Author's reply

Dr Ogundipe is, of course, quite right in pointing out that our conclusion about the greater acceptability of couple therapy compared with antidepressant drugs can only apply to the particular regime used in our trial. However, it is worth noting that similarly high drop-out rates have been recorded in other well-regarded trials of antidepressant treatment. For example, in the National Institute of Mental Health trial, the drop-out rate from imipramine treatment was 33% over 18 months (Jacobson & Hollon, 1996).

Dr Ogundipe's view about intention-to-treat analysis is contrary to current statistical opinion and the recommendations of the British Medical Journal for the reporting of clinical trials (Altman, 1996; Schulz, 1996). The comment about baselines is not relevant here, since subjects who comply may fare differently and in an unpredictable way from those who do not comply. Thus, any observed differences between groups constructed in this manner may be due not to treatment but to factors associated with compliance. In this study, patients who dropped out were younger and had higher depression scores than those who completed the trial. The method of analysis should be consistent with the experimental design of a study. For randomised trials, such consistency requires the preservation of the random treatment assignment. Because methods that violate the principles of randomisation are susceptible to bias, they should not be used.

An analysis of the number needed to treat may be a sensible suggestion in general, although for a number of technical reasons it is not popular among statisticians (see Hutton, 2000). In any case, in this trial the high drop-out rate from the medication group would make the results of such an analysis suspect.

Marital discord was assessed using the Dyadic Adjustment Scale. As shown in Table 1 in the paper, the two treatment groups did not differ on this score, making it unlikely that this variable confounded the results.