3.1 The manufacturer presented direct clinical-effectiveness evidence from 2 randomised controlled trials comparing aripiprazole with placebo and performed a meta-analysis of these trials. The manufacturer also presented a network meta-analysis based on a network containing aripiprazole, olanzapine, risperidone, quetiapine and placebo.
3.2 The manufacturer undertook a systematic review to identify published evidence for aripiprazole for the treatment and prevention of acute manic and mixed episodes in children and adolescents with bipolar disorder. The review identified 3 randomised controlled trials (NCT00110461, NCT00194077 and NCT00116259) that investigated aripiprazole in children and adolescents with bipolar disorder. The NCT00194077 trial was excluded because it only included children under the age of 10, a group outside the licensed indication. The manufacturer considered NCT00110461 to be the main evidence for the use of aripiprazole in children and adolescents. The NCT00116259 trial was not discussed in detail because the manufacturer considered it to be a small trial in a specific population of children and adolescents with bipolar disorder (both types I and II) and attention-deficit hyperactive disorder (ADHD). No relevant non-randomised controlled trial evidence for aripiprazole was identified from the systematic review.
3.3 NCT00110461 was a multicentre, double-blind, randomised, placebo-controlled clinical trial undertaken across 59 sites in the USA between March 2005 and February 2007. The study was designed to test the safety and efficacy of 2 doses of aripiprazole in children and adolescents with bipolar I disorder who experienced manic or mixed episodes with or without psychotic features. There were 296 participants randomised to receive aripiprazole 10 mg/day (n=98), aripiprazole 30 mg/day (n=99) or placebo (n=99). The study duration was 30 weeks with a 4-week acute phase followed by an extension phase of 26 weeks.
3.4 Participants in NCT00110461 were aged between 10 and 17 years with a DSM-IV diagnosis of bipolar I disorder who were experiencing manic or mixed episodes, with or without psychotic features. Comorbid diagnoses were permitted including ADHD, conduct disorder, oppositional defiant disorder and anxiety disorders. All participants had a baseline Young Mania Rating Scale (YMRS) score of more than 20. Exclusion criteria included bipolar II disorder, unspecified bipolar disorder and psychosis caused by other medical conditions or concomitant mediations. Participants with learning disabilities and those who were determined by the investigator to be at risk of suicide were also excluded.
3.5 In the NCT00110461 trial, 72% (296/413) of participants screened were enrolled in the study. Participant characteristics at baseline were comparable between the 3 groups in the trial. Approximately 63% of participants were aged 13 or over and fell within the licensed population. In response to a clarification request, the manufacturer provided post-hoc data on the proportion of participants in mixed and manic states at baseline and on the proportion of 'rapid cyclers' (participants who experienced 4 or more manic, hypomanic or mixed episodes in the previous year). In both cases the proportions were relatively similar between the groups.
3.6 The primary outcome in NCT00110461 was the change from baseline to week 4 on the YMRS total score. Secondary outcomes were changes from baseline scores in the Children's Global Assessment Scale (CGAS), Clinical Global Impressions Scale-Bipolar Version (CGI-BP) severity scores for mania, depression and overall bipolar illness, Children's Depression Rating Scale - Revised (CDRS-R) score, General Behaviour Inventory Scale (GBI) score and Attention-Deficit Hyperactivity Disorders Rating Scale (ADHD-RS-IV) score. A YMRS response rate was based on a responder definition of a 50% or more reduction from the YMRS total score at baseline.
3.7 In the NCT00110461 trial, both aripiprazole doses demonstrated statistically significant improvements over placebo in the YMRS total score at week 4, with treatment differences from placebo of −5.99 (95% confidence interval −8.49 to −3.50; p<0.0001) for the aripiprazole 10 mg group, and −8.26 (95% CI −10.7 to −5.77; p<0.0001) for the aripiprazole 30 mg group. Statistically significant improvements in YMRS total score were demonstrated at week 1 for both aripiprazole doses, and were maintained until week 30. At week 30, the treatment difference in YMRS total score from placebo for the aripiprazole 10 mg dose was −5.89 (95% CI −8.70 to −3.08), and for the aripiprazole 30 mg dose it was −6.73 (95% CI −9.53 to −3.94). The YMRS responder rates were also significantly higher in the aripiprazole arms than in the placebo arm at week 4 and week 30.
3.8 Both aripiprazole doses demonstrated significant improvements at week 4 compared with baseline in the secondary end points: CGAS core, CGI-BP severity scores for mania, depression and overall bipolar illness, GBI – parent/guardian version and subject version mania total score, and the ADHD-RS-IV total score. Significant differences were not observed at week 4 in the CGI-BP severity scores for depression, GBI – patient depression total scores or the CDRS-R score. A significant difference was observed in the 10 mg aripiprazole arm for the GBI-parent/guardian version for depression (p= 0.0430) but not in the 30 mg arm.
3.9 The manufacturer presented results from a post-hoc subgroup analysis based on the change in YMRS score from baseline calculated using both the observed case dataset and the last observation carried forward dataset for age subgroups (10–12, and 13–17 years). Results from the last observation carried forward data showed that participants receiving aripiprazole in both age groups had a statistically significant change in YMRS score from baseline at both weeks 4 and 12 compared with those receiving placebo. Using the observed case dataset, there was a statistically significant change in YMRS score in participants receiving aripiprazole in both age groups at week 4; however by week 12, although the change in YRMS score was still greater in participants receiving aripiprazole compared with those receiving placebo, this was not statistically significant.
3.10 At the end of the acute phase of the trial (4 weeks) the proportion of participants not completing the trial was 14.3% in the aripiprazole 10 mg group, 22.2% in the aripiprazole 30 mg group and 23.2% in the placebo group. At 30 weeks the drop-out rates were 65.3% in the aripiprazole 10 mg group, 77.7% in the aripiprazole 30 mg group and 87.9% in the placebo group. The manufacturer noted the statement in the Committee for Medicinal Products for Human Use (CHMP) assessment report that results based on the observed case analysis dataset failed to show statistical significance for aripiprazole compared with placebo for both doses on all analysed efficacy end points at week 12. The manufacturer stated that the lack of statistically significant differences between aripiprazole and placebo at week 12 along with the high discontinuation rate resulted in the CHMP restricting the treatment length with aripiprazole to 12 weeks.
3.11 The manufacturer also carried out a post-hoc subgroup analysis to investigate if the efficacy of aripiprazole was influenced by the presence or absence of ADHD symptoms. Results presented in 3 age subgroups (10–12, 13–14, and 15–17 years) suggested no change in the treatment effect.
3.12 In NCT00110461, health-related quality of life was measured by the Pediatric Quality of Life Enjoyment and Satisfaction Questionnaire (PQ-LES-Q). The manufacturer reported that although the results did not reach statistical significance, both aripiprazole arms demonstrated a trend for improvement relative to placebo.
3.13 The manufacturer presented adverse events occurring in more than 5% of any group in NCT00110461 over the acute phase and over the full trial duration. There were no deaths or suicides in the study. The manufacturer acknowledged that somnolence and extrapyramidal symptoms occurred more frequently in the aripiprazole arms than in the placebo group. In the acute phase (up to 4 weeks) extrapyramidal disorder occurred in 12.2% (12/98) of the aripiprazole 10 mg group, 27.3% (27/99) of the aripiprazole 30 mg group and 3.1% (3/97) of the placebo group. In the acute phase (up to 4 weeks) somnolence occurred in 19.4% (19/98) of the aripiprazole 10 mg group, 26.3% (26/99) of the aripiprazole 30 mg group and 3.1% (3/97) of the placebo group. In the acute phase (up to 4 weeks) akathisia occurred in 8.2% (8/98) of the aripiprazole 10 mg group, 11.1% (11/99) of the aripiprazole 30 mg group and 2.1% (2/97) of the placebo group. The manufacturer concluded that most of the adverse events occurred in the acute phase of the study and were mild to moderate in severity, and therefore were expected to be manageable.
3.14 The manufacturer presented changes in baseline metabolic parameters at 4 and 30 weeks. Differences in the treatment groups between the incidence of clinically significant weight gain (7% or more weight gain compared with baseline) and change from baseline body mass index (BMI) were not statistically significant at week 4. Data on the incidence of clinically significant weight gain at 30 weeks was considered to be academic in confidence by the manufacturer and therefore cannot be reported in this document. There were no differences between the treatment groups in the proportion of participants with a BMI on or above the 95th percentile for age and sex at week 4. Data on the proportion of participants with a BMI on or above the 95th percentile at week 30 was considered to be academic in confidence.
3.15 The manufacturer noted that the CHMP limited the indication for aripiprazole to adolescents aged 13 and over as a result of safety concerns in younger people. The manufacturer reported results from a post-hoc subgroup analysis to assess the safety profile across age subgroups (10–12 years, and 13–17 years). In the 10–12 years subgroup there were statistically significant increases in mean weight and BMI changes from baseline in the aripiprazole 30 mg treatment arm at weeks 4 and 12. There were also statistically significant increases in weight and BMI measurements in the aripiprazole 10 mg treatment arm at week 12 using the last observation carried forward analysis for the 10–12 years subgroup.
3.16 The NCT00116259 trial was a double-blind randomised placebo-controlled clinical trial undertaken in Brazil. There were 43 participants who were randomised to receive aripiprazole 20 mg per day (n=18) or placebo (n=25). The study duration was 6 weeks. Participants were aged between 8 and 17 years with a DSM-IV diagnosis of bipolar I or II disorder, comorbid with DSM-IV ADHD. All participants were in an acutely manic or mixed state defined as a YMRS score of more than 20 at the baseline visit. Exclusion criteria included learning disabilities and severe risk of suicide.
3.17 Results from NCT00116259 showed that participants receiving aripiprazole had a statistically significantly larger reduction in YMRS total scores from baseline to week 6 than participants receiving placebo (−27.22 versus −19.52, p=0.02). A greater proportion of participants on aripiprazole compared with placebo experienced a response (88.9% versus 52%, p=0.02; number needed to treat [NNT] =2.70) and experienced remission (72% versus 32%, p=0.01; NNT=2.50).
3.18 The manufacturer presented a meta-analysis of NCT00110461 (pooled 10 mg per day and 30 mg per day) and NCT00116259 (20 mg per day). Results from the meta-analysis indicated that aripiprazole was statistically significantly superior to placebo in inducing symptomatic response (as measured as a 50% or more change in YMRS score) at weeks 1, 2 and 4, but not at week 3. Also, results from the meta-analysis showed aripiprazole to be associated with a statistically significant higher rate of extrapyramidal symptoms than placebo, but not of somnolence. The manufacturer commented that the meta-analysis was performed to be transparent rather than to provide meaningful results. The manufacturer stated that the results should be treated with caution because the information from the NCT00116259 trial is limited by the small trial size and the different population compared with NCT00110461 (that is, it included people with bipolar II disorder and was restricted to participants with comorbid ADHD).
3.19 As there are no head-to-head trials comparing aripiprazole with the comparators specified in the final scope issued by NICE, the manufacturer presented a network meta-analysis to determine the relative efficacy of the treatments using placebo as the common comparator. Five randomised controlled trials were identified in addition to the 2 studies including aripiprazole (NCT00110461 and NCT00116259). The 5 studies were as follows: 3 studies for risperidone (risperidone compared with placebo, risperidone compared with divalproex sodium [valproate semisodium] and risperidone compared with divalproex sodium and lithium), 1 study for quetiapine (quetiapine compared with placebo), and 1 study for olanzapine (olanzapine compared with placebo). The network meta-analysis included NCT00110461, one of the studies for risperidone (risperidone compared with placebo), and the studies for quetiapine and olanzapine (both compared with placebo). The manufacturer undertook a sensitivity analysis in which the NCT00116259 trial and the remaining 2 studies for risperidone (risperidone compared with lithium and divalproex sodium) were also included.
3.20 The network analysis used a fixed-effects Bayesian model, because the manufacturer considered there was not enough evidence to support the estimation of a random-effects model. Results from studies that included treatment groups with different intervention doses were pooled to provide an average treatment dose effect. Efficacy outcomes considered in the network meta-analysis were the YMRS response rates at weeks 1, 2 and 3, and discontinuation at week 3. Analyses were also conducted for the following safety outcomes: extrapyramidal symptoms, clinically significant weight gain, clinically significant increase in prolactin and somnolence.
3.21 The manufacturer presented results from the network meta-analysis as relative risks using placebo and aripiprazole (pooled dose) as references. These results indicated that all the antipsychotics considered (aripiprazole, risperidone, quetiapine and olanzapine) were statistically more effective than placebo at achieving YMRS response at weeks 1–3. The results also indicated that there were no statistically significant differences in YMRS response rates at weeks 1–3 between the antipsychotics. No statistically significant differences were found for discontinuation of treatment at week 3 between the interventions considered in the network.
3.22 Participants receiving aripiprazole were found to be statistically significantly more likely to experience extrapyramidal symptoms than those receiving placebo. They were also more likely to experience them than participants receiving risperidone and quetiapine, but these differences were not statistically significant. There was no statistically significant difference between aripiprazole and placebo in the risk of experiencing a clinically significant increase in weight. Aripiprazole (pooled dose) was statistically significantly less likely to cause clinically significant weight gain than olanzapine and quetiapine. Participants receiving aripiprazole were less likely to experience a clinically significant increase in prolactin than those on olanzapine, risperidone or quetiapine. Participants receiving aripiprazole were found to be statistically significantly more likely to experience somnolence than those receiving placebo. They were also more likely to experience somnolence than participants receiving risperidone and quetiapine, but this was not statistically significant.
3.23 The manufacturer undertook a systematic review to identify relevant cost-effectiveness or cost–utility studies. No economic evaluations were identified for the treatment of bipolar I disorder in children and adolescents. The manufacturer developed a de novo economic model to assess the cost effectiveness of aripiprazole compared with the other antipsychotics. The population in the economic evaluation was adolescents with manic episodes of bipolar I disorder between 13 and 17 years as specified in the marketing authorisation. The age of onset used in the model was 15 years and the time horizon was until the person reached adulthood at 18 years.
3.24 The cost-effectiveness model presented by the manufacturer was a Markov cohort model with a weekly cycle length. The treatment pathway in the model was based on a sequence of up to 4 treatment lines. The first 3 related to treatment with an antipsychotic drug and the fourth included lithium treatment for participants whose condition was resistant to previous therapy. The antipsychotic treatment lines were identical in structure and each contained an acute phase of 3 weeks of inpatient treatment, a sub-acute phase of up to 5 weeks of inpatient treatment for participants who experienced a response, a maintenance phase of outpatient treatment for an average of 4 weeks, and then withdrawal of treatment. The 'therapy resistance' phase contained up to 5 weeks inpatient lithium treatment, followed by outpatient lithium treatment and a maintenance phase similar to that in the antipsychotic lines for participants who experienced a response.
3.25 Participants entered the model at the start of the first treatment line. They moved to the next treatment line if they discontinued treatment before response during the acute phase, if they did not respond to current treatment by the end of week 3, or if they relapsed before discharge from hospital. If participants relapsed within the maintenance phase they remained on the same treatment line to which they had responded. If participants did not respond to 3 lines of antipsychotic treatment they entered the 'therapy resistance' treatment line. If participants relapsed on 'therapy resistance' treatment they returned to the inpatient lithium treatment (that is the 'therapy resistance' hospitalised state). The modelling of adverse events was included in the treatment-related health states. Participants could also die in any health state in the model.
3.26 Based on clinical opinion, the manufacturer specified the treatment sequence of risperidone, quetiapine and olanzapine to represent usual care (labelled strategy 1 in the model). The manufacturer considered either quetiapine or olanzapine could be replaced by aripiprazole. For the base-case analyses, olanzapine was replaced with aripiprazole and the position of aripiprazole in the treatment sequences was varied, giving 4 different strategies:
Strategy 1: risperidone, quetiapine, olanzapine.
Strategy 2: risperidone, aripiprazole, quetiapine.
Strategy 3: aripiprazole, risperidone, quetiapine.
Strategy 4: risperidone, quetiapine, aripiprazole.
3.27 Clinical data for the effectiveness for each antipsychotic in the 3‑week acute phase were taken from the results of the network meta-analysis based on pooled dose levels. It was assumed the effectiveness of each antipsychotic intervention was not influenced by its position in the treatment pathway and that the treatment was at a constant dose. Beyond the acute phase, it was assumed that all antipsychotics were equally effective and the common weekly relapse value was based on expert opinion. The manufacturer also assumed an identical mortality rate for all the antipsychotic interventions, based on UK life tables adjusted to reflect the higher rates of mortality observed among participants with bipolar disorder.
3.28 The model included 3 adverse events: extrapyramidal symptoms, somnolence and weight gain. Data for the incidence of these events were based on the results of the network meta-analysis. There were no available data for the incidence of extrapyramidal symptoms or somnolence associated with olanzapine treatment, and so these values were set equal to the lowest incidence of the other antipsychotics.
3.29 The health-related quality of life data collected in NCT00110461 were not based on a preference-based measure. As a result of a lack of available data, for preference-based utility values from bipolar disorder in adolescents, the manufacturer based the utilities in the model on published data from adult populations with bipolar disorder. For the main analysis, the utility values for the bipolar health states were based on EQ-5D data from a study of an adult UK population with bipolar I disorder. A multiplicative model for the utility values was used to take into account the demographic (age and sex) of the population from which the utility was calculated. The calculated multipliers were then applied to an appropriate general population utility value for the adolescent population in the model. The utility values were also further adjusted according to hospitalisation status, adverse events and the ageing of the cohort. The utility value for weight gain was taken from a cohort of children who are overweight or obese and the utility values for somnolence and extrapyramidal symptoms came from participants with schizophrenia.
3.30 In the economic model in-hospital costs were based on NHS reference costs 2010/1155 (code MHIPC1; NHS Trusts Mental Health Inpatients – Children). This cost was assumed to include costs relating to adverse events, but not the cost of antipsychotic treatments. Outpatient resource use was based on expert opinion, with costs taken from the Personal Social Services Research Unit. Non-propriety costs were used for each of the antipsychotics apart from aripiprazole. The base case was based on the pooled dose efficacy and safety results and so an average dose cost was used.
3.31 In the base case, the cost-effectiveness results were similar for the 4 strategies. The use of aripiprazole at any point in the treatment pathway dominated usual care (that is, was more effective and less costly), which had the highest total cost (£75,066) and the lowest total quality-adjusted life years (QALYs) (2.516). Aripiprazole used as second-line therapy after risperidone had the lowest total cost (£74,133) and the highest total QALYs (2.525) and dominated the other base-case strategies.
3.32 The manufacturer undertook a wide range of univariate sensitivity analyses and identified the rates of response applied during the acute treatment phase as the main parameters that influenced the cost-effectiveness estimates of the strategies. The manufacturer noted that this was expected because the response rates were varied by ±30%, and the differences in the base-case response rates between the 4 antipsychotic treatments considered were not substantial. Scatterplots from probabilistic sensitivity analysis were also presented in the submission. From these results the manufacturer noted that there was some uncertainty surrounding the cost-effectiveness results, but that for all 3 strategies containing aripiprazole, the majority of probabilistic sensitivity analysis iterations indicated cost effectiveness or dominance.
3.33 The manufacturer presented various scenario analyses in its submission and in response to a clarification request. These included:
Using 10 mg aripiprazole rather than a pooled dose.
Using results from the network meta-analysis sensitivity analysis based on all trials identified in the manufacturer's review.
Swapping the position of quetiapine and olanzapine in the base-case treatment strategy.
Using utility values from a different source.
Changing the starting age of participants to 13 years and 17 years.
Reducing treatment efficacy between lines 1 and 2 and between lines 2 and 3.
Including the cost of drug-related adverse events in the model.
Extending the acute and euthymic treated phases of the model.
None of these scenarios produced results that differed greatly from the base case, and use of aripiprazole in the treatment pathway always dominated usual care.
3.34 The ERG noted that the searches in the manufacturer's submission were limited to January 2012 and requested that the manufacturer update them. As a result of time constraints the manufacturer provided results from a non-systematic approach and identified 3 further trials. The ERG repeated and updated the searches until January 2013. The ERG found 3 studies not identified by the manufacturer, but none of them were phase III randomised controlled trials.
3.35 The ERG considered that the NCT0011461 trial was of reasonable methodological quality, and measured a range of outcomes that are relevant to the decision problem. However, the ERG expressed concern that the trial population described in the manufacturer's submission is not likely to represent the UK clinical population:
Clinical advisers to the ERG considered that the age of the population in the trial was much younger than that seen in UK clinical practice.
Clinical advisers also raised concerns about the high proportion of participants in the trial with comorbid ADHD.
The ERG considered it likely that many of the participants in the trials were treated as outpatients, which does not reflect current UK practice for this population.
The ERG commented that excluding from the trial participants who were at risk of suicide may have resulted in the trial population having less severe disease than those presenting in UK clinical practice.
The ERG noted the manufacturer's acknowledgement that the inclusion and exclusion criteria for NCT0011461 could have reduced the external validity of the trial population. It also noted that no information was presented on the duration or maintenance of effect after 12 weeks of treatment with aripiprazole.
3.36 The ERG noted the advice in the European Medicines Agency (EMA) guidance for clinical investigation of bipolar disorder, that the occurrence of switching to depression should be investigated. In response to a clarification request, the manufacturer provided depression outcomes at weeks 4 and 30 using the CGI-BP severity depression score, the CDRS-R score, the GBI total score – parent/guardian (depression), and the GBI total scores – patient (depression) score. The ERG considered that although the data presented did not raise concerns about the occurrence of depression associated with aripiprazole treatment, the effect of aripiprazole on depression was not explored in depth in the submission.
3.37 The ERG stated that the key clinical evidence for this appraisal came from the network meta-analysis. The ERG noted that relevant placebo-controlled trials of the antipsychotic comparators were used for the indirect comparison. The ERG agreed with the manufacturer that the NCT00116259 trial should be excluded from the base case because of the different participant population. The ERG also agreed that excluding the study comparing risperidone with lithium and divalproex sodium was appropriate. The ERG did not consider it appropriate for the study comparing risperidone with divalproex sodium to be excluded but acknowledged that including this study would have little impact on the conclusions from the network meta-analysis.
3.38 The ERG questioned the pooling of doses from treatment arms with multiple doses in the network meta-analysis because it considered that different doses are associated with different efficacies and side effects. The manufacturer was requested to conduct a network meta-analysis with doses from multiple treatment arms separated, but was unable to do so in the time available. The ERG considered that caution should be used when interpreting the results from the network meta-analysis based on pooled interventions from multiple treatment arms.
3.39 The ERG considered that a random-effects model was more appropriate for the network meta-analysis than the fixed-effects model used by the manufacturer because of the heterogeneity between the trials. The ERG undertook the network meta-analysis using a random-effects model, and obtained similar results to those from the fixed-effects model; although in all cases the 95% credible intervals were wider, reflecting the increased uncertainty.
3.40 In general the ERG was satisfied that the economic evidence submitted by the manufacturer does not represent a biased assessment of the cost effectiveness of aripiprazole. The ERG considered the manufacturer's model to be robust and transparent and well structured, allowing for analysis of uncertainty in the model inputs.
3.41 The ERG explored the manufacturer's probabilistic sensitivity analysis. The ERG noted that all changes in the total costs and QALYS between the deterministic and mean probabilistic sensitivity analysis results were less than 0.1% of the deterministic value. Although second-line aripiprazole (strategy 2) dominated all the other strategies in the deterministic analysis and mean probabilistic sensitivity analysis results, the probabilistic 95% confidence intervals included the possibility of each strategy dominating this strategy. The ERG summarised the results to show that the strategy that excludes aripiprazole (strategy 1) was dominated by each of the other strategies in over half of the probabilistic sensitivity analysis results: strategy 2 dominated strategy 1 in 72.1% of iterations, strategy 3 dominated strategy 1 in 54.4% of iterations, and strategy 4 dominated strategy 1 in 57.2% of iterations. The cost-effectiveness acceptability curve indicated that the probability that strategy 1 is cost effective is roughly half that of the probabilities for strategy 2 or strategy 3 for all the thresholds explored.
3.42 The ERG highlighted that its clinical advisers and those of the manufacturer stressed the importance of tailoring the treatment sequence to reflect an individual's needs (based on factors such as severity of symptoms, side-effect profile and comorbidities). As there are limited data available to model treatment within subgroups, the ERG conducted an exploratory scenario analysis to assess the possible implications of treatment sequences tailored to reflect an individual's needs. The results showed that only small changes in the modelled results for each treatment sequence were needed for the incremental cost-effectiveness ratio (ICER) for that strategy to be £20,000 or £30,000 per QALY gained. The ERG stated that these results suggest that the actual place of aripiprazole within a treatment sequence is likely to depend on individual circumstances.
3.43 The ERG explored the potential implications of 2 different treatment durations for aripiprazole: 1 use reflected its licensed duration of a maximum of 12 weeks, the other reflected clinical opinion of its real-life use based on an average of 12 months. The ERG amended the manufacturer's model to have maximum treatment duration (for all antipsychotics) of 12 weeks. Responding to a request, the manufacturer provided an amended version of its model to have an average of 12 months of antipsychotic treatment. The ERG considered that although total costs and total QALYs showed a reduction in both of the 2 new models, the substantive conclusions of the manufacturer's base case analysis remained unchanged.
3.44 The ERG considered that a treatment sequence containing all 4 antipsychotics was appropriate. Clinical advisers to the ERG stated that if a person's condition had not responded to the first 3 antipsychotic interventions, clinicians would generally try the remaining antipsychotic rather than conclude that a person's condition was treatment resistant. It was not possible to evaluate this scenario within the time frame of the clarification process. The ERG stated that there is no evidence to suggest that a 4-line treatment sequence would substantially alter the conclusions.
3.45 The ERG made the following changes to the model to explore the plausibility of the manufacturer's results:
it used network meta-analysis results from a random-effects model
it included a half-cycle correction
it adjusted the discounting formula used
it amended the mortality rate calculations
it included a 10% reduction in efficacy between lines 1 and 2, and 15% between lines 2 and 3
it included a logical constraint on the probabilistic sensitivity analysis inputs so that week 3 probability of discontinuing or responding did not exceed 100%.
These amendments were incorporated into a 'licensed duration' model, which reflected the maximum 12 week duration specified in the marketing authorisation, and into a 'real-world' model, which assumed 4 weeks of acute treatment and 12 months of euthymic treatment.
3.46 The deterministic and probabilistic cost-effectiveness results from both models were similar to those obtained in the manufacturer's base-case model:
The probabilistic sensitivity analysis results from the 'licensed duration' model indicated that the use of aripiprazole at any point in the treatment pathway dominated usual care, which had the highest total cost (£72,157) and the lowest total QALYs (2.458). Aripiprazole used as second-line therapy after risperidone had the lowest total cost (£70,707) and the highest total QALYs (2.471).
The probabilistic sensitivity analysis results from the 'real-world' model indicated that aripiprazole used as first-line therapy had the highest total cost (£63,384) and the lowest total QALYs (2.428). Aripiprazole used as second-line therapy had the lowest total cost (£62,138) and the highest total QALYs (2.429).
3.47 The ERG also explored the impact on the cost effectiveness of the amendments (described in section 3.45) separately, and concluded that the key drivers to changes in the cost-effectiveness results were the extension of treatment duration either to 4 weeks acute treatment or 12 months of euthymic treatment.