This is a critical abstract of an economic evaluation that meets the criteria for inclusion on NHS EED. Each abstract contains a brief summary of the methods, the results and conclusions followed by a detailed critical assessment on the reliability of the study and the conclusions drawn.

The study compared the effectiveness and costs of venlafaxine XL (Efexor XL) and diazepam for the treatment of generalised anxiety disorder (GAD) among non-depressed patients. Patients received either 75 mg venlafaxine XL once daily or 5 mg diazepam 3 times daily or placebo for 8 weeks. After 8 weeks, patients who responded to treatment and who had received venlafaxine XL received either 75 mg venlafaxine once daily or placebo for a further 4 months. Those who responded and who had received diazepam for the first 8 weeks were then treated with a placebo for a further 4 months. Those who did not respond to treatment in the initial 8 weeks received no more trial medication.

Treatment.

Cost-effectiveness analysis.

The study population comprised adults (18 years or older) who met the Diagnostic and Statistical Manual (DSM IV) criteria for GAD, had a minimum screening and baseline Hamilton Rating Scale for Anxiety (HAM-A) score of 20, with scores of 2 or higher on item 1 (anxious mood) and item 2 (tension) on the HAM-A, and who did not have a diagnosis of major depressive disorder according to DSM-IV criteria within 6 months of screening.

The setting was primary care. The economic study was conducted in the UK.

The year in which the effectiveness and resource data were collected was not reported. The costs were reported at 2000/2001 prices.

The effectiveness data were derived from a single study.

The costing was undertaken retrospectively on a different sample of patients from that of the single study. General practitioners (GPs) provided estimates of resource use, for which costs were estimated.

Very limited information on the nature of the study sample was provided. Ninety-six patients received 75 mg venlafaxine XL and 70 patients received diazepam. There was no information on the method of sample selection, the numbers of those invited to participate who refused, or the numbers of those excluded from the study. Further details are given in another study (Hackett et al. 2003, see 'Other Publications of Related Interest' below for bibliographic details).

The authors stated that the study was a double-blind randomised controlled trial. There was no information on the method of randomisation, or on methods of blinding. The duration of follow-up was 6 months. No loss to follow-up was reported.

The analysis was conducted on an intention to treat basis. The primary outcome was "successful treatment", as judged by the investigators' clinical global impression (CGI) improvement score. A CGI score of 2 or less indicated that patients had responded to the treatment, while patients were considered to be in remission if the score was 1. There was no information on whether the groups were comparable at baseline. Further details are given in another study (Hackett et al. 2003).

After 8 weeks, for patients receiving venlafaxine XL, the probability of responding to treatment was 0.749. The probability of discontinuing treatment during the 8 weeks was 0.143, while the probability of not responding to treatment was 0.108.

For patients who responded to venlafaxine XL at 8 weeks, there was a probability of 0.405 of remission.

For patients in remission at 8 weeks and who were therefore continuing treatment with venlafaxine XL, there was a probability, at 6 months, of

remaining in remission, 0.45;

relapse at 6 months, 0.1;

discontinuing treatment due to an adverse event, 0.20;

discontinuing treatment due to non-compliance, 0.15; and

discontinuing treatment due to lack of efficacy, 0.10.

For patients responding to venlafaxine XL at 8 weeks and who were therefore continuing treatment with venlafaxine XL, but who were not in remission, there was a probability, at 6 months, of:

remission, 0.313;

continuing to respond but no remission, 0.313;

not responding to treatment, 0.063;

discontinuing due to lack of efficacy, 0.093;

discontinuing due to an adverse event, 0.125; and

discontinuing due to non-compliance, 0.093.

For patients continuing treatment with venlafaxine XL, there was a probability of 0.198 that they would discontinue treatment during a period of 8 weeks to 6 months.

For patients who discontinued treatment of venlafaxine XL during the first 8 weeks, the probability that this was due to a lack of efficacy was 0.20. The probability that this was due to an adverse event was 0.48, while the probability that it was due to non-compliance was 0.32.

After 8 weeks, for patients receiving diazepam, the probability of responding to treatment was 0.743. The probability of discontinuing treatment during the first 4 weeks of treatment was 0.098, while the probability of not responding to treatment after 4 weeks was 0.159.

For patients who responded to treatment after 4 weeks, there was a probability of remission at 6 months of 0.393. Despite no further treatment, the probabilities for those in remission at 6 months to remain in remission or to relapse were 0.42 and 0.580, respectively.

Patients who responded to treatment with diazepam after 4 weeks had a probability of 0.607 that they were not in remission. After being switched to another treatment (probability 1), they had a probability of 0.101 of being in remission at 6 months.

For patients receiving diazepam, but who discontinued treatment within the first 4 weeks, the probability that this was due to a lack of efficacy was 0.125. The probability that this was due to an adverse event was 0.250, while the probability that it was due to non-compliance was 0.625.

Further, 27.6% of venlafaxine XL-treated patients and 16.8% of diazepam-treated patients would be in remission by 6 months, (p=0.07), while 3% (venlafaxine treated) and 16.9% (diazepam treated), respectively, would relapse by 6 months, (p=0.001).

The authors concluded that starting treatment with venlafaxine XL instead of diazepam would significantly increase the expected probability of being in remission at 6 months, and would decrease the expected probability of relapsing at 6 months.

A decision tree model was developed to predict the treatment of patients over 6 months, following their initial treatment with either diazepam 3 times daily or 75 mg venlafaxine XL once daily.

"Successful treatment" was the summary outcome measure used in the economic analysis. This was defined as the percentage of patients who were in remission at 6 months and the percentage of those not relapsing within 6 months. It was derived directly from the study.

Discounting was not carried out since the costs were incurred over during less than 2 years. The resource quantities and the unit costs were reported separately. The cost boundary adopted was that of the UK National Health Service. Included in the analysis were the acquisition costs of venlafaxine XL and diazepam, the cost of GP consultations, the cost of visits (to a psychiatrist, counsellor, a community mental health team member, or psychologist) and the cost of switching medication. The average costs were reported in 2000/2001 prices. The resource data were derived from expert opinion, following a survey of GPs (12 GPs were randomly selected from a pool of 50 GPs in England and Wales who had experience of treating GAD with both venlafaxine and diazepam). It was unclear whether the costing data were also derived through this method or from another source.

The costs were treated deterministically as point estimates.

The indirect costs were not included in the analysis.

UK pounds sterling ().

To address uncertainties within the decision tree model, a univariate sensitivity analysis was undertaken on all the probability values and resource use variables used in the model. In addition, Monte Carlo simulations were undertaken in which the probability values, resource use values and costs were varied simultaneously. The probabilities were randomly varied according to a beta distribution by assuming a 25% standard deviation around the mean. Resource values and cost data were varied randomly according to a normal distribution, also by assuming a 25% standard deviation around the mean.

The expected probability of remission at 6 months was 0.276 for patients who started and responded to venlafaxine XL treatment, compared with 0.168 for those who started and responded to diazepam, (p=0.07).

The expected probability of not relapsing at 6 months was 0.970 for patients who started and responded to venlafaxine XL treatment, compared with 0.831 for those who started and responded to diazepam , (p=0.001).

The expected 6-month cost per patient was 351.54 for patients who started on venlafaxine XL and responded to treatment, and 310.56 for those who started on diazepam and responded to treatment.

Therefore, the additional cost per patient with venlafaxine XL compared with diazepam was 40.98.

The incremental cost-effectiveness ratio of treating a patient with venlafaxine compared with diazepam was calculated. The incremental cost-effectiveness for each additional patient in remission at 6 months was 380. The incremental cost-effectiveness for each additional patient in whom relapse was avoided at 6 months was 295.

The univariate sensitivity analysis showed that, although the magnitude of the cost-effectiveness of venlafaxine XL was sensitive to changes in model variables such as remission, relapse and probability of discontinuing treatment, there was no range in which diazepam would become a cost-effective treatment for GAD in comparison with venlafaxine XL. These findings were confirmed by the Monte Carlo simulation.

Starting treatment for generalised anxiety disorder (GAD) with venlafaxine XL (75 mg per day), compared with diazepam (5 mg three times daily), is considered the clinically more effective and cost-effective strategy for treating non-depressed patients suffering from GAD from the perspective of the UK National Health Service.

A justification was given for the comparator used. It reflected current practice in the authors' setting. The reader should consider whether this comparator reflects current practice in his or her own setting.

The analysis was based on a randomised controlled trial. Few details of the study design were reported, and the authors referred to another paper (Hackett et al 2003). Consequently, it was difficult to judge the internal validity of the study from this paper. In particular, it was unclear how the study sample was selected, whether it was representative of the study population, and whether the two randomised groups were comparable at baseline. Further, twice as many people were randomised to receive venlafaxine XL as those who received diazepam. Although the authors noted this limitation, they did not explain why this occurred. They also gave no explicit justification for their assumption that it would not introduce bias into the findings.

The health benefit measure used two of the effectiveness measures (the percentage of patients in remission at 6 months and the percentage of patients avoiding relapse at 6 months). Therefore, the comments in the 'Validity of estimate of measure of effectiveness' field (above) apply. The authors acknowledged that the study was limited because the model did not include changes in quality of life, improvements in general well-being and patient preferences. The use of quality-adjusted life-years as a measure of benefits would have enabled comparisons across health technologies.

The authors limited their analysis to direct costs and adopted the perspective of the health service. Hence, all the relevant categories of cost were included in the analysis. The unit costs and the resource quantities were reported separately, which enhances the reproducibility of the results. The resource use quantities were based on expert opinion, following a survey of GPs. However, a sensitivity analysis on these assumptions was undertaken, using ranges that appear to have been appropriate. The source of the price data was unclear. A sensitivity analysis of the prices was conducted. The price year was reported, which increases the generalisability of the results. Discounting was not carried out, which was appropriate given the short follow-up period.

Following a search of the literature, the authors were unable to find any other studies that compared pharmacological treatments for GAD, and very few studies that discussed the socio-economic impact of GAD. Thus, the authors were unable to make appropriate comparisons with other studies. They commented that their findings are applicable to those with GAD who are not suffering from depression. It is unclear whether these results would apply to those with co-existing co-morbidities such as depression or obsessive-compulsive disorder.

The authors noted further limitations of their study. First, the limited published data on the resources used to treat GAD in the UK and how patients comply with their medication. Therefore, resource use was estimated through GP opinion rather than published data. Second, the study was based on a decision model that used probabilities obtained from a controlled clinical trial, and as such, may not reflect naturalistic clinical practice. Third, it was assumed that patients who discontinued trial medication due to adverse events, lack of efficacy or non-compliance, and patients who were not offered trial medication beyond 8 weeks because of poor-response, would not achieve remission during the 6-month study period. This might not reflect outcomes seen in clinical practice, as this assumption was based on the expert opinions of GPs.

The authors stated that their study supports the use of venlafaxine XL compared with diazepam as the first-line treatment of GAD in order to maximise the probability of patients achieving remission. They also suggested that further prospective studies should be undertaken to determine the cost-effectiveness associated with venlafaxine XL from a societal perspective, and to determine whether better treatment of GAD improves productivity or allows the unemployed or economically inactive to work.

Sponsored by Wyeth Pharmaceuticals.

Hackett D, Haudiquet V, Salinas E. A method for controlling for a high placebo response rate in a comparison of venlafaxine XR and diazepam in the short-term treatment of patients with generalised anxiety disorder. Eur Psychiatry 2003;18:182-7.