I shared a critical review of Das-Munshi et al's paper ^{Reference Das-Munshi, Béwares, Boydell, Dewey, Morgan and Stansfeld1} with colleagues at the Ealing Journal Club, West London Mental Health Trust, highlighting issues of considerable relevance. My observations below have therefore been enriched by the reflections of a group of doctors working in psychiatry.

The first observation is that clinical applicability is limited by the study's use of a screening test and its cross-sectional survey design. A screening test for psychotic experiences rather than a clinical assessment tool for psychotic illness distances the findings from the clinical domain. As Das-Munshi et al indicate, there is reason to believe that psychotic experiences and illness are correlated; and the statistical properties of Bebbington's Psychosis Screening Questionnaire (PSQ) ^{Reference Bebbington and Nayani2} v.agoldstandard ^{Reference Wing, Babor, Brugha, Burke, Cooper and Giel3} look impressive. However, especially with the symptom domain of delusions, there seems room to doubt how effectively a member of the general public will self-assess for presence/absence of symptoms; and, with the symptom domain of hypomania, how well this taps into the core concept of psychosis (construct validity).

As the authors acknowledge, the cross-sectional design brings broader limitations, precluding inferences of temporality. Even if an association between psychosis and ethnic density is granted, it remains unclear which way the arrow is pointing. It may even be that ethnic density is just another factor in a nexus of contributors to overall risk of psychosis (multifactorial model).

A second observation is that, although the associations between psychosis and psychosocial factors (racism/discrimination, chronic strains and social support) are quite convincing, those between psychosis and ethnic density are less robust. Indeed, statistical significance is only reached for the Indian group and the combined minority sample. From here, it seems a substantial jump to claim that ‘the general trend was supportive of similar associations for all minority groups’.

This connects to the third observation: that, in grouping together the different ethnic groups (thereby generating sufficient power for a significant effect), the authors may have fallen foul of the same ‘over-homogenising’ they rightly deem a ‘limitation of previous work’.

Finally, we were intrigued to note the ‘cross-over’ effects in the interactional models (Figs 3 and 4). It appears that, at very low ethnic densities (‘ethnic isolation’), practical/confiding support from the nominated closest person may actually increase risk of psychotic experiences. A chance finding cannot be excluded, but there are other possibilities. First, more supportive relationships may have elements of enmeshment, possibly related to high expressed emotion. Second, a more unwell person might elicit higher levels of support from those closest to them. Third, in the context of ethnic isolation, a close and supportive relationship may actually limit chances of encountering disconfirmatory evidence from outside the ‘relationship bubble’ for any psychotic-spectrum symptoms emerging.

We would like to thank Das-Munshi and colleagues for their thought-provoking paper, and would welcome their own thoughts on the above observations.