The Appraisal Committee considered evidence submitted by AstraZeneca and a review of this submission by the Evidence Review Group (ERG). See the Committee papers for full details of the evidence.
3.1 The key clinical evidence came from a phase II trial (Study 19) that was an international, multicentre, double‑blind, randomised, placebo‑controlled trial comparing olaparib with routine surveillance (a 'watch and wait' strategy) in patients with platinum‑sensitive epithelial ovarian cancer (including fallopian tube and peritoneal cancer). Study 19 included 265 patients and took place at 82 study centres in 16 countries. The patients started olaparib or matching placebo within 8 weeks of their last dose of a platinum‑containing regimen. Patients were included in the trial only if their disease was platinum‑sensitive, which was defined as disease progression more than 6 months after completing their penultimate platinum regimen. The mean age of the trial population was 57 years and 96% of patients were white.
3.2 The company's submission focused on a subgroup of 136 patients with BRCA gene mutations, because this is the population covered by the marketing authorisation. In Study 19, knowledge of BRCA‑mutation status was not a requirement for entry into the study. BRCA‑mutation status was determined for almost all patients in the trial, but this was largely done retrospectively. In the original clinical study report, the subgroup analysis by BRCA‑mutation status was based on germline‑BRCA status at entry to the study. In the analysis presented in the submission, and on which the licensed indication is based, patients were classified as having the BRCA mutation (referred to as BRCAm) if the mutation was identified in a sample of either their blood (germline mutation) or their tumour (somatic or tumour mutation).
3.3 The company explained that the demographic characteristics of the BRCAm subgroup were generally consistent with the whole trial population and the 2 groups were well balanced in terms of age and ethnicity. There were fewer people aged over 65 years in the BRCAm subgroup than the whole trial population, but the company explained that this is consistent with inherited germline BRCA mutations in relapsed, high‑grade serous epithelial ovarian cancer.
3.4 Olaparib was associated with a statistically significant improvement in the primary outcome of progression‑free survival for the whole trial population and also for the BRCAm subgroup. For the whole population, median progression-free survival was 8.4 months in the olaparib group and 4.8 months in the placebo group (hazard ratio [HR] 0.35; 95% confidence interval [CI] 0.25 to 0.49). For the BRCAm subgroup, median progression-free survival was 11.2 months in the olaparib group and 4.3 months in the placebo group (HR 0.18; 95% CI 0.10 to 0.31).
3.5 The company explained that Study 19 was not powered to assess overall survival and that survival data were immature. An interim analysis of overall survival was done in November 2012. Crossover to olaparib was not permitted during the treatment period of the study, but after completing the study 23% of patients in the placebo arm had a poly‑ADP‑ribose polymerase (PARP) inhibitor (a second PARP inhibitor was not permitted in the olaparib group). The company specified that this was likely to have a confounding effect on the overall survival results, in favour of the placebo group. Therefore, the company performed an additional post‑hoc analysis of overall survival to investigate the effect of crossover to olaparib from the placebo arm. A restriction method was used that excluded data from centres in which some patients had received PARP inhibitors after disease progression. This excluded 25% (67/265) of patients in the whole trial population and 29% (39/136) of patients in the BRCAm population.
3.6 For the whole trial population, overall survival analysis was done when 58% of the patients had died (November 2012). Median overall survival was 29.8 months (95% CI 27.2 to 35.7) in the olaparib arm and 27.8 months (95% CI 24.4 to 34.0) in the placebo arm. A statistically significant difference in median overall survival was not identified in the whole trial population at this point (HR 0.88; 95% CI 0.64 to 1.21; p=0.44).
3.7 For the BRCAm subgroup, overall survival analysis was done when 52% of the patients had died. A statistically significant difference in overall survival was not identified at this point (median 34.9 months in the olaparib arm and 31.9 months in the placebo arm; HR for death 0.73; 95% CI 0.45 to 1.17; p=0.19). In the analysis that excluded patients who had a PARP inhibitor after disease progression, the median survival was also 34.9 months in the olaparib group, but fell to 26.6 months in the placebo group. This gave a statistically significant difference in overall survival between the groups of 8.3 months (HR for death 0.52; 95% CI 0.28 to 0.97; p=0.039).
3.8 For the whole population, there was a statistically significant improvement in the time from randomisation to treatment discontinuation or death for olaparib compared with placebo (HR 0.39; 95% CI 0.30 to 0.51). For the BRCAmsubgroup, median time to treatment discontinuation or death was 11.0 months in the olaparib arm and 4.6 months in the placebo arm (HR 0.36; 95% CI 0.24 to 0.53).
3.9 There was also a statistically significant improvement in time to first subsequent therapy or death with olaparib, defined as the time from randomisation to the start of the first cancer therapy after discontinuation of olaparib or placebo, or death. In the whole population, the HR was 0.41 for olaparib compared with placebo (95% CI 0.31 to 0.54). For the BRCAm subgroup, the median time to first subsequent therapy or death was 15.6 months in the olaparib group and 6.2 months in the placebo group (HR 0.33; 95% CI 0.22 to 0.50; p<0.00001). The difference in median time to first subsequent therapy or death between placebo and olaparib was 9.4 months, compared with a median progression‑free survival difference of 6.9 months.
3.10 Olaparib was associated with a statistically significant improvement in time to second subsequent therapy or death, defined as the time from randomisation to the start of the second line of cancer therapy after discontinuing olaparib or placebo, or death. In the whole population, the HR for time to second subsequent therapy or death for olaparib compared with placebo was 0.54 (95% CI 0.41 to 0.72). For the BRCAm subgroup, the median time to second subsequent therapy or death was 23.8 months in the olaparib group and 15.2 months in the placebo group, a difference of 8.6 months (HR 0.44; 95% CI 0.29 to 0.67; p=0.00013).
3.11 Study 19 collected adverse event data from the time of consent to treatment to 30 days after the last dose of treatment. The most common adverse events in the whole trial population were nausea (71%), fatigue (52%), vomiting (34%), diarrhoea (27%) and abdominal pain (25%). The most common adverse events in the BRCAm subgroup for olaparib were nausea (73%), fatigue (54%), vomiting (36%), diarrhoea (30%) and abdominal pain (23%).The most common grade 3 or higher adverse events for olaparib in the BRCAm subgroup were fatigue (7%), anaemia (5%) and neutropenia (4%). In the olaparib treatment group, 6 people (4.4%) in the whole trial population and 5 people (6.8%) in the BRCAm subgroup stopped treatment because of adverse events.
3.12 Health-related quality of life was assessed in Study 19 using the Trial Outcome Index (TOI), the Functional assessment of Cancer Therapy/National Comprehensive Cancer Network Ovarian Symptom Index (FOSI), and the Functional assessment of Cancer Therapy Ovarian (FACT‑O) questionnaires. There were no statistically significant differences in the average change in health‑related quality of life for olaparib compared with placebo.
3.13 The company initially submitted 2 economic analyses; 1 that excluded the cost of BRCA testing, and an additional 1 that included the cost of germline BRCA testing (blood test) and the costs and benefits of expanding testing to the relatives of people identified as having germline mutations. Following consultation, the company submitted further analyses that included the cost of tumour testing.
3.14 The company's model compared olaparib with routine surveillance in patients withBRCA mutation‑positive, platinum‑sensitive relapsed ovarian cancer. There were 4 states in the model: progression‑free (with or without maintenance treatment); first subsequent treatment; second subsequent treatment; and death. All patients entered the model in the progression‑free health state. The model had a fixed treatment regimen lasting a maximum of 6 cycles and a time horizon of 15 years. A discount rate of 3.5% was applied to costs and health benefits and the analysis was done from an NHS and personal social services perspective.
3.15 The proportion of the hypothetical cohort in each health state in the company's model was estimated using semi‑Markov‑state transitions, in which time in each health state depends on the time since entry into that state. The company stated that this approach was more flexible than the partitioned survival approach, traditionally used to evaluate fixed chemotherapy regimens. It also stated that this approach was preferable when survival data were immature, and had been used in NICE's technology appraisal guidance on bevacizumab in combination with paclitaxel and carboplatin for first-line treatment of advanced ovarian cancer. Transition probabilities for the BRCAm subgroup were estimated by fitting parametric survivor functions to time-to-event data from Study 19. The parametric models were adjusted for Ashkenazi Jewish ancestry (a risk factor for BRCA gene mutation), time to progression on penultimate platinum therapy, and full compared with partial response to the last platinum therapy before study enrolment. The company stated that a log-normal distribution had been used for estimating the time to first subsequent treatment or death because the goodness‑of‑fit Bayesian information criterion values showed this model to be the most appropriate. The company also stated that visual inspection of the cumulative survival plot for time to first subsequent treatment or death showed that the proportional hazards assumption had been met.
3.16 In the base‑case analysis, the confounding effect from patients in the placebo group having a PARP inhibitor after disease progression was adjusted for. This was done by assuming that the probability of moving from the first subsequent treatment state to the second subsequent treatment state in the placebo group was the same as in the olaparib treatment group, in which further PARP inhibitors were not allowed after treatment with olaparib was stopped. The predicted mean and median times to first subsequent treatment and time to death were based on the results of the analysis that adjusted for use of a PARP inhibitor in the routine surveillance group (see section 3.5). The predicted median overall survival in the model was 38 months for the olaparib population and 26 months for the routine surveillance population.
3.17 Study 19 did not include a generic measure of health‑related quality of life (such as the EQ‑5D), which could have been used to estimate utilities. For the progression‑free health states, utilities were obtained by mapping FACT‑O data from Study 19 to EQ‑5D by applying a published algorithm. Utility estimates for the subsequent‑therapy health states were obtained from the estimates in the OVA‑301 trial of trabectedin plus pegylated liposomal doxorubicin hydrochloride (PLDH) compared with PLDH alone in patients with relapsed advanced ovarian cancer in NICE's technology appraisal guidance on trabectedin for the treatment of relapsed ovarian cancer. Health‑related quality of life decrements associated with adverse events were not included in the base‑case analysis because the company stated that they were not found to be associated with a statistically significant or clinically meaningful change in utility values.
3.18 The company's model included costs for post‑study chemotherapy, monitoring (such as CT scans) and adverse events. The monthly cost of olaparib was based on the UK list price at the time of developing the model (that is, £3950 per pack). The list price has subsequently decreased to £3550 per pack (see section 2.3). The mean daily dose of olaparib used in the model was based on the actual mean daily dose used in Study 19. A unit cost of £127 was applied for an outpatient visit, £3 for a blood test and £90 for a CT scan. It was assumed that all patients with advanced ovarian cancer would have regular scheduled follow-up consultations as part of their ongoing care. Therefore no additional administration costs were applied for treatment with olaparib. The rates of treatment with subsequent chemotherapy after disease progression were based on data from Study 19. The acquisition costs of subsequent chemotherapy were obtained from the Electronic Marketing Information Tool (eMit) for generic drugs and from the British national formulary (BNF).
3.19 The deterministic incremental cost-effectiveness ratio (ICER) estimated by the company's model for olaparib compared with routine surveillance was £49,826 per quality‑adjusted life year (QALY) gained. The probabilistic ICER was £49,146 per QALY gained.
3.20 The company's probabilistic analysis showed that at a maximum acceptable amount for an additional QALY of £30,000, olaparib had a 2% probability of being cost effective compared with routine surveillance. If the maximum acceptable amount for an additional QALY was £50,000, then olaparib would have a 52% probability of being cost effective compared with routine surveillance.
3.21 The company did a series of deterministic one‑way sensitivity analyses to assess the effect of varying the costs, utility estimates and clinical data in the model. The company varied the input values for key parameters by 20% of the value used in the deterministic base case. The lowest ICER reported in the company's one‑way sensitivity analysis was £38,975 per QALY gained (utility for olaparib, progression‑free [on maintenance therapy] state, 0.92). The highest ICER reported in the company's one-way sensitivity analysis was £69,051 per QALY gained (utility for olaparib, progression‑free [on maintenance therapy] state, 0.61).
3.22 The company did a series of scenario analyses. A key driver of the cost-effectiveness results was the estimate of overall survival used in the model (that is, whether it was derived from the model or based on trial data) and the parametric survival curves applied to the distribution of time from first subsequent treatment or death (generalised gamma or log normal). When trial data were used, applying the generalised gamma distribution increased the base-case ICER to £80,715 per QALY gained, and applying the log‑normal distribution increased the base‑case ICER to £68,812 per QALY gained. When the cost of BRCA‑mutation testing was included, the cost of olaparib treatment increased by approximately £2900. This resulted in an ICER of £53,089 per QALY gained for olaparib compared with routine surveillance.
3.23 The company did additional economic analyses comparing olaparib with routine surveillance in people with BRCAm, platinum‑sensitive relapsed ovarian cancer, which included the costs and benefits of extending BRCA‑mutation testing to their relatives. This analysis combined the results of the company's first cost‑effectiveness analysis with results from the cost–utility analysis of genetic testing for women with a family history of breast cancer, which was developed as part of NICE's guideline on familial breast cancer.
3.24 The model was a semi‑Markov design with a number of health states including incidence of new cancers, survival and death. The Breast and Ovarian Analysis of Disease Incidence and Carrier Estimation Algorithm (BOADICEA) was used to analyse the pattern of inheritance of a specific genetic trait (such as the BRCA1 or BRCA2 mutation). This identified 5 family pedigrees, with different risk‑of‑disease profiles. The model had a 50‑year time horizon. A discount rate of 3.5% was applied to costs and health benefits and the analysis was done from an NHS perspective.
3.25 The results were obtained by adding the incremental costs and QALYs from the company's first analysis to the costs and QALYs for BRCA‑mutation testing and comparing these with no testing (from the economic model used in NICE's guideline on familial breast cancer). Across the 5 individual pedigrees, the ICER for olaparib compared with routine surveillance ranged from £33,069 per QALY gained to £41,716 per QALY gained, with an average ICER of £39,343 per QALY gained.
3.26 The ERG was satisfied that all relevant clinical studies were included in the company's submission. The ERG questioned the extent to which Study 19 reflected clinical practice in England, because it used blood (for inherited germline mutations) and tumour (for acquired somatic mutations) BRCA‑mutation testing to select patients. It commented that these tests may not be routinely performed in England and that it is unclear if tumour testing would be possible in England on a large scale. Therefore, the ERG considered that the population in the trial may differ from the population treated in clinical practice in England.
3.27 The ERG commented that the clinical study report for Study 19 lacked clarity as to when and under which circumstances patients had subsequent chemotherapy after disease progression. Therefore the outcomes of time to first subsequent therapy or death and time to second subsequent therapy or death may not reflect routine clinical practice in England. Clinical advisers for the ERG suggested that this would be most likely to shorten estimates and may affect comparative estimates between study groups.
3.28 The ERG considered that the clinical effectiveness evidence from Study 19 was weak and at a high risk of bias because of a number of problems:
Subgroup analysis: the BRCAm subgroup was identified post hoc and had a small sample size (n=136), leading to potential bias and uncertainty as to the true size of the treatment effect in that group. An interaction test to assess whether there was evidence that the subgroup was statistically significantly different to the rest of the trial population was not presented in the company's submission, but it was included in the clinical study report; the results of this test were inconclusive.
Post‑hoc identification of outcomes: the outcomes of time to treatment discontinuation or death, time to first subsequent therapy or death, time to second subsequent therapy or death and long‑term overall survival were all identified post hoc.
Continuation of study drug after progression: some patients continued olaparib treatment after disease progression, which may not be in line with the marketing authorisation and may not reflect clinical practice. This could alter treatment effect as measured by progression‑free survival, time to first subsequent therapy or death, time to second subsequent therapy or death and overall survival because patients may have had olaparib for a different length of time than would occur in clinical practice.
Crossover: crossover of patients in the placebo arm to PARP inhibitor treatment, after disease progression, may have caused confounding and bias in the overall survival results.
Immaturity of overall survival results: there was a lack of mature and conclusive evidence that treatment with olaparib increases overall survival.
3.29 The ERG considered overall survival, rather than progression‑free survival, to be the most clinically relevant measure to assess the effect of a treatment on survival.
3.30 The ERG commented that time to first subsequent therapy or death and time to second subsequent therapy or death could be considered more clinically relevant than progression‑free survival. This is because, in England, patients with ovarian cancer often have treatment after disease progression if their symptoms cause problems. However, the ERG considered that there were issues with time to first subsequent therapy or death and time to second subsequent therapy or death in the context of Study 19, including that the analyses were post hoc.
3.31 The ERG highlighted that it had concerns about the validity of the treatment pathway in the model. It noted that all patients who survived the first subsequent therapy event and all patients who survived the second subsequent therapy event were assumed to go on to have another course of active chemotherapy. The ERG stated that it may not be realistic to assume that all patients with advanced ovarian cancer would be well enough to have active chemotherapy. The ERG added that its clinical advisers had suggested that it was more clinically realistic to assume that some patients would opt for best supportive care. It also noted that the company's model limited the number of subsequent chemotherapy treatments to a maximum of 2, but 32% of patients in Study 19 had 3 or more subsequent chemotherapy treatments.
3.32 The ERG had concerns about the structure of the company's model and considered that it would have been more appropriate to use a partitioned survival model approach to estimate transition probabilities.
3.33 The ERG was concerned by the exclusion of progression‑free survival (the primary end point in Study 19) from the model, and that overall survival data from Study 19 were not directly included. It also expressed concern that the time to first subsequent therapy or death outcome data used in the model were collected post hoc so may have been biased, and that the continued use of olaparib beyond disease progression was permitted, which may not reflect the marketing authorisation.
3.34 The ERG had concerns about the company's approach to modelling the time‑to‑event outcomes. It was unclear to the ERG why certain covariates such as Ashkenazi Jewish ancestry had been adjusted for, but others such as patients' age and performance status had not. It was also unclear how covariates had been included in the extrapolated Kaplan–Meier curves. The ERG was also concerned that the company's curve-fitting process did not appear to have included any external data, or expert subjective judgement on the plausibility of the extrapolated curves. It noted that curve fitting appeared to have been based only on visual inspection of how well the curves fit the observed data, and goodness of fit according to Akaike information criterion and Bayesian information criterion statistics. The ERG also noted that the company's submission stated that the assumption of proportional hazards was met for the time‑to‑event outcomes of time to treatment discontinuation or death and for time to first subsequent therapy or death. However the ERG noted that the log–log survival plots for each of these outcomes showed that the curves for each treatment group crossed, indicating that the proportional hazards assumption may not be appropriate.
3.35 The ERG was concerned that the model overestimated the results that were seen in the trial, and that the degree of overestimation was consistently greater in the olaparib group. It considered that the median time to first subsequent therapy or death predicted in the model was upwardly biased by 4.4 months, the median time to second subsequent therapy or death was upwardly biased by 3.2 months and the median overall survival was upwardly biased by 3.1 months. The ERG noted that there was also a degree of overestimation in the routine surveillance group. The median time to first subsequent therapy or death predicted in the model was upwardly biased by 0.8 months and the median time to second subsequent therapy or death was upwardly biased by 0.7 months. By contrast, the median overall survival estimate in the model for the routine surveillance group was 5.9 months lower than the results from the trial. Therefore the ERG did not have confidence in the overall survival gains or QALY benefits predicted for olaparib in the company's model.
3.36 The ERG had concerns about the utility estimates used for the first subsequent therapy (0.72) and second subsequent therapy (0.65) health states, noting that these were derived from estimates for progression‑free survival and progressed disease from the EVO‑301 trial in NICE's technology appraisal guidance on trabectedin for the treatment of relapsed ovarian cancer. The ERG suggested that the company's model implicitly assumed that patients in the first subsequent therapy state have a health‑related quality of life comparable to that of patients with stable disease, and that patients in the second subsequent therapy state have a health‑related quality of life comparable to that of patients with progressive disease. The ERG also noted that the company's model assumed that adverse events do not have any additional effect on patients' health‑related quality of life. It highlighted that the validity of these assumptions was not discussed in the company's submission.
3.37 The ERG did not agree with excluding the cost of BRCA‑mutation testing from the company's base‑case analysis. It highlighted that the NICE guide to the methods of technology appraisal 2013 states that if a diagnostic test to establish the presence or absence of a biomarker is carried out solely to support the treatment decision for the specific technology, the associated costs of the diagnostic test should be incorporated into the assessments of clinical and cost effectiveness.
3.38 The ERG amended the company's model by correcting 2 errors and including the cost of BRCA‑mutation testing. This increased the probabilistic base-case ICER to £53,374 per QALY gained. However, the ERG did not believe that the company's model provided robust estimates of overall survival or QALY gains. Therefore it advised caution in interpreting any results produced using the model.
3.39 The ERG did not consider it appropriate to combine the results of the company's model with the model developed for NICE's guideline on familial breast cancer. It noted that the 2 models were different in terms of the treatment pathway assumed for ovarian cancer. The ERG pointed out that the model in NICE's guideline on familial breast cancer does not include olaparib as a treatment option for ovarian cancer. Therefore the company's analysis reflected a scenario in which BRCA‑mutation testing and olaparib treatment were available for the patient but not for their relatives. The ERG suggested that the 5 family pedigrees identified may not reflect the range of possible family structures seen in the population with advanced ovarian cancer. It therefore did not consider that the ICERs from the analysis that included BRCA testing were meaningful.
3.40 The ERG did not accept that all relevant comparisons had been included in the analysis such as: no testing and no drug; testing and no drug; and testing and drug treatment for people who test positive for a BRCA mutation. The ERG commented that the analysis may therefore have led to inappropriate conclusions, because the benefit of the joint intervention was driven by the costs and benefits of BRCA testing rather than the costs and benefits of olaparib treatment.
3.41 The company included the cost of somatic testing as a sensitivity analysis in its model. The results of this additional sensitivity analysis incorporating the costs of tumour testing and corrections to 2 errors in the model identified by the ERG (see section 3.38) produced a probabilistic ICER of £51,587 (incremental costs £47,032; incremental QALYs 0.91) and a deterministic ICER of £51,552 (incremental costs £45,879; incremental QALYs 0.89) for olaparib compared with routine surveillance.
3.42 To address concerns raised by the Committee in the first appraisal consultation document about the structure of the model and the plausibility of the projected survival benefits associated with olaparib, the company conducted an analysis of projected survival outcomes for the population that excluded sites where patients in the placebo group were able to crossover to a PARP inhibitor in the Study 19 BRCAm subgroup. This used an alternative modelling approach by fitting overall survival from Study 19. The company reported that the results of these analyses provided reassurance that the projected survival and QALY benefits associated with olaparib as well as the resulting ICERs compared with routine surveillance had not been overestimated when compared with the original submission. The results of the analyses, including the costs of somatic testing, produced ICERs ranging from £37,917 to £66,491 using independent fitting models and £54,618 to £57,349 using a treatment‑adjusted model.
3.43 The ERG had serious concerns about the validity of the company's new survival analysis. In particular, it highlighted that the modelled estimates of overall survival provided by the company were very different to those produced by the ERG in their exploratory analyses. Other comments included that the company provided no details of the parameter estimation procedure, or the statistical software used, and that the methods and rationale for covariate adjustment in the survival modelling were unclear. Therefore, the ERG believed that the company's new survival analysis did not produce reliable estimates of overall survival. Full details of all the evidence are in the Committee papers.
Subgroup of patients who received 3 or more lines of platinum‑based chemotherapy before randomisation
3.44 The company submitted clinical evidence for the subgroup of patients in Study 19 who received 3 or more lines of platinum‑based chemotherapy before randomisation. The company reported that the demographic characteristics of this subgroup were generally consistent with the overall BRCAm population for both arms of the study and that there may be a clinically relevant imbalance in prognostic factors between the 2 treatment arms, potentially favouring placebo. Patients randomised to the placebo arm were more likely to be younger than 50 years (29.4% compared with 14.9% in the olaparib arm), have achieved a complete response to their previous chemotherapy (61.8% compared with 44.7%) and have favourable Federation of Gynaecological Oncologists (FIGO) staging (61.8% compared with 85.1% were diagnosed with FIGO stage IIIC or IV disease).The olaparib arm had a higher proportion of patients who were fully platinum‑sensitive (63.8% compared with 47.1%) and who had an ECOG performance status of 0 (85.1% compared with 73.5%).
3.45 The clinical evidence showed statistically significant benefits for olaparib compared with placebo for progression‑free survival, time to first subsequent treatment or death, and time to secondary subsequent treatment or death. Median progression‑free survival was 11.2 months in the olaparib group and 4.3 months in the placebo group (HR adjusted for stratification factors 0.11; 95% CI 0.05 to 0.23). Median time to first subsequent treatment or death was 13.6 months in the olaparib group and 5.6 months in the placebo group (HR adjusted for stratification factors 0.28; 95% CI 0.16 to 0.49). Median time to secondary subsequent treatment or death was 20.3 months in the olaparib group and 14.3 months in the placebo group (HR adjusted for stratification factors 0.41; 95% CI 0.24 to 0.70). No statistically significant improvement in overall survival was reported. Median time to death was 31.3 months in the olaparib arm and 26.5 months in the placebo arm without adjustment for crossover (HR adjusted for stratification factors 0.69; 95% CI 0.38 to 1.27) and 32.9 months for olaparib and 20.6 months for placebo with adjustment for crossover (HR adjusted for stratification factors 0.56; 95% CI 0.26 to 1.20).
3.46 The company submitted 2 economic models that assessed the cost effectiveness of olaparib in the subgroup of patients who received 3 or more previous lines of platinum‑based chemotherapy before randomisation: its semi‑Markov model based on 4 health states (see section 3.14) adapted for this subgroup, and a more standard partitioned survival model with 3 health states (using extrapolated overall survival estimates for the population of Study 19 from sites where crossover was not permitted) to support the robustness of the model. Both models included the costs of somatic testing. They also incorporated a reduced list price of £3550 per pack, and a revised patient access scheme that the company had agreed with the Department of Health (see section 2.3).
3.47 In the company's 4 health‑state model, the base case analysis used the best fitting curves (based on AIC/BIC statistics and visual inspection) for the independently fitted parametric curves for time to first subsequent therapy or death (log normal); time to treatment discontinuation or death (log logistic); first subsequent treatment to second subsequent treatment (log normal) and second subsequent treatment to death (Weibull). The base case deterministic ICER for olaparib compared with routine surveillance was £37,583 per QALY gained and the probabilistic ICER was £37,864 per QALY gained. The company did a series of scenario analyses to assess the impact on model outcomes of using alternative parametric distributions for the time to first subsequent therapy or death. The results of the analyses, excluding the generalised gamma distribution which the company reported was unstable, produced ICERs ranging from £37,583 to £42,876 using independent fitting models and £39,036 to £49,244 using a treatment‑adjusted model.
3.48 In the company's 3 health‑state model, the base case analysis used the best fitting parametric survival curves from the independently fitted models for overall survival (log normal); progression‑free survival/time to first subsequent treatment (log normal) and time to treatment discontinuation (log logistic). The base case deterministic ICER for olaparib compared with routine surveillance was £46,806 per QALY gained and the probabilistic ICER was £45,343 per QALY gained. Scenario analyses to assess the impact of using alternative parametric distributions on model outcomes produced ICERs ranging from £40,000 to £59,664 using independent fitting models and £49,290 to £70,826 using a treatment‑adjusted model (excluding the generalised gamma distribution).
ERG's comments on the evidence for the subgroup of patients who received 3 or more lines of platinum‑based chemotherapy before randomisation
3.49 The ERG considered that the clinical evidence for the subgroup of patients who received 3 or more previous lines of platinum‑based chemotherapy before randomisation should be interpreted with caution because of the small sample size and a potential imbalance between the 2 treatment groups, caused by known and unknown confounders.
3.50 The ERG reiterated its concerns about the appropriateness of the company's 4 health‑state model structure which the ERG considered may skew the available evidence, producing estimates of incremental survival for olaparib that are exaggerated.
3.51 The ERG considered the fit of the company's modelled overall survival predictions against the empirical Kaplan–Meier curves. It was concerned that the company's 4 health‑state model overestimated the probability of survival for patients receiving olaparib beyond 2 years, and underestimated the probability of survival for patients on routine surveillance beyond 2.5 years. In addition, the survival curves indicated no survival benefit for olaparib compared with placebo beyond 3.2 years whereas the model suggested that the greatest difference in survival happened beyond this time. The ERG also found that the company's 4 health‑state model produced a higher estimate of incremental health gain (1.43 life years) than the highest estimates from the 3 health‑state model (1.10 life years) and Study 19 (0.347 years without adjustment for treatment crossover, and 0.507 years with adjustment for treatment crossover).