Predicting the development of schizophrenia

The aim of Chuna & Mahadun's study ¹ is compelling and of great clinical interest for preventive interventions in psychosis. Unluckily, the results presented by the authors are not reliable as they are undermined by severe methodological caveats.

First, the systematic research of the literature failed to uncover the majority of studies reporting follow-up transition rates in a sample of individuals at high risk (HR) for psychosis. The authors included only 12 studies but many more were available in the electronic databases (references supplied). Second, the authors did not check for potential overlapping between samples including studies enrolling the same individuals. For example, Yung et al (2003) ² and Yung et al (2004) ³ were both retrieved despite the authors of these studies clearly stating in their manuscript that the ‘current paper [2004] continues that research [2003; n = 49] by expanding the sample size to 104’. The same applies to Yung et al (2005) ⁴ and Yung et al (2008), ⁵ while Woods et al (2009) ⁶ is a revised analysis of Cannon et al (2008). ⁷ Third, the authors stated in the inclusion criteria that the included studies ‘had a clearly specified population, from which a prodromal criterion was administered to identify clearly those with prodromal symptoms [HR +] from those without [HR–]’. There is no such a ‘clearly specified population’ from which the high-risk individuals are sampled. The sampling is based on help-seeking behaviours and does not epidemiologically represent the local population; in fact, the prevalence of high-risk symptoms in the general population is unknown. The second requirement to be included in the Chuma & Mahadun's meta-analysis was that ‘the two groups [HR+ and HR–] were then followed up for a number of months and assessed again with a diagnostic instrument to determine those who had converted to schizophrenia’. This is really surprising as the vast majority of the longitudinal studies did not follow-up the help-seeking individuals who underwent the clinical assessment at the prodromal services but were not considered at risk for psychosis (HR–). Consequently, it is completely obscure how the authors may have estimated the correct prevalence of false negatives (HR–, who developed psychosis over time) in their analysis. Given all the above concerns, I feel the results of this meta-analysis should be considered carefully as pilot data strongly undermined by significant methodological biases.