We read with great interest the paper by Gainotti et al (Reference Gainotti, Azzoni and Marra1999) and we wish to raise some concerns about their study.

Description of the statistical analysis was omitted from the paper and it is impossible for the reader to know how the data were analysed. The authors did not mention whether the patients were administered drugs usually given to stroke patients (e.g. steroids, beta-blockers and anticonvulsant drugs) and which can induce some depressive symptoms. In the section entitled “ Criteria used to make a diagnosis of major depression”, the authors state that the validity of their diagnostic criterion (Hamilton Rating Scale for Depression (HAM-D) score >17) has been documented by others (Reference Salzman, Schneider and AlexopoulosSalzman et al, 1994). This assertion is false - we find no scientific support in the Salzman et al paper for Gainotti et al's assertion. Also, information on the clinical features of Gainotti et al's sample is limited by their omission to report standard deviations and the mean age of patients with endogenous depression (see Table 1, p. 164).

The most important problem, however, is with regard to the interpretation of HAM-D scores across groups. It is not made clear whether the mean HAM-D scores reported in Table 1 are calculated across the whole sample or only within the group with depression. In the psychiatric literature a HAM-D score of <12 is not generally considered clinically significant, so if Gainotti et al have calculated mean HAM-D score only within the group with depression, it is difficult to understand why the mean is so low (11.8 at <2 months post-stroke). Alternatively, if mean HAM-D scores were calculated across the whole sample (i.e. patients with and without depression), the increase in the mean with increasing time post-stroke may be due simply to the increase in the relative number of people with depression (27% at <2 and 2-4 months post-stroke v. 40% at >4 months post-stroke). In other words, the increase in the mean HAM-D score does not necessarily imply that the severity of depression in the whole sample increases from the acute to the later post-stroke period, but this tendency may be due simply to an increased proportion of patients with depression within the sample.

Finally, Cohen's κ for diagnostic concordance is not given by Gainotti et al. This index is routinely calculated when different diagnostic criteria are adopted for patient classification. Inspection of Table 1 shows that nine well subjects out of 43 were misclassified as having depression using the quantitative criterion of HAM-D score. This begs the question, how many patients with depression were classified as being well?