3 The company's submission

The Appraisal Committee (section 7) considered evidence submitted by the company of nalmefene and a review of this submission by the Evidence Review Group (ERG; section 8).

Clinical effectiveness

Nalmefene compared with psychological intervention

3.1 The company identified 3 randomised controlled trials (ESENSE1, ESENSE2 and SENSE) in adults with alcohol dependence, comparing 18 mg nalmefene (on an as‑needed basis) plus psychosocial support with placebo plus psychosocial support. ESENSE1 (n=604) and ESENSE2 (n=718) were identical efficacy studies with a follow‑up period of 24 weeks. SENSE (n=675) was primarily designed to collect safety data for up to 12 months on nalmefene, but after the study had started the protocol was amended to include efficacy analyses. SENSE had a follow‑up period of 12 months.

3.2 Psychosocial support (in the form of BRENDA), focusing on treatment adherence and reduction of alcohol consumption, was provided to all treatment groups in the 3 studies. The first part comprised a biopsychosocial evaluation, followed by sharing the results with the patient. The next stage involved expressing empathy for the patient and together identifying their needs, providing direct advice to the patient to meet those needs, assessing patient reaction to advice and adjusting the treatment plan as needed. All sessions were provided by trained professionals and were delivered at weekly intervals for the first 2 weeks and then monthly. Sessions lasted for 15–30 minutes except for the first longer session, which was 30–40 minutes.

3.3 Alcohol dependence was diagnosed using the Diagnostic and Statistical Manual of Mental Disorders (DSM‑IV‑TR). To be included in the studies, patients must have had 14 or fewer days of abstinence in the 28 days preceding the screening visit, and have an average daily alcohol consumption of medium risk or higher: equivalent to more than 40 g per day (equivalent to more than 5 units) for men and more than 20 g per day (equivalent to more than 2.5 units) for women. Patients had at least 6 heavy drinking days in the 28 days prior to enrolment. A heavy drinking day was defined, in line with the World Health Organization classification of drinking risk levels, as alcohol consumption of more than 60 g per day (equivalent to more than 7.5 units) for men and more than 40 g per day (equivalent to more than 5 units) for women. People with severe medical comorbidities were excluded from all 3 studies, and those with severe psychiatric comorbidities were excluded from the 2 ESENSE trials. The 3 studies were conducted across different regions of Europe. In total, there were 156  sites; 5 sites in the UK were included in the SENSE trial.

3.4 The ESENSE trials contained 4 study periods. The first was a 1–2 week screening period, after which all patients were randomised 1:1 to either the nalmefene plus BRENDA group or placebo plus BRENDA group for 24 weeks. Patients were then instructed to take 1 tablet (the maximum daily dose) on an 'as‑needed' basis, preferably 1–2 hours before they perceived a risk of drinking. If the patients started to drink without taking a tablet, they were advised to take a tablet as soon as possible. The patients who completed the 24‑week trial entered a 4‑week, double‑blind, run‑out period to evaluate any treatment discontinuation effects. Those who had been initially randomised to nalmefene were re‑randomised to receive either nalmefene or placebo, and patients originally in the placebo group continued on placebo. A safety follow‑up visit was scheduled for 4 weeks after completion of the run‑out period or after withdrawal from the study.

3.5 Similar to the ESENSE studies, the SENSE study also began with a 1–2 week screening period, after which patients were randomised 3:1 to receive 52 weeks of as‑needed treatment with nalmefene plus BRENDA or placebo plus BRENDA. A safety follow‑up period was scheduled for 4 weeks after completion of the study or after withdrawal from the study.

3.6 The primary outcomes in ESENSE1 and ESENSE2 measured changes from baseline in the number of heavy drinking days per month and total alcohol consumption at month 6. The company highlighted that the primary end points of number of heavy drinking days and total alcohol consumption were in accordance with the recommendations in the European Medicines Agency guideline on the development of medicinal products for the treatment of alcohol dependence. Total alcohol consumption was defined as mean daily alcohol consumption in grams per day, over a month (28 days). Patients self‑reported their daily alcohol consumption using the timeline follow‑back method at monthly intervals. This provided retrospective estimates of the number of standard drinks consumed each day in the previous month, which were subsequently converted into grams of alcohol per day. Secondary outcomes included the effect of nalmefene on: proportion of people whose alcohol dependence responded to treatment based on different drinking measures, alcohol dependence symptoms and clinical status, liver function and other clinical safety laboratory tests, pharmaco‑economic outcomes, treatment withdrawal effects after 24 weeks, safety and tolerability of nalmefene and quality‑of‑life measures (SF‑36 and EQ‑5D).

3.7 Similar to ESENSE1 and ESENSE2, the primary outcomes for the SENSE study were change from baseline in the number of heavy drinking days per month and total alcohol consumption at month 6. These outcomes were added as an amendment to the protocol while the study was ongoing. No protocol amendments were made to outcomes to assess the safety and tolerability of nalmefene.

3.8 In ESENSE1 and ESENSE2, approximately 78% of all patients enrolled had a high or very high drinking risk level at baseline. In SENSE, 52% of the enrolled patients had a high or very high drinking risk level at baseline. In ESENSE1, ESENSE2 and SENSE, 74%, 57% and 52% respectively continued drinking at this level at randomisation. After an agreement with the Scientific Advisory Group to the European Medicines Agency, the company performed a post hoc analysis in the subgroup of patients in the 3 studies who had a high or very high drinking risk level both at baseline and at randomisation. The company stated that the Scientific Advisory Group recognised the validity of the post hoc subgroup analyses and that these analyses form the basis of the marketing authorisation for nalmefene.

3.9 Results of the post hoc analyses in the licensed population (that is, people who had a high or very high drinking risk level at baseline and maintained such a level at randomisation) showed that there were greater reductions in the number of heavy drinking days and total alcohol consumption in patients treated with nalmefene plus BRENDA, than with placebo plus BRENDA. The treatment difference in the changes from baseline to 6 months in the number of heavy drinking days, using mixed model repeated measures analysis, was −3.7 days per month (95% confidence interval [CI] −5.9 to −1.5, p=0.001) in ESENSE1, and −2.7 days per month (95% CI −5.0 to −0.3, p=0.025) in ESENSE2. The treatment difference in the changes from baseline to 6 months in total alcohol consumption was −18.3 g per day (95% CI −26.9 to −9.7, p<0.001) in ESENSE1, and −10.3 g per day (95% CI −20.2 to −0.5, p=0.040) in ESENSE2. In the SENSE study, the treatment difference in the changes from baseline to 6 months in the number of heavy drinking days was −2.6  days per month (95% CI −5.5 to 0.2, p=0.071) at 6 months, and −3.6 days per month (95% CI −6.5 to −0.7, p=0.016) at month 13. The difference in total alcohol consumption at month 6 was −15.3 g per day (95% CI −29.1 to −1.5, p=0.031) and at month 13 was −17.3 g per day (95% CI −30.9 to −3.8, p=0.013).

3.10 The company did not perform a meta‑analysis of the efficacy data for the ESENSE1, ESENSE2 and SENSE studies but pooled the primary outcomes, the change from baseline to month 6 in monthly heavy drinking days, and total alcohol consumption from ESENSE1 and ESENSE2. In the ESENSE1 and ESENSE2 studies there were 23 heavy drinking days per month at baseline in the nalmefene plus BRENDA group with a reduction to 10 heavy drinking days per month at month 6 (a reduction of 55%). In the placebo plus BRENDA group there were 22 heavy drinking days per month at baseline with a reduction to 13 heavy drinking days per month at month 6 (a reduction of 42%). At 6 months the number of heavy drinking days had been reduced by 3.01 days per month (95% CI −4.36 to −1.66, p<0.0001) and total alcohol consumption had been reduced by 14.22 g per day (95% CI −19.96 to −8.47, p<0.0001). In the ESENSE1 and ESENSE2 studies there was a total alcohol consumption of 107.7 g per day in the nalmefene plus BRENDA group, which reduced to 49.0 g per day at month 6 (a reduction of 61%). In the placebo plus BRENDA group there was a total alcohol consumption of 103.3 g per day, which reduced to 51.9 g per day at month 6 (a reduction of 50%). The odds ratio for the pooled response of drinking risk level for the ESENSE1 and ESENSE2 trials was 1.87 (95% CI 1.35 to 2.59, p<0.001).

3.11 The company reported the results for a number of secondary outcomes in the 3 nalmefene studies. Secondary outcomes included response at month 6 (response of drinking risk level defined as a downward shift from baseline in drinking risk level by 2 risk categories). The odds ratio for nalmefene for response of drinking risk level was 2.15 (95% CI 1.38 to 3.36, p<0.001) in the ESENSE1 study and 1.59 (95% CI 0.98 to 2.59, p=0.062) in the ESENSE2 study. In ESENSE1 and ESENSE2, the EQ‑5D health state and utility index score in the licensed population increased more from baseline to month 6 in the nalmefene plus BRENDA group than in the placebo plus BRENDA group. This was statistically significantly in favour of nalmefene for the health state score in ESENSE1 only. Pooled analysis of the EQ‑5D (a quality of life questionnaire) results in ESENSE1 and ESENSE2 in the licensed population produced a mean change from baseline, for the health state score and the utility index score, of 3.46 points (p=0.0124) for the health state score and 0.03 points (p=0.0445) for the utility index score. The EQ‑5D health state and utility index score in the licensed population increased more from baseline to month 6 in the nalmefene group than in the placebo group with a mean change in utility index score from baseline to month 6 of 0.03±0.02 (95% CI 0.00 to 0.06, p=0.0445) and a mean change in health state score from baseline to month 6 of 3.46±1.38 (95% CI 0.75 to 6.17, p=0.0124).

Nalmefene compared with naltrexone

3.12 Because there were no direct head‑to‑head studies comparing nalmefene plus BRENDA with naltrexone (comparator) plus psychosocial intervention, the company investigated whether a network meta‑analysis or indirect comparison could be conducted. The company carried out a systematic review to identify studies evaluating nalmefene and naltrexone for the reduction of alcohol consumption in people who were actively drinking and had alcohol dependence. The review identified 3 randomised controlled studies that compared oral naltrexone (50 mg per day) plus psychosocial intervention, with placebo plus psychosocial intervention in actively drinking adults with alcohol dependence. The company stated that all the studies had limitations in the data reported, meaning that an indirect comparison could not be performed. These differences included study design, inclusion and exclusion criteria, study objective and end points as well as a lack of reporting of data from the naltrexone studies.

BRENDA (psychosocial support in ESENSE1, ESENSE2 and SENSE) compared with other types of psychological interventions

3.13 To determine which types of psychosocial intervention should be included in the systematic review, the company carried out a survey of 20 primary care practices and experts and concluded that the following types of psychosocial intervention should be incorporated: cognitive behavioural therapies, behavioural therapies, social network and environment therapies, brief interventions and motivational enhancement therapy.

3.14 The company carried out a literature search and identified 7 studies on psychosocial intervention that met the inclusion criteria and which the company added to the 43 studies identified in the NICE guideline on alcohol-use disorders. The company did not carry out a meta‑analysis of these studies (no explicit reasons were provided in the company's submission) but it did provide a summary of the absolute reductions in drinking that were provided in the psychosocial intervention trials. These trials showed that absolute reduction in total alcohol consumption from these studies ranged from 9.3–50.7 g per day, with a median value of 18.3 g per day (range of follow‑up time: 6–12 months). For the absolute reduction in number of monthly heavy drinking days, the range was 1.3–19, with a median value of 5.7 days (range of follow‑up time: 3–12 months). In the nalmefene studies, the absolute reduction in total alcohol consumption in the nalmefene plus BRENDA group ranged from 58.3–70.4 g per day, whereas in the placebo plus BRENDA group, the absolute reduction ranged from 40.0–60.1 g per day. The absolute reduction in the number of monthly heavy drinking days in the nalmefene plus BRENDA group ranged from 11.6–12.9 days, whereas in the placebo plus BRENDA group the absolute reduction ranged from 8.0–10.2 days (range of follow‑up time: 6–12 months).

3.15 The frequency of treatment‑emergent adverse events was recorded for all 3 nalmefene trials for both the total and licensed population. The percentage of adverse events was slightly higher in the licensed population than in the total population. The adverse events observed with the highest incidences in the nalmefene group as compared with the placebo group were nausea, dizziness, insomnia and headache. The incidence of nausea (22%) and dizziness (18%) were high in the first month of treatment but decreased to approximately 1–2% in subsequent months. Treatment‑emergent psychiatric events that included confusion, abnormal thinking and hallucinations were approximately 3 times more common with nalmefene, with an incidence of 2.9%.

Cost effectiveness

3.16 The company developed a de novo analysis to estimate the cost effectiveness of as needed nalmefene plus psychosocial support compared with psychosocial support alone for treating alcohol dependence. The company used a Markov model, which consisted of a short‑term model (1 year based on the nalmefene studies) with 1 month cycles, and a long‑term model (up to 5 years using extrapolated trial results) with 1 year cycles. The model with 1 month cycles aimed to take account of treatment efficacy and patient adherence, observed treatment discontinuation, incidence of alcohol‑attributed harmful events and deaths. It also reduced the number of assumptions and uncertainties needed by the company. The 1 month cycle length was used to align with the patient follow‑up in the nalmefene studies (number of heavy drinking days and total alcohol consumption over 28 days). Half‑cycle correction was not incorporated because the company considered these to be negligible, because the initial cycles were 1 month long. The model was developed based on the nalmefene studies that used BRENDA as the psychosocial support.

3.17 The population in the model consisted of a cohort with alcohol dependence and defined drinking levels according to the World Health Organization's definition of drinking risk levels (see table 1). In accordance with the pooled data from ESENSE1, ESENSE2 and the SENSE studies, the company assumed that on entry to the model, 57.5% of those patients who met the criteria specified in the marketing authorisation for nalmefene, would be in the very high risk drinking level and 42.5% would be in the high risk drinking level.

Table 1 World Health Organization definition of drinking risk levels

Drinking risk level (applies to a single day)

Total consumption (g/day)

Men

Women

Very high risk

>100

>60

High risk

>60–100

>40–60

Medium risk

>40–60

>20–40

Low risk

1–40

1–20

Abstinent

0

0

3.18 The short‑term time horizon of 1 year contained 5 drinking level health states as shown in table 1. Patients entered the model in either the high or very high drinking level state in line with the marketing authorisation for nalmefene. After the first year, 3 yearly health states were considered: controlled drinking, medium risk drinking, and high or very high risk drinking. Patients in the controlled drinking health state were assumed to be of a low risk drinking level or abstinent after 12 months and therefore these patients stopped all treatments.

3.19 To account for the possibility that patients with controlled drinking may become heavy drinkers again, 19% were modelled to relapse at the end of the year and due to have a second round of treatment. Patients who relapsed returned to the same treatment in which they were initially successful in controlling their alcohol intake. The proportion of patients who relapsed was also distributed among the drinking levels in the same way as the initial patient cohort in the model. The same transition probabilities were also applied. It was assumed that treatment was effective in patients in the medium risk drinking level group after 12 months, and patients continued on treatment but this only applied to approximately 10% of patients in the model. These patients could transition to either controlled drinking or high or very high risk drinking level, leading to a second‑line treatment option. After 12 months, it was presumed that treatment was not effective in patients in the high or very high risk drinking group and their current treatment was stopped. They were modelled to change treatment strategy to an abstinence‑orientated or second‑line approach, which would include assisted alcohol withdrawal followed by acamprosate or oral naltrexone plus psychosocial intervention, to prevent relapse.

3.20 Transition probabilities for patients changing drinking state in the first year were obtained from pooled data from the ESENSE1, ESENSE2 and SENSE studies. Transition probabilities for the subsequent years were obtained from different sources, depending on the drinking risk level. The abstinent or low drinking risk levels were based on data reported by Taylor et al. (1985), with the transition probabilities for those in the medium drinking risk level calculated from the last 6 months of the SENSE study.

3.21 The risk of a patient experiencing a serious or temporary harmful event was related to their World Health Organization drinking risk level. The serious harmful events included by the company were based firstly on those events that were costly to the healthcare system and had a strong evidence base. The company also modelled temporary events using tunnel states including costs and quality‑adjusted life year (QALY) decrements but no long‑term effects were accounted for when the person survived the tunnel state. Temporary events comprised of lower respiratory tract infections, transport‑related injuries and injuries not related to transport. Patients who experienced a serious event stayed in that state for the remaining duration of the model. Patients who experienced a temporary event stayed in a tunnel health state for 1 month before returning to the pre‑tunnel health state. In a tunnel state, the proportion of patients passing through the state (or event) acquired costs and an immediate decrement in utility, in addition to other costs (alcohol treatment costs) and utilities incurred by the drinking level health states. However, the state or event will not produce any long‑term effects as long as the patient survives the tunnel state.

3.22 To take account of the risks of crime in the first year of treatment, the company applied relative risks for each drinking risk level to an underlying general population value, which is assumed to be those patients that are abstinent. The company assumed a number of probabilities of committing crime based on gender in the first year.

3.23 The company's model allowed patients to move from any health state to the death state over the time horizon. Patients could die either from alcohol‑attributed harmful events or all‑cause mortality.

3.24 The model also incorporated risks of dropping out because of harmful events from nalmefene or other reasons. This was based on data from the 3 pooled nalmefene clinical trials in the model. An adverse event could cause the patient to change or stop their treatment, depending on the treatment and the source of the adverse event. If a patient dropped out because of nalmefene‑related adverse events, they stayed in the nalmefene treatment arm but their treatment changed to psychosocial support only. Patients who changed treatment because of nalmefene‑related adverse events transitioned to their corresponding drinking level for the psychosocial support treatment. For both the nalmefene plus psychosocial support treatment and the psychosocial support alone, patients who dropped out because of other reasons had their treatment changed to 'no treatment' and transitioned immediately to high or very high World Health Organization drinking risk level with the same distribution as at entry into the model.

3.25 A number of cost parameters were used in the model, with the cost of a visit to the GP or expert care being the same for both nalmefene plus psychosocial support and psychosocial support alone. For both these groups, the proportion of patients receiving treatment at a GP practice and at expert level was set at 75% and 25% respectively.

3.26 In the model, the costs of second‑line treatment with naltrexone or acamprosate were taken from the NICE guideline on alcohol-use disorders. The second‑line treatment for assisted withdrawal using naltrexone or acamprosate had several costs attached, depending on the location of treatment: home‑based assisted withdrawal (£596), secondary care outpatient‑assisted withdrawal (£606) or secondary care inpatient‑assisted withdrawal (£4145). The company then used a weighted average of £1044 per patient having medically assisted withdrawal. The model also took into account societal costs related to both crime and productivity as specified in the remit to NICE from the Department of Health. The inclusion of a societal perspective was taken account of in scenario analyses and was not included in the company's base case.

3.27 Utility weights were obtained from the EQ‑5D questionnaire, used to assess patients' health‑related quality of life in the 3 nalmefene trials. The EQ‑5D data were used to model the effect of a reduction in alcohol consumption. The results from the 3 trials were pooled to estimate utility values for the cost‑effectiveness model (see section 3.11 for results).

3.28 The company's base‑case results showed that nalmefene plus psychosocial support dominated psychosocial support alone (that is, it is more effective and less costly). The company carried out a number of sensitivity analyses. The parameters that had the most effect on the cost effectiveness results were the number of medical visits per month (for both treatments), the proportion of people having treatment following relapse, the utility values used and the cost of nalmefene. Nalmefene plus psychosocial support still dominated when all parameters were varied, except for when the number of medical visits per month was doubled. When applying the upper bound for this parameter, the incremental cost‑effectiveness ratio (ICER) increased to £6274 per QALY gained.

3.29 The company also tested 8 different scenarios observing the impact of varying the time horizon, perspective on cost, assuming nalmefene intake on every day that the patient was in the model, source of utility data used and removing the second‑line treatment option (results in brackets after each scenario).

  • Scenario 1: Time horizon reduced to 1 year (ICER was £24,684 per QALY gained for nalmefene plus psychosocial support compared with psychosocial support).

  • Scenario 2: Societal perspective included (nalmefene plus psychosocial support continued to dominate psychosocial support).

  • Scenario 3: Time horizon reduced to 1 year and societal perspective included (nalmefene plus psychosocial support continued to dominate psychosocial support).

  • Scenario 4: Nalmefene intake assumed to be every day rather than as needed (ICER was £289 per QALY gained for nalmefene plus psychosocial support compared with psychosocial support).

  • Scenario 5: No second‑line treatment options are allowed (ICER was £5090 per QALY gained for nalmefene plus psychosocial support compared with psychosocial support).

  • Scenario 6: Using utility values from the STREAM study (nalmefene plus psychosocial support continued to dominate psychosocial support).

  • Scenario 7: A threshold analysis increasing the treatment effect of psychosocial support relative to nalmefene plus psychosocial support to identify the level of efficacy needed to have an ICER of £20,000 and of £30,000 per QALY gained.

  • Scenario 8: An assumption that psychosocial support was associated with zero costs (£8088 cost per QALY gained for nalmefene plus psychosocial support compared with psychosocial support).

3.30 After a clarification request, the company corrected a minor error in the model and presented 2 further scenarios (termed scenarios 9 and 10 by the ERG). Scenario 9 provided an ICER for the use of psychosocial intervention as suggested by the NICE guideline on alcohol-use disorders, with 1 session of psychosocial intervention lasting 60 minutes per week for 12 weeks. Scenario 9A increased the costs of psychosocial support in the psychosocial support alone arm, whereas scenario 9B assumed the cost increase for psychosocial support applied to both nalmefene plus psychosocial support arm and psychosocial support alone arm. In both situations (9A and 9B), nalmefene plus psychosocial support dominated psychosocial support alone. Scenario 10 assessed alternative assumptions for the treatment pathway of patients at a medium risk level after 12 months. Three scenarios were explored: the first assumed that patients relapse after 12 months to high or very high drinking risk level; the second assumed that treatment was effective and was modelled in line with other patients in whom treatment was effective; the third scenario assumed that treatment was not effective in patients in the nalmefene plus psychosocial support arm but that it was for patients in the psychosocial support alone arm. For the first 2 scenarios, nalmefene plus psychosocial support still dominated psychosocial support alone, whereas for the third scenario the ICER was £6280 per QALY gained when comparing nalmefene plus psychosocial support with psychosocial support alone.

Evidence Review Group's comments

3.31 The ERG commented that the company had carried out a comprehensive systematic review and all relevant studies for nalmefene plus psychosocial support were included. It was unsure if all relevant naltrexone data had been included. The ERG also commented that the company's model was generally well constructed and had few errors.

3.32 The ERG indicated that the post hoc subgroup analyses of patients who had high or very high drinking risk level in the 3 nalmefene studies may cause the efficacy and safety data to be less robust because they were not powered for this analysis. The robustness may also be affected by the high dropout rates in the nalmefene trials. The company carried out sensitivity analyses to account for the missing data but there were some inconsistencies as to whether statistical significance was achieved or not. The ERG also indicated that patient self‑reporting of alcohol intake could bias the results.

3.33 The ERG indicated that the uncertainties in the clinical evidence related to the types and frequencies of psychosocial intervention, along with its treatment duration and generalisability to England. Psychosocial support in the form of BRENDA was used in the nalmefene trials but was delivered at different intervals to the psychosocial intervention (including behavioural therapies, cognitive behavioural therapy and behavioural couples therapies) recommended in the NICE guideline on alcohol-use disorders. The ERG stated that the evaluation carried out in the model does not meet that specified in the final scope and that it was difficult to know how the results would apply to people receiving different forms and frequencies of psychosocial intervention.

3.34 The ERG had concerns about the generalisability of the population in the 3 nalmefene studies to clinical practice in England. People with severe psychiatric comorbidities were excluded from all 3 nalmefene trials, and those with severe medical comorbidities were excluded from the ESENSE trials. The company commented in its submission that many people with alcohol dependence also have diagnosed medical conditions and/or psychiatric comorbidities. Patients were also excluded from the nalmefene trials if they were taking certain medication, such as drugs for angina, anticoagulants, anticonvulsants, insulin, sedatives and systemic steroids. The ERG stated that the safety and efficacy of nalmefene in people taking these drugs was therefore uncertain. Only a small number of trial patients were from the UK (SENSE trial only, 5 sites out of a total of 156) and the company did not provide any data on the variability of the outcomes for different European countries. The ERG stated that the generalisability of this data for England was unknown.

3.35 The ERG noted that naltrexone was not formally modelled as a comparator in the economic analysis even though it was included in the final scope issued by NICE. The model assumed that if patients stopped nalmefene treatment because of adverse events, they would switch to psychosocial support alone, but it did not account for switching to naltrexone. The ERG commented that it was unsure whether this assumption could be favourable or unfavourable to nalmefene.

3.36 The ERG stated that its clinical advisers did not agree with the assumption that people would remain on treatment (regardless of drinking level) for the full year. The ERG commented that its clinical advisers believed that GPs would not let patients drink at very high risk levels for more than 6 months without recommending intensification of psychosocial intervention and additional expert input, and that 3 months might be a more likely cut‑off point.

Evidence Review Group's exploratory analyses

3.37 The ERG formulated 4 comparisons in its exploratory analysis (see table 2).

Table 2 The 4 comparisons formulated by the Evidence Review Group

Comparison

Definition

Comparison 1

The analysis of the cost effectiveness of adding nalmefene to a psychosocial intervention of lower intensity than recommended in the NICE guideline on alcohol-use disorders.

Comparison 2

Threshold analyses that estimates the reduction in the benefit associated with nalmefene necessary to reach cost per quality‑adjusted life years (QALYs) of £20,000 and £30,000.

Comparison 3

The company did not comment on the likely cost effectiveness of delayed initiation of nalmefene for people whose alcohol dependence did not respond to psychosocial intervention as recommended in the NICE guideline on alcohol‑use disorders, compared with immediate initiation of nalmefene for all patients. Delayed use of nalmefene would be aligned with the recommendation for pharmacotherapy in the NICE guideline on alcohol‑use disorders, although this guideline was written before nalmefene was licensed.

Comparison 4

The company did not comment on the likely cost effectiveness of nalmefene use (delayed or immediate) with the use of off‑label naltrexone, following informed consent being obtained, as recommended in the NICE guideline on alcohol‑use disorders.

3.38 For comparison 1, the ERG carried out a number of exploratory analyses including:

  • Analysis 1: Impact of patients withdrawing from nalmefene because of adverse events, also withdrawing from psychosocial support – 2 scenarios were run, the first assumed that all patients withdrawing from nalmefene also withdrew from psychosocial support, and the second assumed that 50% of the patients also withdrew from psychosocial support.

  • Analysis 2: 50% of patients received outpatient medically assisted withdrawal and 50% had this treatment at home.

  • Analysis 3: The costs for serious and temporary events were zero and the utility was the same as the very high risk level, although the ERG did not deem this plausible.

  • Analysis 4: The cost of an expert psychosocial support appointment was £119 rather than £94, according to more recent data.

  • Analysis 5: The utility for patients on nalmefene plus psychosocial support and for psychosocial support alone were equal in the first year, although the ERG did not deem this plausible.

3.39 The ERG's base case included assumptions 1, 2 and 5, with the additional assumption that 50% of people withdrawing from nalmefene would also withdraw from psychosocial support treatment. In the ERG base case, nalmefene plus psychosocial support still dominated psychosocial support alone. The ERG carried out a second analysis using their base case assumption but also presumed no second‑line treatment options were allowed and the ICER was £5166 per QALY gained when comparing nalmefene plus psychosocial support with psychosocial support alone. Although the ERG was critical of the fact that the company did not conduct a half‑cycle correction, the model was not adapted by the ERG to allow this for 2 reasons: the first was the time needed to carry out this adaptation and the second because after the first year (in which monthly cycles were used), there was no differential efficacy between the 2 arms apart from people drinking at medium drink risk levels. Also, any potential inaccuracy was relatively small compared with the uncertainty explored in comparisons 2 and 3.

3.40 For comparison 2, the ERG suggested that it was unlikely for people at medium risk drinking level to have treatment indefinitely and assumed in comparison 1 that these people would relapse to high and very high risk levels. The ERG was unable to carry out a threshold analysis altering the variable treatment options because this part of the model was not functioning, and also given that the impact in the ICER was small, the ERG left the assumption as it was. The threshold analysis carried out by the company in scenario 7 was reassessed in the ERG's comparison 2 (with the exception that those at a medium risk drinking level were assumed to remain on treatment). The results produced by the ERG were similar to the company's results. If the efficacy of nalmefene and psychosocial support compared with psychosocial support alone were reduced by 62.8%, then the ICER would become £20,000 per QALY gained. The reduction would have to be 71.5% for the ICER to reach £30,000 per QALY gained. When additional factors accounting for the potential cost of crime and loss of productivity were considered, the efficacy of nalmefene and psychosocial support compared with psychosocial support alone would need to be reduced by 80.4% and 83.1% for the ICER to be £20,000 and £30,000 per QALY gained respectively.

3.41 For comparison 3, the ERG highlighted that there were few data to assess the cost effectiveness of nalmefene with psychosocial intervention when using the psychosocial intervention as described in the NICE guideline on alcohol-use disorders. The time point at which psychosocial intervention alone was not successful was also unknown but the nalmefene trials indicated that when patients were treated with BRENDA alone, approximately 20% were either abstinent or of low risk drinking level at month 3. The ERG suggested a greater response may be seen with higher‑intensity psychosocial intervention and that the costs of nalmefene can be saved without incurring health losses particularly if nalmefene use was delayed. The ERG did caution that there would be uncertainty about the efficacy of nalmefene in patients whose alcohol dependence had not responded to psychosocial support alone.

3.42 For comparison 4, again the ERG suggested there were few data available and therefore did not feel comfortable estimating an ICER for this comparison.

3.43 Full details of all the evidence are available.