We would like to clarify some aspects of our paper in reply to the points highlighted by Di Michele & Bolino.

First, we would stress the fact that the preliminary analysis of our data had been extensive, but that only data relevant to the specific scope of our study, which consisted in checking the most recent version of Robinson and co-workers' biological theory of post-stroke depression, were included in the manuscript. Our data are clearly inconsistent with this theory.

Regarding our data analysis, continuous data were treated using one-way analysis of variance, whereas frequencies of distribution were analysed by means of χ² tests.

The possible influence of drugs was checked in our study by excluding all patients who were taking at the time of examination (or had taken in the previous four weeks) antidepressant drugs. We did not check other drugs (such as steroids, beta-blockers or anticonvulsants) which could induce depressive symptoms, since they were not considered relevant for the specific scope of our study.

The patients with endogenous depression were matched as for age (60.1 years) and educational level (7.9) with the three groups of post-stroke patients. Only a slight difference in gender distribution was observed between stroke patients and those with endogenous depression (a preponderance of females (20 : 10) among the group with endogenous depression). This not unexpected difference was not considered relevant with respect to the scope of the study.

Regarding interpretation of the HAM-D scores across groups, the main scope of our study consisted in determining whether the nature of post-stroke depression is different in the acute and in more chronic post-stroke periods. From this point of view, it was important to evaluate at various time intervals from stroke the qualitative aspects of depression and their anatomical—clinical correlates, whereas the severity of depression in patients with major depression was much less relevant. For this reason, the HAM-D scores were calculated, as the authors of the letter correctly argue, in each group as a whole (including subjects with and without depression) and the increment of the mean depression score across groups mainly reflected the relative increment of subjects with depression. Though this fact is not very relevant to the aim of our research, we must add that even considering only the patients with major post-stroke depression, we could observe a non-significant trend towards an increase in the mean HAM-D score from the acute (20.2) to the post-acute (21.8) and to the more chronic (23.5) post-stroke period.

Concerning the HAM-D score criterion for diagnosis of major depression and correlations between clinical and psychometric criteria, although different cut-off scores have been proposed in the literature, a score of 18 on the HAM-D is the most currently used (Reference Endicott, Cohen and NeeEndicott et al, 1981; Reference Rapp, Smith and BrittRapp et al, 1990). Furthermore, good concordance exists in our study between clinical (DSM-III-R) and psychometric (score >17 on the HAM-D) criteria. We have measured this concordance on our data by the κ statistic (Reference Holman, James and HelmanHolman et al, 1982), which gives a numerical measure of chance-corrected categorical agreement. According to this index, which results from the ratio between the chance-corrected observed agreement and the chance-corrected perfect agreement, the perfect agreement corresponds to +1, the complete disagreement corresponds to ‒1 and the chance level is 0. The chance-corrected level of agreement between DSM-III-R criteria and HAM-D score >17 was quite satisfactory in our study (κ=0.84).