We have discussed the paper by Brannigan et al^{Reference Brannigan, Tanskanen, Huttunen, Cannon, Leacy and Clarke1} at our journal club, and we wish to raise some of the points that arose from a critical appraisal of this paper, and the panel discussion that ensued.

First, the figures presented in the abstract appear to be inaccurate. The figure 3.28 (odds ratio for any stress exposure) has been taken from the fully adjusted results; the figure 3.13 (odds ratio for exposure to moderate stress) has been taken from the unadjusted results, and the figure 7.02 (odds ratio for those exposed to severe stress) does not appear at all in the results. There appears to be an error in Table 1 – under the heading ‘maternal psychopathology’ the figures 3402 (93.8%) and 224 (6.2%) should be the other way around.

However, the main concerns raised were around the statistical methods used. ‘There was some variability in the number of prenatal questionnaires returned. Therefore, we adjusted for the numbers of questionnaires returned to account for this.’ This provides no information on how many questionnaires were returned by each participant and how this was adjusted for. Does this not mean that, for some participants, ‘prenatal stress' could actually refer to ‘self-reported stress during a single month of pregnancy’?

‘A modal measure of stress was used as it best represented the individual scores when compared with the mean, which was less accurate due to variability in the number of returned prenatal questionnaires.’ From this we were struggling to understand the reasoning behind use of the mode; we felt it possible that single highly stressful events would not have been captured in the final data.

By using diagnosis of personality disorder on a hospital discharge register as the primary outcome of interest, there were only 40 positive cases, and only 9 with no comorbid psychiatric diagnosis. This is a very small sample to compare with the 3586 without a diagnosis. Moreover, some of the covariates that the authors controlled for had groups of participants as small as one individual, which made the overall results too unstable to interpret.

Various psychosocial mechanisms were suggested, including ‘early life separation from parents, childhood trauma and parenting styles’, however, there was no suggestion that these variables were related directly to levels of prenatal stress.

Mental health support in the perinatal period is clearly of huge importance for the well-being of mother, baby and the wider family, and research into this area is needed. However, there is a responsibility to ensure that the statistics are robust, and conclusions justified, particularly in view of the extensive media coverage generated by this paper.