Pharmacological and Medical Interventions for Acute Episodes

National Collaborating Centre for Mental Health (UK)

NCBI Bookshelf. A service of the National Library of Medicine, National Institutes of Health.

National Collaborating Centre for Mental Health (UK). Bipolar Disorder: The NICE Guideline on the Assessment and Management of Bipolar Disorder in Adults, Children and Young People in Primary and Secondary Care. London: The British Psychological Society and The Royal College of Psychiatrists; 2014 Sep. (NICE Clinical Guidelines, No. 185.)

See 2018 Partial Update

April 2018: Footnotes and cautions have been added and amended to link to the MHRA's latest advice and resources on sodium valproate. Sodium valproate must not be used in pregnancy, and only used in girls and women when there is no alternative and a pregnancy prevention plan is in place. This is because of the risk of malformations and developmental abnormalities in the baby. November 2017: Footnotes for some recommendations were updated with current UK marketing authorisations and MHRA advice. Links to other guidelines have also been updated. Some research recommendations have been stood down. See these changes in the short version of the guideline.

April 2018: Footnotes and cautions have been added and amended to link to the MHRA's latest advice and resources on sodium valproate. Sodium valproate must not be used in pregnancy, and only used in girls and women when there is no alternative and a pregnancy prevention plan is in place. This is because of the risk of malformations and developmental abnormalities in the baby. November 2017: Footnotes for some recommendations were updated with current UK marketing authorisations and MHRA advice. Links to other guidelines have also been updated. Some research recommendations have been stood down. See these changes in the short version of the guideline.

Bipolar Disorder: The NICE Guideline on the Assessment and Management of Bipolar Disorder in Adults, Children and Young People in Primary and Secondary Care.

Show details

Contents

< Prev Next >

6Pharmacological and Medical Interventions for Acute Episodes

6.1. Introduction

Pharmacological interventions are commonly used to manage acute episodes in bipolar disorder. Acute episodes may carry significant risk of suicide, neglect, disinhibition, recklessness, irritability and, sometimes, threats to others. Therefore the settings in which pharmacological interventions are carried out, and the wishes and abilities of service users and families to manage episodes safely, require careful consideration in relation to risk assessment.

On average, people with bipolar disorder experience more depressive than manic episodes, and depressive episodes last longer than mania (Judd et al., 2002a; Judd et al., 2003a; Morriss et al., 2013). The effective treatment of bipolar depression is therefore a clinical priority for the NHS. The main aims of the treatment of bipolar depression are response (that is, resolution of symptoms) and return to a premorbid level of social functioning.

The management of mania in the community can be particularly challenging for carers. During a manic episode, the service user may sleep for only a few hours and be driven to move from one activity to another. Mania involving high levels of restlessness, irritability and insomnia often requires inpatient admission. Similarly, agitated episodes of depression or mixed affective episodes, particularly in people expressing suicidal intent or with a history of self-harm, may require inpatient admission.

The management of acute bipolar episodes is complex because of the propensity to be highly changeable in both the severity of symptoms and the polarity of the episode (mania, hypomania, mixed affective or depression). Practitioners often consider all mental states displayed within recent days, not just the one displayed at the time of interview, in making a risk assessment. Furthermore, bipolar disorder tends to be associated with other comorbid mental disorders, and medication may be associated with physical side effects. The management of acute episodes should also consider the risk of switching into a different episode in the short to medium term. Most people who have an acute episode will have another within 12 months, so treatment of acute episodes should consider long-term management as well.

6.1.1. Definitions

Lithium

Lithium is an element that is present in a normal diet, and is handled by the body in a similar way to sodium. The ubiquitous nature of sodium in the human body, its involvement in a wide range of biological processes and the potential for lithium to alter these processes have made it extremely difficult to ascertain the key mechanism(s) of lithium in regulating mood (for a review, see Marmol [2008]).

Lithium is licensed for the treatment of mania and recurrent depression, and for the prevention of further mood episodes in people with bipolar disorder. A meta-analysis and at least two large database studies have concluded that lithium treatment is associated with a reduced risk of suicide (Cipriani et al., 2013c; Collins & McFarland, 2008; Goodwin et al., 2003).

Lithium has a narrow therapeutic range, meaning that levels below 0.4 mmol per litre are unlikely to be effective in the majority of patients and levels above 1.0 mmol per litre are associated with increasing toxicity (muscle weakness, coarse tremor, disorientation, seizures and loss of consciousness). Some commonly used medicines, such as non-steroidal anti-inflammatory drugs, diuretics and ACE (angiotensin-converting-enzyme) inhibitors, can increase lithium levels in the blood and therefore cause toxicity. Lithium has adverse effects on the kidneys, thyroid and parathyroid (McKnight et al., 2012). Lithium is a known human teratogen, that is, it is potentially harmful to an unborn child.

Antipsychotics

Antipsychotic medication is thought to exert its effects by blocking dopamine (D₂) receptors in the brain. These drugs have been in common use to treat schizophrenia and mania for over 60 years, although few were originally licensed for the latter indication. Over the past 10 years or so, there have been an increasing number of studies examining the efficacy and tolerability of newer antipsychotic drugs in the treatment of both mania and bipolar depression, resulting in some being specifically licensed for these indications. Antipsychotics have long been used to prevent or reduce the severity of new mood episodes in people with bipolar disorder, although the relative effectiveness of these drugs against each pole of the illness is thought to differ (Gitlin & Frye, 2012). The use of antipsychotics in people with bipolar disorder has increased significantly in the UK over recent years (Hayes et al., 2011).

Antipsychotic drugs are variably associated with a range of side effects, the most problematic of which is probably weight gain. Other side effects include dry mouth, blurred vision, sedation, sexual dysfunction, extrapyramidal side effects (tremor, stiffness, restlessness and abnormal movements) and dizziness.

Anticonvulsants

Valproate is a simple branched-chain fatty acid that is commonly used for the treatment of epilepsy. Although it is known to exert a large range of effects on brain functioning, its exact mechanism of action in bipolar disorder remains unclear. For a review, see Rosenberg (2007).

Valproate is available in various forms including sodium valproate, valproic acid and valproate semi sodium, although only valproate semi-sodium has UK marketing authorisation for the treatment of manic episodes in the context of bipolar disorder. This guideline uses the generic term ‘valproate’, as it is the active element in all formulations.

Valproate in all formulations is used for the treatment of mania and bipolar depression, and for the prevention of new mood episodes. Valproate is associated with a number of side effects including tremor, weight gain and, rarely, liver damage. It can interact with a number of commonly prescribed medicines and notably is known to decrease plasma levels of olanzapine (Haslemo et al., 2012), an antipsychotic drug that is commonly prescribed in people with bipolar disorder. Valproate is a known major human teratogen. There are significant risks associated with taking valproate during pregnancy for the unborn child, including risk of autism (Christensen et al., 2013; NICE, 2014) and its use is best avoided completely in women of child-bearing age.

Carbamazepine is structurally related to the tricyclic antidepressants. It has been used as an anticonvulsant in people with epilepsy since 1974 (Israel & Beaudry, 1988), and it is licensed for the treatment of people with bipolar disorder who are intolerant of lithium or in whom lithium is ineffective.

Although carbamazepine is known to reduce both neuronal firing and the release of excitatory neurotransmitters in the brain, the exact mechanism by which it exerts its effects in people with bipolar disorder is not understood.

The main side effects associated with carbamazepine are dizziness, drowsiness, nausea and headaches, and it can cause a low white blood cell count, hyponatraemia (low level of sodium in the blood) and rarely, liver damage. Carbamazepine is a potent inducer of hepatic cytochrome enzymes and this can lead to increased metabolism, so lower plasma levels of a number of commonly prescribed medicines. For example standard dose combined oral contraceptives can be rendered ineffective due to the increased metabolism of oestrogen. Carbamazepine is also a known human teratogen.

Lamotrigine is another anticonvulsant that is commonly used in people with bipolar disorder, where it is licensed for the prevention of episodes of depression. Its mechanism of action in people with bipolar disorder is not fully understood.

Lamotrigine is associated with a rash which can be serious so, to minimise the risk of this occurring, the dose of lamotrigine has to be increased very slowly at the start of treatment. Lamotrigine can also cause drowsiness, dizziness and blurred vision, and it can depress the bone marrow. Lamotrigine, too, is a known human teratogen, although it is considerably safer in pregnancy than valproate.

Dosage recommendations are complex, particularly when lamotrigine is used with other anticonvulsant drugs.

Anticonvulsant drugs can interact with each other and if more than one of these drugs is prescribed, the British National Formulary (BNF) should be checked to ensure doses are adjusted if required.

Antidepressants

Antidepressants all exert their effect by increasing levels of one or more of serotonin, noradrenaline and dopamine within the brain.

Despite having a relatively modest effect size in the treatment of unipolar depression (NICE, 2009), antidepressants are widely prescribed for this indication. Antidepressants are also commonly prescribed for people with bipolar depression (Sidor & McQueen, 2011), but their use is controversial for two reasons. First, there is considerable doubt about whether antidepressants have any efficacy in bipolar depression (Sachs et al., 2007; Sidor & McQueen, 2012) and, second, there are concerns that these drugs could induce switching into mania (Tondo et al., 2010) or accelerate cycling so that the time to the next relapse decreases and the time spent in relapse increases. However, there is considerable uncertainty whether antidepressants do in fact cause such switching or cycle acceleration given the natural propensity for bipolar disorder to be highly changeable (Altshuler et al., 2004).

There are a number of different types of antidepressants and, of these, selective SSRIs are the most frequently prescribed. These drugs are generally well tolerated although they can cause headache, gastrointestinal upset and sexual dysfunction. SSRIs can also cause hyponatraemia (low blood sodium) and they increase the risk of bleeds, particularly in the gastrointestinal tract. Further background information about the different types of antidepressants and their relative side effects can be found in the NICE guideline for the management of depression (NICE, 2009) or the BNF¹⁵.

Nutritional interventions

Adequate intake of dietary omega-3 fatty acids (eicosapentaenoic acid [EPA] and docosahexaenoic acid) is essential for the maintenance of good physical health. Western diets may contain insufficient quantities of these fatty acids. Supplements containing omega-3 fatty acids are widely available from health food shops and are commonly taken for their perceived health benefits. The majority of those who take such complementary therapies have mental health problems (Werneke, 2009). This suggests that these treatments are considered to be acceptable by many patients.

Fatty acids are essential components of cell membranes, and omega-3 fatty acids are known to be anti-inflammatory. There is also some evidence to suggest that they alter the structure and function of cell membranes, which in turn impacts on the functioning of monoamine neurotransmitters (Chalon, 2006). These properties have led to widespread interest in the use of omega-3 fatty acids in a wide range of psychiatric conditions, including mood disorders (Bloch & Hannestad, 2012; Sarris et al., 2012).

Herbal preparations

Herbal preparations are rarely recommended for bipolar depression. It is likely that St John’s wort, a treatment for unipolar depression, is being used by a small proportion of people with bipolar disorder, but there is no evidence concerning its efficacy and it can have some potentially toxic interactions with some medicines with high serotonergic activity such as antidepressants or anticoagulants such as warfarin. Other herbal preparations (such as valerian) are also often used as hypnotics during depression, again with little evidence of efficacy but there is less concern about interactions with prescribed drugs.

6.2. Pharmacological and Nutritional Interventions for Mania, Hypomania and Mixed Episodes

6.2.1. Introduction

The main aim in treating mania, hypomania and mixed episodes (a mood state in which manic and depressive symptoms are both exhibited) is to achieve rapid control of affective symptoms. More commonly, mania may cause people to act in a disinhibited manner, and such behaviour may have long-term adverse repercussions for the individual’s career and relationships. Mixed episodes are reported to be associated with an increased risk of suicide. As indicated above, an important treatment aim is to prevent further affective episodes occurring immediately after the current episode, including switching into a depressive episode, when the risk of suicide is greater. Service users may have long stays in hospital if their mood repeatedly switches from mania into depression and back again. Therefore, the management of manic, hypomanic and mixed affective episodes needs to consider the risk of further episodes within days, weeks or months after improvement in the acute phase.

6.2.2. Clinical review protocol

The review protocol summary, including the review question and the eligibility criteria used for this section of the guideline, can be found in Table 10 (a complete list of review questions and protocols can be found in Appendix 7; further information about the search strategy can be found in Appendix 8).

Table 10

Clinical review protocol summary for the review of pharmacological and nutritional interventions for mania, hypomania and mixed episodes.

6.2.3. Studies considered¹⁶

The search for systematic reviews identified a recent review that included a network meta-analysis of pharmacological interventions for mania: CIPRIANI2011 (Cipriani et al., 2011). The review reported the critical outcomes identified by the GDG, and the results were directly relevant to treatment of bipolar mania in the UK. To determine if new studies could change the conclusions of the review, the GDG conducted a search.

The search for new studies identified five RCTs: ASTRAZENECA2011 (Astrazeneca, [unpublished] 2011b), BEHZADI2009 (Behzadi et al., 2009), CHIU2005 (Chiu et al., 2005), KANBA2012 (Kanba et al., 2012), SZEGEDI2012 (Szegedi et al., 2012). Two studies about ‘bipolar anxiety’ were excluded from all reviews: SHEEHAN2009 (Sheehan et al., 2009), SHEEHAN2013 (Sheehan et al., 2013). Two open-label studies: SCHAFFER2013 (Schaffer et al., 2013), SINGH2013 (Singh et al., 2013); and three trials of medications neither routinely used nor licensed for the treatment of mental health problems: ZHANG2007 (Zhang et al., 2007), KULKARNI2006 (Kulkarni et al., 2005; Kulkarni et al., 2006), MCELROY2011 (McElroy et al., 2011) were also excluded from this review. Results could not be obtained for five studies: BOSE2012 (Bose et al., 2012), BRISTOLMYERSSQUIBB2011 (Bristol-Myers Squibb, [unpublished] 2011), FOREST2012 (Forest, 2012), KNESEVICH2009 (Knesivich et al., 2009), YANG2009 (Yang, 2009); although they have published several papers about the drug, the manufacturer of cariprazine has not reported the results of clinical trials and they refused requests from the NCCMH for data.

Of the five new RCTs, three (N = 940; ASTRAZENECA2011, KANBA2012, SZEGEDI2012) could have been considered for the network meta-analysis (had they been available at the time the analysis was conducted). The new studies were analysed and their results compared with the results of the network meta-analysis for the critical outcomes. Two additional RCTs (N = 103) which did not meet inclusion criteria for the network meta-analysis were also identified. These were a trial of folic acid added to valproate (BEHZADI2009) and a trial of omega-3 polyunsaturated fatty acids added to valproate (CHIU2005).

Further information about both included and excluded studies can be found in Appendix 16 and 34.

6.2.4. Clinical evidence review

The GDG considered the findings of CIPRIANI2011 alongside new trials (see Table 11). The review assessed the effects of all antimanic drugs for the treatment of mania. These included 14 treatments: aripiprazole, asenapine, carbamazepine, valproate, gabapentin, haloperidol, lamotrigine, lithium, olanzapine, paliperidone, quetiapine, risperidone, topiramate, ziprasidone and placebo. All studies included participants within the same target population, which was clearly defined; most included studies recruited patients rated as having moderate to severe manic symptoms and 76% of trials were conducted in inpatient clinics. The search strategy was technically adequate. The authors searched Medline, Embase, Cumulative Index to Nursing and Allied Health (CINAHL), PsycINFO, the Cochrane Central Register of Controlled Trials and the trial databases of the main regulatory agencies to identify relevant studies published between 1st January 1980 and 25th November 2010. In addition, all relevant authors and principal manufacturers were contacted to supplement incomplete reports of the original papers or to provide new data for unpublished studies. The overall quality of studies was rated as good (using the Cochrane Risk of Bias tool), with only three studies assessed as high risk on one item. Nevertheless, many studies were rated as unclear in term of allocation concealment and selective reporting.

Table 11

Comparison between new studies and network meta-analysis (all results compared with placebo).

The network meta-analysis found robust evidence that several pharmacological interventions are efficacious. Furthermore, there was evidence of differential effectiveness among medications, which is a unique strength of network meta-analysis. Haloperidol, risperidone, olanzapine, lithium, quetiapine, aripiprazole, carbamazepine, asenapine, valproate and ziprasidone were statistically significantly more effective than placebo, while gabapentin, lamotrigine and topiramate were not. For discontinuation, olanzapine, risperidone and quetiapine were significantly better than placebo. On the dichotomous outcome for efficacy (50% reduction in manic symptoms) the results were consistent with continuous outcomes, but less clear cut and with wider confidence intervals. Asenapine, ziprasidone, lamotrigine and topiramate were not significantly more effective than placebo and no binary efficacy data were available for gabapentin. The few data made it difficult to draw clear conclusions for this outcome. In head-to-head comparisons, haloperidol had the highest number of significant differences compared with other antimanic drugs, partly because it was often used as an active comparator. It was significantly more effective than lithium, quetiapine, aripiprazole, carbamazepine, asenapine, valproate, ziprasidone, lamotrigine, topiramate and gabapentin. Risperidone and olanzapine had a very similar profile of comparative efficacy, being more effective than valproate, ziprasidone, lamotrigine, topiramate and gabapentin. Topiramate and gabapentin were significantly less effective than all the other antimanic drugs. In terms of discontinuation, haloperidol was significantly inferior to olanzapine; lithium inferior to olanzapine, risperidone and quetiapine; lamotrigine inferior to olanzapine and risperidone; gabapentin inferior to olanzapine; topiramate inferior to many other antimanic treatments, such as haloperidol, olanzapine, risperidone, quetiapine, aripiprazole, carbamazepine and valproate. Statistical heterogeneity was moderate overall. However, for most comparisons 95% CIs were wide and included values indicating very high or no heterogeneity, which portrayed the small number of studies available for every pairwise comparison. In the meta-analyses of direct comparisons for efficacy, I² values higher than 75% were recorded for the comparisons ziprasidone versus placebo (I² = 76.6%) and olanzapine versus lithium (I² = 89.2%), with five and three studies, respectively. For acceptability, I² values higher than 75% were recorded for the comparisons aripiprazole versus haloperidol (I² = 84.1%) and lithium versus lamotrigine (I² = 82.0%), with two and three studies in the meta-analysis, respectively. Most loops (networks of three comparisons that arise when collating studies involving different selections of competing treatments) were consistent because their 95% CIs included 0 (that is, the direct estimate of the summary effect does not differentiate from the indirect estimate) according to the forest plots. Analysis of inconsistency indicated that there was inconsistency in three of the total 33 loops for efficacy measured as a continuous outcome (aripiprazole-placebo-haloperidol; olanzapine-placebo-risperidone; quetiapine-placebo-haloperidol), but none for acceptability (34 loops) or binary efficacy (18 loops). The authors could not identify any important variables that differed across comparison in those loops, but the number of included studies was very small in the three inconsistent loops.

Examining the results of several trials reported after the publication of the network meta-analysis, the GDG concluded that the most recent evidence is consistent with the results of the network meta-analysis and that the inclusion of new studies would not change the conclusions of that review. One study of folic acid added to valproate reported effects that the GDG considered implausibly large and insufficient to lead to a recommendation (BEHZADI2009). In one study of omega-3 polyunsaturated fatty acids, it was not possible to extract outcomes, however the authors reported no effect of the intervention on manic symptoms. For these reasons, the GDG used the results of the network meta-analysis when considering what recommendations to make.

Of the drugs included in the network meta-analysis (CIPRIANI2011) without new evidence, seven were shown on the primary outcome to have an advantage over placebo: carbamazepine (SMD −0.36, 95% credible interval [CrI] −0.60 to −0.11), valproate (SMD −0.20; 95% CrI, −0.37 to −0.04), haloperidol (SMD −0.56; 95% CrI, −0.68 to −0.43), lithium (SMD −0.37; 95% CrI, −0.50 to −0.25), olanzapine (SMD −0.43; 95% CrI, −0.54 to −0.32), quetiapine (SMD −0.37; 95% CrI, −0.51 to −0.23), risperidone (SMD −0.50; 95% CrI, −0.63 to −0.38). A further three we shown on the primary outcome to be little better than placebo: gabapentin (SMD 0.32; 95% CrI, −0.18 to 0.82), lamotrigine (SMD −0.08; 95% CrI, −0.34 to 0.18), topiramate (SMD 0.07; 95% CrI, −0.09 to 0.24), ziprasidone (SMD −0.19; 95% CrI, −0.37 to −0.03).

According to the same network meta-analysis, haloperidol was significantly more effective than lithium (SMD −0.19, 95% CrI = −0.36 to −0.01), quetiapine (SMD −0.19, 95% CrI = −0.37 to −0.01), aripiprazole (SMD −0.19, 95% CrI = −0.36 to −0.02), carbamazepine (SMD −0.20, 95% CrI = −0.36 to −0.01), asenapine (SMD −0.26, 95% CrI = −0.52 to −0.01), valproate (SMD −0.36, 95% CrI = −0.56 to −0.15), ziprasidone (SMD −0.36, 95% CrI = −0.56 to −0.15), lamotrigine (SMD −0.48, 95% CrI = −0.77 to −0.19), topiramate (SMD −0.63, 95% CrI = −0.84 to −0.43), and gabapentin (SMD −0.88, 95% CrI = −1.40 to −0.36). Risperidone and olanzapine had a very similar profile of comparative efficacy, being more effective than valproate, ziprasidone, lamotrigine, topiramate and gabapentin. Olanzapine, risperidone and quetiapine led to significantly fewer discontinuations than did lithium, lamotrigine, placebo, topiramate and gabapentin. Ranking of drugs included in the network meta-analysis by their overall probability to be the best treatment in terms of their combined efficacy and acceptability (reflected in their dropout rate) resulted in the following order (from highest to lowest probability of being best treatment): risperidone, olanzapine, haloperidol, quetiapine, carbamazepine, aripiprazole, valproate, lithium, ziprasidone, asenapine, placebo, lamotrigine, topiramate, gabapentin.

6.2.5. Health economics evidence

Systematic literature review

The systematic search of the economic literature undertaken for the guideline identified no study on the cost effectiveness of nutritional interventions and four eligible studies on the cost effectiveness of pharmacological treatments for adults with bipolar disorder in a manic, hypomanic or mixed episode (Bridle et al., 2004; Caro et al., 2006; Revicki et al., 2003; Zhu et al., 2005). Of these, only Bridle and colleagues’ study was conducted in the UK, while the rest studies were conducted in the US. References to included studies and evidence tables for all economic evaluations included in the systematic literature review are provided in Appendix 32. Completed methodology checklists of the studies are provided in Appendix 31. Economic evidence profiles of studies considered during guideline development (that is, studies that fully or partly met the applicability and quality criteria) are presented in Appendix 33.

Olanzapine versus valproate semisodium

Revicki and colleagues (2003) evaluated the cost effectiveness of valproate semisodium versus olanzapine in adults with bipolar I disorder in a manic episode in the US. The economic analysis was conducted alongside a multicentre RCT (ZAJECKA2002). The study was a cost consequence analysis; the RCT outcomes considered in the analysis were the participants’ clinical improvement based on the Mania Rating Scale (MRS) from the Schedule for Affective Disorders and Schizophrenia Change Version and the Hamilton Rating Scale for Depression, and the participants’ health-related quality of life (HRQoL) measured by the Quality of Life Enjoyment and Satisfaction Questionnaire and the number of days with restricted activity. The perspective of the analysis was that of a third-party payer. Costs included hospitalisation costs, physicians’ fees, costs of emergency room, costs of psychiatric, physician, psychologist or other mental health provider visits, home health service visit costs and medication costs. HRQoL and resource use data were collected via telephone interviews; a number of resource use data, such as the number of inpatient physician visits and type of outpatient visits, were based on assumptions. National unit costs were used. The time horizon of the analysis was 12 weeks. Participants in the RCT discontinued treatment if they did not improve after 3 weeks, but data were still collected for a total period of 12 weeks.

The results of the analysis showed that there were no significant differences between the two drugs in terms of clinical, HRQoL and economic outcomes over the 12-week period. Valproate semisodium was associated with significantly lower outpatient costs compared with olanzapine; nevertheless, total direct medical costs associated with the two drugs were similar (mean total cost per person US$13,703 for valproate semisodium and US$15,180 for olanzapine, p = 0.88, cost year not stated). The study is partially applicable to the UK context as it was conducted in the US. Moreover, it is characterised by potentially serious limitations, relating to the short time horizon of the analysis (12 weeks), the use of assumptions for some resource use data and potential conflicts of interest.

Zhu and colleagues (2005) also conducted a cost consequence analysis alongside a multicentre RCT (TOHEN2002) to evaluate the cost effectiveness of olanzapine versus valproate semisodium in adults with bipolar I disorder who were hospitalised for a manic or mixed episode in the US. The time horizon of this analysis was 47 weeks, comprising 3 weeks of acute phase and 44 weeks of maintenance phase. Only participants who entered the maintenance phase of the RCT were included in the economic analysis (59% of the initial study sample). The clinical outcomes considered were the clinical improvement based on the Young Mania Rating Scale (YMRS) and the rate of symptom remission (defined as YMRS score ≤ 12) at 3 weeks, and the median time to remission of manic symptoms. The perspective of the analysis was that of a third-party payer. Cost elements included hospitalisation (full and partial), outpatient psychiatric physician and other mental health provider visits, emergency room visits, home visits by healthcare professionals, medication and laboratory tests. Effectiveness and resource use data were taken from the RCT; resource use data were collected from hospital and other medical records and family reports. National unit costs were used.

According to the analysis, total costs were similar between the two drugs (mean total cost per person US$14,967 for olanzapine, US$15,801 for valproate semisodium; p > 0.05, cost year 2000). Olanzapine was found to be significantly better than valproate semisodium in improving manic symptoms at 3 weeks and in the percentage of people achieving remission (54.4% versus 42.3%, respectively). The median time to remission was 14 days for olanzapine and 62 days for valproate semisodium. The results of the analysis suggest that olanzapine is a more effective treatment option that valproate semisodium for people with bipolar disorder experiencing mania at no extra cost. The study is partially applicable to the NHS context as it was conducted in the US. Moreover, it is characterised by potentially serious limitations including the design of the study regarding collection of resource use data and potential conflicts of interest.

Quetiapine versus usual care

Caro and colleagues (2006) developed a discrete event simulation model to evaluate the cost effectiveness of quetiapine versus usual care in adults with bipolar I disorder experiencing a manic episode in the US. Usual care comprised 45% monotherapy with lithium, 25% lithium plus risperidone, 25% lithium plus olanzapine and 5% lithium plus quetiapine. The time horizon of the analysis was 100 days. The analysis adopted a third-party payer perspective. Cost elements consisted of hospitalisation and physician fees, emergency room and intensive care units, routine physician and psychiatrist visits, laboratory tests, medication and management of side effects. The outcome measures used were the percentage of people responding at 21 days and the percentage of people remitting at 84 days. Clinical data for the economic model were taken from a literature review, whereas resource use data were derived from administrative databases; national unit costs were used.

Quetiapine was found to be overall less costly than usual care (mean total cost per person US$5,525 for quetiapine and US$6,912 for quetiapine in 2004 prices). It was also found to be more effective than usual care: the percentage of people responding at 21 days was 54% for quetiapine and 43% for usual care; the percentage of people remitting at 84 days was 80% for quetiapine and 74% for usual care. Consequently quetiapine was the dominant treatment option. Results were sensitive to drug prices, discharge criteria and side-effect management costs. The study is partially applicable to the UK context as it was conducted in the US; the definition of usual care may not reflect usual care in the UK. The analysis is characterised by a number of potentially serious limitations including the source of cost and effectiveness data and potential conflicts of interest.

Antipsychotic drugs (olanzapine, quetiapine and haloperidol) compared with lithium and valproate semisodium

The economic analysis by Bridle and colleagues (2004) was the only study undertaken in the UK. The objective of the study, which informed a previous NICE Technology Appraisal on the use of newer anti-manic drugs (NICE, 2003a), was to evaluate the cost effectiveness of quetiapine, olanzapine and valproate semisodium in the treatment of adults with bipolar disorder experiencing a manic episode. The study was based on decision-analytic modelling. Effectiveness data were derived from a systematic review and network meta-analysis. The availability of effectiveness data in the network meta-analysis determined the choice of drugs included in the economic analysis. The following drugs were thus considered in the analysis: quetiapine, olanzapine, valproate semisodium, haloperidol and lithium.

The primary measure of outcome was the number of responders to treatment; response was defined as ≥ 50% improvement in manic symptoms, expressed in changes in YMRS scores. The time horizon was equal to 3 weeks in the base-case analysis, to reflect the most commonly reported length of follow-up for which effectiveness data were provided in the clinical trials. Estimated costs, expressed in 2001–2002 prices, included direct medical costs from the NHS perspective; these consisted of hospitalisation and drug-acquisition costs, as well as costs of diagnostic and laboratory tests required for monitoring. Resource use data were based on expert opinion, information from manufacturers and further assumptions. Unit costs were taken from national sources. Costs of treating adverse events were not included in the analysis, because of lack of relevant data reported in the literature. However, the authors’ opinion was that the majority of adverse events associated with the drugs compared were unlikely to have significant resource use implications in the 3-week time horizon of the model. Hospitalisation costs were estimated to be the same for all drug treatment options because all people experiencing a manic episode were assumed to be hospitalised at the start of the model and to remain hospitalised for the total 3-week period, regardless of response to treatment.

The base-case results of the analysis showed that mean response rates for olanzapine (0.54) and haloperidol (0.52) were higher than for lithium (0.50), quetiapine (0.47) and valproate semisodium (0.45). Haloperidol had the lowest mean total costs per person (£3,047) in comparison to valproate semisodium (£3,139), olanzapine (£3,161), lithium (£3,162) and quetiapine (£3,165). In terms of cost effectiveness, lithium, valproate semisodium and quetiapine were dominated by haloperidol because they were all less effective and more costly than haloperidol. Compared with haloperidol, olanzapine was more effective and resulted in higher total costs, demonstrating an incremental cost effectiveness ratio (ICER) equal to £7,179 per additional responder. This means that if decision-makers are prepared to pay less than £7,179 per additional responder, then haloperidol is the optimal decision; however, if they are prepared to pay at least £7,179 per additional responder, then olanzapine is the most cost-effective option.

One-way sensitivity analyses showed that results relating to dominance of haloperidol were robust to alternative assumptions tested, such as discharge of non-responders at a later time than responders, treatment of non-responders with second and third-line pharmacological therapies, reductions in diagnostic and laboratory costs, inclusion of effectiveness data for people initially excluded from analysis according to a modified intention-to-treat approach, and inclusion of treatment costs for extrapyramidal symptoms because of haloperidol use. Under these scenarios, the ICER of olanzapine compared with haloperidol ranged between £1,236 (when longer hospitalisation was assumed for non-responders) and £7,165 (when second and third-line treatment was assumed for non-responders) per additional responder. Base-case results were sensitive only to the entire exclusion of diagnostic and laboratory costs from the analysis, which constituted a rather extreme scenario.

Probabilistic analysis demonstrated that, for a willingness to pay (WTP) equal to £20,000 per additional responder, the probabilities of each drug being cost-effective were: olanzapine 0.44, haloperidol 0.37, lithium 0.16, quetiapine 0.02 and valproate semisodium 0.01. The probability that olanzapine was cost-effective increased as the WTP increased: for a maximum WTP £10,000 per additional responder this probability reached 0.42, increasing to 0.45 if the maximum WTP rose to £40,000. When the WTP for an additional responder was zero, haloperidol was the most cost-effective option (with probability equalling 1), as this was the least costly option of those assessed.

Although the study was conducted in the UK, it is only partially applicable to the NICE context because its primary measure of outcome was the rates of response and not the quality-adjusted life year (QALY), which is the preferred outcome measure by NICE, due to lack of appropriate utility data. As a result, the reported ICERs are difficult to interpret because there is no set threshold for the WTP per additional responder to anti-manic therapy. In addition, although the study was well conducted it is characterised by potentially serious limitations: first of all, the model had a very short time horizon of 3 weeks, which was nevertheless dictated by the time horizon of the RCTs included in the network meta-analysis. This means that potential differences across drugs regarding benefits and resource use, including the overall length of hospitalisation (beyond 3 weeks), were not taken into account. However, potential differences in the length of hospitalisation among drugs may affect significantly their relative cost effectiveness because inpatient care is the major driver of total medical costs associated with treatment of mania. Cost differences between drugs were found to be very small and were attributed exclusively to differences in acquisition and monitoring costs because hospitalisation costs were assumed to be the same across drugs over the time period of 3 weeks. Finally, omission of costs and HRQoL aspects of side effects from the analysis were also acknowledged by the authors as a further limitation of their study.

Overall conclusions from existing economic evidence

The existing economic evidence on drugs for the treatment of mania in people with bipolar disorder is rather limited and not directly applicable to the NICE decision-making context. All studies included in the review are characterised by potentially serious limitations. Evidence from the US suggests that olanzapine and valproate semisodium are associated with similar overall costs; in terms of effectiveness, one study showed superiority of olanzapine and the other study found no difference in effectiveness. An additional US study indicated that quetiapine was dominant (more effective and less costly) than usual care. The only UK study included in the review showed that haloperidol was dominant over lithium, valproate semisodium and quetiapine. Olanzapine was more effective and more costly than haloperidol, with an ICER equal to £7,179 per additional responder. However, the study is characterised by potentially serious limitations and its results are not easy to interpret due to the lack of use of QALYs as a measure of outcome.

It should be noted that quetiapine and olanzapine are now available in generic form, and therefore their acquisition cost is lower than the cost of the patented forms evaluated in the studies included in the systematic review. Thus their relative cost effectiveness is likely higher than that suggested in the literature.

Economic modelling

Introduction – objective of economic modelling

The cost effectiveness of pharmacological interventions for the treatment of adults with bipolar disorder experiencing a manic episode was identified by the GDG as an area with potentially major resource use implications that should be addressed by economic modelling. However, the availability of clinical and cost data did not allow the development of a model with a time horizon longer than 3 weeks that would overcome the limitations characterising the study by Bridle and colleagues (2004). Therefore, a simple economic analysis was attempted which updated the costs and clinical data reported by Bridle and colleagues (2004) and allowed the GDG to consider the costs associated with pharmacological interventions for mania alongside their clinical effectiveness as reported in Cipriani and colleagues (2011). In addition, a cost-utility analysis was conducted, using available utility data that allowed outcomes to be expressed in the form of QALYs.

Economic modelling methods

Interventions assessed

The interventions that were assessed in this economic analysis were determined by the availability of data reported in the network meta-analysis by Cipriani and colleagues (2011). Only drugs that were found to be effective in this study and licensed in the UK were considered in the economic analysis. Cipriani and colleagues (2011) evaluated the following drugs: aripiprazole, asenapine, carbamazepine, valproate, gabapentin, haloperidol, lamotrigine, lithium, olanzapine, quetiapine, risperidone, topiramate and ziprasidone. Paliperidone was not assessed separately, but relevant data were pooled with risperidone data because paliperidone is the main active metabolite of risperidone. The economic analysis did not consider ziprasidone because it is not licensed in the UK. Moreover, gabapentin, lamotrigine and topiramate were found to be not significantly better than placebo in the network meta-analysis and were therefore excluded from the economic analysis. Thus the economic analysis assessed the costs and outcomes of the following nine drugs: aripiprazole, asenapine, carbamazepine, valproate, haloperidol, lithium, olanzapine, quetiapine and risperidone.

Costs and outcomes considered in the analysis

The economic analysis adopted the NHS and personal social services perspective, as recommended by NICE (2012). Costs included hospitalisation costs, drug acquisition costs and costs of laboratory testing. The measures of effectiveness were determined by the outcome measures reported in Cipriani and colleagues (2011), which included the change scores on the YMRS as a primary outcome and the proportion of people who responded to treatment as a secondary outcome. Moreover, the economic analysis estimated the number of QALYs gained associated with each pharmacological treatment.

Time horizon of the analysis

The time horizon of the economic analysis was 3 weeks, the same as in the study by Bridle and colleagues (2004), which reflected the time horizons of the RCTs included in the network meta-analysis that provided the effectiveness data.

Clinical input parameters

All clinical input parameters were taken from the study by Cipriani and colleagues (2011). These included the SMDs of YMRS scores and the ORs of response rates, as well as the baseline probability of response for placebo. The latter was estimated by pooling the data from all placebo arms included in the network meta-analysis and found to equal 31.1%. This baseline probability of response was used to estimate the probability of response for each drug using the following formulae:

p_{x} = {odds}_{x} / (1 + {odds}_{x})

and

{odds}_{x} = (1 / {OR}_{b,x}) * p_{b} / (1 - p_{b})

where p_b is the probability of response for placebo (baseline), OR_b,x is the odds ratio for response of placebo versus each drug as reported in Cipriani and colleagues (2011) and odds_x the odds of each drug to achieve response.

Utility data and estimation of quality-adjusted life years

In order to express outcomes in the form of QALYs, the health states of the economic model need to be linked to appropriate utility scores. Utility scores represent the HRQoL associated with specific health states on a scale from 0 (death) to 1 (perfect health). More details on the estimation of utility scores, the NICE criteria on selection of available utility data and on the systematic review of the literature that aimed to identify utility scores associated with distinct health states experienced by adults with bipolar disorder are provided in section 6.4.5. This analysis considered utility scores corresponding to the health states of ‘mania’ equalling 0.44, and ‘full response – euthymia’ equalling 0.90, as reported in Table 21; the difference in utility between these states (0.46) was estimated using data reported in Revicki and colleagues (2005a). The utility score for mania was used for all people at the start of the model and for people not responding to treatment; the utility score for euthymia was used for people responding to treatment. The model assumed linear increase in utility in those responding to treatment between the start of the model and the point where response was achieved.

Table 21

Input parameters and utility data used to populate the economic model of pharmacological interventions for acute depression in adults with bipolar disorder.

Cost data

Similar to the economic analysis by Bridle and colleagues (2004), people in all arms of the economic model were assumed to be hospitalised over the 3-week time horizon of the analysis. Therefore, hospitalisation costs were the same across all drugs and were excluded from the guideline analysis.

The drug daily dosage was determined according to optimal levels of administration (based on the BNF and the GDG expert opinion) and was consistent with the dosage range reported in the RCTs included in the network meta-analysis by Cipriani and colleagues (2011). Drug acquisition costs were taken from the NHS Electronic Drug Tariff, February 2014¹⁷ (NHS Business Services Authority, 2014).

Required laboratory testing was determined by the GDG expert opinion. It was agreed that at initiation of all drugs a number of tests should be undertaken, including electrocardiogram (ECG), assessment of renal function (creatinine, blood urea and electrolytes), glucose, lipid profile and thyroid function tests. The costs of these tests were not included in the analysis because they were common to all arms of the model. In addition to these tests, the GDG expressed the opinion that liver function should be tested at initiation of all drugs except lithium; for lithium, three tests of plasma lithium concentration were required to determine optimal dose. The cost of liver function testing was taken from data reported in the economic analysis described in the previous NICE guideline (NCCMH, 2006). The cost of plasma lithium concentration testing was taken from the Newcastle upon Tyne Hospitals NHS Trust biochemistry laboratory services tariff for 2006/07.

All costs were uplifted to 2014 prices using the hospital and community health services (HCHS) pay and prices inflation index (Curtis, 2013). The inflation index for the year 2014 was estimated using the average value of the HCHS pay and prices indices of the previous 3 years.

The drug daily dosages and the associated acquisition costs, as well the laboratory testing costs that were utilised in the model are reported in Table 12.

Table 12

Average daily dosage, daily and 3-week acquisition costs, and additional required laboratory testing costs of pharmacological interventions for the treatment of adults with bipolar disorder experiencing a manic episode included in the economic analysis (more...)

Data analysis

Estimated costs of pharmacological interventions are presented alongside effectiveness data – SMDs of YMRS scores and ORs of response as reported in Cipriani and colleagues (2011) – and the mean QALY gain per person. Formal synthesis of costs and SMDs in an ICER was not attempted because the resulting figures would be difficult to interpret and therefore would not be useful in decision-making. On the other hand, ICERs expressing cost per additional responder were estimated despite the fact that they were difficult to interpret, to enable comparisons with the results reported in Bridle and colleagues (2004). In addition, incremental analysis where the ICER was expressed as cost per QALY was undertaken. Probabilistic analysis was not possible to undertake using the summarised efficacy data (mean and 95% CIs) that were reported in Cipriani and colleagues (2011). The cost data used in this analysis were very limited and were not subject to uncertainty because the drug and laboratory testing unit prices are determined. Therefore, other sensitivity analysis was not attempted.

Economic modelling results

Results of the economic analysis using the SMDs and the ORs of response of each drug versus placebo are presented in Table 13 and Table 14, respectively.

Table 13

Results of the economic analysis of pharmacological interventions for the treatment of adults with bipolar disorder experiencing a manic episode: effectiveness expressed by the standardised mean difference (SMD) of YMRS scores compared with placebo and (more...)

Table 14

Results of the economic analysis of pharmacological interventions for the treatment of adults with bipolar disorder experiencing a manic episode: effectiveness expressed by the odds ratios of response rates of placebo versus each drug, quality-adjusted (more...)

Table 14 also presents the QALY gains per person associated with each drug. In both tables, drugs have been ordered from the most to the least effective. As shown in Table 13, the three most effective drugs in terms of the magnitude of the SMD are haloperidol, risperidone and olanzapine; these drugs also have the lowest costs, all below £10 per person. These drugs are followed by quetiapine and lithium which have comparable costs, as well as aripiprazole which, however, has a total acquisition and laboratory testing cost of £76.

In terms of ORs of response and QALYs, the four most effective drugs were carbamazepine, haloperidol, olanzapine and risperidone, all with comparable costs. These are followed by quetiapine, which also has comparable costs, valproate, which has somewhat higher costs, and aripiprazole, the most costly drug of the analysis. According to formal incremental analysis, all drugs below the four most effective drugs are dominated by absolute dominance because they are less effective and more costly than one of more of the four most effective drugs. Haloperidol and olanzapine are dominated by rules of extended dominance (the latter occurs when an option is less effective and more costly than a linear combination of two alternative options). The ICER of carbamazepine versus risperidone is £149 per additional responder or £3,842 per QALY. It needs to be noted that carbamazepine was not among the most effective drugs in the analysis of YMRS change scores, which was the primary analysis of efficacy data in Cipriani and colleagues (2011). If carbamazepine is excluded from incremental analysis then haloperidol and olanzapine are no longer dominated. The ICER of haloperidol versus olanzapine is £283 per additional responder or £7,333 per QALY and the ICER of olanzapine versus risperidone is £151 per additional responder or £3,918 per QALY. Using the NICE cost effectiveness threshold of £20,000-£30,000 per QALY, haloperidol becomes the most cost-effective option if carbamazepine is excluded from analysis. This is followed by olanzapine then risperidone. Quetiapine is the next most cost-effective option, dominating all the remaining drugs in the analysis.

The ICERs expressing cost per additional responder are difficult to interpret because there is no set threshold for the WTP per additional responder to treatment for mania. Nevertheless costs were estimated, to enable comparison with the respective ICERs reported in Bridle and colleagues (2004). The comparison reveals that the ICERs estimated in this analysis are much lower than those reported by Bridle and colleagues, who estimated an ICER of olanzapine versus haloperidol equal to £7,179 per additional responder; this discrepancy may be attributable to the very different drug acquisition costs between the guideline analysis and the analysis by Bridle and colleagues (2004) because the latter, many of the drugs considered have become available in generic form. It should also be noted that the total costs reported in this analysis are substantially lower than those reported by Bridle and colleagues (2004), because this analysis did not include costs of hospitalisation which, in both analyses, were assumed to be common across all arms and were thus cancelled out.

The methodology checklist and the economic evidence profile of the analysis are provided in Appendices 31 and 33, respectively.

Discussion – limitations of the analysis

The results of the economic analysis suggest that haloperidol, olanzapine, risperidone and quetiapine may be more cost-effective than the other drugs assessed in the analysis. Carbamazepine was shown to be the most effective (and cost-effective) option when ORs of response and QALYs were used, but not in the analysis that utilised SMDs. After excluding carbamazepine from the cost-utility analysis, haloperidol became the most cost-effective treatment option, followed by olanzapine, risperidone and quetiapine. However, the efficacy and cost differences between haloperidol, olanzapine, risperidone and quetiapine were overall shown to be rather small.

The economic analysis is very simplistic and has only taken into account the costs associated with drug acquisition and the additional laboratory tests required for each drug over a period of 3 weeks. This short time horizon was imposed by the short time horizons of the RCTs that were included in the meta-analysis that provided the effectiveness data. Side effects and their impact on costs and HRQoL were not considered in the analysis, due to the short time horizon and the lack of relevant data. Hospitalisation costs were assumed to be the same for all drugs over 3 weeks because all people with bipolar disorder experiencing an acute episode were estimated to be hospitalised over the first 3 weeks of acute treatment. However, the total length of hospitalisation and outcomes of drugs beyond the 3-week period were not taken into account in the analysis due to lack of relevant data. If some drugs result in better outcomes beyond the period of the 3 weeks and reduce the total length of hospitalisation, then they are expected to be more cost-effective because hospitalisation is the most substantial driver of costs in the treatment of mania; the mean cost of Mental Health Care Clusters per bed-day was £344 in 2013, according to NHS reference costs (NHS Department of Health, 2013).

Another limitation of the analysis is the use of utility data from Revicki and colleagues (2005a) owing to the lack of more relevant utility data for the state of mania. The study described hypothetical health states using vignettes, which were valued by stable outpatients with bipolar disorder in the US. As discussed in section 6.3.8, these utility values do not meet NICE criteria on use of utility values and do not reflect the UK general population’s preferences. The results of the cost-utility analysis should be therefore interpreted with caution.

In conclusion, the analysis has not overcome many of the limitations characterising previous studies. Nevertheless, the results indicate that haloperidol, olanzapine, risperidone and quetiapine may be more cost-effective options for the treatment of mania than other drugs assessed. Factors such as acceptability, rate and type of side effects associated with each drug should also be considered when making recommendations. Moreover, it needs to be noted that a number of drugs included in the economic analysis are currently patented and are associated with high acquisition costs. Once these drugs become available in generic form their price is expected to reduce, and their relative cost effectiveness is then likely to change thus needing re-assessment.

Economic evidence statement

The existing economic evidence is rather limited and not directly applicable to the NICE decision-making context; all reviewed studies are characterised by potentially serious limitations. In the economic analysis conducted for this guideline, haloperidol, olanzapine, risperidone and quetiapine appear to be more cost-effective options than other drugs included in the analysis. However, this analysis is also characterised by potentially serious limitations.

6.3. Pharmacological and Nutritional Interventions for Acute Episodes of Bipolar Depression

6.3.1. Introduction

People with bipolar disorder spend considerably more time depressed than manic; for those with bipolar I disorder it has been estimated that for two-thirds of the time that they are unwell it is with depression (Judd et al., 2003a; Judd et al., 2002a). For those with bipolar II disorder, over 90% of unwell days are due to depression. Bipolar disorder is associated with a high prevalence of suicide with most of these occurring during the depressed phase (Novick et al., 2010). A number of medications have been used for bipolar depression, alone and in combination, including antidepressants used for unipolar depression (SSRIs, tricyclics, monoamine oxidase inhibitors) as well as antipsychotics, anticonvulsants and lithium.

6.3.2. Clinical review protocol

The review protocol summary, including the review question and the eligibility criteria used for this section of the guideline, can be found in Table 15 (a complete list of review questions and protocols can be found in Appendix 7; further information about the search strategy can be found in Appendix 8).

Table 15

Clinical review protocol summary for the review of pharmacological and nutritional interventions for acute episodes of bipolar depression.

6.3.3. Studies considered

Twenty-seven RCTs (N = 9,006) published between 1999 and 2012 compared eligible interventions and reported outcomes that could be used for network meta-analysis: BRISTOLMYERSSQUIB2006 (Bristol-Myers Squibb, [unpublished] 2006), BRISTOLMYERSSQUIB2007 (Bristol-Myers Squibb, [unpublished] 2007), BROWN2006 (Brown et al., 2006), CALABRESE1999 (Calabrese et al., 1999), CALABRESE2005 (Calabrese et al., 2005a), CALABRESE2008a (Calabrese et al., 2008), CALABRESE2008b (Calabrese et al., 2008), CALABRESE2008c (Calabrese et al., 2008), CALABRESE2008d (Calabrese et al., 2008), DAVIS2005 (Davis et al., 2005), GHAEMI2007 (Ghaemi et al., 2007), MCELROY2010 (McElroy et al., 2010), MUZINA2011 (Muzina et al., 2011), NEMEROFF2001 (Nemeroff et al., 2001), PFIZER2009a (Pfizer, [unpublished] 2009a), PFIZER2009b (Pfizer, [unpublished] 2009b), QUANTE2010 (Quante et al., 2010), SACHS2011 (Sachs et al., 2011), SILVERSTONE2001 (Silverstone, 2001), SUNOVION2012a (Sunovion Pharmaceuticals Inc., [unpublished] 2012a), SUNOVION2012b (Sunovion Pharmaceuticals Inc., [unpublished] 2012b), SUPPES2010 (Suppes et al., 2010), THASE2006 (Thase et al., 2006), TOHEN2003 (Tohen et al., 2003), TOHEN2012 (Tohen et al., 2012a), VANDERLOOS2009 (Van der Loos et al., 2009), YOUNG2010 (Young et al., 2010). Six of these were unpublished (BRISTOLMYERSSQUIB2006, BRISTOLMYERSSQUIB2007, PFIZER2009a, PFIZER2009b, SUNOVION2012a, SUNOVION2012b). Studies included in the network meta-analysis were analysed by comparing discontinuation (for any reason) and response, given not discontinued.

A joint network meta-analysis on discontinuation and number of responders given not discontinued was carried out by subtracting the number of patients who had discontinued from the total number of patients randomised. A separate network meta-analysis to estimate relative effects of response out of all randomised patients (that is, not conditional on discontinuation) was also carried out.

All studies reported the number of patients discontinuing out of the total number randomised, but only 25 studies reported a useable measure of response on a dichotomous or continuous scale (BRISTOLMYERSSQUIB2006 and BRISTOLMYERSSQUIB2007 did not report response).

Data on response were reported in different formats. The relative effect of interest was the odds ratio of response, so the following approach was taken to incorporate as much of the available data as possible:

(1): For studies reporting the number of responders on only one of the Hamilton Depression Rating Scale (HAMD) or Montgomery Ǻsberg Depression Rating Scale (MADRS) scales, those data were used in the analysis.
(2): For studies reporting the number of responders on both the HAMD and MADRS the log-odds ratio of response, given not discontinued and given by each measure was averaged and the standard error of the log-odds ratios was calculated as the average of the standard errors on each scale.
(3): For studies not reporting the number of responders but reporting the mean and standard deviation (SD) on one of the scales (HAMD or MADRS), the within-study standardised mean difference (SMD) and its variance were calculated according to the Hedges’ g formula and used in the analysis.
(4): For studies not reporting the number of responders but reporting the mean and SD on both the HAMD and MADRS scales, the within-study SMD on each scale and their standard errors were calculated as above, and then averaged. This combined SMD and its variance (the standard error squared) were used in the analysis.

One additional three-arm study (N = 174; POST2006 [Post et al., 2006) was a comparison of three drugs that could not be connected to the network, so the pairwise comparisons are reported separately below.

An additional 29 studies were excluded; eight were open-label studies: AMSTERDAM2009 (Amsterdam & Shults, 2009), ASTRAZENECA2012a (Astrazeneca, [unpublished] 2012a), ASTRAZENECA2012b (Astrazeneca, [unpublished] 2012b), NIERENBERG2006 (Nierenberg et al., 2006), NOLEN2007 (Nolen et al., 2007), TAMAYO2009 (Tamayo et al., 2009), WANG2010 (Wang et al., 2010), YONGNING2005 (Yong Ning & Hui, 2005); seven trials were of medications neither routinely used nor licensed for the treatment of mental health problems: CHENGAPPA2000 (Chengappa et al., 2000), DENICOFF2005 (Denicoff et al., 2005) DIAZGRANADOS2010 (Diazgranados et al., 2010), FUREY2013 (Furey & Zarate, 2013), STAMM2011 (Stamm et al., 2011), SZUBA2005 (Szuba et al., 2005), WATSON2012 (Watson et al., 2012), YOUNG2004 (Young et al., 2004), ZARATE2012 (Zarate et al., 2012); and four trials included people who did not have bipolar disorder: FIEVE1968 (Fieve et al., 1968), KESSELL1975 (Kessell & Holt, 1975), SMITH1978 (Smith et al., 1978), SPEER2009 (Speer et al., 2009). Three studies were excluded because did not include a sufficient number of participants to be included; one was a study of pramipexole as a second-line intervention: GOLDBERG2004 (Goldberg et al., 2004); one was a study of pramipexole: ZARATE2004B (Zarate et al., 2004); one was a study of paroxetine and mood stabilisers: YOUNG2000; and one was a study of risperidone and paroxetine: SHELTON2004 (Shelton & Stahl, 2004). One study was excluded because it involved a comparison of antidepressants as a class (rather a specific drug) with placebo: SACHS2007. One study of tranylcypromine was excluded because it did not report response on an accepted measure: HIMMELHOCH1991 (Himmelhoch et al., 1991). Two studies were excluded because they did not report usable outcomes; one compared olanzapine and fluoxetine alone or in combination: AMSTERDAM2005a (Amsterdam & Shults, 2005a); one compared valproate with lithium: OQUENDO2011 (Oquendo et al., 2011). One study of eicosapentaenoic acid was excluded because there were only six participants in each group: OSHER2005 (Osher et al., 2005). One was excluded because participants were not acutely depressed: FRANGOU2006 (Frangou et al., 2006). Results could not be obtained for eight studies: AHUJA2011 (Ahuja et al., 2011), COLOMBO2000 (Colombo et al., 2000), FOREST2010 (Forest, 2010), FRYE2000 (Frye et al, 2000) GAO2008 (Gao et al., 2008), MCELROY2013 (McElroy et al., 2013), PATKAR2012 (Patkar et al., 2012), SACHS2002 (Sachs et al., 2002); although they have published several papers about the drug, the manufacturer of cariprazine has not reported the results of clinical trials, and they refused requests for data.

Further information about both included and excluded studies can be found in Appendices 16 and 34.

6.3.4. Network meta-analysis of pharmacological interventions for acute episodes of bipolar depression

Trials included in the network meta-analysis included between 19 and 833 participants at baseline (median 298). Where known, participants were on average (median of means) aged 40 years and about 58% of them were female. Fourteen trials included only participants with bipolar I disorder; one trial included only participants with bipolar II disorder (CALABRESE2008c), and only 37% of participants in another had bipolar II disorder (MUZINA2011).

Studies of medication alone or as an addition to another treatment were included. All participants were taking a mood stabiliser in six studies (QUANTE2010, SACHS2011, NEMEROFF2001, SUNOVION2012a, SUNOVION2012b, VANDERLOOS2009). Twelve studies reported that participants were not taking mood stabilisers at baseline (BRISTOLMYERSSQUIB2006, BRISTOLMYERSSQUIB2007, CALABRESE1999, CALABRESE2005, CALABRESE2008a, CALABRESE2008b, CALABRESE2008c, CALABRESE2008d, DAVIS2005, GHAEMI2007, MCELROY2010, MUZINA2011, PFIZER2009a, PFIZER2009b, SUPPES2010, THASE2006, TOHEN2003, YOUNG2010), though participants in some of these studies could be taking other medications including anxiolytics or hypnotics. Nine studies included a mix of participants taking or not taking mood stabilisers, or did not report their use.

Quality of the evidence

To rate the quality of evidence, guidelines may use GRADE profiles for critical outcomes. However, GRADE has not yet been adapted for use in network meta-analyses. To evaluate the quality of the evidence from the network meta-analysis, information about the factors that would normally be included in a GRADE profile will be reported (that is, risk of bias, publication bias, imprecision, inconsistency and indirectness).

6.3.5. Risk of bias

All included trials were assessed for risk of bias (Appendix 17). Of those in the network meta-analysis, 21 were at low risk for sequence generation and nine of these were at low risk of bias for allocation concealment. Allocation concealment was unclear in 18 trials. All trials were double-blind and were rated as low risk of bias for participant and provider blinding, although effects of medication, including side effects, may make it difficult to maintain participant and provider blinding, particularly at higher doses. Assessor blinding was considered separately for all trials; seven were at low risk of bias and assessors were aware of treatment conditions in one trial. For incomplete outcome data, response was analysed assuming that participants who discontinued treatment did not respond. Because of the high rate of missing data and/or the handling of missing data, continuous outcomes were at high risk of bias in 22 trials.

Selective outcome reporting and publication bias

Several methods were employed to minimise risk of selective outcome reporting and publication bias. The NCCMH review team wrote to all authors to request trial registrations and unpublished outcomes, and all authors of included trials, all stakeholders, and pharmaceutical manufacturers were asked to provide unpublished trials. Nonetheless, only six were at low risk of selective outcome reporting bias, the remaining 14 and seven were at unclear and high risk of bias (see Figure 5).

Figure 5

Risk of bias summary.

Inconsistency

Inconsistency was assessed by fitting an unrelated mean effects model (Dias et al., 2013) and comparing the fit of this model to the fit of the full network meta-analysis model using the residual deviance (Dias et al., 2013). The posterior mean of the residual deviance for discontinuation was 63.5, very close to the respective 64 data points of the model; the posterior mean of the total residual deviance for response was 58.44, moderately high compared with the respective 51 data points. This finding may be attributable to one study (THASE2006) that did not fit the model well regarding response. Only one loop in the network had the potential for inconsistency, and there was no evidence of inconsistency for response and for discontinuation.

Indirectness

All evidence in the network meta-analysis is direct insofar as it relates to the population, interventions and outcomes of interest.

Effects of interventions

In the network meta-analysis, all interventions except aripiprazole ranked higher than placebo for response given no discontinuation, but only six were statistically superior to placebo (lurasidone, valproate, quetiapine, the combination of fluoxetine and olanzapine, olanzapine alone, and lamotrigine) (see Table 16). Quetiapine and lurasidone were less well tolerated than placebo; for discontinuation, the combination of fluoxetine and olanzapine, valproate, olanzapine alone and lamotrigine ranked higher than placebo. When responses for all randomised participants (that is, assuming the dropouts did not respond) were compared, moclobemide and ziprasidone were also ranked below placebo. Other interventions that were included in the network but were not statistically superior to placebo were imipramine, lithium, moclobemide, paroxetine and ziprasidone. Excluding valproate, which only 48 people received, the five efficacious interventions were received by 292 to 1,867 participants.

Table 16

Pharmacological interventions for acute episodes of bipolar depression (results from network meta-analysis).

6.3.6. Pharmacological interventions for acute episodes of bipolar depression that could not be included in the network meta-analysis

One RCT (N = 174; POST2006) published in 2006 compared bupropion, sertraline and venlafaxine in outpatients. In the total sample, mean age was 42 years, 50% were female and 73% were diagnosed with bipolar I disorder. Little difference was found between any of the groups on response and discontinuation.

6.3.7. Nutritional interventions for acute episodes of bipolar depression

One RCT (N = 116) published in 2006 compared medication as usual with or without eicosapentaenoic acid supplementation (KECK2006b [Keck et al., 2006b]). There was very low quality evidence that eicosapentaenoic acid supplementation was not associated with a reduction in depressive symptoms (see Appendix 16).