the model were taken from the EQ-5D questionnaire used in the RCT. In the progression-free health state, a utility value of 0.731 was assigned to people receiving sunitinib plus best supportive care and a utility value of 0.781 was assigned to people receiving placebo plus best supportive care. In the progressive disease health state, a utility value of 0.577 was assigned to both arms. The manufacturer did not model the effect of adverse events on utility, and stated that the reduced utility values assigned to the sunitinib plus best supportive care arm would account for disutility from adverse events.

Resource use was not measured directly in the RCT, although the drug use and relative dose intensity estimates were derived from the RCT. In the model, the manufacturer assumed a relative dose intensity of 88.6% for sunitinib and cost data were taken from the BNF 56. The manufacturer had agreed a patient access scheme with the Department of Health, in which the first cycle of sunitinib is free to the NHS.

With discounting at 3.5% per year, sunitinib compared with best supportive care produced a base-case incremental cost-effectiveness ratio (ICER) of £27,365 per QALY gained. One-way sensitivity analyses demonstrated that the ICER was most sensitive to the source of overall survival data for the best supportive care arm. When the ITT data were used to model the placebo plus best supportive care overall survival curve, the ICER increased to £77,107 per QALY gained. Probabilistic sensitivity analyses suggested that sunitinib had a 50% probability of being cost effective compared with best supportive care at a willingness-to-pay threshold of £30,000 per QALY gained.

The ERG stated that the manufacturer's submission was generally of good quality and appropriate to the decision problem. Although the clinical-effectiveness evidence was derived from only 1 RCT, this RCT was of good quality and demonstrated that sunitinib plus best supportive care significantly improved time to tumour progression compared with placebo plus best supportive care. The ERG stated that the economic model developed by the manufacturer was appropriate for the decision problem and appeared to contain no logical errors or internal inconsistencies.

The ERG highlighted the following key areas of concern with the manufacturer's submission:

• the use of the RPSFT method

• the uncertainty surrounding the cost-effectiveness estimate provided by the manufacturer

• the cost of sunitinib for people who continued to receive it after disease progression.

The ERG stated that the RPSFT method used by the manufacturer to control for the effects of crossover was uncommon and could not determine whether the method had been applied correctly. The ERG also highlighted a number of errors and omissions in the probabilistic sensitivity analyses. In particular, the ERG stated that the uncertainties in the progression-free and overall survival states were not modelled fully, as only the certainty of fit of the Weibull curves was assessed in sensitivity analyses for these parameters. The ERG also noted that the manufacturer had used standard deviations rather than standard errors for the utility values in the sensitivity analyses. The ERG was concerned, given the wide confidence interval around the estimate for the overall survival hazard ratio using the RPSFT method, that the uncertainty in the base-case ICER was substantial and likely to be higher than that presented by the manufacturer in the probabilistic sensitivity analyses. The ERG also highlighted a number of other, more minor, errors and omissions in the manufacturer's probabilistic sensitivity analyses.

The ERG noted that the economic model developed by the manufacturer assumed that sunitinib was only given until disease progression. In the RCT, from which the effectiveness of sunitinib was derived, 54 people (22%) received sunitinib after disease progression. The additional cost of sunitinib for these people, as estimated by the ERG, was £2,237. This increased the base-case ICER from £27,400 to £31,800 per QALY gained. When the additional costs of sunitinib were incorporated into the sensitivity analyses that used the ITT data, the ICER increased from £77,100 to £90,500 per QALY gained. The ERG also noted that median progression-free survival in the RCT (and thus the progression-free survival used in the manufacturer's economic model) and the EAP differed markedly. When the ERG increased the median progression-free survival to equal that of the EAP (41 weeks), the ICER increased the base-case from £27,400 to £46,300 per QALY gained. The ERG noted that the ICER increased substantially when the progression-free survival was taken from the EAP. It stated that this was because people who experienced longer progression-free survival received more sunitinib, which increased the acquisition costs.

After the first appraisal committee meeting, the manufacturer presented updated cost-effectiveness analyses incorporating the sunitinib costs after disease progression, as requested by the committee. The manufacturer confirmed that in the base-case cost-effectiveness estimate, only sunitinib costs incurred in the progression-free health state were taken into account, but that in the RCT a total of 22% of participants randomised to the sunitinib arm did receive sunitinib after disease progression. The manufacturer highlighted that there was insufficient evidence to know whether people with unresectable and/or metastatic malignant GIST would continue to receive sunitinib after disease progression in clinical practice. The manufacturer accepted the ERG's estimate of the additional sunitinib costs incurred after disease progression, and agreed that including these costs increased the base-case ICER from £27,365 to £31,817 per QALY gained.

The manufacturer also presented an updated cost-effectiveness analysis incorporating the sunitinib costs based on the sunitinib treatment duration in the EAP, as requested by the committee. The manufacturer stated that the main value of the EAP was that it confirmed the safety and efficacy of sunitinib as demonstrated in the RCT. The manufacturer highlighted that the rationale for the difference in treatment duration between the EAP and RCT was unclear, pointing to the following factors:

• The people in the EAP had not been followed up for as long as the people in the RCT.

• Tumour measurements in the EAP were performed according to local standards of care, and treatment was continued for as long as there was evidence of disease control, unlike the study protocol for the RCT.

• The EAP included participants who were ineligible for the RCT.
  
  The manufacturer's updated cost-effectiveness estimates incorporating the sunitinib treatment duration and costs from the EAP differed slightly from the ERG and increased the base-case ICER from £27,365 to £47,628 per QALY gained.

The manufacturer clarified the RPSFT method used to control for crossover and presented updated cost-effectiveness estimates using censoring to control for crossover, as requested by the committee. The manufacturer restated that the RPSFT method was appropriate as it preserved the randomisation of the trial. However, as the RPSFT method was based on randomisation, it did not change the level of evidence against the null hypothesis and so the 95% confidence interval around the revised RPSFT hazard ratio was wide. The manufacturer also stated that the methods used had been corroborated by an independent statistical expert. The manufacturer highlighted that because of the high level of crossover, which occurred very early in the trial, and the fact that crossover was informative (that is, participants who did not crossover were likely to be different from those who did crossover), traditional approaches to account for crossover were inappropriate. The manufacturer highlighted that censoring the participants at the point at which they crossed over was unreliable because there were only 15 participants who did not crossover to receive sunitinib. Censoring the participants who crossed over to receive sunitinib resulted in an ICER in which best supportive care dominated sunitinib. An additional analysis incorporating data from the 15 participants who did not crossover to sunitinib resulted in an ICER of £20,618 per QALY gained.

The manufacturer also presented updated probabilistic sensitivity analyses on all scenarios; these corrected for the errors and omissions that were identified by the ERG. At willingness-to-pay thresholds of £20,000 and £30,000 per QALY gained, the manufacturer's base-case ICER of £27,365 had a 17% and 57% probability of being cost effective, respectively. Incorporating the costs of sunitinib incurred after disease progression from the RCT resulted in a 7% and 42% probability of it being cost effective at willingness-to-pay thresholds of £20,000 and £30,000 per QALY gained, respectively. All the other updated cost-effectiveness analyses (except the £20,618) had a 0% probability of being cost effective at willingness-to-pay thresholds of £20,000 and £30,000 per QALY gained.

The ERG considered the updated analyses and clarification provided by the manufacturer. It agreed that the RPSFT method appeared appropriate. The ERG confirmed that the RPSFT method applied a multiplicative factor to the time spent after crossover rather than to the whole of overall survival. The ERG highlighted that the hazard ratio for overall survival produced using the RPSFT method (0.505) was similar to the hazard ratio for overall survival produced at the interim ITT analyses before crossover had occurred (0.49). It stated that this strengthened the confidence it had in the results derived using the RPSFT method. Additionally, the ERG agreed with the manufacturer that censoring the participants at crossover in this instance was an unreliable method for controlling for crossover. The ERG also noted that the first 4 months of the overall survival curve for people who received best supportive care, who were then censored at the point at which they crossed over, was similar to the RPSFT overall survival curve. The ERG stated that this gave further credibility to the results derived using the RPSFT method because there would have been minimal censoring during the first 4 months.

Full details of all the evidence are in the manufacturer's submission and the ERG report.