In an important editorial, Mitchell (Reference Mitchell2006) marshalls evidence to show that we do not have to wait 2 weeks for antidepressants to work.

Why has it been so difficult so far to show that they work in the first few days? In addition to the reasons that Mitchell sets out, I should like to mention a further problem. If you analyse the results on a day-by-day basis, it is hard to obtain sufficient statistical power to distinguish the early response to the drug from the response to the placebo, since you have just the scores for that day.

In 1996 my colleagues and I published evidence that the fall in scores on the Hamilton Rating Scale for Depression followed an exponential decay curve with a correlation coefficient of 0.99 (Reference Priest, Hawley and KibelPriest et al, 1996; Reference Livingston and ClarkLivingston & Clark, 1997). This observation corresponds with Mitchell's remarks on the steep fall in scores in the first 2 weeks. A comparison of the slope of the curve for the active drug with the placebo, using all of the data, gives a very sensitive way of testing for efficacy.

By plotting the log of the depression scores against time, a straight line is obtained. Thus the recovery from depression is of one piece, with a smooth process throughout. The clear implication is that there is no delay in the onset of action, either of the active drug or of the placebo. By using the slope of the graph, one can use all of the trial results, not just those on a particular day. The statistical power is greatly increased and the distinction between drug and placebo enhanced.