Letters

Antidepressants and counselling for major depression in primary care

Authors' conclusions were not justified by findings

Measuring preference in primary care studies could be improved

Counselling is not demonstrably as effective as drug treatment for depression

Authors' reply

Authors' conclusions were not justified by findings

EDITORIn their randomised trial of antidepressant drugs and generic counselling for treating depression, Chilvers et al concluded that generic counselling is as effective as antidepressants and that general practitioners should allow patients to have their preferred treatment.¹ Their findings do not, however, support these conclusions.

The authors based their sample size calculation on a difference in mean Beck scores of 5 points as the outcome and found that 44 patients in each arm were required for a power of 80%. This sample size was not achieved in the randomised arms. They did not calculate the sample sizes required for global outcome or remission, but they are likely to be much larger as these outcome variables are categorical. Therefore, the only finding which achieved a power of 80% was related to Beck scores in the combined group of randomised patients and patients expressing preference.

Both general practitioner's rating and the score for research diagnostic criteria in table 1 show that patients choosing counselling were objectively significantly less depressed than the other groups, although their Beck inventory scores were similar. In other words, compared with the other groups, patients choosing counselling were comparatively more depressed subjectively than objectively. These patients were less depressed objectively and might respond more readily than other groups to interventions. Therefore, Chilvers et al should not have combined randomised patients with patients who expressed a preference. Furthermore, they cannot conclude that generic counselling is as effective as antidepressants simply from the apparent lack of differences in Beck scores in the combined patients who expressed a preference.

Chilvers et al further concluded that general practitioners should allow patients to have their preferred treatment. While this recommendation might be appropriate, it does not follow from their findings. To draw this conclusion, the authors would need to compare the outcomes of patients who chose a specific treatment and were offered it with those who requested the same treatment but were offered another treatment instead.

Wai-Ching Leung, honorary lecturer in public health medicine. 
University of East Anglia, Norwich NR4 7TJ w-c.leung{at}uea.ac.uk

Competing interests: None declared.

Measuring preference in primary care studies could be improved

EDITORThe study by Chilvers et al is one of a few supporting a relation between receipt of preferred treatment and improved outcome in treating depression.¹ The patients who chose counselling did better than those randomised to counselling, although the 95% confidence interval reached zero but but did not cross it.

Many patients express a preference for psychological compared with drug treatments.² However, being allowed to choose treatment does not improve short term outcome in depressed patients in primary care given either antidepressants or counselling³ or non-directive counselling, cognitive-behaviour therapy, or usual general practitioner care.⁴

This difficulty in showing the effects of preference may be methodological. As in the current study, preference has been defined as refusal to be randomised within a trial. However, many patients might be prepared to allow themselves to be randomly allocated treatment but would still prefer not to receive the treatment to which they are allocated, diluting the beneficial effects inherent in the preference arm.

An alternative method would be to randomise the entire population and then allow patients to accept or decline the allocated treatment. In this procedure the consent process would be split in two, with patients initially consenting to take part in the study on the understanding that a treatment will be offered but does not have to be accepted, followed by a second stage in which they accept or decline the treatment. Those who decline remain in the study but are treated as the general practitioner believes is clinically appropriate. This gives three groups that can be comparedthe whole cohort, those who accept randomisation, and those who decline itand it allows the effects of preference to be described in more detail.

Andrew Martyn Thornett, clinical research fellow. 
Department of Psychiatry, University of Southampton, Southampton SO14 0YG eanador{at}soton.ac.uk

Competing interests: None declared.

Counselling is not demonstrably as effective as drug treatment for depression

EDITORThe study by Chilvers et al investigating the effect of antidepressants and generic counselling in depression has flaws in its design and interpretation.¹ Its recommendations are not supported by its findings.

The main outcome measure discussed is based on the Beck depression score at 12 months. However, many people who start off being depressed will not be so 12 months later even without treatment, and the main effect of antidepressants is to accelerate what will often be a spontaneous recovery. Hence outcome at 12 months is insensitive as a guide to the effectiveness of any treatment for depression. Also, a substantial proportion of patients will improve fairly quickly with placebo, but Chilvers et al did not provide a placebo for either the drug or the psychological treatment. There were not even "waiting list controls." Hence it is impossible to know whether, in the context of this study, either treatment has any effect whatsoever, either in accelerating recovery or in producing a good outcome at 12 months.

Chilvers et al state that both counselling and antidepressant drugs are effective. For the above reasons, their study provides no evidence at all to support this assertion. One of the bullet points in the box entitled "What this paper adds" states: "12 months after starting treatment, generic counselling is as effective as antidepressants." Again, there is no evidence for this. Presumably the authors are making the classic mistake of equating the failure to show a difference with showing no difference. In fact, of the randomised patients who were followed up, 78% who received drugs were no longer depressed compared with 47% who received counselling. Another bullet point states: "Patients treated with antidepressants may recover more quickly" [my italics], but the text simply states that they did recover more quickly. However, no data relating to time to remission appear anywhere in the results section. Results showing the superiority of antidepressants seem not to have been presented.

Overall, much evidence suggests that antidepressants and some psychological treatments are effective in alleviating depression, but this is not the case for generic counselling. The study by Chilvers et al provides no useful information, and the authors have no business recommending that "general practitioners should allow patients to have their choice of treatment." Following this recommendation would be expected to lead to an avoidable increase in morbidity and mortality from depressive illness while squandering public resources on providing counselling, which is of no proved benefit for this condition.

David Curtis, consultant psychiatrist. 
East London and City Mental Health NHS Trust, Royal London Hospital, London E1 1BB

Competing interests: None declared.

Authors' reply

EDITORLeung rightly points out that our calculation for sample size was based on Beck scores, the primary outcome measure in our trial of antidepressant drugs and generic counselling for the treatment of depression. The main results based on the Beck scores shown in table 2 of our paper are adjusted for baseline scores for research diagnostic criteria, as well as for patient preference or randomised group. Leung seems to suggest that the doctor's report of depression is to be preferred to the patient's. We remain to be convinced.

Both Leung and Thornett draw attention to the difficulty of assessing the effect of patient preference on outcome. Although Leung's suggestion may have some theoretical justification, there would be practical problems in carrying it out. Patients would have to agree to express a preference and then accept the treatment that they did not prefer. We would argue that those consenting to enter such a trial would not have strong preferences and we probably would be no further forward. In Thornett's design the group accepting the allocated treatment would consist of those preferring the allocated treatment and those who were indifferent, thus diluting the effect of preference.

Curtis considers that our trial should have included a placebo arm. He also believes that there is plenty of evidence that antidepressants and some psychological treatments are effective. We did not consider it ethical to include a placebo arm.

With reference to our statement that patients taking antidepressants recover more quickly, the median times are given in the electronic version but not in the paper one. Median time to remission was three months in all groups except the group randomised to antidepressants, where the median time to remission was two months (comparing randomised groups log rank statistic 2.74, P=0.1; pooled log rank statistic for randomised and patient preference trials 0.82, P=0.36). Thirty three (15%) of the 221 patients had a relapse. There were no differences between the groups.

Having shown that generic counselling is as effective as antidepressant treatment, we recommend that patients should be allowed to choose between two effective treatments, thus allocating a scarce resource (counselling) to those who find it most acceptable. Further analyses (in preparation) suggest that the costs of the two treatments are similar.

Clair Chilvers, professor. 
Michael Dewey, senior lecturer. 
michael.dewey{at}nottingham.ac.uk Trent Institute for Health Services Research, University of Nottingham Medical School, Queen's Medical Centre, Nottingham NG7 2UH

Also signed by the 11 other authors: Katherine Fielding (lecturer), Virginia Gretton (research assistant), Paul Miller (lecturer in health economics), Ben Palmer (research associate), Trent Institute for Health Services Research; David Weller (professor), University of Edinburgh; Richard Churchill (lecturer), Idris Williams (professor), University of Nottingham Medical School; Navjot Bedi (specialist registrar in psychiatry), Nottingham Healthcare NHS Trust; Conor Duggan (professor), University of Leicester; Alan Lee (consultant psychiatrist and special senior lecturer), University Hospital, Queen's Medical Centre; and Glynn Harrison (professor), University of Bristol.

On behalf of the Counselling versus Antidepressants in Primary Care Study Group.