Rifaximin for preventing episodes of overt hepatic encephalopathy

Clinical effectiveness

4.3 The Committee discussed the pivotal RFHE3001 trial. The Committee was satisfied that the trial was well conducted and that relevant outcomes were assessed in line with the scope of the appraisal, including health‑related quality of life using the Chronic Liver Disease Questionnaire (CLDQ) and SF‑36 questionnaire. However, the Committee noted that mortality was not reported because the data were not considered mature enough to assess the effect of rifaximin on mortality. The Committee noted that people with more severe liver disease (Model End Stage Liver Disease [MELD] score of 25 or more) were excluded from the trial, but that the company had suggested the trial results may apply to this group. It noted the Evidence Review Group's (ERG) concerns that this was unlikely because the effectiveness of rifaximin in the subgroup of people with MELD scores of 19 to 24 was not statistically significantly better than placebo, although the Committee recognised that this subgroup comprised only 26 of the 299 people in RFHE3001. The Committee was aware that 91.4% of people in the rifaximin arm and 91.2% of people in the placebo arm had concomitant lactulose and queried whether this reflected normal practice. The clinical experts confirmed that this was in line with UK clinical practice and there was limited evidence on the efficacy of rifaximin monotherapy (8.6% of people in RFHE3001). The Committee concluded that RFHE3001 was appropriately conducted and relevant to UK clinical practice.

4.4 The Committee examined the clinical‑effectiveness data from RFHE3001, which compared rifaximin with placebo. The Committee noted that there was a statistically significant reduction in the risk of a breakthrough episode of overt hepatic encephalopathy compared with placebo for the intention‑to‑treat (ITT) population (see section 3.6). It also noted that rifaximin was associated with statistically significant reductions in the risks of hepatic encephalopathy‑related hospital admission and any increase from baseline in Conn scores. However, the improvements in asterixis score and venous ammonia levels were not statistically significant, and the differences between the rifaximin and placebo arms in changes in CLDQ fatigue scores, SF‑36 scores and Epworth Sleepiness Scale scores were minimal (see section 3.6). The Committee considered the mode of action of rifaximin. It heard from the clinical experts that apart from reducing blood ammonia levels, rifaximin also influenced endotoxin production and reduced systemic inflammation, which plays an important role in the development of hepatic encephalopathy. The Committee questioned whether this was in line with the proposed mode of action of rifaximin, but the company explained that the pathogenesis of hepatic encephalopathy remains speculative and is an area of ongoing research. The clinical experts stated that both the ammonia‑lowering and endotoxin‑lowering properties of rifaximin were relevant in producing a treatment effect. The Committee also discussed RFHE3002, which was an open‑label, follow‑up study to assess the long‑term safety and tolerability of rifaximin. It noted that not all people from RFHE3001 continued in RFHE3002, which may be a potential source of selection bias. RFHE3002 provided only exploratory effectiveness data. The Committee concluded that rifaximin was effective in preventing episodes of overt hepatic encephalopathy in the trial population.

4.5 The Committee considered whether the effectiveness of rifaximin was maintained during long‑term treatment. It understood that there is limited evidence on the long‑term effects of rifaximin. The Committee noted that although RFHE3002 was of 2‑years' duration, the lack of a control arm made interpretation of the effect of rifaximin difficult. However, it saw that the time between hepatic encephalopathy episodes progressively decreased as the number of episodes increased. The Committee considered that there was no evidence in RFHE3001 and RFHE3002 that rifaximin alters the natural history of liver disease or hepatic encephalopathy and noted comments received during consultation from the clinical expert that the greatest benefit of the drug may be obtained in the first 3 months of treatment. It noted that further studies exploring the long‑term efficacy of rifaximin were presented in the company's November 2014 submission, and considered that these studies provided limited additional evidence of the long‑term effects of the drug. The Committee heard from the clinical experts that although there is little trial evidence, in their clinical experience (based on 5 years of rifaximin use in clinical practice) rifaximin may provide long‑term benefits. It concluded that the long‑term benefits associated with rifaximin were uncertain, but that the greatest benefits would be expected in the early stages of treatment.

4.6 The Committee considered the adverse event profile associated with rifaximin in RFHE3001. It noted that hepatic encephalopathy adverse events leading to study discontinuation occurred less frequently in the rifaximin group than in the placebo group. However, anaemia, peripheral oedema, pyrexia, arthralgia and dizziness occurred more frequently in the rifaximin group than in the placebo group (see section 3.8). The Committee noted that approximately 56% of people had severe adverse events in RFHE3002. The Committee also noted from the Medicines and Healthcare Products Regulatory Agency (MHRA) public assessment report that cases of diarrhoeaassociatedwith Clostridium difficile have been reported with the use of rifaximin. It understood that this was a potential ongoing safety concern with the use of antibacterial agents including rifaximin. It heard from the company that the MHRA public assessment report stated that there were no new safety concerns with rifaximin and the benefit–risk profile was considered to be positive. It noted the patient expert's statement that the potential side effects of rifaximin were considered to be more tolerable than the physical and psychological side effects associated with current treatment (such as diarrhoea associated with lactulose treatment), which have a detrimental effect on quality of life for people with the condition and their carers. The Committee concluded that the current evidence indicates that rifaximin has an acceptable adverse event profile.

4.7 The Committee considered evidence from clinical audits of rifaximin, provided by the clinical expert in response to the second consultation and by the company in its November 2014 submission. The Committee highlighted that these retrospective observational studies were challenging to interpret. It noted that outcomes were compared before and after starting treatment with rifaximin and therefore could be influenced by many factors. In particular, when rifaximin treatment was started any of a number of factors may also have changed, which could have contributed to the benefits seen with rifaximin; for example:

4.8 In its review of the evidence from clinical audits, the Committee considered the effect of rifaximin on hospital admissions. It understood that the analyses of hospital admissions included all emergency admissions, and therefore it noted that the results may be influenced by admissions unrelated to hepatic encephalopathy. In light of this and the possible confounding factors (see section 4.7), the Committee considered that the results were very uncertain. The Committee noted that the results suggested that treatment with rifaximin reduces the duration of hospital stay, compared with treatment without rifaximin. It heard from the clinical expert that this may be because people admitted with an episode of hepatic encephalopathy while taking rifaximin tend to have less severe symptoms than people taking lactulose. The Committee acknowledged that the results from the audits were subject to uncertainty, but concluded that rifaximin is likely to reduce hospital admissions and may shorten the length of hospital stay.

Cost effectiveness

4.9 The Committee considered the company's economic model and the ERG's critiques of the company's submissions. It noted that the company presented 4 iterations of its model over the course of the appraisal (referred to in this document as the 'original', 'October 2013', 'December 2013' and 'November 2014' analyses; see section 3.16). It accepted the exclusion of neomycin from the analysis because it was not routinely used in clinical practice. The Committee noted that in the October 2013 analysis the company amended the way it estimated the risk of subsequent episodes and that this model assumed that the risk of subsequent hepatic encephalopathy episodes depended on time since the first episode. The Committee was aware that all subsequent episodes were combined in 1 health state (the subsequent overt state), thereby not reflecting the number of episodes. However, it noted that including more health states for subsequent episodes would increase the complexity of the modeland would be difficult to populate with the current evidence base. The Committee therefore concluded that, although the number of episodes was not considered and the company's original analysis oversimplified the nature and course of the disease, the October 2013 model was an improvement on the original analysis and was appropriate for decision‑making.

4.10 The Committee discussed the most appropriate time horizon for the model. It noted that in the company's submissions, a number of different time horizons were presented, including 5‑year, 10‑year and lifetime. It also noted that in the October 2013 analysis 52% of people treated with rifaximin and lactulose (referred to in this document as the rifaximin arm) and 45% of people treated with placebo and lactulose (referred to in this document as the lactulose arm) were predicted as being still alive in the model after 5 years. The Committee was aware that the NICE reference case (Guide to the methods of technology appraisal 2013) indicates a preference for a lifetime time horizon when alternative technologies lead to differences in survival or benefits that persist for the remainder of a person's life. It noted that the ERG had conducted an exploratory analysis based on the company's October 2013 model, incorporating a lifetime time horizon of 40 years. It also noted that the company incorporated a lifetime time horizon (of 42 years) in its December 2013 and November 2014 analyses, in response to a request from the Committee. The Committee heard from the clinical expert that people with hepatic encephalopathy would not survive for up to 40 years because of their underlying liver cirrhosis; the Committee noted that a time horizon of 40 years did not imply that all people survived for 40 years, and heard from the company that most people in the model died within the first 5 to 10 years. It considered that the prolonged survival in the model could be a result of extrapolating outcomes using the log normal distribution, which has a long tail and could lead to implausible survival results, although few people remained alive in the later stages of the model. The Committee concluded that the company's choice of a 5‑year time horizon in its October 2013 analysis was not in line with the NICE reference case, and that the lifetime time horizon presented in the December 2013 and November 2014 analyses was appropriate.

4.11 The Committee discussed the assumptions used to model clinical outcomes. It noted that, in its October 2013 analysis (and all subsequent analyses), the company used data from RFHE3001 only to model breakthrough overt hepatic encephalopathy episodes for both arms of the model, rather than the combined data set from RFHE3001 and RFHE3002 used to model the rifaximin arm in the original analysis. The Committee also noted the ERG's comment that data censoring when modelling time to first breakthrough overt episode was inconsistent between the October 2013 analysis and the original analyses, that is, at day 170 and day 168 respectively. However, it considered the October 2013 analysis at day 170 to be more appropriate given that the clinical study report for RFHE3001 reports data censoring at day 170. It also noted that the company included only newly enrolled people from RFHE3002 when modelling subsequent overt episodes to avoid potential enrichment bias from rollover and crossover people who did not have an overt episode in RFHE3001. The Committee was satisfied with this approach given the evidence presented by the company showing that time to first breakthrough overt episode was not statistically significantly different between the RFHE3002 subgroups (rollover, crossover and newly enrolled people; see section 3.19). The Committee noted that there was less variation in the incremental cost‑effectiveness ratios (ICERs) from the October 2013 analysis (see section 3.31) than those of the original analysis when different distributions were used to model time to first breakthrough and subsequent overt episodes. The Committee concluded that the company's modelling of hepatic encephalopathy episodes in the October 2013 and subsequent submissions was appropriate.

4.12 The Committee considered the effect of rifaximin on hospital admissions in the model. It understood that the company's model predicted a reduced rate of hospital admission with rifaximin compared with lactulose, and noted that comments from consultation supported this. The Committee was aware that, in the October 2013 and December 2013 submissions, the company assumed that 53% of episodes of hepatic encephalopathy led to admission, with a length of hospital stay of 5 days per admission, in both arms of the model. It also saw that, in the November 2014 submission and ERG review, these assumptions were varied to match the number of hospital admissions and bed days seen in clinical audits. To do this, the company identified values for the length of stay in hospital and the probability of hospital admission so that the model simulated the results of the audits. The Committee therefore discussed the validity of using the audit data, and the most appropriate estimates for the length of stay and probability of hospital admission:

4.13 The Committee considered how the company included adverse events in the model. The Committee noted that adverse events were excluded from the cost calculation on the basis that there were no statistically significant differences between the treatment arms in RFHE3001 and there was limited evidence for disutilities associated with the adverse events. It heard from the clinical experts that adverse events were minimal and primarily related to lactulose use. The ERG stated that including adverse events was not likely to have a large impact on the cost‑effectiveness results. The Committee was uncertain of the impact on the ICER and concluded that an attempt should have been made to capture some of the differences in the costs of adverse events between the rifaximin and lactulose arms.

4.14 The Committee considered the approaches taken to estimate the utility scores for use in the model. It noted that in the October 2013 analysis, utilities were based on quality‑of‑life data collected from RFHE3001, derived using both SF‑36 and CLDQ scores. It expressed several concerns about this approach. In particular, it considered that including CLDQ scores in the mapping process unnecessarily introduced uncertainties, was associated with missing data, and was based on a post‑hoc analysis of CLDQ. It highlighted that the post‑hoc analysis identified statistically significant differences that were not seen in the per‑protocol analysis. The Committee also expressed further concerns relating to the choice of a single, linear regression equation, the imputation of missing data, and the use of overall scores rather than individual domains. The Committee did not accept the company's views that the high variance in SF‑36 scores supported the use of the CLDQ, or that the SF‑36 was necessarily influenced substantially more than the CLDQ by issues with recall. The Committee noted the company's scenario analysis in which utilities were derived from SF‑36 data only (December 2013 submission). Although it was aware that the company retained a preference for the analysis that used both SF‑36 and CLDQ data, the Committee considered that the approach based on SF‑36 data only was more appropriate. The Committee concluded that the most appropriate approach for deriving the utility scores was the analysis based on SF‑36 data only.

4.15 The Committee considered the utility scores that were used in the model. The Committee noted that the estimates in the October 2013 and December 2013 analyses implied a quality‑of‑life improvement (utility increment) associated with rifaximin in the remission states. It understood that scenario analyses presented by the ERG in which the utility increment for rifaximin was omitted showed that the model results were highly sensitive to this value. The Committee heard from the clinical expert that remission (in particular, Conn score 1) is also known as minimal hepatic encephalopathy and people still have mild neurological abnormalities, reversal of the sleep‑wake cycle and reduced quality of life. It heard that lower Conn scores do not indicate normal quality of life, although a Conn score of 0 is defined as 'no personality or behavioural abnormality detected'. The Committee heard from the clinical experts that rifaximin improves quality of life and could enable people to return to their normal activities with less dependence on carers. The Committee considered that it was plausible that rifaximin could improve quality of life in the remission states, and understood that the model results were highly sensitive to the size of improvement.

4.16 The Committee reviewed in detail the estimates for the utility increment associated with rifaximin presented by the company. The Committee expressed concerns about the face validity of the utility increment estimated using both CLDQ and SF‑36 data (0.106), noting that this was larger than the difference between the overt and remission states in the October 2013 analysis. The Committee recognised that rifaximin may improve quality of life in the remission states (see section 4.15), and given that this improvement could be seen in clinical practice, the associated utility increment could represent a minimum important difference (MID) in utility value. The Committee noted evidence presented by the company in its November 2014 analyses (which explored MIDs in EQ‑5D values across a range of conditions), from which it had suggested applying a utility increment of 0.074 to match the MID. The Committee noted the wide range of estimates for MID presented, and understood that that these were from a number of different conditions (none of which were related to hepatic encephalopathy) and were therefore difficult to interpret. Taking into account limitations of the evidence on MIDs and the analysis based on both CLDQ and SF‑36 data (see section 4.14), the Committee considered that a utility benefit of 0.106 or 0.074 in the remission states was not reliable. The Committee considered that the estimated utility increment derived using SF‑36 data only (0.032) was more appropriate, but was aware of the uncertainties surrounding this value. The Committee understood that the differences between placebo and rifaximin seen in the baseline‑adjusted analysis of SF‑36 were not statistically significant, and that the utility increment used in the company's economic model reflected the highest value for the difference between rifaximin and placebo that had been seen in this analysis. Given that it had heard that the greatest benefits associated with rifaximin may be obtained in the first 3 months, the Committee considered that incorporating a more conservative value such as that seen at month 3 in the baseline‑adjusted analysis might have been informative. Furthermore, noting its conclusion that the long‑term effectiveness of rifaximin is uncertain (see section 4.5), the Committee considered that even if people did initially experience the maximal utility increment as modelled by the company, it was uncertain whether this increment would be maintained at that value throughout the full 42‑year time horizon of the economic model. The Committee concluded that it was plausible that rifaximin could be associated with a utility increment in the remission states and that there are substantial challenges in identifying the most plausible value for this utility increment. It further concluded that the increment could be lower than the highest value for the difference between rifaximin and placebo in the baseline‑adjusted analysis of SF‑36 (0.032) and could plausibly decrease further during long‑term treatment, but that the most appropriate utility increment for decision‑making was 0.032.

4.17 The Committee considered the other utility estimates derived using SF‑36 data only, presented in the company's December 2013 submission. It noted that the ERG was unclear of the origin of the utility value for the remission states in the lactulose arm (0.568), and heard from the company that this was the average baseline utility score for all people in the analysis. The Committee also noted that the difference in utility between the remission and overt states (0.286) was taken from a survey of the general population using standard gamble methods, whereas the time trade‑off method was used previously. However, it understood from the ERG that the results of the model were not sensitive to this and it considered the company's utility estimate to be broadly reasonable. The Committee understood that there are substantial challenges in capturing quality‑of‑life evidence in people with hepatic encephalopathy, but concluded that the utility estimates had captured the reduced quality of life associated with hepatic encephalopathy.

4.18 The Committee considered the plausibility of the mortality estimates in the company's models. It noted that all of the models presented by the company predicted that rifaximin was associated with a mortality benefit. It recognised that this mortality benefit came from avoiding overt episodes (which were associated with a higher mortality rate than the remission states) rather than an explicit survival benefit applied to the rifaximin arm. The Committee considered whether this was clinically plausible. It noted that no differences in mortality were seen between rifaximin and placebo in RFHE3001. The Committee noted that data showing a mortality benefit associated with rifaximin were presented by the company in its October 2013 submission. However, it was concerned that these data may not provide robust evidence because they were based on a retrospective observational study (Neff et al. 2012) and that mortality was substantially higher than seen in RFHE3001. The Committee also considered that the finding that mortality was higher with rifaximin plus lactulose than with rifaximin alone was unexpected. However, the Committee heard from the clinical expert that it was plausible for rifaximin to affect mortality and that the modelled difference between rifaximin and lactulose seemed reasonable; the clinical expert reported positive experiences with people treated for up to 5 years. The Committee concluded that although a mortality benefit had not been seen in RFHE3001, it was willing to accept that an initial mortality benefit with rifaximin resulting from a reduction in overt hepatic encephalopathy episodes was plausible. However, it considered that the magnitude of the effect is uncertain and that it is likely to decrease over time.

4.19 The Committee considered how mortality had been modelled by the company, noting from ERG scenario analyses (see section 3.44) that the model was sensitive to mortality estimates:

4.20 The Committee discussed the effects of the uncertainties about the mortality and utility assumptions and hospital admissions on the results of the economic model. It noted that none of the ICERs presented took into account all of the uncertainties and all of the Committee's preferred assumptions. It noted that in the scenario based on its preferred method for calculating utility, the ICER was £29,000 per QALY gained for rifaximin compared with lactulose (see section 3.32). However, it noted that if the utility increment associated with rifaximin in the remission states were lower than the estimate of 0.032, or reduced over time, the ICER for rifaximin would increase (see section 3.48). In addition, the Committee noted that the mortality benefit of rifaximin in the model was likely to be overestimated, and that accounting for this by reducing the mortality benefit for rifaximin would increase the ICER (see section 4.19). However, the Committee also noted that incorporating a reduced length of hospital stay for rifaximin, compared with lactulose, would reduce the ICER. The Committee concluded that, on balance, the most plausible ICER was likely to be close to the top end of the range normally considered cost effective.

4.21 The Committee acknowledged that with an ICER close to the top end of the range normally considered cost effective, it needs to identify an increasingly strong case for supporting the technology as an effective use of NHS resources, taking into account:

4.22 With these factors in mind, the Committee considered whether rifaximin reflected a cost‑effective use of NHS resources. Although it understood that the most plausible ICER was at the top end of the range that is normally considered acceptable and was subject to a number of uncertainties, it was also aware of the important unmet medical need in this group of vulnerable people for whom there are few treatment options. It also acknowledged the innovative aspects of this treatment. Consequently, the Committee concluded that rifaximin could be considered a cost‑effective use of NHS resources for reducing the recurrence of episodes of overt hepatic encephalopathy in people aged 18 years or older.

4.23 The Committee considered whether the appraisal might be affected by any issues relating to equality. It noted comments received during consultation from the clinical expert that people with hepatic encephalopathy should be considered vulnerable adults. The Committee understood that this condition can have a substantial disabling effect, but considered that its recommendations do not discriminate on the basis of any characteristics protected under the equalities legislation.

Summary of Appraisal Committee's key conclusions