Author's reply

We are grateful to Dr Jagadheesan and colleagues for allowing us to reiterate the methodological strength of our trial design. If our control group had been as Dr Jagadheesan proposes (continued antidepressant treatment only), then we would wrongly have concluded that exercise is highly effective as an adjunct to drug treatment in old age depression. In fact when compared with the effects of a structured social intervention, group exercise offered only a modest additional benefit. Our non-exercise control crucially allowed us to disentangle the psychosocial effects of coming together as a group from the effects of exercise itself. This novel use of a non-exercise control intervention which matched the exercise intervention in duration, frequency and social contact represents an important methodological advance which future researchers will wish to consider (Lawlor & Hopker, 2001).

Perhaps Salmon may shed some light on the low mean attendance rate by his comment that advocacy of exercise as a treatment for depression ‘must puzzle clinicians, who in treating depressed people, often have to contend with an absence of motivation to tackle much less strenuous features of life's routine’ (Salmon, 1990).

The Results section of our paper is succinct, in part because of editorial constraints on article length. A typical finding (such as a group comparison of reduction in Hamilton Rating Scale for Depression score at 10 weeks) gives only a comparison of proportions and the associated P value, with the statement ‘we used Fisher's exact test in our comparison on the two groups’ implied, but unwritten. In fact, we prided ourselves on the explicit statement of numerator and denominator here and elsewhere (23/42 v. 14/43) when many authors might simply have said ‘55% v. 33%, P=0.05’.

We are surprised that Dr Jagadheesan et al advocate use of the Wilcoxon test for paired comparisons: this would have resulted in a highly undesirable plethora of comparisons (10 weeks v. baseline; 34 weeks v. baseline; 34 weeks v. 10 weeks). Statistical propriety necessitates the use of an approach which recognises the temporal ordering of the trio of results for each subject in each group; hence our appropriate use of repeated measures.

The analysis which resulted in the stated findings for secondary outcome measures was a one-sample t-test on the logarithms of the ratios of outcome:baseline scores.

There is a desperate need for better-quality research in the area of depression, and we believe that our trial design offers an important methodological advance.