Martin M. Katz: Onset of Antidepressant Effect
Don Klein’s response raises several questions about my statements in the original essay that probably could have been answered with more careful reading of the referenced sources.
For example, regarding his first query about why the need for a large prospective study to confirm findings from our study reported in Katz et al 2004, when such “large sample” studies of Szegedi et al 2009 and Stassen et al 1997 have already been conducted.
The answer is that the referenced studies used only the Ham-D as an outcome measure, whereas our 2004 study reports results with an array of dimensional measures of depression that reinforce the predictability of the early changes. Since definitive demonstration of one or more of these dimensions as a predictor increases the power and breadth of prediction, the results of the large sample study can be useful in determining the range of effects of which a new drug might be capable, if included in such studies.
Klein further questions whether Szegedi et al (2009) and Stassen et al (1997) are “large” studies, since he was only able to retrieve and study the abstracts. But even the Szegedi abstract (and title) retrieved from PubMed, notes that the sample size included 6,000+ patients and the Stassen abstract refers to some 1,500 patients, figures hard to ignore in reading the abstracts. Then, Klein diverts the discussion to our 2004 study, noting that the sample was 82 patients. He takes issue with the “multiple measures” and the lack of correction for multiple analyses. He has a point here, but we were clear in the text to indicate that we confined the analyses to “specific hypotheses, focused on five constructs and one of the severity dimensions”, thus, limiting the number of analyses. Further, many of the primary findings had p-values less than the 0.05 level and were consistent across several methods of statistical analysis. So that if one had questions here, an even more conservative approach, such as accepting only those findings at the p<0.01 level, findings that could clearly not be due to chance variations, would strongly reinforce the validity of the major study findings.
Regarding the algorithm for prediction referred to in the 2004 study, he questions the “definitiveness” of this finding. I do not see where we, at any point in the article, justified the algorithm as “definitive”. Our major point was that that was the best model that could be achieved by combining variables from a sample of this size. In view of the way it was applied, one would consider the algorithms, designed to provide the best combination of sensitivity and specificity, exploratory, in nature, another reason, to recommend a prospective large sample study on the prediction issue.
To achieve a reasonable estimate of when onset of clinical effects occurred in each drug and placebo group, we chose to apply the “median” time of onset approach, i.e., the time at which >50% of patients within a group, showed >20% change or significant improvement on that dimension. It seemed to us and to Stassen and Szegedi to be a highly defensible criterion for estimating the time at which a drug initiates significant improvement. Don Klein is welcome to differ with us on that but it is not clear on what basis. The fact that individual patients will vary on this measure of onset is of course, obvious, but here we are simply seeking an adequate estimate of the time at which most patients (>50%) in this treatment group, show a significant amount of change; the median measure provided a simple and accurate estimate.
The “analysis of onset” paragraph on pg 569 of Katz et al (2004) referred to was to determine when improvement that leads to clinical response at outcome for the treatment-responder group, begins. The statistics were aimed at determining the initial (at two weeks) indicators for those behavioral variables that distinguished the responders from the non-responders to the specified treatment. Effect size was then used to see whether the behavioral changes at two weeks that distinguished the treatment responders, were not only statistically significant, but likely to be “visible” to the observer.
Klein ends his queries with a quote from Stassen et al that clearly describes how their group arrived at > 20% change as a reasonable, statistically based definition of onset of improvement. After Klein claims not to have found the demonstration that I refer to as drug effects at 1 and 2 weeks (despite much of the original essay prepared on “onset” dedicated to establishing that the evidence was cumulative and quite compelling on this critical issue), he concludes that “clearly the problem of therapeutic onset has not been solved”. Maybe not completely, but we wonder: what is the basis, the studies that support his unacceptance of the voluminous evidence compiled over the last three decades that lead to the very logical, and well supported conclusion that the established antidepressant drugs begin their clinical effects within the first two weeks of treatment?
Regarding the table from our 2004 study, that Klein modified: that query was answered in my previous response.
1. Katz MM, Tekell J, Bowden CL Brannan S,Houston JP, Berman N, Frazer A. Onset and early behavioral effects of pharmacologically different antidepressants and placebo in depression. Neuropsychopharmacology 2004; 29: 566-79.
2. Stassen HH, Angst J, Delini-Stula A. Delayed onset of action of antidepressant drugs? Survey of recent results. Eur Psychiatry 1997: 12: 166-76.
3. Szegedi A, Jansen WT, van Wugenburg AP. Early improvement in the first two weeks as predictors of treatment outcome in patients with major depressive disorder: a meta-analysis including 6,562 patients. J Clin Psychiatry 2009; 70: 344-53.
Martin M. Katz
October 1, 2015