Tyrer's ‘From the Editor's desk’ lyrically asserted that in relation to the paper by Fergusson et al ^{Reference Fergusson, Horwood and Boden1} and other studies, ‘In the parched desert of ignorance and prejudice every established fact becomes an oasis. By “established fact” I mean one that defines the field, the one that all the related and restlessly inchoate facts gather round and say “I belong here”, and then fall into line behind it.’ ^{Reference Tyrer2}

Fergusson et al ^{Reference Fergusson, Horwood and Boden1} conclude that there is evidence that abortion may be associated with a small increase in risk of mental disorders and in comparison, other pregnancy outcomes were not associated with increased risk. Although we acknowledge that aspects of their analytic design are strong and carefully implemented, we believe that the analyses have not maximised the potential of the data-set and that therefore, your editor's rhetorical confidence is not yet justified. We advance the following reasons.

First, Fergusson et al dichotomised each pregnancy exposure. Of 534 women in the Christchurch cohort, 284 had had pregnancies. Women making decisions about terminating pregnancies may have prior pregnancy events and potentially cumulative losses will have different mental health impacts compared with termination as the outcome of a first pregnancy. Pregnancy variables are not independent and mutual adjustment in models for other outcomes will not account for the interactions between pregnancy outcomes. A more useful analysis would have been with a composite variable with never having had a pregnancy event as the reference category.

Second, the combining of therapeutic abortion for fetal malformation with abortion by choice is inappropriate. Most abortions are first trimester. There is an argument for separating termination of pregnancy by gestational age, so that the mental health impact of those in the second or third trimester are visible and separate. It is possible that terminating a wanted pregnancy because of fetal abnormality would be more distressing than an early unwanted pregnancy.

Third, many authors (including Fergusson et al) have found strong relationships between intimate partner violence and poor mental health, and between intimate partner violence and increased association with reporting terminations. ^{Reference Taft and Watson3–Reference Gazmararian, Adams, Saltzman, Johnson, Bruce and Marks6} Despite the potential to include the much more rigorous measure from their previous study of partner violence among this cohort, the authors have excluded their strongest measures of partner violence in this analysis, leaving a major covariate poorly measured.

Fergusson et al conclude that the evidence for abortion impact is small but clear – even causal. Yet there is no evidence that the risks associated with other pregnancy outcomes, particularly loss, are different from those estimated for abortion (see Charles et al ^{Reference Charles, Polis, Sridhara and Blum7} ), nor that mental health disorders are incident after an abortion. This could have been statistically tested using logistic regression among the range of statistical tests already carried out.

It is a pity that such a good cohort study has not been better analysed. With the above adjustments, the authors would be better placed to more clearly identify the vulnerable groups they are wisely seeking to identify.