Abstract
Keywords
Generalized anxiety disorder (GAD) contributes significantly to the burden of disease, ranking twelfth in women and twenty-fourth in men as a cause of morbidity in Australia [1]. Panic disorder (PD) is less common than GAD, but the most common diagnosis among people seeking treatment for an anxiety problem [2]. Using the ICD-10 definitions [3], the National Survey of Mental Health and Wellbeing (NSMHW) [4] gives a 12-month prevalence estimate of 3.0% for GAD and 1.4% for panic disorder (defined as panic disorder with or without agoraphobia).
Treatment guidelines [5] suggest the primary treatments for GAD should be non-pharmacological. Where pharmacological intervention is required, they recommend the antidepressant venlafaxine (a serotonin and noradrenaline reuptake inhibitor [SNRI]) or buspirone. However, buspirone is rarely prescribed in Australia. Benzodiazepines are recommended for acute exacerbations [5], but are not considered here as there is inadequate evidence to support long-term use. However, note that the national BEACH program [6] tells us that 44% of GP encounters for ‘anxiety’ result in a prescription for benzodiazepines (additional analysis of BEACH data: April 2000 to March 2001).
While the guidelines specifically recommend venlafaxine, the Cochrane review ‘Antidepressants for GAD’ found that venlafaxine, paroxetine (a selective serotonin reuptake inhibitor [SSRI]) and imipramine (a tricyclic antidepressant [TCA]) have comparable efficacy compared to placebo [7]. The Cochrane review measures efficacy using dichotomous outcome measures. The current analysis relies on continuous outcome measures reporting means and standard deviations to calculate efficacy. Therefore, many RCTs referred to in the Cochrane review are excluded [8–14]. Neither paroxetine nor venlafaxine are listed on the Pharmaceutical Benefits Scheme (PBS) for GAD [15]. Imipramine is listed without restrictions.
For PD, clinical practice guidelines recommend cognitive behavioural therapy (CBT) as the most consistently efficacious treatment and the treatment that should be attempted initially [5]. Some people with panic disorder may require additional pharmacological treatment, or may not respond to CBT. In this case, the Therapeutic Guidelines [5] recommend SSRIs, and then benzodiazepines, TCAs and irreversible non-selective monoamine oxidase inhibitors (MAOIs). By contrast, guidelines by the Royal Australian College of Physicians recommend TCAs before benzodiazapines and SSRIs [16], noting that medication with benzodiazapines may result in dependence and long-term adverse effects, and that evidence for treating panic with MAOIs is currently insufficient. Paroxetine is the only SSRI listed on the PBS for PD, and imipramine is the only TCA that is listed without restrictions for PD [15]. The interventions analyzed for PD in this report are CBT, SSRIs (paroxetine) and TCAs (imipramine).
The aim of this study is to assess the incremental costeffectiveness of CBT and SNRIs for the treatment of GAD and CBT, SSRIs and TCAs for PD, for those aged 18 years and older. The analysis of interventions for GAD and PD contribute to the state and Commonwealth government-funded Assessing Cost-Effectiveness – Mental Health (ACE-MH) project, in which economic evaluations are performed for a range of interventions for mental disorders, using common methods [17].
Method
The ICER is calculated as the cost (A$) per disability-adjusted life year (DALY) saved. The population eligible for the interventions are adults (aged 18 years and older) with ICD-10 defined GAD/PD in the Australian population in the year 2000 who sought health care for their mental health disorder but would not have received evidence-based medicine (EBM) under current practice. All interventions involve the provision of treatments for which there is evidence of efficacy.
These analyses are performed from the perspective of the health care sector; therefore, both government and patient expenses for services and pharmaceuticals are included. As the tracking of costs and benefits is for one year, discounting is not applied.
The interventions
The number of doctor consultations is assumed to be nine per year for all drug interventions: weekly visits for the first month, monthly visits for the next two months and three-monthly visits thereafter. We derive the proportion of consumers consulting with a GP or referred to a psychiatrist for treatment from the NSMHW, assuming this reflects current practice (approximately 10% referred to a psychiatrist, 90% managed by a GP [4]). For those consulting a psychiatrist, a GP visit for referral is also included.
Current practice
Current practice for the treatment of GAD and PD is determined from the NSMHW by identifying service utilization patterns for those with the disorder who consulted health services for a mental health problem. All prevalent cases of GAD and PD in the year 2000 are included. Consulting is defined as seeking care for a mental health problem during the past 12 months from a general practitioner, psychiatrist, psychologist, physician, surgeon, social worker, mental health team worker or an admission to hospital [4]. It is assumed that those consulting are receiving EBM if they have had three or more consultations with a GP, psychiatrist or psychologist plus CBT and/or drug treatment.
Of those with GAD, 55% consulted; these were further split into treatment with EBM (27%) or non-EBM (28%). Those who consulted and received non-EBM under current practice each averaged 2.5 GP consultations, 0.2 psychiatrist consultations and 0.6 psychologist consultations within the past 12 months. For PD, 65% consulted, with 47% receiving EBM and 19% receiving non-EBM. Those who consulted and received non-EBM under current practice each averaged 2.8 GP consultations, 0.3 psychiatrist consultations and 1.8 psychologist consultations within the past 12 months.
Assessment of benefit
Benefits are assessed by a two-stage process. In the first stage, DALYs are employed to estimate the health benefit from the interventions. The second stage involves the assessment of issues that either influence the degree of confidence that can be placed in the ICERs (such as the level of available evidence), or broader issues that need to be taken into account in decision-making about resource allocation (such as equity and acceptability to stakeholders).
Stage one: measurement of the health gain
The health benefit is measured in DALYs. There is no evidence in the literature that SNRIs, SSRIs, TCAs or CBT can cause or prevent death, so only a change in the years lived with disability (YLD) component of the DALY is modelled.
YLD for the current practice comparator
The one-year prevalence estimates from the NSMHW [4], [20] are 437 417 prevalent cases of GAD and 193 641 prevalent cases of PD in Australia (Table 1).
Critical parameter values, uncertainty distributions and sources of information for determining health benefits and costs
Because GAD and panic disorder are chronic conditions with periods of remission and relapse evident for up to 20 years [21–23], the interventions will not be producing health gain at all times. Therefore, the health gain is multiplied by the proportion of time symptomatic, which is estimated from the NSMHW as the current prevalence (symptomatic in the past 2 weeks for GAD, and past 4 weeks for PD) divided by the prevalence in the past 12 months. The percentage of time symptomatic over 12 months is estimated to be 63% for GAD and 48% for PD.
The disability weights (DWs) used to calculate YLDs are based on the Dutch weights. The DWs for GAD are 0.17 for mild and moderate cases, and 0.60 for severe cases. Those for PD are 0.16 for mild and moderate, and 0.69 for severe [24]. Composite DWs were calculated separately for those who did not consult, those who consulted and received EBM, and those who consulted and received non-EBM. We determine the spread of severity of the disorders using the NSMHW, assuming this reflects the spread of severity in the Australian population. Severity is classified using the Mental Component Score (MCS) of the SF-12, which has a mean population value of 50 and a standard deviation (SD) of 10. We classify cases into: severe disability (<2.5 SD below the mean, i.e. MCS < 25); moderate disability (<1.5–2.5 SD below the mean, i.e. MCS = 25–34.9); and mild disability (< 0.5–1.5 SD below the mean, i.e. MCS = 35–44.9). The proportion of cases in each severity category is multiplied by the appropriate DW for the category to get an average DW for those in each group. The resulting baseline DWs for GAD are: consulted = 0.22; consulted and received non-EBM = 0.20; and did not consult = 0.19. For panic disorder, these are: consulted = 0.28; consulted and received non-EBM = 0.21; and did not consult = 0.12. Note that the higher baseline disability weight in the population consulting for anxiety disorders is merely a result of the fact that more severely affected individuals are more likely to seek care.
Determining the reduction in YLD with treatment
The reduction in disability severity is modelled using the effect size and both the ‘conversion factor method’ and the ‘survey severity method’ to translate the effect size into a reduction in the DW. For the ‘conversion factor method’ we multiply the effect size by the DW conversion factor for panic disorder/agoraphobia. This conversion factor is an average change in the DALY disability weights for the equivalent of a standard deviation change in severity for the particular mental disorder [25]. For the ‘survey severity method’ the effect size is applied directly to the Mental Component Score, which was used to determine the average DW at baseline. The severity of respondents is then reclassified and a new average DW calculated. The difference in average DW is the change attributed to treatment [17].
The effect size for PD interventions is derived from a meta-analysis that reports effect sizes for CBT, SSRIs and TCAs [16]. The effect size for TCAs (0.61) differed markedly from a previous meta-analysis (0.47) [26], so we decided to pool results from RCTs in both metaanalyses to obtain an overall effect size (Table 1). For cases not adherent with treatment, no reduction in DW has been modelled (although they do incur costs of the treatment provided).
Meta-analysis for GAD
For the meta-analysis of RCTs for GAD, the inclusion criteria were subjects aged 18 years and older, a DSM-III-R or DSM-IV diagnosis of GAD [27], and reporting of continuous measures (with both means and standard deviations). Five small RCTs (total intervention n = 84 participants) met the inclusion criteria for CBT [28–32] and two larger RCTs (total intervention n = 435) met the inclusion criteria for venlafaxine [18], [19].
We first calculate an effect size for each study by averaging across relevant continuous outcome measures related to anxiety and depression and health-related quality of life. Within each study, the effect size (standardized mean difference) is calculated using Hedges’ g and pooled across studies using the random effects method of DerSimonian and Laird [33]. The intervention effect sizes (for GAD and PD) are presented in Table 1.
Adherence
It was assumed that the completion rate of the treatment group in the RCTs for GAD [18], [19],[28–32] and panic [16], [26] reflects the best possible adherence with treatment. No longitudinal studies measuring adherence to the interventions were available for GAD or PD, so a minimum rate of 50% was used in the uncertainty analysis (Table 1). This was done to better reflect what could be expected under routine health service conditions where results may vary due to the effect of comorbidity, the motivation of clinicians and patients, the availability of skilled clinicians and the capacity to vary the intervention to suit the needs of the patient.
Stage two: the second stage filter criteria
The filters chosen for assessment in the ACE study were: ‘strength of evidence’, ‘equity’ ‘feasibility’ and ‘acceptability to stakeholders’. The filters are described in Table 2.
The second stage filter criteria
Assessment of costs
Unit costs and data sources are shown in Table 3. Costs that would have been incurred under current practice are subtracted from the intervention (and nonadherence) costs to obtain the incremental cost.
Summary of unit cost information, data sources and assumptions
Health-care seeking behaviour for those not adherent with treatment is unknown; as a consequence, the cost of nonadherence must be estimated. It is assumed that the cost of nonadherence is (on average) the same as the cost of non-EBM. The cost of nonadherence is important to the ICER because those who do not adhere to treatment would be expected to incur some cost but receive no health benefit.
Uncertainty analysis
Simulation modelling techniques are used to allow the presentation of an uncertainty range around the health benefits, costs and ICERS (Table 1) [34]. @RISK software [35] was used to conduct Monte Carlo simulations, which allow multiple recalculations of a spreadsheet, each time choosing a value from the specified distribution for each input variable (shown in Table 1). We use 2000 iterations for each of the two methods for translating effect size into a change in the DW (i.e. the conversion factor method and the survey severity method). Thus, the final results (Tables 4,5) are based on the 2000 ++ 2000 iterations. Median values were calculated because results are not normally distributed. The ranges presented can be interpreted as the range within which the true result lies with 95% certainty.
The incremental benefits, costs and incremental cost-effectiveness ratio of cognitive behaviour therapy and the serotonin and noradrenaline reuptake inhibitors for the management of generalized anxiety disorder, compared with current practice
The incremental benefits, costs and cost-effectiveness of cognitive behaviour therapy and antidepressants for the management of panic disorder, compared with current practice
Results
Cognitive behavioural therapy provided by a public psychologist is the most cost-effective intervention for treating both GAD and PD. Treatment of GAD with venlafaxine provides a similar ICER as CBT by other providers (Table 4). Tricyclic antidepressants are the second most cost-effective option for panic disorder, followed by CBT by other providers and SSRIs (Table 5). Within a 95% uncertainty range, all of the intervention options have ICERs not greater than A$40 000 for GAD and A$55 000 for PD.
For both GAD and PD, the major contributors to uncertainty around the ICERs for all interventions (CBT, SNRI, SSRI and TCAs) are: the effect size; the reduction in disability weight; and (for CBT only) the variation factor around the cost of consulting private psychologists, private psychiatrists and general practitioners.
The results of the second-filter criteria assessment are presented in Table 6. A main outcome was the elucidation of issues surrounding the availability and distribution of an adequate workforce for CBT interventions. The drug interventions are likely to be more feasible, although possibly less acceptable, due to concerns about side-effects.
Results from assessment of second filters
Discussion
Economic analysis raises important issues as to what constitutes ‘value-for-money’. It is not uncommon for a threshold ICER (or ‘shadow price’) to be set as a guide to assist decision-making. In ACE–MH, for example, an ICER of A$50 000 per DALY has been used. However, this should not be over-interpreted or taken out of context. It is important to reflect, for example, on how well the ICER captures the various dimensions of ‘benefit’ in mental health. The second stage filters are designed to allow the ICERs to be placed within a broader decision context.
Cognitive behavioural therapy is a cost-effective intervention for both GAD and PD, particularly when provided by publicly funded psychologists. It is likely (≥ 73% chance) that the ICERs for all interventions will be below our threshold of A$50 000 per DALY saved. The SNRI for GAD has a similar ICER as CBT (by providers other than public psychologists). However, it has lower efficacy than CBT, resulting in lower total health benefit.
For both disorders, the assumption was made that drugs would be obtained on the PBS. However, GAD is not an indication for obtaining venlafaxine on the PBS (although depression, highly comorbid with GAD, is). Additionally, it is possible that some PD patients will purchase other antidepressants not listed on the PBS (such as the SSRIs sertraline or fluvoxamine, or the TCA clomipramine). Had the model costed drug interventions assuming no PBS benefit, overall cost-effectiveness would remain unchanged, although the government would not contribute to payment for drugs.
A lack of economic evaluations analyzing interventions for anxiety disorders limits comparison of the results of this analysis. The only economic evaluation comparing interventions for panic disorder was a costbenefit analysis by Otto et al. [36] in the US, with treatment benefits and costs based on a trial undertaken by the authors. Costs were calculated for the services and/or drugs required to maintain for one year a onepoint increase in the Clinician Global Impression of Severity scale. Pharmacotherapy was considerably more expensive (US$1153) than individual CBT (US$646) and particularly group CBT (US$248). The authors concluded that, based on available evidence of cost, acceptability and tolerability, and treatment outcome, CBT should be the initial mode of treatment. However, as this was a non-randomised, non-controlled trial with no uncertainty analysis, the results are not directly comparable with the current study.
A recent cost-effectiveness analysis undertaken by Issakidis et al. [37] examines disability averted by providing a mix of ‘optimal therapies’ to those receiving ‘current therapies’ for a mix of anxiety disorders. Optimal therapies were actually treatment ‘packages’ which emphasized CBT as optimum treatment for most patients, but also included pharmacological treatment for around one-third of patients. The cost-effectiveness of stand-alone CBT or pharmacotherapy was not analyzed. Also in contrast to the current study, the comparator was ‘no treatment’, so the placebo effect was added to the effect size of these interventions, making the health gains much greater. Therefore, the study results are not comparable with this analysis of specific interventions for GAD/PD.
The strengths and limitations of the methods common across the ACE–MH project are presented in detail in Haby et al. [17]. As discussed in that introductory paper, the calculation of health gain is the most difficult methodological issue in the ACE–MH project. One limiting factor specific to this analysis is the absence of longitudinal community surveys of the Australian population, which would provide a greater understanding of the nature and duration of GAD and PD and associated longterm treatment patterns and health service utilization. The available treatment research fails to measure the effect of comorbidity. People with comorbidities are often excluded and/or the papers do not present the results stratified by subgroups according to comorbidity. Therefore, use of these results may not be generalizable to all those with comorbidities. However, the results are likely to be a good reflection of what would happen ‘on average’ and allow meaningful comparisons between treatments.
Particular to GAD, the lack of studies examining the effectiveness of interventions over the long-term makes it impossible to know whether pharmaceutical or psychological interventions are the most appropriate way to manage the chronic nature of this disorder. A recent review of studies exploring long-term pharmacological treatment of GAD concluded that evidence is extremely limited; the few studies identified were methodologically weak, with no placebo control and based on pre-DSM-III-R-defined GAD [38].
By contrast, there is no evidence for long-term gains following drug treatment for PD. In the Gould metaanalysis, which mostly analyzed follow-up trials for imipramine, it was reported that the effect of pharmacological treatments wore off once administration ceased. Moreover, the few available long-term and follow-up studies for SSRIs report an overall worsening of symptoms following the tapering off of medication [39–41].
The GAD and PD literature fails to explore the efficacy of combined pharmaceutical and psychological interventions [42], [43]. Factorial design studies are required to estimate the effect of CBT relative to pharmacological interventions alone and in combination with psychological interventions. In addition, there is a clear need for long-term studies comparing SNRIs against similar antidepressants such as the SSRIs (e.g. paroxetine), the older and cheaper TCAs (e.g. imipramine) and buspirone. However, funding would have to come from public sources, as pharmaceutical companies would not be expected to benefit commercially from performing such research.
While CBT by a public psychologist was the most cost-effective intervention for both anxiety disorders, the second filter analysis revealed issues that may impact on its implementation. Greater use of publicly funded psychologists will require attention to ensuring an adequate workforce, particularly in outer metropolitan and rural regions. It is important to note here that our costeffectiveness analyses assume steady-state operation so implementation costs are not included in the analyses.
Another possibility for the CBT intervention is to fund a mix of providers including other suitably trained health professionals (social workers, nurses, GPs) as currently being piloted in the ‘Better outcomes in mental health care’ initiative [44]. However, use of providers that are not adequately trained may decrease the effectiveness, and therefore cost-effectiveness, of the intervention. Thus, attention to training and accreditation will be required if similar effectiveness is to be achieved.
A change in the cost of accessing different providers will also change the cost-effectiveness of the intervention. We have modelled the same effectiveness for the different providers but different costs, ranging from $47 per session for a psychologist on a public salary to $133 for a private psychiatrist (which includes the Medicare rebate and out-of-pocket cost to the patient). A mix of providers would result in a cost-effectiveness ratio somewhere between that for a public psychologist and that for a private psychiatrist.
Access for outer metropolitan and rural regions is an important consideration for most providers but the use of computer-based CBT may overcome this problem if found to be as effective for anxiety disorders as for depression [45]. In comparison to CBT, the health system provides no barriers to accessing drug treatment for GAD or PD, although the fact that drug therapy is more accessible (by both cost and availability of providers) than psychological therapies, could be unacceptable to consumers. However, an exception is venlafaxine for GAD, which is not currently available on the PBS, which presents an equity issue for access to this intervention.
Caveat
The Assessing Cost-Effectiveness – Mental Health (ACE–MH) project was jointly funded by the Australian Department of Health and Ageing, Mental Health and Suicide Prevention Branch and the Department of Human Services, Mental Health Branch, Victoria in recognition of the importance of research into the costeffectiveness of interventions in mental health treatment and care. This work draws upon, but is also limited by the available research and the assumptions necessary to complete the work.
The results of the analyses provide valuable material, likely to contribute to future policy deliberations by all service providers. Conclusions drawn from the economic evaluations should be considered within the context of the second stage filter process, which qualifies the results taking into account issues of equity, feasibility, strength of evidence, and acceptability to stakeholders. This second stage filter process addresses some of the practical considerations required for changes in actual service practice.
Footnotes
Acknowledgements
Principal investigators for the project are: Theo Vos, Rob Carter and Gavin Andrews. Analyses draw on the National Survey of Mental Health and Wellbeing, for which unit record data was obtained from the Australian Bureau of Statistics (ABS) [4] and a revised scoring algorithm to determine ICD-10 and DSM-IV diagnoses was provided by Gavin Andrews. Information contained in this report on current prescribing behaviour of GPs has been drawn from data collected by the General Practice Statistics and Classification Unit, the University of Sydney in collaboration with the Australian Institute of Health and Welfare [
]. The average cost of various types of medical attendances and the various forms of SSRI were obtained from Medicare Benefits Schedule and Pharmaceutical Benefits Scheme data from the Department of Health and Ageing. We thank Gavin Andrews, Kristy Sanderson, Caroline Hunt and Cath Issakidis for advice on various aspects of the analysis.
We thank members of the ACE–MH steering committee for their input into the project: David Barton, Graham Burrows (Chair), Sue Caleo, Vaughan Carr, Dermot Casey, Joy Easton, William Hart, Helen Herrman, Barbara Hocking, Assen Jablensky, Anthony Jorm, Lyn Littlefield, Patrick McGorry, John McGrath, Paul Morgan, Lorna Payne, Deb Podbury, Kristy Sanderson, Suzy Saw, Bruce Singh, Bruce Tonge, Ruth Vine, Harvey Whiteford.