Economic modelling methods
Interventions assessed
The guideline economic analysis assessed those interventions for adults with social anxiety disorder that are available in the UK, and for which there was adequate clinical evidence to indicate their effectiveness along with an acceptable risk-to-benefit ratio. Further to these criteria, pharmacological interventions were included in the economic analysis if they are prescribed in routine UK clinical practice for the management of anxiety disorders. Benzodiazepines were not included in the economic analysis because they are not indicated for use longer than 2 to 4 weeks for the treatment of anxiety (Joint Formulary Committee, 2013). Computerised interventions were included in the economic analysis, despite their current unavailability in UK routine practice, because they are used in other countries and because they are likely to become available in the UK. Moreover, this guideline updates the NICE TA on computerised CBT (CCBT) for depression and anxiety (NICE, 2006), regarding phobias (see Chapter 8). Social skills training was not considered a separate intervention in the analysis because it is an element contained in other psychological interventions that were included in the economic model and the GDG was of the view that it is best provided as part of those therapies rather than as a separate intervention.
Based on the above criteria the following interventions were included in the economic analysis:
Pharmacological interventions: citalopram, escitalopram, fluoxetine, fluvoxamine, mirtazapine, moclobemide, paroxetine, phenelzine, pregabalin, sertraline and venlafaxine; for the latter, extended release formulations were considered, in accordance with the formulations used in the relevant RCTs included in the NMA that informed the economic model.
Psychological interventions: group CBT, individual CBT, group CBT (Heimberg), individual CBT (Heimberg), standard CT (Clark and Wells), CT (Clark and Wells) with shortened sessions, exposure in vivo, mindfulness training, IPT, psychodynamic psychotherapy, self-help (book) with and without support, self-help (internet) with and without support, and supportive therapy.
The model also considered treatment with pill placebo, consisting, in terms of resource use, of GP visits only, as well as waitlist as alternative treatment options, in order to assess the cost effectiveness of active interventions versus a non-specific medical management (represented by pill placebo) and a ‘do nothing’ option (represented by waitlist). Combined interventions (comprising concurrent provision of both pharmacological and psychological interventions) were not considered in the model structure because the GDG judged that the respective evidence was very limited (each combination therapy included in the guideline systematic review was assessed in one single small trial). Moreover, in many trials combined interventions were found to be less effective than their components and were associated with increased side effects and lower tolerability, so they were obviously less cost effective (more intensive and less effective) than single interventions; consequently there was no need for a formal evaluation of their cost effectiveness.
Model structure
A hybrid decision-analytic model consisting of a decision-tree followed by a two-state Markov model was constructed using Microsoft Office Excel 2007. The model estimated the total costs and benefits associated with provision of various interventions to adults with social anxiety disorder. The structure of the model, which aimed to simulate course of illness and relevant clinical practice in the UK, was also driven by the availability of clinical data.
According to the model structure, hypothetical cohorts of adults with social anxiety disorder were initiated on each of the 28 treatment options assessed, including treatment with pill placebo or inclusion in a waitlist. The duration of initial treatment was 12 weeks for drugs and pill placebo; for psychological interventions it varied by intervention (range between 9 and 16 weeks). In order to estimate QALYs it was assumed that psychological interventions lasted 12 weeks as well, which was consistent with the trial data and with clinical practice. Following treatment, people in each cohort either recovered (that is, they did not meet criteria for a diagnosis of social anxiety disorder any longer) or did not recover. Those recovering were given 6 months of maintenance therapy if they had been initiated on pharmacological treatment in order to sustain the treatment effect. No booster sessions were modelled for psychological interventions, as clinical evidence indicated that these are not necessary for sustained treatment effect. People who did not recover were assumed to stop treatment rather than switch to an alternative intervention; according to the expert opinion of the GDG, because of the nature of the disorder people are usually reluctant to keep in contact with health services and try an alternative treatment option.
During the year post-treatment, people who had recovered might experience a relapse, meaning that they again met the criteria for social anxiety disorder. People who had not recovered following treatment were assumed to remain in a state of social anxiety and not to recover spontaneously over this year. From that point on, all people in each cohort – both those who no longer met criteria for social anxiety disorder (that is, those who recovered and did not relapse) and those who met criteria (that is, those who recovered but relapsed as well as those who did not recover following therapy) – were entered into the Markov model. They could remain in the same health state or move between the two states of ‘no social anxiety’ and ‘social anxiety’. The Markov model was run in yearly cycles. A half-cycle correction was applied. Because of lack of long-term comparative clinical data, transitions between the two health states in the Markov model were assumed to be independent of intervention received at the start of the model.
The analysis considered two time horizons in order to explore short and longer-term costs and benefits: (1) intervention time (12 weeks) plus 1 year post-treatment (represented by the decision tree); and (2) intervention time (12 weeks) plus 5 years post-treatment (consisting of the decision tree and four yearly cycles of the Markov model). The GDG was interested in the long-term cost, benefits and cost effectiveness of the interventions assessed in the analysis and focused on the 5-year post-treatment results. However, 1-year post-treatment findings were also reviewed in order to explore the changes in the relative cost effectiveness of interventions over time. A schematic diagram of the economic model structure is presented in .
Schematic diagram of the economic model constructed for the assessment of the relative cost effectiveness of interventions for adults with social anxiety disorder.
Costs and outcomes considered in the analysis
The economic analysis adopted the perspective of the NHS and PSS, as recommended by NICE (2009b). Costs consisted of intervention costs (healthcare professional time, drug acquisition and equipment/infrastructure required for self-help interventions) and other health and social care costs incurred by people with social anxiety disorder not recovering following treatment or experiencing a relapse following recovery (including GP consultations, home visits from health and social services, counselling or therapy contacts, and inpatient and outpatient secondary care). A secondary analysis that adopted a wider perspective which, in addition to NHS and PSS costs, considered receipt of social security benefits by people with social anxiety disorder was also undertaken. The measure of outcome was the QALY.
Clinical input parameters and overview of methods employed for evidence synthesis
Clinical model input parameters consisted of the probability of recovery at end of treatment, the probability of relapse following recovery within the first year post-treatment, as well as the probabilities of recovery and relapse in the 4-year Markov model phase.
The probability of recovery at end of treatment for all interventions was derived from the NMA undertaken for this guideline. The clinical effectiveness of all interventions was expressed in the form of SMDs. For the economic analysis, the SMDs were transformed into LORs as described in Chapter 3, and these were further transformed into probabilities of recovery, using as baseline the absolute probability of recovery for waitlist, which was estimated from available recovery data on waitlist arms in RCTs that were included in the guideline systematic review. The 40,000 iterations that were recorded in WinBUGS (as described in Chapter 3) were thinned by 4 so as to obtain 10,000 iterations for use in the economic model. This transformation of SMDs derived from the NMA into LORs and the subsequent indirect estimation of probability of recovery for every intervention assessed in the economic analysis was necessary for three reasons:
The recovery data reported in the RCTs included in the guideline systematic review were sparse and not available for all interventions assessed in the economic analysis: of the 101 studies included in the NMA, only 25 reported recovery data; such data were available for 14 out of the 29 interventions considered in the economic analysis. Consequently, available recovery data were not adequate for populating all arms of the economic model.
The economic analysis needed to reflect (and thus utilise) the same relative treatment effects that were estimated in the NMA, which determined the comparative clinical effectiveness of the interventions considered in this guideline.
The methodology adopted allowed estimation of probability of recovery for every intervention included in the economic analysis while preserving the effect of randomisation, because the probability of recovery of each intervention was ‘linked’ to the relative treatment effect of the intervention as estimated in the NMA.
The results of the NMA that were used to populate the economic model are provided in . The table shows the probability of recovery at end of treatment for each intervention considered in the economic analysis. Treatment options have been ranked from most to least efficacious in terms of mean probability of recovery.
Results of the NMA that were utilised in the economic model – probability of recovery at end of treatment.
The probability of relapse after recovery within the first year following pharmacological intervention was estimated based on relevant data reported in relapse prevention studies included in the guideline systematic review. Five placebo-controlled trials assessed the efficacy of pharmacological interventions in preventing relapse in people with social anxiety disorder: four of them assessed an SSRI (KUMAR1999 and STEIN2002b assessed paroxetine, MONTGOMERY2005 escitalopram, VAN-AMERINGEN2001 sertraline) and one assessed pregabalin (GREIST2011). All five studies reported a 6-month ‘drug’ relapse rate for people with social anxiety disorder who had responded to initial drug treatment (12 weeks) and were maintained on drug treatment during the 6 months of the trial (therefore the 6-month ‘drug’ relapse rate referred to participants who relapsed while taking an active drug as maintenance treatment), as well as a 6-month ‘placebo’ relapse rate for people with social anxiety disorder who had responded to initial 12-week drug treatment and received placebo during the 6 months of the study (therefore the 6-month ‘placebo’ relapse rate referred to participants who had responded to 12-weeks of initial drug treatment but then were discontinued from the drug and were given placebo instead). The economic model structure assumed that within the first year following initial drug treatment people who recovered received 6 months of maintenance treatment. Assuming that drug maintenance treatment does provide a benefit and reduces the risk of relapse (compared with discontinuation of the drugs immediately after the initial 12-week treatment), the risk of relapse in the 6 months following maintenance treatment should be lower than the pooled ‘placebo’ relapse rate from the placebo arms of the relapse prevention studies. Conversely, the risk of relapse after stopping the 6-month maintenance treatment should be higher than the pooled ‘drug’ relapse rate from the active drug arms of the relapse prevention studies, which was recorded while people were still on a drug. It was therefore assumed that over the first year following pharmacological intervention people who recovered were maintained on their initiated drug for 6 months and experienced relapses at the ‘drug’ relapse rate, and, after stopping the drug, for the remaining 6 months, they continued to experience some relapses at an overall (annual) rate that was lower than the 6-month ‘placebo’ relapse rate (it was assumed that the ‘placebo’ relapse rate did not increase after 6 months following discontinuation, and therefore the annual ‘placebo’ relapse rate should not be different from the 6-month ‘placebo’ relapse rate). For the purposes of simplicity and because of a lack of more suitable data, it was assumed that the probability of relapse after recovery within the first year following pharmacological treatment equalled the midpoint between the pooled ‘drug’ relapse rate and the pooled ‘placebo’ relapse rate reported in the relapse prevention studies included in the guideline systematic review. This estimate was utilised in all decision nodes of the model that involved drug treatment because relapse data for drugs were sparse and not available for the majority of pharmacological interventions considered in the economic analysis.
The probability of relapse following recovery in the pill placebo arm of the model was assumed to equal that for drug arms. This probability was deliberately not set to equal the ‘placebo’ relapse rate because people who had recovered in this arm had not been initiated on a drug (so they did not experience drug discontinuation that might potentially lead to an increase in the risk of relapse similar to the relapse rate of the placebo arms of the relapse prevention studies).
The probability of relapse after recovery within the first year following psychological intervention was calculated using the respective probability of relapse for drugs, estimated as described above, and the risk ratio (RR) of relapse between drugs and psychological interventions. The latter was derived from an observational study that evaluated the effects of maintenance treatment with phenelzine and group CBT (Liebowitz et al., 1999). The study followed an RCT that compared phenelzine, group CBT, pill placebo and psychological placebo (HEIMBERG1998). People who were initiated on either phenelzine or group CBT and had responded to 12 weeks of treatment (N = 28) were continued on their initial treatment for 6 months, and followed up for another 6 months during which they received no treatment. The study reported relapse rates over the 6-month maintenance treatment period, the 6-month follow-up period, and the combined 12-month period. A risk ratio of relapse for drugs (represented by phenelzine) versus psychological intervention (represented by group CBT) was estimated using the 12-month combined relapse data reported in the study. The probability of relapse after recovery for the psychological arms of the model was subsequently calculated as:
This probability was applied to all psychological intervention arms of the economic model, since no differential relapse data for the range of psychological interventions considered in the model are available in the literature.
The probability of relapse after recovery in the waitlist arm of the model was based on data reported in a prospective naturalistic study that followed people with anxiety disorders over 12 years (Bruce et al., 2005). The study followed 176 people with social anxiety disorder and reported a 12-year probability of recurrence, calculated using standard survival analysis methods. This probability allowed estimation of an annual probability of relapse that was applied to the first year after recovery in the waitlist arm.
The annual probabilities of recovery and relapse for all treatments in the 4-year Markov model phase were assumed to be independent of initial treatment and were also based on data reported in Bruce and colleagues (Bruce et al., 2005). In addition to the 12-year probability of recurrence, the authors also calculated a 12-year probability of recovery using survival analysis, which was used to estimate an annual probability of recovery. The estimated annual probabilities of recovery and relapse were applied to all cohorts in the economic model, regardless of initial treatment, in years 2 to 5 post-treatment (that is, in the Markov phase of the model).
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 needed to be linked to appropriate utility scores. Utility scores represent the Health Related Quality of Life (HRQoL) associated with specific health states on a scale from 0 (death) to 1 (perfect health); they are estimated using preference-based measures that capture people's preferences on the HRQoL experienced in the health states under consideration.
The systematic search of the literature identified one study that reported utility scores for specific health states associated with social anxiety (François et al., 2008) and two studies that reported utility data for adults with social anxiety disorder (and adults without a mental disorder), without differentiating between distinct health states of the condition: (1) Alonso and colleagues (2004), with data analysed and reported in Kaltenthaler and colleagues (2006), and (2) Saarni and colleagues (2007).
François and colleagues (2008) generated utility scores using SF-36 data derived from 517 people with social anxiety disorder who participated in 12 weeks of open-label treatment with escitalopram. Those responding to treatment were entered into a double-blind, placebo-controlled, multinational clinical trial of escitalopram for relapse prevention (MONTGOMERY2005). Participants were included in the open-label phase if they had had a primary diagnosis of generalised social anxiety and a score of 70 or more on the LSAS. Response to treatment was defined as a Clinical Global Impression-Improvement (CGI-I) score of 1 or 2; relapse was defined as either an increase in LSAS total score of 10 or more points or withdrawal of the participant from the study because of lack of efficacy as judged by the investigator. SF-36 data were obtained from participants at baseline, the end of the open-label period, and at 12 and 24 weeks after randomisation. Participants who did not complete the study attended an early discontinuation visit, at which the SF-36 was administered. SF-36 scores were converted into utility scores using the SF-6D algorithm (Brazier et al., 2002). The SF-6D algorithm has been generated using the standard gamble (SG) technique in a representative sample of the UK general population.
Alonso and colleagues (2004) reported EQ-5D and SF-36 data of people participating in a large, community-based mental health European survey, the European Study of the Epidemiology of Mental Disorders. Participants were members of the general population that underwent psychiatric assessments and completed various HRQoL instruments. The authors conducted additional analyses to those reported in their publication and generated EQ-5D and SF-36 utility scores for people who had experienced a wide range of mental disorders over the past 12 months (among whom 219 had social anxiety) and 19,334 people without a mental disorder over the past 12 months. Estimated utility scores were subsequently provided to the research team that conducted the economic analysis for the NICE TA on the use of CCBT for depression and anxiety (Kaltenthaler et al., 2006). Thus, EQ-5D and SF-6D utility scores derived from the European Study of the Epidemiology of Mental Disorders are available in Kaltenthaler and colleagues (2006). Utility scores for EQ-5D have been elicited from the UK general population using the time trade-off technique (TTO) (Dolan et al., 1996; Dolan, 1997). The SF-6D algorithm has been generated using SG in a representative sample of the UK general population (Brazier et al., 2002).
Saarni and colleagues (2007) reported EQ-5D data obtained from people aged 30 years or older, participating in a national health survey in Finland. The survey consisted of a health interview, a health examination and self-report questionnaires. The study reported EQ-5D utility scores for people who had experienced a range of mental disorders over the past 12 months (among whom 60 had social anxiety anxiety, with 14 having pure social anxiety) and 5,279 people with no mental disorder over the last 12 months. The authors used the UK TTO tariff (Dolan, 1997) in order to estimate utility scores from EQ-5D data.
summarises the methods used to derive and value health states associated with social anxiety disorder in the literature and presents the respective utility scores reported in the three utility studies that were identified by the systematic search of the literature.
Summary of studies reporting utility scores for social anxiety.
According to NICE guidance (NICE, 2013) on the selection of utility values for use in cost-utility analysis, the measurement of changes in HRQoL should be reported directly from people with the condition examined, and the valuation of health states should be based on public preferences elicited using a choice-based method, such as the TTO or SG, in a representative sample of the UK population. NICE recommends the EQ-5D (Brooks, 1996; Dolan, 1997) as the preferred measure of HRQoL in adults for use in cost-utility analysis. When EQ-5D scores are not available or are inappropriate for the condition or effects of treatment, NICE recommends that the valuation methods be fully described and comparable to those used for the EQ-5D (NICE, 2013).
The study by François and colleagues (2008) was the only one that reported utility data for different health states of social anxiety disorder. However, the GDG questioned the quality of the data because of the methodological limitations of MONTGOMERY2005, such as the high attrition rates. Moreover, the GDG felt that the utility data reported in the study represented a rather narrow benefit in HRQoL, as the difference in the utility scores between the states of response and non-response was only 0.031; for comparison, a study with similar design that estimated utility scores in responders and non-responders in generalised anxiety reported a respective difference of 0.13 (Allgulander et al., 2007). In addition, the reduction in utility for those relapsing following response was 0.017 in people with social anxiety disorder according to François and colleagues, and 0.03 in people with generalised anxiety according to Allgulander and colleagues. The GDG therefore judged that utility data reported by François and colleagues might have failed to capture the true benefit in HRQoL once a person with social anxiety disorder responds to treatment, and the true loss in HRQoL once the person relapses following response.
It should be noted that François and colleagues compared their findings with those of Allgulander and colleagues and admitted that ‘the effect of escitalopram on HRQoL was somewhat more modest in patients with generalised social anxiety disorder than in those with generalised anxiety disorder’. However, it was not the effect of escitalopram that was responsible for the discrepancies in the utility changes between the two studies and populations because utility changes reflected alterations in HRQoL once a person had/had not experienced response or relapse, with the two states being defined in a similar way in the two studies. Another point for consideration was that the GDG was interested in the utility of the recovery state, whereas the data reported in François and colleagues referred to the state of response. Finally, François and colleagues reported utility values based on the SF-6D, which is not the NICE preferred measure for use in cost-utility analysis. For all the above reasons the GDG decided not to use the utility data reported in François and colleagues, despite these being the only utility data capturing HRQoL in different health states of social anxiety disorder that were identified in the literature.
The GDG then assessed the EQ-5D-based utility data reported in Alonso and colleagues (2004) and the utility data from Saarni and colleagues (2007). The two studies were very similar in terms of design and reported utility data for people with social anxiety disorder over the last 12 months and for people without a mental disorder over the last 12 months. It was agreed that the utility data for the former could be used for the state of non-recovery or relapse (‘social anxiety’) in the guideline economic model; the utility data for people without a mental disorder over the last 12 months could be used as a proxy for the state of recovery (‘no social anxiety’). It was acknowledged that this is probably not a very accurate proxy because people recovering from social anxiety disorder may not reach the HRQoL of a person without a mental disorder over the last 12 months. Another limitation of these data is that the diagnosis of social anxiety disorder referred to a period of 12 months prior to the study, so some participants in both studies might have experienced an improvement in their condition over this period (and actually might not have social anxiety disorder at the point of interview). Nevertheless, the GDG accepted these as reasonable limitations and decided to use the data by Saarni and colleagues (2007) in the base-case analysis (which reflect a greater improvement in HRQoL following recovery), and to use the more conservative data by Alonso and colleagues in sensitivity analysis. Utility data from both studies are based on the EQ-5D UK tariff and therefore are in accordance with NICE guidance on the selection of utility data for use in cost-utility analysis (NICE, 2013).
It was assumed that the improvement in utility for people with social anxiety disorder recovering following treatment occurred linearly over the duration of treatment, starting from the utility value of social anxiety disorder and reaching the utility value of no social anxiety disorder. The duration of all treatments considered in the analysis was assumed to be 12 weeks in order to simplify calculation of utilities in people improving following treatment across cohorts. All changes in utility between the two states of ‘social anxiety’ and ‘no social anxiety’ were assumed to occur linearly over the time period of the change.
Side effects from medication are expected to result in a reduction in utility scores of people with social anxiety disorder. Disutility because of side effects was not considered in the analysis because the model structure did not incorporate side effects. This was due to inconsistent reporting of specific side effect rates across the studies included in the guideline systematic review. Moreover, no studies on people with social anxiety disorder reporting ‘disutility’ because of side effects were identified in the literature. However, Revicki and Wood (1998) examined the effect of side effects from antidepressants in the HRQoL of people with depression. According to the study, people with a side effect reported lower utility scores compared with those not experiencing side effects. The observed mean disutility ranged from 0.01 for dry mouth and nausea to 0.12 for nervousness and light-headedness. However, except for light-headedness and dizziness, the reduction in utility caused by side effects did not reach statistical significance. The GDG felt that it may be reasonable to extrapolate this evidence to the population of people with social anxiety disorder; consequently, it is possible that lack of consideration of disutility because of side effects has not had a great impact on the results of the economic analysis. Nevertheless, omission of the negative impact of drugs on HRQoL of adults with social anxiety disorder is acknowledged as a limitation of the analysis because it may have resulted in an over-estimation of the cost effectiveness of pharmacological interventions relative to psychological interventions considered in the model.
Cost data
Costs considered in the economic model consisted of intervention costs and extra health and social care costs incurred by adults with social anxiety disorder not recovering following treatment or relapsing following recovery. In addition, a secondary analysis considered receipt of social security benefits by adults with social anxiety disorder not recovering or relapsing following recovery.
Pharmacological intervention costs consisted of drug acquisition costs and GP visit costs. Intervention costs of placebo related to GP visit costs only. Costs were calculated by combining resource use estimates with respective national unit costs. Drug acquisition costs were taken from the NHS Electronic Drug Tariff, February 2013 (NHS Business Services Authority Prescription Pricing Division, 2013). For each drug the lowest reported price was selected and used in the analysis; where available, costs of generic forms were considered. The average daily dosage of each drug was determined according to optimal clinical practice (according to GDG expert opinion) and was consistent with the respective average daily dosage reported in the RCTs considered in the NMA that informed the economic model. Initial treatment with drugs was estimated to last 12 weeks, while people recovering following drug treatment received another 6 months (26 weeks) of maintenance treatment at the same daily dosage. All people under any pharmacological treatment (or placebo) were assumed to visit their GP four times over the 12 weeks of initial treatment; in addition, those recovering were assumed to pay three extra GP visits during maintenance treatment. The GP unit cost (£43 per patient contact lasting 11.7 minutes) was taken from Unit Costs of Health and Social Care 2012 (Curtis, 2012). This figure includes direct care staff costs and qualification costs.
Details on the resource use and total intervention costs of pharmacological interventions for adults with social anxiety are presented in .
Average daily dosage, drug acquisition costs and total intervention costs of pharmacological interventions for adults with social anxiety disorder included in the economic model (2012 prices).
Intervention costs of psychological interventions were also calculated by combining resource use estimates with relevant national unit costs. Resource use estimates in terms of therapists' time were based on relevant data reported in RCTs included in the NMA that informed the economic model. For self-help studies the additional cost of a book or a computerised program was considered. All psychological interventions were assumed to be delivered by Band 7 clinical psychologists because this is broadly consistent with the type of therapists who delivered the interventions in the majority of RCTs included in the NMA. The unit cost of a Band 7 clinical psychologist per hour of client contact has been estimated based on the median full-time equivalent basic salary for Agenda for Change Band 7 and includes salary, salary on-costs and overheads, but excludes qualification costs because the latter are not available for clinical psychologists (Curtis, 2010). However, exclusion of qualification costs from the clinical psychologist unit cost would underestimate the total psychological intervention costs and would therefore, in all likelihood, overestimate their cost effectiveness relative to pharmacological interventions. In order to consider the qualification cost for clinical psychologists, a number of mental health professionals with different qualifications and salary bands were selected (for example, consultant psychiatrists and mental health nurses) and the reported unit costs for these professions with and without qualification costs were compared. The rate of unit costs without/with qualification costs was found to be 0.85, and this allowed estimation of a unit cost for Band 7 clinical psychologists at £101 per hour of client contact in 2012 prices, which included qualification costs. This cost was used in the base-case analysis. A one-way sensitivity analysis tested delivery of a self-help intervention by a Band 5 therapist (such as a mental health nurse) and delivery of group therapies by one Band 7 and one Band 6 (for example, trainee in clinical psychology) therapist.
In addition to therapists' time, the intervention costs of all psychological interventions included an initial GP visit for referral to psychological services. Moreover, the intervention costs of self-help programmes included the cost of either a book or a computerised program and related infrastructure/equipment required for the delivery of such a program (licence fee or website hosting, personal computers [PCs] and capital overheads).
The cost of a book for self-help was based on the cost of Rapee's Overcoming Shyness and Social Phobia: A Step by Step Guide (Rapee, 1998) available in the market (£22.95). The website hosting cost of computerised self-help was estimated based on information provided by the GDG, relating to a pilot research internet-based self-help program for people with social anxiety disorder currently tested in England. According to this information, the annual cost of secure internet hosting reached £14,000 (including maintenance and software bug fixing of the program), and was paid at an individual service level. Based on IAPT audit of activity data (information provided by the GDG), an average IAPT service sees about 2,500 people every year, of which 1.5% are estimated to have social anxiety disorder. Assuming 80% of these are offered (and accept) internet-based self-help, this means 30 people with social anxiety disorder use the internet-based self-help program, resulting in a website hosting cost of £467 per person. Since the particular internet-based self-help program was developed for research purposes, no licence fee was considered at the estimation of the intervention cost, although this cost component, which may be considerable, needs to be taken into account in the assessment of cost effectiveness of other computerised self-help packages for social anxiety disorder that may be available in the future. The annual costs of hardware and capital overheads (space around the PC) were based on reported estimates made for the economic analysis undertaken to inform the NICE TA on CCBT for depression and anxiety (Kaltenthaler et al., 2006) and equal £161 and £1,070, respectively (in 2012 prices). Kaltenthaler and colleagues (2006) estimated that one PC can serve around 100 people with a mental disorder treated with computerised programs per year. Assuming that a PC is used to full capacity (that is, it serves no fewer than 100 people annually, considering that it is available for use not only by people with social anxiety disorder, but also by people with other mental disorders, such as depression), the annual cost of hardware and capital overheads was divided by 100 users, leading to a hardware and capital overheads cost per user of £13. It must be noted that if users of such programs can access them from home or a public library, then the cost of hardware and capital overheads to the NHS is zero.
No booster (maintenance) sessions were assumed for psychological interventions. The intervention cost of waitlist was zero. presents the resource use elements and the estimated intervention costs of all psychological interventions considered in the model.
Resource use and estimated intervention costs of psychological interventions (2012 prices).
Costs of treating side effects of drugs were not considered in the economic analysis due to lack of consistency in reporting appropriate side effect data across all drugs. Nevertheless, the GDG estimated that the majority of common side effects, such as nausea, insomnia, sexual problems, dizziness, fatigue, palpitations and tachycardia, would be discussed during GP monitoring, which was considered at the estimation of intervention costs relating to initial and maintenance pharmacological intervention. Regarding less common side effects, such as hypertension (associated with SNRIs) and gastrointestinal bleeding (associated with SSRIs), these were thought to result in higher management costs at an individual level, but given their low frequency they were deemed to entail smaller economic implications at a study population level. Therefore, although the omission of costs associated with management of side effects is acknowledged as a limitation of the analysis, it is not considered to have substantially affected the economic modelling results.
The extra health and social care costs incurred by adults with social anxiety disorder not recovering post-treatment or relapsing following recovery were taken from Patel and colleagues (2002). The authors analysed service use data on 63 people with social anxiety and 8,501 people without psychiatric morbidity derived from the Psychiatric Morbidity Survey conducted in the UK in 1993–1994 (Meltzer et al., 1995). The study combined data on GP consultations, home visits from health and social services, counselling or therapy contacts and inpatient and outpatient secondary care with relevant national unit costs and subsequently estimated an annual total health and social care cost incurred by people with social anxiety disorder and people without psychiatric morbidity. People with social anxiety disorder in the model were estimated to incur the annual total health and social care cost for this population reported in Patel and colleagues, whereas people who recovered and were in the state of ‘no social anxiety’ were assumed to incur the respective cost incurred by people without psychiatric comorbidity reported in the study. People who relapsed following recovery during the first year post-treatment were assumed to incur the ‘social anxiety’ health and social care cost for 6 months and the ‘no social anxiety’ health and social care cost for the remaining 6 months.
Patel and colleagues also reported the mean annual value of social security benefits for people with social anxiety disorder and those without psychiatric comorbidity, and these costs were used in a secondary analysis that adopted a wider perspective in order to capture the broader economic implications of social anxiety disorder.
Health and social care costs as well as social security benefits were assumed to be the same across all arms of the economic model during the period of initial (12-week) treatment and therefore were excluded from further consideration.
All costs were expressed in 2012 prices, uplifted, where necessary, using the Hospital and Community Health Services Pay and Prices Index (Curtis, 2012). Costs and QALYs were discounted at an annual rate of 3.5%, according to NICE guidance (NICE, 2009b).
reports the values of all input parameters utilised in the economic model and provides information on the distributions assigned to specific parameters in probabilistic analysis, as described in the next section.
Input parameters utilised in the economic model of interventions for adults with social anxiety disorder.
Handling uncertainty
Model input parameters were synthesised in a probabilistic analysis. This means that all model input parameters were assigned probability distributions (rather than being expressed as point estimates), to reflect the uncertainty characterising the available clinical and cost data. Subsequently, 10,000 iterations were performed, each drawing random values out of the distributions fitted onto the model input parameters. Results (mean costs and QALYs for each intervention) were averaged across the 10,000 iterations. This exercise provided more accurate estimates than those derived from a deterministic analysis (which utilises the mean value of each input parameter ignoring any uncertainty around the mean), by capturing the non-linearity characterising the economic model structure (Briggs et al., 2006).
The distributions of the probability of recovery following treatment (year 1 of the model), which were obtained from the NMA, were defined directly from values recorded in each of the 10,000 respective iterations performed in WinBUGS and used in the economic analysis, as described earlier. The log-odds of recovery on waitlist was assumed to follow a normal distribution with mean −2.629 and variance 1.235. The LORs of recovery for each treatment relative to waitlist, as estimated by the WinBUGS model (described in Chapter 3), were applied to simulated values of this normal distribution and converted onto the probability scale. This ensured that the full posterior distribution of the relative treatment effects was used to estimate the absolute probabilities of recovery for each treatment.
The distribution of the probability of relapse for drugs was determined by assigning beta distributions to the pooled relapse rates reported for drug arms and placebo arms in the four relapse prevention RCTs included in the guideline systematic review. The risk ratio of relapse of drugs versus psychological interventions was assigned a log-normal distribution. Utility values were assigned beta distributions using the method of moments. The distributions of the annual probabilities of recovery and relapse in years 2–5 of the model were determined by assigning beta distributions to the 12-year respective probabilities that were used to estimate annual probabilities. The estimation of distribution ranges was based on available data in the guideline meta-analysis and the published sources of evidence.
Uncertainty in intervention costs was taken into account by assigning different probabilities in the number of GP visits (pharmacological interventions) or number of sessions (individual psychological interventions) attended by adults with social anxiety disorder. These probabilities were determined by data reported in the respective RCTs included in the NMA, such as completion rates, average number of sessions attended, and so on. Regarding pharmacological interventions, the same completion rate was applied to all drugs due to lack of relevant data specific to each of the drugs considered in the model. Based on data reported in large pharmacological intervention RCTs included in the NMA (N > 100), the completion rate of the 12-week initial treatment with pharmacological interventions was estimated at 75%. It was therefore assumed that 65% of people in each drug arm of the model attended four GP visits (as described in ) and 10% attended either one less or one or two more visits (which might be occasionally required for the management of side effects). The 25% of people discontinuing the 12-week drug treatment were assumed to pay one or two visits to their GP. People discontinuing treatment were assumed to incur only 50% of the 12-week drug acquisition cost; in addition, if they recovered, they were assumed not to continue with the 26-week maintenance treatment. People who recovered and were thus offered 26 weeks of maintenance treatment were assumed to attend three GP visits (as described in ) at a probability of 55%. The remaining 45% were assumed to pay either fewer visits (0 to 2) or one more visit because of side effects. If the number of GP visits during maintenance treatment equalled zero, no 26-week drug acquisition costs were considered in the model.
Regarding individual psychological interventions, based on relevant reported data, the completion rate was estimated at approximately 85% for all interventions except standard CT (Clark and Wells), which reached a 100% completion rate in the respective RCTs. According to the studies, participants were broadly considered as completers if they had missed no more than four sessions in total. Using this information and the average number of sessions in each arm of a trial or in the subgroup of completers, where reported, the following assumptions were made for all individual psychological interventions with the exception of standard CT (Clark and Wells): 70% of people in each individual psychological intervention arm of the model attended the optimal number of sessions (as described in ); another 15% of people completed treatment but attended one to four fewer sessions; the remaining 15% of people in each cohort discontinued treatment and attended randomly a lower number of sessions (missed five or more sessions and at minimum attended only one session of the intervention).
The cost of group psychological intervention was deemed to be stable and not subject to uncertainty, irrespective of compliance with treatment; this is because participants in a group are not replaced by another person when they occasionally miss one or more sessions or discontinue treatment. Therefore the same resources (in terms of healthcare professional time) are consumed and the full cost of treatment is incurred whether people attend the full course or fewer group sessions. Drug acquisition costs are also not subject to uncertainty. Consequently intervention costs of group psychological interventions and drug acquisition costs were not assigned probabilistic distributions. Extra health and social care costs for people not recovering or relapsing following recovery, as well as social security benefit costs, were assigned a gamma distribution, determined by data reported in the source study.
provides details on the types of distributions assigned to each input parameter and the methods employed to define their range.
Extra probabilistic sensitivity analyses were also undertaken to explore the impact on the results of the following alternative scenarios:
Adoption of a wider perspective which, in addition to NHS and PSS costs, considered receipt of social security benefits by people with social anxiety disorder, as reported in
Patel and colleagues (2002).
A change in the healthcare professional unit cost for self-help and group-based interventions: this scenario assumed delivery of self-help interventions by a Band 5 therapist (for example, a mental health nurse) and delivery of group interventions by one Band 7 and one Band 6 therapist (the latter reflecting the salary of a trainee in clinical psychology). The unit cost of a Band 5 mental health nurse was taken from
Unit Costs of Health and Social Care 2012 (
Curtis, 2012). The unit cost of a Band 6 trainee therapist was not available and was therefore assumed to be in the middle between the unit cost of a Band 5 mental health nurse and a Band 7 clinical psychologist.
Presentation of the results
Results of the economic analysis are presented as follows:
For each intervention, mean total costs and QALYs are presented, averaged across 10,000 iterations of the model. An incremental analysis is provided, where all options have been ranked from the most to the least effective (in terms of QALYs gained). Options that are dominated by absolute dominance (that is, they are less effective and more costly than one or more other options) or by extended dominance (that is, they are less effective and more costly than a linear combination of two alternative options) are excluded from further analysis. Subsequently, ICERs are calculated for all pairs of consecutive options remaining in the analysis.
ICERs are calculated by the following formula:
where ΔC is the difference in total costs between two interventions and ΔE the difference in their effectiveness (QALYs). ICERs express the extra cost per extra unit of benefit (QALY in this analysis) associated with one treatment option relative to its comparator. The treatment option with the highest ICER below the NICE lower cost-effectiveness threshold of £20,000 per QALY (NICE, 2008) is the most cost-effective option.
In addition to ICERs, the mean net monetary benefit (NMB) of each intervention is presented. This is defined by the following formula:
where E is the effectiveness (number of QALYs) and C the costs associated with the treatment option, and λ is the level of the willingness-to-pay per unit of effectiveness, set at the NICE lower cost-effectiveness threshold of £20,000 per QALY (NICE, 2008). The intervention with the highest NMB is the most cost-effective option (Fenwick et al., 2001). Moreover, for the most cost-effective intervention, the probability that it is the most cost-effective option is also provided, calculated as the proportion of iterations (out of the 10,000 iterations run) in which the intervention had the highest NMB among all interventions considered in the analysis.
Economic modelling results
The results of the economic analysis for the time horizon of 5 years post-treatment are provided in . This table provides mean QALYs and mean total costs for each intervention assessed in the economic analysis, as well as the results of incremental analysis, the NMB of each intervention, and its ranking by cost effectiveness (with higher NMBs indicating higher cost effectiveness). Interventions have been ordered from the most to the least effective in terms of number of QALYs gained.
Results of economic modelling, 5 years post-treatment - base-case analysis: NHS and PSS perspective.
At 5 years post-treatment, standard CT (Clark and Wells) is the most effective intervention because it produces the highest number of QALYs. This result was not unexpected, given that the intervention had the highest probability of recovery among all interventions in the NMA. At the same time, it is the second most costly intervention, following psychodynamic psychotherapy. According to the NMBs provided in , standard CT (Clark and Wells) produces the highest NMB and therefore appears to be the most cost-effective intervention. Its ICER versus phenelzine (which is the next most effective intervention not dominated by absolute or extended dominance in incremental analysis) equals £8,426 per QALY, which is below the NICE lower cost-effectiveness threshold of £20,000 per QALY. The probability of standard CT (Clark and Wells) being the most cost-effective intervention is 69%, which reflects the proportion of the 10,000 iterations of the economic model in which it had the highest NMB among all interventions. According to the analysis, the second most cost-effective option at 5 years post-treatment is individual CBT. Phenelzine ranks third in terms of cost effectiveness, while book-based self-help without support ranks fourth. Individual CBT (Heimberg) ranks fifth and book-based self-help with support ranks sixth. Of the other individual psychological interventions, CT with shortened sessions ranks ninth, psychodynamic psychotherapy ranks 25th, and IPT ranks 26th, just above waitlist; supportive therapy is the least cost-effective intervention, ranking in 28th place. Group psychological interventions rank in places between 10 and 15, with the exception of mindfulness training, which ranks 23rd. Drugs (with the exception of phenelzine) rank between places 8 and 22, with paroxetine being the most cost-effective drug after phenelzine, followed by venlafaxine, fluvoxamine, sertraline, fluoxetine and escitalopram. Internet-based self-help ranks seventh (with support) and 20th (without support).
provides the cost effectiveness plane of the analysis, 5 years post-treatment. Each intervention is placed on the plane according to its incremental costs and QALYs compared with waitlist (which is placed at the origin).
Cost-effectiveness plane of all interventions for adults with social anxiety disorder assessed in the economic analysis plotted against waitlist - incremental costs and QALYs per 1000 adults with social anxiety disorder, 5 years after treatment.
Detailed results of the base-case economic analysis, with 95% CIs of costs and QALYs and disaggregation of costs are provided in Appendix 23.
Regarding 1 year post-treatment, phenelzine was the most cost-effective intervention among those considered in the analysis because it produced the highest NMB. Its ICER to paroxetine, which was the next most effective non-dominated intervention in incremental analysis, was £4,063 per QALY, while the ICER of standard CT (Clark and Wells) versus phenelzine exceeded £47,000 per QALY, which is well above the NICE cost-effectiveness threshold of £20,000 per QALY. The probability of phenelzine being the most cost-effective intervention at 1 year post-treatment was 55%. The second most cost-effective option at 1 year post-treatment was paroxetine, followed by book-based self-help (without support) and sertraline. Overall, results indicated that in the short term, drugs seemed to be more cost effective than psychological interventions for adults with social anxiety disorder: drugs ranked in the first 13 places, with the exception of places 3 (book-based self-help without support) and 11 (book-based self-help with support). The various forms of individual CBT, including CT (Clark and Wells), seemed to follow drugs and book-based self-help in terms of cost effectiveness. Group psychological interventions, internet-based self-help and other individual psychological interventions were less cost effective compared with drugs, book-based self-help, and individual forms of CBT. Results for 1 year post-treatment, including mean QALYs and costs with 95% CIs, disaggregation of costs, incremental analysis, NMBs, ranking of interventions by cost effectiveness and the cost-effectiveness plan are presented in Appendix 23.
Results were robust under all alternative scenarios examined in sensitivity analyses. Standard CT (Clark and Wells) was the most cost-effective intervention at 5 years post-treatment when a wider perspective that included social security benefits was adopted, alternative unit costs for self-help and group psychological interventions were assumed, and alternative utility values were used. Ranking of interventions in terms of cost effectiveness was broadly the same after using a wider perspective, and alternative unit costs and utility values. Results of secondary and sensitivity analyses can be found in Appendix 23. The economic evidence profile of the guideline economic analysis is provided in Appendix 24.
Discussion – limitations of the analysis
The guideline economic analysis assessed the cost effectiveness of a broad range of pharmacological and psychological interventions for adults with social anxiety disorder over 5 years post-treatment. In addition, 1-year post-treatment results were obtained and compared with the 5-year post-treatment results. This is because the GDG was interested in the potential changes in the relative cost effectiveness of interventions over time. The results of the analysis suggest that, although in the short-term pharmacological interventions appear to be, overall, more cost effective than psychological interventions, at 5 years post-treatment the relative cost effectiveness of individual forms of CBT improves significantly, so that standard CT (Clark and Wells), individual CBT, individual CBT (Heimberg) and CT (Clark and Wells) with shortened sessions rank first, second, fifth and ninth, respectively, in terms of cost effectiveness. The probability of standard CT (Clark and Wells) being the most cost-effective intervention at 5 years is 69%. Phenelzine is the third most cost-effective intervention. Book-based self-help also appears to be cost effective compared with other treatment options, with the two forms of it (with and without support) being among the six most cost-effective interventions of those assessed. Group-based psychological interventions do not appear to be particularly cost effective relative to other available treatments, ranking in places between 10 and 15, with the exception of mindfulness training, which ranks 23rd. Drugs (with the exception of phenelzine) are also not particularly cost effective, ranking between places 8 and 22; following phenelzine, the next most cost-effective drugs are (in order): paroxetine, venlafaxine, fluvoxamine, sertraline, fluoxetine and escitalopram. Internet-based self-help ranks seventh (with support) and 20th (without support). Other individual psychological interventions, such as psychodynamic psychotherapy, IPT and supportive therapy rank 25th, 26th and 28th, respectively.
The emergence of individual psychological interventions in the form of CBT as cost-effective options at 5 years can be attributed to two factors: first, over the 5-year time horizon there is a longer time period to accrue the benefits resulting from the differential relapse rate between psychological and pharmacological interventions, which was applied in the first year of the economic model. Based on the model input parameters, the proportion of people who relapse following post-treatment recovery is substantially lower if they receive a psychological, rather a pharmacological, intervention, and at 5 years post-treatment the benefit of being free from social anxiety has been enjoyed over a longer time period. Second, over a 5-year time horizon the high intervention costs of individual psychological interventions (which are responsible for the relatively low performance of these interventions in terms of cost effectiveness at 1 year post-treatment) are spread over a longer time period and are offset to a greater extent by NHS and PSS cost savings because there are fewer people relapsing and incurring such extra costs.
Results of the economic analysis were overall robust to different scenarios explored through sensitivity analysis. Results were practically unaffected when a wider perspective incorporating social security benefits was adopted, and when self-help and group psychological interventions were assumed to be delivered by less trained therapists. Moreover, using alternative utility data that assumed more conservative utility gains following recovery did not change the overall conclusions.
The clinical effectiveness data utilised in the model were derived from the NMA undertaken for this guideline. This methodology enabled evidence synthesis from both direct and indirect comparisons between interventions, and allowed simultaneous inference on all treatments examined in pairwise trial comparisons while respecting randomisation (Caldwell et al., 2005; Lu & Ades, 2004). The NMA utilised continuous data to estimate the relative treatment effects of interventions, and then transformed the estimated SMDs into probabilities of recovery, using waitlist as baseline, as discussed in Chapter 3. This was necessary in order to populate the economic model, as no comprehensive recovery data were available for the range of interventions assessed in the economic analysis. Moreover, the economic analysis needed to reflect the same relative treatment effects that were estimated in the NMA, which determined the comparative clinical efficacy of the interventions considered in this guideline. Transformation of SMDs into probabilities of recovery is valid as long as the relative treatment effect estimated using continuous data is equal to the treatment effect estimated using recovery data. Such an assumption cannot be checked for all interventions included in the economic analysis (since no recovery data are available for a large number of interventions); however, a comparison between continuous and recovery data indicated a strong relationship between them and therefore this transformation is unlikely to have introduced strong bias in the analysis (more details are provided in Chapter 3).
The assumptions and any limitations of the NMA model, as well as the limitations of individual studies considered in the NMA, have unavoidably impacted on the quality of the economic model clinical input parameters. For example, many of the included studies were not registered and both the clinical and economic results may be vulnerable to reporting and publication bias. The assumptions underlying the NMA model have been described in detail in Chapter 3; the characteristics and any limitations of the individual studies and the NMA model have been described in Section 6.5.2.
Treatment discontinuation because of side effects or other reasons was not considered in the model structure because no relevant data were systematically reported in the trials considered in the guideline systematic literature review. However, the probabilistic model did assume that a percentage of people might have not completed treatment or there might have been less than perfect compliance. In addition, most clinical efficacy data were analysed on an intention-to-treat basis and implicitly accounted for discontinuation.
One limitation of the model is the relapse data used to populate the model. Relapse data for pharmacological interventions are very sparse in the literature. Ideally, the economic model required drug-specific data on the probability of relapse after 6 months of maintenance treatment for adults with social anxiety disorder who have recovered following initial 12-week drug treatment. However, no such data were identified in the literature. Because of the lack of relevant relapse data specific to each drug considered in the analysis, the probability of relapse for all pharmacological interventions was assumed to be the same, and was estimated as the midpoint of the pooled relapse rates reported for drug and placebo arms in the relapse prevention RCTs included in the guideline systematic review. These rates referred to relapse during pharmacological maintenance treatment and relapse after discontinuation of initial (12-week) pharmacological treatment without maintenance, respectively. Moreover, relapse prevention studies measured relapse following response to treatment rather than recovery, which was the modelled outcome in the economic analysis. It is possible that the probability of relapse following recovery is lower than that following response to treatment and therefore the economic analysis may have potentially overestimated relapse following treatment. Furthermore, in reality, different drugs are likely to be associated with different risks for relapse, and this possibility has not been reflected in the economic model due to lack of drug-specific relapse data in the literature.
The RR of relapse of pharmacological versus psychological interventions was adopted from a small observational study (N = 28) that evaluated the effects of maintenance treatment with phenelzine and group CBT (Liebowitz et al., 1999), due to lack of other relevant data. Subsequently, as with pharmacological interventions, all psychological interventions were assumed to have the same risk of relapse due to lack of intervention-specific data, but, as in the case of drugs, this assumption may not hold. Nevertheless, the mean probabilities of relapse for pharmacological and psychological interventions estimated for the economic model (42% versus 14%, respectively) are very close to respective relapse rates reported for people with OCD (45% versus 12%, respectively) (Simpson & Fallon, 2000) and broadly consistent with respective figures reported for panic disorder (40% versus 5%, respectively) (Clark et al., 1994).
The RR of relapse of pharmacological versus psychological interventions was applied to the first year of the model only. For years 2 to 5 the model conservatively assumed that the same probability of relapse applied to all interventions, both psychological and pharmacological. This assumption may have favoured drugs, if the beneficial effect of psychological interventions relative to drugs in terms of reduced relapse rates, as indicated by Liebowitz and colleagues (1999), persists beyond 1 year.
Utility data used in the economic model were taken from a study that analysed survey data on people who had experienced social anxiety disorder (or other mental disorders) and people without a mental disorder over the 12 months prior to the survey interview. A limitation of these data is that the diagnosis of social anxiety disorder referred to a period of 12 months prior to the survey, so some participants might have experienced an improvement in their condition over this period, and might have actually recovered at the point of interview. Therefore, it is not certain that the HRQoL of this mixed group of people accurately reflects the HRQoL of the study population in the model (that is, people with a current diagnosis of social anxiety disorder). Moreover, the HRQoL of people without a mental disorder over the last 12 months may be higher than the HRQoL of people recovering from social anxiety disorder. However, after reviewing relevant literature, the GDG decided that these utility data were most appropriate to use in the economic model, because compared with other available utility data, they were judged to reflect more closely the HRQoL of adults with social anxiety disorder and those recovering following treatment, and also met the NICE criteria for the selection of utility data for cost-utility analysis.
Owing to the lack of comprehensive overall and specific side effect rates across all interventions, (dis)utility data due to side effects associated with drug treatment, and costs of treating these side effects, the model did not consider these parameters. Nevertheless, probabilistic analysis did take into account that a small proportion of people receiving pharmacological interventions may attend a higher number of GP visits for the management of side effects. In any case, omission of side effects from the model structure may have potentially led to overestimation of the cost effectiveness of drugs relative to psychological interventions, and may have had an impact on the relative cost effectiveness between different drugs.
Extra NHS and PSS costs incurred by people with social anxiety disorder not recovering or relapsing following recovery were taken from a study that utilised service use data from a national survey (Patel et al., 2002). The survey was conducted in 1993–1994 and is therefore outdated. However, it was not possible to identify recent data specific to UK service use of people with social anxiety disorder in the literature. The recent Adult Psychiatric Morbidity Survey (McManus et al., 2009b) did not report data specific to people with social anxiety disorder. More recent service use data for people with social anxiety disorder have been reported in a US study (Wang et al., 2005) and a study conducted in the Netherlands (Acarturk et al., 2008) but these refer to different healthcare settings and do not necessarily reflect UK relevant resource use. Therefore, the study by Patel and colleagues (2002) was the best source for obtaining this cost parameter for the economic model.
A secondary analysis that adopted a wider perspective incorporating social security benefits was undertaken. The relative cost effectiveness of interventions was practically unaffected by inclusion of such benefits. However, it must be noted that, because of a lack of more specific data, the model assumed that people recovering from social anxiety disorder received reduced benefits (equalling benefits received by people without a mental disorder), and then returned to receipt of higher social benefits (equalling benefits received by people with social anxiety disorder) if they relapsed. However, receipt of social benefits is a long-term process that is not necessarily directly related to events characterising the clinical course of social anxiety disorder, such as recovery or relapse, within a short period of time, such as the 5 years of the model time horizon. Thus this secondary analysis may have overestimated the reduction in social benefits received by people recovering following treatment.
