3 The company's submission
The Appraisal Committee (section 6) considered evidence submitted by Eli Lilly and Company and a review of this submission by the Evidence Review Group (ERG; section 7). See the Committee papers for full details of the evidence.
Clinical effectiveness
RAINBOW (Ramucirumab combination therapy)
3.1 RAINBOW was a global, randomised, placebo‑controlled, double‑blind, phase 3 study in which ramucirumab plus paclitaxel was compared with placebo plus paclitaxel. The study, which started in 2010, recruited adults with advanced gastric cancer or gastro–oesophageal junction adenocarcinoma who had disease progression on or within 4 months after treatment with platinum‑containing and fluoropyrimidine‑containing chemotherapeutic regimens with or without an anthracycline.
3.2 The trial randomised 665 adults to have either ramucirumab 8 mg/kg (n=330) or placebo (n=335) intravenously on days 1 and 15, plus paclitaxel 80 mg/m2 intravenously on days 1, 8 and 15, over a 28‑day cycle. Randomisation was stratified according to geographic region (region 1 was Europe, Israel, USA and Australia; region 2 was Argentina, Brazil, Chile and Mexico; and region 3 was Hong Kong, Japan, South Korea, Singapore and Taiwan), time to progression from the start of first‑line therapy, and disease measurability. The study was carried out in 170 centres in 27 countries in North and South America, Europe, Asia and Australia.
3.3 People in the trial had metastatic or non‑resectable locally advanced disease, an Eastern Cooperative Oncology Group (ECOG) performance status score of 0 or 1, and adequate haematologic, hepatic, coagulation and renal function. People with squamous cell or undifferentiated gastric cancer were excluded from the trial. People who previously had any chemotherapy other than platinum and fluoropyrimidine, with or without an anthracycline, were also excluded from the trial. Prior treatment with trastuzumab (which has a marketing authorisation in combination with capecitabine or 5‑fluorouracil and cisplatin for HER2‑positive metastatic adenocarcinoma of the stomach or gastro–oesophageal junction, and which has been recommended in NICE's technology appraisal guidance on trastuzumab for the treatment of HER2-positive metastatic gastric cancer as a first‑line treatment) was permitted.
3.4 At baseline, most characteristics were balanced between the treatment groups in RAINBOW. These characteristics included: age; sex; ethnic origin; geographic region; disease measurability; time to progression from the start of first‑line therapy; weight loss in the preceding 3 months; and presence and location of the primary tumour. There was a difference between the treatment groups in ECOG performance status: 35% of people in the ramucirumab plus paclitaxel arm had an ECOG performance score of 0 compared with 43% in the placebo plus paclitaxel arm. A high proportion of people in RAINBOW were male (71%) and most were white (61% white, 35% Asian, 4% black and other). Most people (79%) had gastric cancer and those remaining had gastro–oesophageal junction adenocarcinoma. Previous trastuzumab therapy was had by 20 people in the ramucirumab plus paclitaxel arm compared with 19 people in the placebo plus paclitaxel arm.
3.5 The primary endpoint of RAINBOW was overall survival. Primary and secondary endpoints were analysed using the intention‑to‑treat population (that is, the full population of 665 people who were randomised to the trial). At the date of data cut‑off (12 July 2013), 256 (77.6%) people had died in the ramucirumab plus paclitaxel arm compared with 260 (77.6%) in the placebo plus paclitaxel arm. The data for people who had not died (22.4%) were censored on the last date that the person was known to be alive (on or before the data cut‑off date or lost to follow‑up). Median overall survival was 9.63 months for ramucirumab plus paclitaxel and 7.36 months for placebo plus paclitaxel (2.27‑month improvement in survival; hazard ratio [HR] 0.81; 95% confidence interval [CI] 0.68 to 0.96; p=0.0169). Median progression‑free survival was 4.40 months for ramucirumab plus paclitaxel and 2.86 months for placebo plus paclitaxel (1.54‑month improvement in progression‑free survival; HR 0.64; 95% CI 0.54 to 0.75; p=0.0001).
3.6 Quality of life was assessed in RAINBOW using the European Organisation for Research and Treatment of Cancer (EORTC) QLQ‑C30 (global health status, functioning and symptoms) instrument. Ramucirumab plus paclitaxel was associated with improved outcomes for 14 of the 15 symptom scales compared with placebo plus paclitaxel, although statistical significance was only reached in 2 of the symptom scales: emotional function and nausea and vomiting.
3.7 In RAINBOW, a similar percentage of people in both study arms stopped treatment because of adverse events (11.8% in the ramucirumab plus paclitaxel arm and 11.3% in the placebo plus paclitaxel arm). The most frequently reported treatment‑emergent adverse event was neutropenia, which had a higher all‑grade incidence in the ramucirumab plus paclitaxel arm (54.4%) than the placebo plus paclitaxel arm (31.0%).
Geographic region subgroup results
3.8 The company presented a pre‑specified analysis according to geographic region for RAINBOW. The proportion of people in each geographic region was:
-
60% from region 1 (Europe, Israel, USA and Australia)
-
7% from region 2 (Argentina, Brazil, Chile and Mexico)
-
33% from region 3 (Hong Kong, Japan, South Korea, Singapore and Taiwan).
3.9 The company stated that region 1 had characteristics most representative of patients in England. The company presented the outcomes for region 1 showing that in this subgroup there was a 2.66‑month greater median overall survival (p=0.0050), and 1.41‑month greater median progression‑free survival (p<0.0001) for ramucirumab plus paclitaxel compared with placebo plus paclitaxel. The median survival times for both treatment arms in the intention‑to‑treat population of RAINBOW were longer compared with those for region 1, which the company attributed to the higher rates of third- and fourth‑line chemotherapy use among people in region 3 (Asia) after stopping treatment with ramucirumab.
REGARD (ramucirumab monotherapy)
3.10 REGARD was an international, randomised, double‑blind, placebo‑controlled, phase 3 trial in which ramucirumab plus best supportive care was compared with placebo plus best supportive care. The study, which started in 2009, involved adults with advanced gastric cancer or gastro–oesophageal junction adenocarcinoma who had disease progression on or within 4 months after the last dose of treatment with first‑line, platinum‑containing or fluoropyrimidine‑containing chemotherapy, or on or within 6 months after the last dose of adjuvant therapy.
3.11 The trial randomised 355 adults in a 2:1 ratio to have ramucirumab 8 mg/kg (n=238) or placebo (n=117) intravenously once every 2 weeks. Treatment was given until there was evidence of progressive disease or unacceptable toxicity. Randomisation was stratified by geographic region, weight loss over the previous 3 months, and location of the primary tumour (gastric or gastro–oesophageal junction). The study was done across 119 centres in 29 countries in North, Central and South America, Europe, Asia, Australia and Africa.
3.12 People in the trial had metastatic disease or locally recurrent, unresectable disease, a life expectancy of 12 weeks or more, and an ECOG performance status score of 0 or 1.
3.13 The primary endpoint was overall survival. Efficacy analysis was by intention to treat. Median overall survival was 5.2 months for ramucirumab plus best supportive care and 3.8 months for placebo plus best supportive care (1.4‑month improvement in median survival; HR 0.78; 95% CI 0.60 to 1.0; p=0.047). Median progression‑free survival was 2.1 months for ramucirumab plus best supportive care and 1.3 months for placebo plus best supportive care (0.8‑month improvement in median progression-free survival; HR 0.48; 95% CI 0.38 to 0.62; p<0.0001).
3.14 Health-related quality of life in the REGARD trial was assessed using the EORTC‑QLQ‑C30 instrument. At 6 weeks, the proportion of patients with improved or stable quality of life was higher for the ramucirumab arm (34.1%) than the placebo arm (13.7%); but the difference between those people for whom quality‑of‑life data were available was not statistically significant (p=0.23).
3.15 Overall safety results for the REGARD trial showed similar percentages of people in each group had at least 1 serious adverse event: 45% in the ramucirumab group compared with 44% in the placebo group. There was a greater proportion of people who stopped treatment in the ramucirumab group (10.5%) compared with the placebo group (6.0%).
Network meta-analysis
3.16 The company carried out a network meta‑analysis to compare ramucirumab plus paclitaxel with best supportive care and docetaxel. The company identified 23 trials for inclusion in the network, but only 5 trials were included in the analyses of overall survival in the original company submission. The meta‑analysis incorporated evidence for ramucirumab plus paclitaxel (RAINBOW), docetaxel (COUGAR‑02), irinotecan (Hironaka et al. 2013; Roy et al. 2013; Thuss‑Patience et al. 2011), paclitaxel (RAINBOW; Hironaka et al. 2013) and placebo or best supportive care (COUGAR‑02; Thuss‑Patience et al. 2011). Roy et al. (2013) was a non‑randomised multinational study comparing second‑line irinotecan with docetaxel. The company did not include the Roy et al. study to estimate overall survival in the base case, but it did include this study in a sensitivity analysis. It was not the company's preference to include FOLFIRI (a regimen made up of folinic acid, irinotecan and fluorouracil) in the network, but in the company's response to clarification it incorporated FOLFIRI using the trial by Sym et al. (2011).
3.17 Results from the indirect comparison suggested that ramucirumab plus paclitaxel was associated with a statistically significantly improved overall survival compared with best supportive care (HR 0.34; 95% CI 0.17 to 0.71), paclitaxel (HR 0.81; 95% CI 0.68 to 0.96) and irinotecan (HR 0.72; 95% CI 0.52 to 0.99), and with a numerically (but not statistically significant) improved overall survival compared with docetaxel (HR 0.51; 95% CI 0.23 to 1.13) and FOLFIRI (HR 0.86; 95% CI 0.45 to 1.65).
Cost effectiveness
3.18 The company submitted 2 separate 3‑state partitioned survival models to assess the cost effectiveness of ramucirumab as monotherapy and in combination with paclitaxel; the structures of both models were the same. The 3 states included pre‑progression, post‑progression and death, with all patients entering in the pre‑progression health state. The models used a cycle length of 1 week, and a half‑cycle correction was applied to all calculations. A lifetime horizon was used in both models (equating to about 7 years). Both costs and benefits were discounted at a rate of 3.5%. In the post‑progression health state, the company stated that a minority of patients had a third‑line treatment (47.9%, 46%, 30.3% and 37.6% in the ramucirumab plus paclitaxel, paclitaxel alone, ramucirumab monotherapy and best supportive care plus placebo arms of the trials respectively). One‑way sensitivity analyses were used to explore the uncertainty around utility values, survival analysis, unit costs, choice of third‑line therapy and various resource‑use assumptions. Probabilistic sensitivity analysis was also used to explore parameter uncertainty in the model.
3.19 The primary comparator in the combination‑therapy model was best supportive care; docetaxel was also included in the model because the company stated that the COUGAR study was UK based and was important in shaping clinical practice. Paclitaxel was also included in the combination‑therapy model, although only as a means of validating clinical evidence. The company stated that, based on the results of its survey of UK treatment patterns, irinotecan and FOLFIRI were not used sufficiently in clinical practice to warrant their inclusion in the economic model.
3.20 The comparator included in the monotherapy economic model was best supportive care, which the company justified by noting the licence for ramucirumab that specifies its use for patients in whom treatment with paclitaxel is not appropriate. The company claimed that in clinical practice, people who are not eligible for paclitaxel can be more broadly characterised as not eligible for cytotoxic chemotherapy.
3.21 Transition probabilities between the health states were determined from parametric survival functions fitted to the data from the RAINBOW (combination therapy) and REGARD (monotherapy) trials. Time in the pre‑progression state was estimated directly from the progression‑free‑survival curves, and time in the post‑progression state was estimated from the difference between the progression‑free‑survival and the overall‑survival curves at each time point. Transition probabilities for the comparators docetaxel and best supportive care were estimated using results from the network meta‑analysis.
3.22 For the combination‑therapy model, the company modelled overall survival using Kaplan–Meier data from the RAINBOW trial until the end of the trial period and then extrapolated with an exponential distribution from 22.14 months (the point at which the last event was seen in the placebo plus paclitaxel arm) to 53.5 months (the time point at which survival was 0.1% in the placebo plus paclitaxel arm when extrapolated using the Weibull distribution). The company stated that this represented the most conservative approach and used the trial data to the fullest extent. The overall‑survival data were relatively mature with survival in both arms of about 10% by the end of follow‑up. The company did not use the Kaplan–Meier data from the trial for progression‑free survival because the 6‑weekly assessments caused a stepped curve, so parametric curves were used to incorporate the interval censoring. The Weibull distribution was chosen and the company stated that this provided a more plausible fit to the trial data. Progression‑free‑survival data were mature with less than 4% whose disease had not progressed at the end of the trial in both arms. The hazard ratio estimates from the network meta-analysis for best supportive care and docetaxel compared with ramucirumab plus paclitaxel were applied to the baseline curves for ramucirumab plus paclitaxel.
3.23 In the monotherapy model, the company used the gamma distribution to model overall survival and the interval‑censored log‑normal distribution to model progression‑free survival. Log‑normal for progression‑free survival was stated to be better than other distributions because of the shorter tails of extrapolation, which made it more conservative.
3.24 Utility values for the pre‑progression and the post‑progression health states were taken from EQ‑5D data from the RAINBOW trial. The company stated that for the monotherapy model it used EQ‑5D data from the RAINBOW trial because utility data in the REGARD trial were only collected with the EORTC‑QLQ‑C‑30 instrument, which would need to be mapped to the EQ‑5D. In addition, the company stated that this would need data to be inputted, because there were insufficient post‑baseline data available as a result of the rapid disease progression in both arms.
3.25 Baseline utility values were adjusted with utility decrements applied for treatment‑related adverse events in both the combination‑therapy model and monotherapy model. The types of adverse events included in the models were based on those that were grade 3 and 4 and occurred in more than 5% of people for each relevant trial. The values of the utility decrements were taken from the literature. A utility increment was applied in the combination‑therapy model to the proportion of people whose disease responded to ramucirumab plus paclitaxel in the RAINBOW trial (27.9%). The company assumed that the response rate for docetaxel was the same as that seen for placebo plus paclitaxel in the RAINBOW trial (16.1%). No response rate was applied to best supportive care in the combination‑therapy model because the response rate seen in REGARD was very low (2.6%). In addition, no utility increments were applied to people whose disease responded in the monotherapy model because of the low response rates seen in the REGARD trial.
3.26 The costs of the intervention and comparators included the drug acquisition, administration and monitoring costs as well as the costs of tests. Any leftover drug in opened vials was assumed to be discarded (no vial sharing). Drug acquisition costs depended on the cost of the drug, average dose needed, treatment duration, relative dose intensity and pre‑medication needed. The costs of available generic chemotherapies were taken from the electronic market information tool (eMIT), which uses the actual price paid by hospitals over the last 12 months. The drug dosages for each regimen were based on estimates of body weight and body surface area, which were taken from the baseline patient characteristics of the RAINBOW and REGARD studies for use in the combination‑therapy and monotherapy models respectively. Treatment duration was estimated using parametric curves to determine the time on treatment from trial data. The trials confirmed progression by radiological assessment and patients in the trial (and therefore also in the model) were assessed every 6 weeks. Time on treatment for docetaxel was taken from the literature. Rates for the tests and monitoring were based on expert clinical input. The cost components of best supportive care were identified from a review of hospital medical records.
3.27 Costs further consisted of follow‑up, adverse event, hospitalisation, third‑line therapy (including drug acquisition, administration and follow‑up care) and terminal care costs. The company included hospitalisation costs taken from trial data as well as adverse events, because it stated that people may be admitted to hospital because of factors other than treatment‑related adverse events. Costs of adverse events were included in the models based on their incidence and impact. Grade 3 or 4 adverse events with an incidence of 5% or more and adverse events that had a significant impact on cost- and health‑related quality of life were determinants for inclusion in the model.
3.28 The base‑case incremental cost‑effectiveness ratio (ICER) for the combination‑therapy model was £118,209 per quality‑adjusted life year (QALY) gained for ramucirumab plus paclitaxel compared with best supportive care. The company estimated a probabilistic ICER from the combination-therapy model of £116,820 per QALY gained for ramucirumab plus paclitaxel compared with best supportive care. The deterministic sensitivity analysis of the combination‑therapy model showed that the ICER was most sensitive to the source of drug prices (eMIT compared with BNF), length of hospital stay, dose intensity and the body surface area or body weight source data (all trial patients compared with region 1 trial patients).
3.29 The base‑case ICER for the monotherapy model was £188,640 per QALY gained for best supportive care compared with ramucirumab monotherapy. The probabilistic ICER from the monotherapy model was £189,232 per QALY gained for ramucirumab compared with best supportive care. The deterministic sensitivity analysis of the monotherapy model showed that the ICER was most sensitive to the hospital admission rates, length of hospital stay, assumptions on waste (vial waste compared with vial sharing) and extrapolation of post‑progression survival.
3.30 For the combination‑therapy model, the company did a scenario analysis using the region 1 geographical subgroup (Europe, Israel, USA and Australia). In this analysis, it adjusted overall survival, progression‑free survival and time on treatment. The company used log‑logistic and Weibull distributions. Costs per QALY for ramucirumab plus paclitaxel compared with best supportive care were £114,474 for the Weibull distribution and £95,618 for the log‑logistic distribution.
3.31 For modelling overall survival, the company's base case for the combination‑therapy model used Kaplan–Meier data until the end of the trial period and then extrapolated with an exponential distribution. Independently fitted overall‑survival curves showed that the Weibull distribution followed by the log‑logistic distribution had the best fit to the trial data seen for ramucirumab plus paclitaxel. The log‑logistic distribution was the best fit for the placebo plus paclitaxel trial data and the Weibull was the second worst fitting distribution. The company explored alternative approaches to modelling overall survival such as scenario analyses using the Weibull and log‑logistic distributions. For ramucirumab plus paclitaxel compared with placebo plus paclitaxel, the Weibull distribution gave similar results to the base‑case analysis (ICER of £117,236 per QALY gained), whereas the log‑logistic distribution reduced the ICER to £96,103 per QALY gained.
3.32 For the monotherapy model, the company modelled overall survival in the base case using the gamma distribution. In a scenario analysis, the company used the log‑normal distribution (the distribution with a better fit using the goodness‑of‑fit diagnostic tests), which reduced the ICER to £174,485 per QALY gained.
ERG critique
3.33 The ERG stated that the RAINBOW trial was a good‑quality randomised controlled trial including more than 300 patients in each treatment group, and uncertainty about long‑term follow‑up is likely to be small because both overall survival and progression‑free survival were mature. It also stated that the direction of the imbalances in baseline characteristics in the RAINBOW trial was in favour of the comparator group (that is, paclitaxel alone). The ERG stated that overall the treatment arms for region 1 participants were reasonably balanced, although it noted that it included very few UK patients.
3.34 The ERG noted that in the REGARD trial there was an imbalance in histological subtype, percentage of peritoneal metastases, number of metastatic sites and previous anticancer treatment. It commented that most of the imbalances in baseline characteristics in the REGARD trial favoured the intervention group (that is, ramucirumab monotherapy). The ERG stated that the main issue with the evidence for ramucirumab monotherapy was that the REGARD trial's inclusion criteria did not specify whether patients were suitable for treatment in combination with paclitaxel. Given that eligibility criteria for RAINBOW and REGARD were almost the same and that all patients in the RAINBOW trial had paclitaxel, the ERG stated it was possible that all patients in the REGARD trial could have been eligible for paclitaxel.
3.35 The ERG considered that the network meta-analysis results should be interpreted with caution. Because of significant differences in countries in Europe and North, Central and South America compared with Asian countries in the incidence of gastric cancer, histology, and screening and treatment approaches, the inclusion of at least 1 trial in an Asian population would lead to a high level of heterogeneity. In addition, the ERG was particularly concerned at the reliance of the network on a study that was carried out in an entirely Japanese population (Hironaka et al. 2013); all comparisons with ramucirumab plus paclitaxel used this link in the evidence network. The company included a study by Thuss‑Patience et al. (2011), which included an irinotecan arm. The ERG noted that this study closed prematurely because of poor recruitment, and only included 40 patients meaning it was underpowered. The ERG stated that the network meta‑analysis would have been more reliable if it had included results from Roy et al. (2013), which also included an irinotecan arm. The ERG noted that the inclusion of Roy et al. made the hazard ratios for the comparator treatments more favourable.
3.36 For the combination‑therapy model, the ERG agreed with the company that best supportive care and docetaxel were relevant comparators for ramucirumab plus paclitaxel. However, the ERG did not agree with the company's decision to exclude comparators, which were included in the final scope, based on the 'established use' criterion for 3 reasons:
-
Established NHS practice is already incorporated as a criterion for defining the most appropriate scope.
-
The inclusion criterion used by the company of at least 10% usage in the NHS is not a formal rule.
-
The proportion of treated patients is very low and therefore the proportion of patients having certain comparators will always be low when calculated over all patients whose disease progressed after chemotherapy.
3.37 According to the company's survey of real‑world treatment patterns, paclitaxel was used for 3% of patients, which included all people whose disease progressed after chemotherapy; the ERG noted that this proportion would be 10.5% if the number of people who had paclitaxel was divided by the total number of people who had second‑line therapy. The ERG also considered it plausible that the proportion of people having paclitaxel may increase if NICE was to recommend ramucirumab plus paclitaxel for this indication in the NHS. The ERG also commented that the company's survey of real‑world treatment patterns was based on data from June to July 2013, and that since then favourable results for docetaxel from the COUGAR II study have been published, which may have resulted in increased real‑world use of taxanes in general (paclitaxel as well as docetaxel). The ERG stated that the use of irinotecan and FOLFIRI could also increase as a result, and so the inclusion of these treatments in the comparison could also be considered relevant.
3.38 In additional exploratory analyses, the ERG included the comparators defined in the final scope. These analyses were presented using the company's base‑case assumptions (with the exception of correcting confirmed programming errors – see section 3.45). The results of these exploratory analyses are presented in table 1. The ERG commented that these analyses should be interpreted with caution because they relied on the network meta‑analysis that was associated with significant uncertainty as a result of heterogeneity between the studies. For this reason, the ERG presented results of the exploratory analyses as ICERs for ramucirumab plus paclitaxel compared with each treatment separately (pairwise), rather than in an incremental analysis.
Table 1 Pairwise base‑case results for additional comparators compared with ramucirumab plus paclitaxel using the company's base‑case assumptions*
|
Intervention |
Comparator |
Hazard ratio |
Incremental QALY |
Incremental cost |
ICER |
|
Ramucirumab plus paclitaxel |
Best supportive care |
3.70 |
0.33 |
£39,584 |
£118,174 |
|
Docetaxel |
1.79 |
0.24 |
£34,153 |
£145,302 |
|
|
Irinotecan |
Not reported |
0.15 |
£31,238 |
£213,015 |
|
|
Paclitaxel |
1.59 |
0.1 |
£26,790 |
£273,657 |
|
|
FOLFIRI |
Not reported |
0.1 |
£28,166 |
£294,362 |
|
|
Abbreviations: QALY, quality‑adjusted life year; ICER, incremental cost‑effectiveness ratio. *Company's base‑case assumptions were used except for a corrected programming error. |
|||||
3.39 The ERG stated that comparing ramucirumab with best supportive care was sufficient. It noted the comparison was in line with the final scope of this guidance, if it is accepted that 'not suitable for paclitaxel' means the same as 'not suitable for further cytotoxic chemotherapy'. If this is not accepted, the ERG stated that comparisons with cytotoxic chemotherapy, other than paclitaxel (docetaxel, irinotecan and FOLFIRI), were missing.
3.40 The ERG stated that, in general, the process for the extrapolation of survival curves was clear; but the choice of the survival modelling did not follow the same procedure for all progression‑free‑survival and overall‑survival curves in the combination‑therapy and monotherapy models. The ERG agreed that for the combination‑therapy model, the Kaplan–Meier overall‑survival curve with exponential extrapolation was the most plausible approach because of the poor fit of parametric functions from independent modelling approaches. The ERG stated that, although it understood the reasons for interval‑censoring adjustments in the modelling for progression‑free survival, this approach appeared to slightly underestimate progression‑free survival for the paclitaxel plus placebo arm to a greater extent than in the ramucirumab plus paclitaxel arm. The ERG also noted that the company had justified using the Weibull‑distribution model because the proportional hazards assumption was held; but it stated that there was evidence suggesting violation such as censoring in the tails, overlapping of Kaplan–Meier curves in the first month and interval censoring. The ERG also noted that the proportional hazards assumption was only assessed between the paclitaxel plus placebo and ramucirumab plus paclitaxel arms; it was assumed to hold for the progression‑free‑survival curves of best supportive care and docetaxel. According to the ERG, choosing the Weibull‑distribution model for progression‑free survival over the log‑logistic (with a better fit) for the sake of the proportional hazards assumption was unnecessary as well as conflicting with the approach taken for modelling overall‑survival curves.
3.41 For the monotherapy model, the ERG commented that it was not clear which approach had been followed in interval-censoring adjustments. In addition, the ERG commented that considering Akaike Information Criteria/Bayesian Information Criteria fit and Cox–Snell residuals, the log‑logistic distribution might have been a more appropriate choice for modelling progression‑free survival, but that the log‑normal and log‑logistic parametric estimates were almost the same. Overall, the ERG concluded that the interval‑censored log‑normal distribution for progression‑free‑survival modelling and the Gamma distribution for overall‑survival modelling were plausible.
3.42 The ERG stated that it would expect the average weight of UK patients to be higher than the RAINBOW baseline patient population (about one‑third of all patients were Asian). The ERG considered it more appropriate to use region 1 data for body surface area and body weight because it is believed that region 1 data better reflected the UK population. The ERG considered the company's scenario analysis in which it adjusted the analysis for region 1 was plausible, but it stated that this was more relevant for body surface area, body weight and hospitalisations. Therefore, the ERG considered that the company's scenario analysis, which only adjusted for overall survival, progression‑free survival and time on treatment, was not appropriate as a new base case.
3.43 The ERG commented that using an incidence‑based threshold criterion (5% in each relevant treatment arm) for the inclusion of adverse events resulted in a different selection of adverse events for best supportive care in the combination‑therapy and monotherapy models. According to the ERG, this approach was inconsistent.
3.44 The ERG indicated a potential double counting of hospitalisation costs because Health Resource Groups' (HRGs) codes referring to adverse events also take hospitalisations into account. In the response to the clarification letter, the company provided a scenario that reduced the rate of hospitalisations by an estimate of the proportion of hospitalisations due to adverse events. The ERG used these adjusted hospitalisation rates in its exploratory analyses for its base case. The ERG found an error in the half‑cycle correction of the model submitted by the company. The impact of this correction on the ICER was negligible. The ERG also found a technical error in the costs for docetaxel (both in the second and third line). Furthermore, according to the ERG, the drug acquisition costs for ramucirumab plus paclitaxel were underestimated because these were based on the average weight of the patients in the RAINBOW trial (one‑third of patients in RAINBOW were Asian).
3.45 The ERG did an exploratory analysis and in its base case it included the following adjustments:
-
removal of programming errors
-
correction of programming errors relating to docetaxel price
-
hospitalisation stratification based on treatment and region
-
body surface area and weight based on region 1
-
correction of double counting of hospitalisations due to adverse events.
3.46 The ERG analysis resulted in ICERs compared with best supportive care of £129,431 and £188,055 per QALY gained for the combination‑therapy and monotherapy models respectively. For the combination therapy model, the ERG also applied these amendments to analyses in which the comparator was first, docetaxel, and second, paclitaxel. When the comparator treatment was docetaxel, the ICER increased to £168,164 per QALY gained. When the comparator treatment was paclitaxel, the ICER increased substantially to £359,794 per QALY gained. In both of these analyses, the increase in the ICER was largely due to removing the double counting of hospitalisations for adverse events and using body surface or weight based on region 1. However, the increase was much greater when the comparator was paclitaxel. This was because the impact on the costs of docetaxel when the body surface area was based on region 1 was much lower than for ramucirumab and paclitaxel.
3.47 In addition, the ERG explored 3 different scenarios in the combination‑therapy model:
-
The study of Roy et al. (2013) was included in the overall‑survival network meta‑analysis, which showed that the ICER was sensitive to its inclusion (increase of about £14,000 per QALY gained for ramucirumab plus paclitaxel compared with best supportive care).
-
An analysis was carried out in which the efficacy data were only based on direct evidence from the RAINBOW trial (that is, not using the estimates of treatment effectiveness results from the network meta‑analysis). This showed that the ICER for ramucirumab plus paclitaxel compared with paclitaxel increased from the ERG's base case of £359,794 per QALY gained to £392,108 per QALY gained.
-
In addition to using the efficacy data, the utility values from the RAINBOW trial were also directly implemented. In this scenario analysis, the amount of time each utility value is applied in the pre‑progression state was taken into consideration. This resulted in an ICER for ramucirumab plus paclitaxel compared with paclitaxel of £408,223 per QALY gained.