Mannitol dry powder for inhalation for treating cystic fibrosis

The manufacturer presented clinical-effectiveness data from 2 randomised multinational double-blind controlled trials (DPM-CF-301 and DPM-CF-302). The trials were designed to assess the effectiveness of twice-daily mannitol at a dose of 400 mg compared with a 50 mg dose of mannitol twice daily, assumed to be sub-therapeutic. Patients in both arms also received best supportive care with or without rhDNase. Best supportive care included, but was not limited to, inhaled antibiotics, anti-inflammatory agents, bronchodilators, vitamin supplements, pancreatic enzymes, and antidiabetic agents for people with diabetes. The trials had 26-week double-blind phases, followed by further unblinded phases of 26 to 52 weeks. The inclusion and exclusion criteria for the 2 trials were similar but differed in the lower cut-off for 'FEV₁% predicted' (FEV₁% of the patient adjusted for the average FEV₁% in the population for any person without cystic fibrosis of similar age, sex and body composition); this was 30% in DPM-CF-301 and 40% in DPM-CF-302.

DPM-CF-301 included 295 patients (190 adults) and took place at 40 centres in Australia, New Zealand, the UK and Ireland. The manufacturer presented results for adults only, in line with the marketing authorisation. There were 114 adults in the mannitol arm, of whom 58 used rhDNase and 56 did not use rhDNase, and 76 adults in the control arm, of whom 44 used rhDNase and 32 did not use rhDNase. There were 30 adults in the mannitol arm and 13 in the control arm (43 in total) who could not take rhDNase because of ineligibility, intolerance, or inadequate response to rhDNase. The remaining 45 patients did not use rhDNase for other reasons that were not recorded.

DPM-CF-302 included 305 patients (151 adults) and took place at 53 centres in the USA, Canada, Argentina, Germany, Belgium, France and the Netherlands. The manufacturer presented results for adults only. There were 93 adults in the mannitol arm, of whom 64 used rhDNase and 29 did not use rhDNase, and 58 adults in the control arm, of whom 41 used rhDNase and 17 did not use rhDNase. There were 15 adults in the mannitol arm and 7 in the control arm (22 in total) who could not take rhDNase because of ineligibility, intolerance, or inadequate response to rhDNase. The remaining 24 patients did not use rhDNase for other reasons that were not recorded. After a request by the ERG for clarification, the manufacturer submitted information on 2 groups defined by their use of rhDNase: (1) people using rhDNase and (2) people who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase.

The trial protocols for DPM-CF-301 and DPM-CF-302 were similar. Because mannitol and hypertonic saline have similar modes of actions, patients taking nebulised hypertonic saline were excluded from DPM-CF-301 and, in DPM-CF-302, patients could use nebulised hypertonic saline at initial assessment but had to stop 4 weeks before the baseline assessment. Potential participants were screened for bronchial hyperresponsiveness to mannitol, and those with hyperresponsiveness were excluded before randomisation. Randomisation was stratified by region and rhDNase use. The studies were powered to show an improvement in FEV₁ in both patients who took rhDNase and also the total trial population. There were 4 follow-up visits after the screening visit, at week 0 (start of treatment with mannitol or control) and at weeks 6, 14, and 26. In both trials, patients were offered the opportunity to continue or start mannitol treatment in an open-label phase for a further 26 weeks to gain more information on adverse reactions; in DPM-CF-301, there was an additional open-label extension phase of 26 weeks, giving a total of 78 weeks.

The primary outcome in both trials was the absolute FEV₁ as measured in millilitres over 26 weeks. Both trials reported changes in FEV₁ from baseline in the mannitol group compared with the control group.

Secondary outcomes included the proportion of patients who responded by FEV₁ criteria defined as achieving an increase from baseline of at least 100 ml in FEV₁, or at least a 5% increase in absolute FEV₁ in millilitres, or at least a 5 percentage point increase in FEV₁% predicted. Protocol-defined pulmonary exacerbations (PDPE) were defined as pulmonary events with 4 or more pre-defined symptoms or signs needing intravenous antibiotics. Reductions in the frequency of both PDPE and hospital care were measured in both trials.

Both trials measured quality of life using the Cystic Fibrosis Questionnaire – Revised (CFQ-R); DPM-CF-302 also used the Health Utility Index 2 (HUI2). The CFQ-R was administered to patients at week 0 and then at weeks 14 and 26. Antibiotic use and adverse events were also measured in both trials.

In the manufacturer's original submission, lung function was reported only for the subgroup of adults using rhDNase, in line with the fact that only 1 subgroup was pre-specified in the statistical plan of the study protocol. After a request by the ERG for clarification, the manufacturer submitted data on change in FEV₁ and exacerbations for adults using rhDNase and also for adults who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase.

In the DPM-CF-301 trial, mannitol statistically significantly improved lung function over 26 weeks compared with the control, as measured by a change in FEV₁ from baseline of 109.3 ml (95% confidence interval [CI] 52.8 to 165.8; p<0.001) in adults using rhDNase. This difference was evident at 6 weeks of treatment and was maintained over the 26-week double-blind phase. For the other measures of lung function, the mean difference from baseline between adults randomised to receive mannitol plus rhDNase compared with the control group were: mean percentage change in FEV₁ from baseline of 4.2% (95% CI 0.3 to 8.1); change in FEV₁% predicted of 2.7% (95% CI 0.6 to 4.7) and change in forced vital capacity (FVC) of 117.4 ml (95% CI 1.0 to 233.9). After a request for clarification, the manufacturer provided the change in FEV₁ over 26 weeks for the 43 adults who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase; this was 147.0 ml, a difference that was statistically significant between the mannitol and control group (95% CI 23.2 to 270.7, p=0.02).

In the DPM-CF-301 trial, for people using rhDNase, the incidence of PDPE (over the 26-week time horizon of the trial) was 27.6% for adults randomised to mannitol compared with 36.4% in the control group. For adults who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, the incidence of PDPE was 16.7% in the mannitol group compared with 30.8% in the control group. The rate of PDPE per year was 1.41 for adults receiving mannitol plus rhDNase compared with 1.58 in the control group. The estimated PDPE rate per year was 0.41 for adults randomised to receive mannitol who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase compared with an estimated 0.64 in the control group. There was a 36.5% reduction in the rate of exacerbations in the mannitol group compared with the control group for patients who could not use rhDNase, but this was not statistically significant. The trials were not powered to demonstrate statistically significant differences for PDPE outcomes in these subgroups.

In the DPM-CF-302 trial, the mean change in FEV₁ over 26 weeks for adults using rhDNase and randomised to receive mannitol was not statistically significantly different from the control group (88.5 ml, 95% CI -8.5 to 185.4). For the other measures of lung function in adults using rhDNase randomised to receive mannitol compared with the control group, the mean difference in the changes from baseline were 5.4% (95% CI -0.4 to 11.3) for FEV₁, 3.0% (95% CI -0.6 to 6.5) for change in FEV₁% predicted and 96.9 ml (95% CI -7.7 to 201.6) for changes in FVC. After clarification, the manufacturer provided the difference in the change in FEV₁ over 26 weeks for mannitol in 22 adults who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase compared with the control group; this was 208.6 ml (95% CI -9.3 to 426.5, p=0.061).

In the DPM-CF-302 trial, the incidence of PDPE (over the 26 weeks of the trial) was 18.8% in adults using rhDNase alone compared with 9.8% for adults receiving mannitol plus rhDNase. For adults who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, the PDPE incidence was 13.3% for adults randomised to mannitol compared with 42.9% in the control group. The estimated PDPE rate per year was 0.83 for adults randomised to mannitol plus rhDNase compared with 0.19 in the control group. The estimated PDPE rate per year was 0.26 for adults randomised to receive mannitol who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase compared with an estimated 0.86 in the control group. However, the trials were not powered to demonstrate statistically significant differences for these outcomes in patients not receiving rhDNase or patients who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase.

In response to the appraisal consultation document (ACD), the manufacturer presented analyses pooling the results of the DPM-CF-301 and DPM-CF-302 trials. The outcomes in the pooled analyses did not include the primary outcome, absolute FEV₁, but instead were FEV₁% predicted, the proportion responding according to FEV₁ criteria, and the estimated rate of PDPE per patient per year. For the pooled adult population of people using rhDNase, the mean change in FEV₁ over 26 weeks for patients receiving mannitol plus rhDNase was 94.1 ml (95% CI 29.7 to 158.42). For people who received mannitol but did not take rhDNase (irrespective of the reason), the change was 110.3 ml (95% CI not given, p<0.005). The change was 166.7 ml (95% CI 52.0 to 280.6) for the subgroup of adults receiving mannitol who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase. The pooled results for both trials for adults who could not use rhDNase were 0.27 PDPE per year in the mannitol group compared with 0.96 PDPE per year in the control group (not statistically significant).

The manufacturer reported on adverse events for the whole adult population, but not for the rhDNase subgroups based on rhDNase use. Overall, 87% of all adults experienced at least 1 adverse event, the most common adverse event being cough (the only adverse event that occurred in more than 10% of adults). However, the manufacturer considered productive cough a desired effect of treatment with mannitol. Other adverse events with an incidence of between 1% and 10% over 26 weeks were decreased appetite, headache, haemoptysis, bronchospasm, wheezing, asthma, condition aggravated, pharyngolaryngeal pain, and chest discomfort.

Haemoptysis was the most clinically significant adverse event in both studies and was observed in 11.9% of adults on treatment with mannitol and 8.5% in the control group in the DPM-CF-301 trial, and in 7.1% in the mannitol group and 2.5% in the control group in the DPM-CF-302 trial. In response to the ACD, the manufacturer submitted pooled analyses on haemoptysis events based on the protocol-defined criteria for PDPE. For all adults using rhDNase, 16.4% experienced haemoptysis in the mannitol group and 20.0% in the control group. For adults not using rhDNase, these figures were 14.1% in the mannitol group and 14.3% in the control group.

Health-related quality of life was only presented for the whole adult population in the original manufacturer's submission. There were no statistically significant changes in the CFQ-R domains in either trial for adults randomised to receive mannitol relative to the control group. The results suggested some improvement in the respiratory, physical and vitality domains of CFQ-R, but these did not achieve statistical significance. In DPM-CF-302, there was no statistically significant difference in HUI2 measurements between adults randomised to receive mannitol and the control group.

In response to the ACD, the manufacturer submitted the proportion of adults whose condition was considered to respond to treatment based on FEV₁ criteria for both trials. In the DPM-CF-301 trial, for adults using rhDNase, the condition responded in 41.4% in the mannitol group and 27.3% in the control group; for adults not using rhDNase, these figures were 60.7% in the mannitol group and 40.6% in the control group. In the DPM-CF-302 trial, for adults using rhDNase, the condition responded in 45.3% in the mannitol group and 39.0% in the control group; for adults not using rhDNase, these figures were 44.8% in the mannitol group and 44.8% in the control group.

The manufacturer developed a Markov health-state transition model, implemented as a patient-level simulation model evolving over the lifetime of the patient, and modelling 2 treatment options: treatment with inhaled mannitol and treatment without inhaled mannitol. The manufacturer did not model inhaled hypertonic saline as a treatment that a patient may use with, or instead of, mannitol. The model assumed treatment with mannitol for lifetime or until drop-out (according to the rate modelled on the trials). The analysis had a time horizon of 100 years, at which point all patients would have died. The cycle lengths were taken from the time between visits in the 2 trials, and were 6 weeks for the first cycle, 8 weeks for the second cycle and 12 weeks for each subsequent cycle. The transition parameters between the health states depended on characteristics derived from the clinical trial such as age, history of pulmonary exacerbations and use of mannitol.

The health states in the model include cystic fibrosis, cystic fibrosis with improved respiratory symptoms, lung transplantation, death from cystic fibrosis, and death from an unrelated cause. At baseline, all patients enter the cystic fibrosis health state. As patients progress, if their FEV₁% predicted falls below 30%, they enter the lung transplantation state in which they have a probability of receiving a transplant in subsequent cycles. The model includes a discontinuation rule under which patients whose condition does not respond to mannitol treatment within 6 weeks stop mannitol and switch to best standard of care. Modelled to mirror the clinical trial, the definition of a response is either a relative increase of 5% or more in absolute FEV₁ or an absolute increase of 100 ml or more in FEV₁ at week 6 from baseline. In subsequent cycles, a patient may switch between the health states of cystic fibrosis and cystic fibrosis with improved respiratory symptoms and back again, and patients in either state may experience a pulmonary exacerbation. Patient characteristics such as body mass index (BMI), age and FEV₁% predicted are updated.

The manufacturer used clinical-effectiveness data from the DPM-CF-301 and DPM-CF-302 trials to obtain baseline values and some, but not all, transition parameters used in the model, such as FEV₁% predicted at week 26, the probability of being a 'responder' at 26 weeks and the relative risk of pulmonary exacerbations for 'responders' to treatment. Other transition parameters were derived from the literature and from the commissioned BioGrid study using regression analysis, such as FEV₁% predicted over time, the rate of exacerbations and mortality after lung transplant. The baseline characteristics (age, sex, BMI and FEV₁% predicted) were taken from the pooled adult population from the DPM-CF-301 and DPM-CF-302 trials. The manufacturer also used data from the trials to estimate the probability of response to mannitol, FEV₁% predicted after 26 weeks of treatment, the effect of treatment on pulmonary exacerbations, and the probability of improvement in respiratory symptoms. The manufacturer estimated changes in FEV₁% predicted and the risk of an exacerbation after baseline from the BioGrid retrospective observational study of disease progression in cystic fibrosis, which used data from Australia (the BioGrid data) and was commissioned by the manufacturer. The decline over time in FEV₁% predicted was modelled dependent on age, age above 30 years, and pulmonary exacerbations (using hospital admissions as a proxy). The manufacturer estimated the relationship between FEV₁% predicted and mortality rate from the BioGrid data using survival analyses. In the model, mortality depended on FEV₁, exacerbations, age, sex, concurrent infection with Burkholderia cepacia and lung transplantation. After a request for clarification from the ERG, data on these variables were provided to update the model for patients who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase.

Utility values were drawn largely from HUI2 data collected during the DPM-CF-302 trial; the manufacturer also included values from the literature for lung transplantation and pulmonary exacerbations. The baseline utility was taken as the mean overall HUI2 global utility score at baseline (0.899). The manufacturer calculated the change in utility between baseline and visit 3 or in week 14, or between baseline and the last visit in the case of early withdrawal. The manufacturer calculated the HUI2 global utility scores for each health state by adding the average change to the baseline utility. The increase in utility value for patients with improved symptoms was 0.009 in the control group and 0.019 in the mannitol group. The decrease in utility for patients without improved symptoms was 0.046 for patients in the control group and 0.022 in the mannitol group. Utility values were not linked directly to lung function. The HUI2 questionnaire was administered in the trial at weeks 0, 12 and 26, but had a recall period of 1 week, and so did not necessarily capture the effect of PDPEs on health-related quality of life at the time they occurred. Therefore, the manufacturer took utility data for PDPEs and post-lung transplantation from the literature. Adverse events had a negative impact on CFQ-R data in both trials.

The manufacturer calculated the costs of treatment with or without mannitol accumulated up to 26 weeks, but made no distinction between patients whose lung function improved and those whose did not. The model included costs related to pulmonary exacerbation and for the time periods before and after lung transplantation. Costs were taken from national reference costs. The manufacturer included costs for concomitant medications (mostly antibiotics) for both groups, and used a mean cost of £3,253 in the mannitol group and £2,972 in the control group (with a cost of £0 for the subclinical trial dose). In the trials, most patients were admitted to hospital at least once, and approximately 40% had a community visit during the 26-week randomised phase of the trial. Costs of pulmonary exacerbation were taken from the trial data. For patients receiving mannitol, the mean total cost of medications, community visits and hospitalisations without a PDPE in the 26-week trial period was £4,391, and taking into account PDPE the cost was £12,852. From the trial, for patients in the control group, the mean total costs without PDPE were £4,664 and with at least 1 PDPE were £10,354. The manufacturer used peri-transplant costs from the UK literature and resource use from the trial and patient records. The manufacturer applied a discount rate of 3.5% to both costs and benefits.

The manufacturer's base-case results indicated an incremental cost-effectiveness ratio (ICER) for mannitol compared with treatment without mannitol (best supportive care) of £47,095 per quality-adjusted life year (QALY) gained in adults using rhDNase and £41,074 per QALY gained in adults who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase.

The manufacturer undertook extensive scenario analyses and deterministic sensitivity analyses for the treatment of adults using rhDNase and for adults not using rhDNase (irrespective of the reason for non-use). The parameters that changed the ICER by more than 10% were:

• FEV₁% predicted at baseline

• the regression parameter estimate for mannitol treatment used to predict the FEV₁% predicted after 26 weeks of treatment (that is, the effect of treatment)

• the relative risk of an exacerbation with mannitol relative to not using mannitol for people whose condition was considered to respond to treatment

• the relative risk of a subsequent exacerbation if there had been an exacerbation in the previous year

• hazard rate of death for the FEV₁% predicted

• the utility decrement associated with an exacerbation

• the utility associated with no improvement in respiratory symptoms among patients using or not using mannitol

• the cost of an exacerbation.

The manufacturer also performed sensitivity analyses showing the effect of several parameters, including treatment failure after 1, 5, 10 and 20 years. The base-case model assumed that patients using mannitol maintained the difference in lung function over their lifetime. Not maintaining the improvements in FEV₁% predicted over the long term had a large effect on the ICER. If the improvements in FEV₁% predicted were maintained for only 1 year the ICER was £149,587 per QALY gained; if improvements were maintained for 5, 10 and 20 years, the ICERs were £86,981, £63,539 and £49,907 per QALY gained respectively. Other factors that had an impact on the ICER were the rate ratio of pulmonary exacerbations between people receiving and those not receiving mannitol, whether the discontinuation rule was applied, the relative risk of a PDPE if the patient experienced an exacerbation in the previous year, costs and utilities. The manufacturer concluded that the main factors affecting the ICER were:

• the cost of mannitol

• the relative risk of pulmonary exacerbations in the mannitol group

• the impact of pulmonary exacerbations on quality of life

• the FEV₁% predicted when starting mannitol

• the improvement in FEV₁% predicted on mannitol treatment

• the hazard rate of death for FEV₁% predicted

• utility for patients whose symptoms do not improve.

After a request from the ERG, the manufacturer provided scenario analyses taking into account reduced adherence to treatments, which reduced the costs in the mannitol group. Using a lower mean adherence gave an ICER of £37,387 per QALY gained for mannitol compared with the control in adults using rhDNase, and £33,934 per QALY gained for mannitol compared with the control in adults not using rhDNase.

There was uncertainty around the relative risk of PDPE, and the model was sensitive to fluctuation in this parameter. Using the relative risk of exacerbation of 0.7 associated with treatment with mannitol for the total adult population (provided by the manufacturer in response to a request for clarification from the ERG), the ICER for mannitol compared with not using mannitol was £54,329 per QALY gained in adults using rhDNase and £27,673 per QALY gained in adults who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase. Using a relative risk of exacerbations based on adults who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, the ICER per QALY gained was £19,828.

In the manufacturer's probabilistic sensitivity analyses, as an add-on therapy to best standard of care, mannitol had a 16.4% probability of being cost effective at an ICER of £30,000 per QALY gained and a 7.4% probability at an ICER of £20,000 per QALY gained. For non-users of rhDNase, mannitol had a 25.8% probability of being cost effective at an ICER of £30,000 per QALY gained and a 10.9% probability at an ICER of £20,000 per QALY gained. In response to a request from the ERG to estimate the cost effectiveness separately for adults using rhDNase and adults who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, the manufacturer re-ran the probabilistic sensitivity analysis. This resulted in mean ICERs of £53,796 per QALY gained for adults using rhDNase and £30,080 per QALY gained for adults ineligible for rhDNase.

The manufacturer conducted 2 further subgroup analyses, 1 by baseline FEV₁% predicted and the other among patients whose condition responded to treatment with mannitol by 6 weeks. The analyses showed that, as baseline FEV₁% predicted declines, the ICER decreases. For FEV₁% predicted 80% or more, the ICERs were £56,228 per QALY gained for adults using rhDNase and £50,688 per QALY gained for adults not using rhDNase. For FEV₁% predicted less than 40%, the corresponding ICERs were £30,746 per QALY gained for adults using rhDNase and £23,704 per QALY gained for adults not using rhDNase.

In response to the ACD, the manufacturer submitted a revised model for people with cystic fibrosis not using rhDNase, including people whose reason for not using rhDNase was not reported in the trials and is not known. In addition, the manufacturer changed several key parameters in the model. The health states in the model more closely model health states rather than treatment states. The costs and utility values in the revised model no longer depend on treatment, but rather on whether the simulated patient has improved respiratory symptoms or not. The utility values are no longer directly tied to lung function. The manufacturer also included a new stopping rule, centred on the Committee's concerns that a stopping rule based on an FEV₁ improvement as defined would be unlikely to be implemented in practice. In the new stopping rule, people are permitted to continue using mannitol if their FEV₁ improves by more than 0%, that is, if their FEV₁ improves at all. The manufacturer continued to base its model on the BioGrid data, but submitted evidence in an effort to show that the BioGrid population was similar to the UK population with cystic fibrosis. For people not using rhDNase, the manufacturer used a revised utility value of 0.896. In the original model, the change in utility value was 0.015 for people using mannitol who had improved respiratory symptoms, and 0.031 for people using mannitol who did not have improved respiratory symptoms. In the revised model, the change in utility value for people not using rhDNase and who took mannitol and had improved respiratory symptoms increased to 0.025 and for people who did not use rhDNase who took mannitol and had no improvement it decreased to 0.001. In the ERG's analyses, treatment for people who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase had costs of £3,885 if they improved and £4,385 if they did not improve, a difference of £500. In the model developed in response to the ACD, treatment for people who did not use rhDNase had costs of £2,307 if they improved and £3,255 if they did not improve, a difference of £948. The manufacturer chose a baseline PDPE rate of 1.01, based on the ERG's critique of the original model.

The manufacturer did not provide revised estimates of cost effectiveness for the whole population. The revised ICER for mannitol compared with best supportive care was £19,993 per QALY gained in people not using rhDNase, and had an 82.2% probability of being cost effective at an ICER of £30,000 per QALY gained and a 46.5% probability at an ICER of £20,000 per QALY gained. Furthermore, the manufacturer provided a new analysis based on adherence rates of 30% and 70%. Keeping the QALYs constant, the ICER per QALY gained was £6,327 at a compliance of 30% and £14,137 at a compliance of 70%.

The manufacturer submitted additional clinical evidence indicating that the relative benefit with respect to FEV₁ associated with taking mannitol was maintained over 78 weeks. This evidence derived from an extension of the 2 trials. The manufacturer also provided scenario analyses showing the ICERs for shortening the model's time horizon from lifetime (100 years) to 5, 10 and 50 years to be £45,329, £25,151 and £20,018 per QALY gained respectively.

In response to the ACD, the manufacturer also conducted a survey defining the treatment pathway for managing sputum in cystic fibrosis. The manufacturer noted that this survey showed 82% of people with cystic fibrosis have trialled hypertonic saline before age 18 years. The manufacturer stated that clinicians would be reluctant to change treatments for people with well-controlled cystic fibrosis. They also noted that approximately a third of people currently using hypertonic saline did not have well-controlled cystic fibrosis and may be able to benefit from mannitol.

The ERG regarded DPM-CF-301 and DPM-CF-302 as well designed, high-quality trials, with a large combined study population. The ERG noted the change in therapeutic indication of mannitol limiting it to adults, which reduced the combined study population to 341 and consequently reduced the statistical power of all the analyses.

The ERG conducted pooled analyses on the results of the DPM-CF-301 and DPM-CF-302 trials. These showed statistically significant differences between mannitol and the control with all outcomes related to lung function. Among adults using rhDNase, the differences between mannitol and control over 26 weeks were as follows: 91.8 ml (95% CI 30.9 to 152.7) for change in FEV₁, 4.6 (95% CI 1.3 to 7.8) for percentage change in FEV₁, 2.7 (95% CI 0.9 to 4.5) for FEV₁% predicted and 106.1 ml (95% CI 28.3 to 183.9) for FVC. The ERG analysed data from adults who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, rather than from the broader group of adults not using rhDNase because that reflected the anticipated marketing authorisation at that time, but the marketing authorisation was eventually not restricted to this group. For adults who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, the change in FEV₁ from baseline was 162.3 ml (95% CI 51.8 to 272.9).

The ERG's pooled analyses showed that, for the adults using rhDNase, there were no statistically significant differences between the mannitol and control arm in incident PDPE over the 26-week trial period (relative risk [RR] 1.00, 95% CI 0.61 to 1.66), and no statistically significant difference in the estimated rate of PDPE per year (RR 1.14, 95% CI 0.75 to 1.73). In the group of people who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, there were also no statistically significant differences between mannitol and the control in the incidence of PDPE (RR 0.44, 95% CI 0.18 to 1.10) over the course of the trial. The ERG stated that restriction of the therapeutic indication to adults meant that the analysis was under-powered, a problem compounded by the post-hoc subgroup analyses of the pooled trial data, and contributed to the uncertainty around the results.

The ERG conducted an indirect comparison of mannitol with hypertonic saline, in line with the scope. The 2 measures common to the identified study of hypertonic saline (0.9% saline; Elkins et al. 2006) and the DPM-CF-301 and DPM-CF-302 trials were measurements of FEV₁ and pulmonary exacerbations. The ERG found that mannitol improved FEV₁ compared with hypertonic saline, although this was statistically significant only for the subgroup that could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase. FEV₁ was 94.3 ml (95% CI 29.0 to 159.6) higher with mannitol than with hypertonic saline, regardless of rhDNase use, and for adults using rhDNase FEV₁ was 23.8 ml (95% CI -65.0 to 112.5) higher with mannitol than with hypertonic saline.

In the ERG's view, the basic structure of the manufacturer's Markov model was appropriate for the research question, sufficiently inclusive and diverse to model the complexities of cystic fibrosis, but the ERG expressed concerns about the cost-effectiveness model.

The ERG questioned the assumption by the manufacturer in the model that mannitol use was completely independent of rhDNase use (that is, that any benefit of mannitol did not depend on whether a patient used, or did not use, rhDNase). This led the ERG to re-analyse the data according to rhDNase use and to divide the group not using rhDNase into those who could not use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase and those who could use rhDNase but did not do so for unknown reasons.

The ERG indicated that there was statistically significant heterogeneity in the overall group of people not using rhDNase. Patients who were able to take rhDNase but did not do so had different characteristics than those who were unable to take rhDNase because of ineligibility, intolerance or inadequate response to rhDNase. The ERG noted that mannitol is more likely to provide effective treatment to people who cannot rather than do not take rhDNase and who do not take hypertonic saline.

The ERG noted that the manufacturer had not used the results of the trials in the model, but instead had developed regression equations to estimate lung function. The ERG felt that the use of regression was appropriate for this Markov patient-level model, and noted that the manufacturer had consulted with experts on cystic fibrosis and modelling. The manufacturer also ran a microsimulation (100,000 trials) to compare the model output with the pooled results of the DPM-CF-301 and DPM-CF-302 trials. The ERG found small mistakes in the manufacturer's model, but noted that the validation checks matched the results of the clinical studies at a time point of 26 weeks.

The ERG noted that 1 of the most important assumptions made by the manufacturer was that any absolute improvement in FEV₁% predicted relative to patients not using mannitol would be maintained throughout the lifetime of the patient, and would directly translate into lower rates of morbidity and mortality. That is, a patient on mannitol would experience a decline in lung function over time consistent with the natural history of disease but, on cessation of mannitol treatment, would start the decline at a higher level of FEV₁% predicted than a patient not taking mannitol. The ERG was concerned that there were no long-term data to support this assumption. The ERG questioned the manufacturer's use of Australian BioGrid data for transition parameters, which may not be generalisable to the UK.

The ERG was concerned about several assumptions made by the manufacturer in the original model about pulmonary exacerbations, namely the narrow confidence intervals around the baseline rate based on the BioGrid data used in the deterministic sensitivity analysis, and questioned whether it applied to a UK population.

The ERG was concerned about assumptions made by the manufacturer in the original model that HUI2 utility and cost parameters depended on treatment, but not on health state. The ERG questioned that there was no direct link between lung function and utility and suggested that a model linking lung function to utility could have altered the cost effectiveness and could have decreased the ICER.

The ERG conducted exploratory analyses to examine the effect on the ICER of varying the model assumptions and the input parameters, including the difference in costs and utilities associated with respiratory symptoms and exacerbations, and the mortality rate of cystic fibrosis by varying the FEV₁% predicted. However, because of a lack of data, the ERG could not investigate the manufacturer's assumption that the probability of moving between health states remained the same over the lifetime of the patient.

The ERG amended the model to include treatment-independent and improvement-specific values for costs and utilities; using rhDNase subgroup-specific relative risks associating treatment with exacerbations, changing the cost of rhDNase from £16.88 to the most recent price of £16.55 (BNF 61); and adjusting model parameters, probabilities and distributions.

The ERG's exploratory cost-effectiveness analysis included the treatment options of best standard of care, rhDNase and mannitol, but not hypertonic saline. The ERG compared best standard of care with mannitol plus best standard of care. In people using rhDNase, best standard of care included rhDNase, and in people who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, best standard of care did not include rhDNase. These amendments resulted in the ERG's ICER for mannitol plus best supportive care compared with best supportive care of £80,098 per QALY gained in adults using rhDNase and £29,883 per QALY gained in adults who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase. The main reasons for the changes to the ICERs were the use of health-state specific costs and utilities used by the ERG rather than treatment specific costs and utilities used by the manufacturer, and the population specific relative risks for exacerbations.

The ERG investigated the relationship between improvements in FEV₁% predicted and survival, and found evidence to support the assumption that a 1 percentage point improvement in FEV₁% predicted was related to an approximate 5% reduction in mortality.

The ERG examined the assumption that the improvement in FEV₁% predicted caused by mannitol would be maintained over the lifetime of the patient by reducing the time horizon of the model as a proxy for a shorter duration of effectiveness. This was similar to a scenario analysis conducted by the manufacturer. The ERG's analyses in people who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase resulted in ICERs for a time horizon of 5 years of £90,126 per QALY gained. For a time horizon of 10 years, the ICER was £49,854 per QALY gained for people who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase.

The ERG pointed out that the manufacturer had generated cost data based on whether a patient received mannitol or not, rather than whether the patient was in a given health state. The manufacturer divided costs according to respiratory symptoms or according to rhDNase use, but did not estimate costs by both factors simultaneously. However, the ERG acknowledged that, in its revised model in response to the ACD, the manufacturer derived costs using patient-level data. The ERG used the information available to calculate the ratio of the improvement-specific costs to the overall mean costs as an estimate of the difference in costs by health state. The ERG calculated that patients with improved respiratory symptoms have 93% of the overall costs, whereas patients without improved symptoms have 105% of the overall costs. The ERG assumed these percentages also applied to mean costs with rhDNase. The ERG estimated 6-month treatment costs for improved and not improved respiratory symptoms in people who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase. The ERG decided that health-state specific costs should be used rather than treatment specific costs. The ERG confirmed that the manufacturer used treatment specific values in its revised analysis of people not using rhDNase.

The ERG re-ran the probabilistic sensitivity analyses with assumptions based on its exploratory analyses, varying the exacerbation rate in the control group, making the costs and utilities improvement specific rather than treatment specific, and using shorter time horizons. The ERG calculated that there was a zero probability that the ICER for mannitol would lie below £30,000 per QALY gained for people using rhDNase. For those who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase, the probability that the ICER would be below £20,000 per QALY gained was 5%, and the probability it would be below £30,000 per QALY gained was 50%.

The ERG considered the health-related quality-of-life data provided by the manufacturer in the form of HUI2 data collected in the DPM-CF-302 trial, and questioned the use of treatment-dependent values for utility. For its own sensitivity analyses, the ERG used values for utilities received from the manufacturer in response to a request for clarification and assumed that these values were independent of treatment. The ERG did not identify any other substantial health-related benefits not included in the QALY.

The factors identified by the ERG as causing substantial differences in the ICERs generated by the ERG and the manufacturer included whether or not someone took rhDNase alongside mannitol, the assumption that any improvement in FEV₁% predicted caused by mannitol would be sustained over the patient's lifetime, the assumption that patients whose condition did not respond to mannitol would discontinue therapy, and the effect of pulmonary exacerbations on utility. The manufacturer addressed these in their response to the ACD and in the second Committee meeting by changing these assumptions to be in line with those used by the ERG.

In their response to the ACD, the manufacturer provided evidence to suggest that the BioGrid data were similar to the UK population with cystic fibrosis. The ERG explained to the Committee that there was a clinically meaningful difference in the FEV₁% predicted values for the BioGrid data and the UK data (of 60.2% for the BioGrid data and 66.3% for the UK data) because every percentage point decrease in predicted FEV₁% predicted has an impact on mortality.

The ERG highlighted the additional analyses in the manufacturer's response to the ACD, which supported its assumption that improvements in FEV₁% predicted would be maintained throughout the lifetime of the patient. In the second Committee meeting, the ERG stated that there was uncertainty about whether the benefit of mannitol would persist over time, decrease at the same rate as that of the control group, or decrease at a slower rate. The ERG commented that a time horizon of 50 years was likely to accurately represent the lifetime horizon of the adult UK population with cystic fibrosis.

In examining the manufacturer's revised analysis in people who do not use rhDNase, the ERG identified 3 drivers that decreased the ICER from that in people who cannot use rhDNase because of ineligibility, intolerance or inadequate response to rhDNase (as originally modelled): including the exacerbation rate chosen for the model; acknowledging that some people treated with mannitol stop taking it ('drop-outs'); and the change in the estimated price for best supportive care, with the difference between best supportive care and mannitol being smaller in the original model (£500) than in the revised model (£948).

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