Vortioxetine for treating major depressive episodes

The company conducted a systematic review of the literature to identify studies evaluating the clinical effectiveness and safety of vortioxetine for treating adults having a moderate‑to‑severe major depressive episode. These adults included those who had not tolerated initial antidepressant treatment or whose condition had responded inadequately to it, and who needed further antidepressant therapy (hereafter referred to as the 'second‑line population'). Therefore, in its original submission, the company did not include in its analyses all adults with major depressive disorder, as specified in NICE's final scope and vortioxetine's marketing authorisation. In its original submission, it identified 2 phase III randomised controlled trials, REVIVE and TAK318.

REVIVE was an international (14 European countries including the UK), double‑blind, randomised, active‑control trial. It included 501 adults with a single episode of moderate‑to‑severe major depressive disorder or recurrent major depressive disorder whose condition had inadequately responded to monotherapy with a selective serotonin reuptake inhibitor (SSRI) or a serotonin–norepinephrine reuptake inhibitor (SNRI). Patients were randomised 1:1 to flexible doses of vortioxetine (10–20 mg daily; starting dose 10 mg daily), or agomelatine (25–50 mg daily; starting dose 25 mg daily). Patients were assessed weekly during the first 4 weeks of treatment and then every 4 weeks until the end of the 12‑week treatment period. A further safety assessment was scheduled 4 weeks after completing or withdrawing from the study. Most patients enrolled into REVIVE were women (74.7%), most were white (99.8%), the mean age was 46.3 years and they had a mean of 2.5 previous major depressive episodes. The company stated that both groups had comparable baseline Montgomery–Åsberg Depression Rating Scale (MADRS) scores and previous antidepressant use. Most patients received the maximum dosage of vortioxetine (20 mg, 64.7%) and agomelatine (50 mg, 71.7%) from weeks 4 to 12.

The primary outcome measure in REVIVE was change in MADRS score from baseline to week 8 (MADRS is a rating scale consisting of 10 items, each rated 0 [no symptom] to 6 [severe symptom], contributing to a total score from 0 to 60; the higher the score, the more severe the condition). A 'full analysis set' population (that is, people who were randomised into the study and had a baseline assessment and at least 1 further assessment) was used to analyse the efficacy outcomes. The company tested a primary hypothesis of non‑inferiority. Non‑inferiority was considered established if the upper bound of the two‑sided 95% confidence interval of the difference between treatment groups in MADRS total score at week 8 did not exceed +2 MADRS units compared with agomelatine. The mean change from baseline in MADRS total scores at week 8 were −16.5 and −14.4 points in the vortioxetine group and the agomelatine group respectively. This resulted in a mean difference of −2.16 points in favour of vortioxetine (95% confidence interval [CI] −3.51 to −0.81; see table 1).

Pre-specified subgroup analyses of the primary outcome were carried out by the company for sex, age, baseline severity, baseline anxiety and class of prior antidepressant. The company stated that these analyses suggested that vortioxetine improved the MADRS score compared with agomelatine across all pre‑specified subgroups.

The company stated that vortioxetine statistically significantly improved outcomes compared with agomelatine across the analyses of outcomes reflecting response and remission measured by MADRS score (see table 2). Response is defined as a 50% or more decrease from baseline in the MADRS. Remission is defined as a MADRS total score of 10 or less.

The company did not identify any published studies of the cost effectiveness of vortioxetine for treating the second‑line population. It submitted a decision tree model with a Markov component to include subsequent treatment switches to third and later lines. It assumed that a patient can be offered 1 of 5 treatments: vortioxetine, agomelatine, citalopram, sertraline and venlafaxine. The company conducted the economic analysis from an NHS and personal social services perspective and chose a time horizon of 12 months so did not discount costs and health effects. A half‑cycle correction was applied to the health effects but not the costs in the Markov part of the model (cycle length 2 months).

The company stated its economic model represented a single major depressive episode. Hypothetical patients entered the model with major depressive disorder that had not responded to initial therapy. The decision tree included:

• an acute phase of treatment of 8 weeks (months 0–2)

• a maintenance phase of 6 months (months 2–8) and

• a recovery phase of 4 months' duration (months 8–12).
  
  The time that patients spent in the decision tree varied and depended on whether treatment was successful in each phase. If treatment succeeded in all 3 phases, with remission achieved and sustained to recovery, a hypothetical patient spent the entire 12 months in the decision tree model. The company's economic model also included events in which treatment was not successful (lack of remission or adverse events). These events led to a further treatment, that is, to third and subsequent lines of treatment. Patients who did not complete the acute or maintenance phase left the decision tree model and entered the Markov component. In a given cycle of the Markov component, a patient's condition could either remit or not. In its original economic model, the company assumed that patients remained on treatment for 6 months after their condition had remitted in the acute phase unless they experienced a long‑term adverse reaction (insomnia, sexual dysfunction or weight gain).

The company took data on the probability of remission after 8 weeks of treatment (acute phase) from its indirect treatment comparison (see table 4). The company assumed that a person's probability of relapse depended on the line of treatment rather than a specific drug: initial second‑line treatment (14.2%, from Limosin et al. 2004), third‑line treatment (25.0%, from STAR*D), and fourth- plus fifth‑line treatment (42.6%, from STAR*D). STAR*D was a prospective, sequentially randomised controlled trial of outpatients with nonpsychotic major depressive disorder who received 1 (n=3671) to 4 (n=123) successive acute treatment steps, including treatment combinations and augmented therapies. Patients who relapsed during the maintenance phase (which the company assumed occurred halfway through this phase) could switch to third and subsequent lines of treatment. The company assumed that clinicians then assessed these patients for remission 2 months after starting third‑line treatment. It took the data reflecting the proportion of patients whose condition was in remission after each line of treatment from STAR*D: third- (13.7%), fourth- (13.0%) and fifth‑line treatment (13.0%). The company considered that patients who had not relapsed after 6 months of maintenance treatment had recovered. These patients stopped treatment and the company assumed that they could not experience recurrent depression.

Resource use and costs in the company's economic model included those for treatment (drug), adverse events and each health state (that is, monitoring, inpatient and outpatient admissions). The company based drug costs on the list prices from the 'Monthly Index of Medical Specialities'. Dosages in the acute phases were based on the World Health Organization Defined Daily Dose (for example, 10 mg daily for vortioxetine), and dosages in the maintenance phase were based on the mean dose reported at the end of trials included in the company's indirect treatment comparison. The company took data for health state resource use for the acute phases from an unpublished interim analysis of the PERFORM study (n=226, which included people previously untreated) and, for the maintenance phase, from Byford et al. (2011; the General Practice Research Database 2001/06 – 88,935 people with depression and at least 2 antidepressant prescriptions). The company took data for the health state costs from Unit Costs of Health and Social Care (2013) and NHS Reference Costs. The company assumed that no treatments were prescribed to manage adverse events, but that around one‑third of people would incur an additional GP visit. Therefore, the company costed all adverse events based on an assumed 0.3 GP visits per patient per adverse event (£13.50).

To estimate health‑related quality of life in the acute phase, the company used EQ‑5D data from REVIVE (see table 6). However, in its original economic model for the maintenance phase, the company used EQ‑5D data from Sapin et al. (2004). Sapin was a French study that included 250 people with major depressive disorder in primary care, and assessed health‑related quality of life at baseline and after 8 weeks of treatment. The company noted that the mean MADRS score at baseline was 32.7 in Sapin compared with 29.1 in REVIVE, which may explain why the baseline EQ‑5D score from Sapin was lower than that in REVIVE. The company included disutility values associated with adverse events from Sullivan et al. (2004), and applied them for 3 weeks in the company's base case analysis.

The ERG stated that the reporting of the company's searches were clear and appropriate. The ERG noted that the company presented no evidence to suggest that the relative efficacy between non‑SSRIs may vary between first- and second‑line use (and beyond). It stated that it would be more appropriate to include the full evidence base for vortioxetine and its comparators, rather than restricting the evidence base from the outset to the second‑line population, so excluding 22 of the 24 completed studies of vortioxetine.

The ERG commented that REVIVE and TAK318 appeared well conducted but raised the following concerns:

• Both trials included comparators of limited relevance to clinical practice in England (NICE has not issued any guidance for agomelatine).

• Both trials were short considering the duration of treatment recommended by NICE to achieve and consolidate remission, so evidence of long‑term efficacy was uncertain.

• Both trials evaluated the efficacy of vortioxetine 10–20 mg daily, so the efficacy of the licensed 5 mg daily regimen was uncertain.

The ERG commented that the population enrolled into REVIVE was broadly representative of the second‑line population in England. For example, baseline MADRS scores ranged from 22–43 points, which is consistent with people with moderate‑to‑severe major depressive disorder. However, the ERG noted that:

• most patients were white (99.8%), which is unlikely to be reflective of the second‑line population in England

• 23% of patients had received an SNRI as initial treatment, which is not reflective of clinical practice in England, where first‑line SNRI use is negligible

• most patients were recruited from an outpatient psychiatric setting (97.2%)

• the proportion of patients from the UK was small (about 7%).

The ERG noted that, although the efficacy analyses in REVIVE and TAK318 used a modified intention‑to‑treat analysis (that is, full analysis set rather than including all randomised patients), the risk of bias was likely to be low because relatively few patients randomised were excluded.

The ERG commented that the results from the company's analysis of the primary and secondary outcomes from REVIVE had relatively wide confidence intervals, so the size of the difference in efficacy between vortioxetine and agomelatine was uncertain (see tables 1 and 2).

The ERG agreed with the company's assessment of bias for Rosso et al. (2012), so considered it was reasonable to exclude it, but noted it was the only trial that indirectly compared vortioxetine with duloxetine. The ERG stated that it was questionable whether Kasper et al. (2013) was suitable for inclusion in the indirect treatment comparison. It stated that it was unclear whether the population consisted entirely of patients receiving second‑line treatment, or whether it also included those who had been treated for a previous depressive episode in the last 12 months but were starting first‑line treatment for a current major depressive episode.

The ERG stated that it had significant concerns over the validity of the company's indirect treatment comparison because of the differences in the baseline patient characteristics and severity of depression of the populations across the 4 trials. It also stated that time of outcome assessment between trials (varying from 6–14 weeks) may also affect the results because rates of remission and withdrawal are likely to be time-dependent. The ERG concluded that the heterogeneous nature of data included in the network meant that the results may not be reliable.

The ERG highlighted that there was little evidence of a statistically significant improvement in the efficacy for vortioxetine compared with the comparators, given that the results from the company's indirect treatment comparison had wide confidence intervals. It stated that the findings in each specific trial drove the results of the company's indirect treatment comparison because of the sparse evidence network (that is, each arm of the network was informed by 1 trial). The ERG noted that basing results on risk differences was potentially inappropriate because they may be sensitive to the heterogeneity across trials (see table 4). However, it acknowledged that the company's results based on odds ratios were largely consistent (see table 5). The ERG also commented that the results from the company's sensitivity analysis including the 2 placebo‑controlled trials were broadly similar to those that excluded them.

The ERG stated that there was no evidence to suggest the relative efficacy between drugs that were not classified as SSRIs (for example, SNRIs) vary between first- and second‑line treatment (and beyond) (see section 3.21). Therefore, it sought further evidence from the company on a first‑line population during the clarification stage:

• The ERG re-analysed data from a published meta‑analysis of placebo‑controlled trials with active reference treatment arms (Pae et al. 2015). Pae compared vortioxetine with agomelatine (1 trial), duloxetine (5 trials) and venlafaxine (1 trial). The ERG noted that both the European Medicines Agency and the company have criticised the use of trials including active references because they are not true randomised comparisons, given that patients whose condition is known to be non‑responsive to the reference treatment are excluded, possibly biasing results in favour of the active reference. The ERG accepted the potential for such bias, but did not consider it substantial enough to exclude these trials. The ERG stated that Pae found no evidence of any difference in efficacy between vortioxetine and venlafaxine, and that vortioxetine was less effective than duloxetine in reducing depression scores, or achieving response and remission.

• Llorca et al. (2014) published an indirect treatment comparison that included 57 placebo controlled trials of the following drugs: vortioxetine, agomelatine, desvenlafaxine, duloxetine, escitalopram, sertraline, venlafaxine, vilazodone. Llorca found no evidence of any difference in efficacy between vortioxetine and its comparators. The ERG commented that there was evidence to suggest fewer people stop vortioxetine because of adverse events than other treatments, including sertraline and venlafaxine. The ERG considered that Llorca may represent the most reliable evidence for comparing vortioxetine with other treatments.

The ERG concluded that, based on all the evidence, vortioxetine is likely to be similar in efficacy to other antidepressants, but may be superior to agomelatine and inferior to duloxetine.

The ERG agreed that vortioxetine appears generally safe and tolerable in people with major depressive disorder. The ERG stated that, although the incidence of adverse events was high in people taking vortioxetine, most were mild to moderate in nature, and there was no conclusive evidence that they were dose‑dependent.

The ERG also concluded that vortioxetine may have a better overall safety profile than other antidepressants, but sparse comparative data for adverse events prevented the ERG making a firm conclusion.

The ERG stated that the company developed an unnecessarily complicated model structure, and that it was unclear why:

• The company used different modelling approaches in the maintenance and recovery periods, rather than an initial decision tree for the acute phase and then a separate Markov component for all people in the subsequent 10‑month period.

• The company assumed different time-points for relapse (after 3 months) and stopping treatment in the maintenance phase because of adverse reactions (after 1 month), which favoured those treatments with higher acquisition costs. The ERG noted that this introduced inconsistency between the timing of relapse for people within the decision tree and Markov components.

The ERG commented that basing the decision to change treatments solely on remission data at 8 weeks was an important limitation of the company's original economic model. It stated that the company's model therefore excluded people whose condition responded to treatment partially but had not remitted and that, in clinical practice, clinicians use response in deciding whether to continue treatments. The ERG commented that the company also used the 8‑week remission data in its original economic model to inform decisions to change treatment at 4 weeks in the model. The ERG explained that this ignored the costs of additional treatment for people whose disease responded but did not remit. It also explained that it may have overestimated health benefits for people whose disease remits because it assigned a utility value based on health improvements demonstrated over 8 weeks rather than 4 weeks. The ERG concluded that the company's base case may have underestimated vortioxetine's costs and overestimated vortioxetine's benefits.

The ERG noted that:

• Because the company had assumed that a person's depression was not at risk of recurrence in the recovery phase, it introduced a potential bias in favour of the most effective initial treatment. The ERG agreed that the risk of relapse (or recurrence) may be different in later phases than in earlier phases of the condition, but that assuming no risk of recurrence seemed overly optimistic.

• The company had assumed that people remain on treatment for 6 months after remission in the maintenance phase. The ERG considered that this was reasonable and consistent with NICE's guideline on depression in adults, but was aware that NICE recommends 2 years of continued treatment in people considered to be at high risk of relapse.
  
  The ERG acknowledged that the company explored both of these assumptions in the response to clarification by varying the time‑horizon of the model from 8 months (no recovery period) up to 2 years (treatment and monitoring costs continued in the recovery period). The ERG concluded that, although the company's base‑case analysis was robust to these scenarios, the ICER for vortioxetine compared with the next most cost‑effective treatment was higher than in the company's base‑case analysis, suggesting that including the original assumptions had potentially favoured vortioxetine.

The ERG stated that a half‑cycle correction for both costs and utility values would have been appropriate, rather than for utility values only, because different health states are associated with different costs for consultation or hospitalisation.

The ERG highlighted that using a 12‑month time horizon was reasonable for the 'average' patient because an untreated major depressive episode is estimated to last 5–6 months (World Health Organization 2008). However, the ERG noted that some people may be treated for longer than 12 months and therefore 12 months may not have been sufficient to capture all of the relevant costs and benefits.

The ERG highlighted that there was uncertainty around whether STAR*D was an appropriate study to inform the prognosis of people with depression whose condition had not remitted after second‑line treatment. The ERG considered that STAR*D included treatments that did not reflect the comparators in the model, and that the population of STAR*D was different from the population of REVIVE. It explained that using data from STAR*D for third- and later lines of treatment imposed a poorer prognosis (that is, lower remission rates and higher relapse rates) than expected for a population with the same characteristics as in REVIVE. The ERG stated that using STAR*D may have made the most effective second‑line treatment look even better (that is, vortioxetine in the company's base case analysis).

The ERG disagreed with the company's decision to use the same utility value for relapse, and for people whose condition was not in remission after 3 or more treatments. This was because they are very different health states. It highlighted that the utility value from Sapin et al. (2004), used by the company for people whose condition had not remitted, was lower than the utility value reported for people whose condition had not remitted at week 8 in the REVIVE trial. The ERG considered that it was not necessary to use a different source for the utility values in the maintenance phase, and that using these 2 sources (REVIVE and Sapin et al. 2004) favoured vortioxetine in the company's base‑case analysis. It also felt that the relapse health state should have reflected the recurrence of moderate‑to‑severe major depression and so the baseline level of utility (that is, 0.54). The ERG proposed alternative utility values for the company's model, see table 9.

The company provided additional evidence in its response to the appraisal consultation document. The company focused its additional evidence submission on people who had not tolerated, or whose major depressive episode had responded inadequately to, 2 antidepressants (hereafter referred to as the 'third-line population'). The company stated that there was no clinical-effectiveness evidence available for vortioxetine in people having third-line treatment. The company did not carry out any additional searches for comparators, and so the available clinical‑effectiveness evidence for vortioxetine and its comparators included: its original indirect treatment comparison for people having second‑line treatment (see section 3.11), Pae et al. (2015), Llorca et al. (2014) and the SOLUTION trial provided with the response to the appraisal consultation document.

SOLUTION was an international, double‑blind, randomised, active‑control trial. It included 410 East Asian adults with recurrent moderate‑to‑severe major depressive disorder and did not exclude any people based on the line of treatment used for their current major depressive episode. Patients were randomised 1:1 to fixed doses of vortioxetine (10 mg daily) or venlafaxine (150 mg daily). They were assessed weekly during the first 2 weeks of treatment and then every 2 weeks until the end of the 8‑week treatment period. The primary outcome measure in SOLUTION was change from baseline in MADRS score at week 8. A 'full analysis set' population was used to test a primary hypothesis of non‑inferiority. Non‑inferiority was considered established if the upper bound of the two‑sided 95% confidence interval of the difference between treatment groups in MADRS total score at week 8 did not exceed +2.5 MADRS units compared with venlafaxine. The mean change from baseline in MADRS total scores at week 8 were −19.4 points in the vortioxetine group and −18.2 points in the venlafaxine group. This resulted in a mean difference of −1.2 points in favour of vortioxetine (95% CI −3.0 to 0.6). At week 8, 43.1 and 41.4% of the people's major depressive episode had remitted in the vortioxetine group and venlafaxine group respectively. The company considered that the SOLUTION trial was relevant to the decision problem because it directly compared vortioxetine with venlafaxine.

The company noted that, as a third‑line treatment, SSRIs were not offered in clinical practice in England and so were not relevant comparators for vortioxetine. The company revised its economic model structure so that it:

• defined treatment success, and decisions to switch treatment, by remission and response (rather than remission alone)

• used the time point when patients changed to another treatment because their condition did not respond to treatment from the trials for the time point in the model (8 weeks rather than 4 weeks)

• included a risk of relapse or recurrence at all stages of depression (rather than only in the acute or maintentance phase)

• used utility values from REVIVE (rather than using utility values from REVIVE for the acute phase and utility values from Sapin et al. 2004 for the maintenance and recovery phases)

• included a 24‑month time horizon (with discounting of costs and health effects after 12 months).

The company adjusted the second‑line remission rates used in its original economic model to reflect third‑line remission rates used in its revised economic model using the proportional reduction observed in STAR*D between second- and third‑line treatment. For fourth and subsequent lines of treatment, the company used the absolute remission reported for third- and fourth‑line treatment in STAR*D.

The company presented probabilistic pairwise ICERs, as well as a fully incremental analysis, for several scenarios using its revised economic model:

• Scenario 1a (see table 15): Primary care setting, up to 6 months' maintenance treatment and assuming equal use of healthcare resources for people whose condition was in remission and for people whose condition responded but was not in remission.

• Scenario 1b (see table 16): Secondary care setting, up to 6 months' maintenance treatment and assuming equal use of healthcare resources for people whose condition was in remission and for people whose condition responded but was not in remission.

• Scenario 1c (see table 17): Primary care setting, up to 6 months maintenance treatment and assuming that use of healthcare resources is 30% higher between weeks 8 and 12 in people whose condition was not in remission but responded compared with people whose condition was in remission.

• Scenario 2a (see table 18): Primary care setting, up to 22 months' maintenance treatment and assuming equal use of healthcare resources for people whose condition was in remission and for people whose condition responded but was not in remission.

• Scenario 2b (see table 19): Secondary care setting, up to 22 months' maintenance treatment and assuming equal use of healthcare resources for people whose condition was in remission and for people whose condition responded but was not in remission.

• Scenario 2c (see table 20): Primary care setting, up to 22 months' maintenance treatment and assuming that use of healthcare resources is 30% higher between weeks 8 and 12 in people whose condition was not in remission but responded compared with people whose condition was in remission.
  
  In the company's revised base case, it assumed that all treatments were equally effective. The company also presented cost‑effectiveness results for scenarios using alternative sources of data on efficacy (see section 3.43).

The ERG did not agree with the company that the results of Llorca et al. (2014) were biased because the rates of response and remission from several trials were not reported. The ERG noted that the results for the mean change in depression score from Llorca (no missing data) were consistent with the results for the rates of response and remission.

The ERG stated that SOLUTION was a well‑conducted randomised controlled trial but did not reflect the population in England. However, the ERG stated that the relative effectiveness between vortioxetine and venlafaxine was unlikely to differ substantially between people treated in East Asia and England. The ERG considered that the results from SOLUTION supported the ERG's original conclusions (and of Llorca et al. 2014) that vortioxetine is similarly effective to other non‑SSRIs, but may be better tolerated.

The ERG stated that the company's revised economic model more accurately reflected whether a person should continue or change treatment for their major depressive disorder in clinical practice in England (see section 3.45). The ERG considered that the company's revised economic model had used the most appropriate available data.

The ERG explained that, in most cases, the conclusions about vortioxetine's cost effectiveness depends on which source of efficacy data is chosen (for example, assumption of equal efficacy, Llorca et al. 2014, Pae et al. 2015, company's indirect treatment comparison in people having second‑line treatment, SOLUTION) rather than the scenario chosen (for example, primary or secondary care setting, length of maintenance treatment).