Dr Fusar-Poli identified a number of studies reporting follow-up transition rates, which is not the same as predictive the validity of the tests or criteria. Most, if not all, of the studies he identified did not have information on predictive attributes of the tests or criteria, such as sensitivity and specificity. However, they had useful information on transition rates. From these it is impossible to know how good the tests/criteria were in ruling in or out the risk of developing schizophrenia from prodromal symptoms, since these studies were not systematically following up those who tested negative to the test.

Dr Fusar-Poli raised another important issue regarding overlapping of samples. We checked for double publication, but not necessarily overlapping of samples. We were interested in knowing how good the test is in predicting schizophrenia in high-risk populations. We therefore were interested in diagnostic attributes of a test in each study/subsample. The values for sensitivity and specificity for Yung et al (2003) ^{Reference Yung, Phillips, Yuen, Francey, McFarlane and Hallgren1} and Yung et al (2004) ^{Reference Yung, Phillips, Yuen and McGorry2} were not identical. For the purposes of predictive validity of a test, these are two different studies. Yung et al (2005) ^{Reference Yung, Phillips, Yuen, McGorry, Kelly and Dell'olio3} had a follow-up of 6 months (n = 105) and Yung et al (2008) ^{Reference Yung, Nelson, Stanford, Simmons, Cosgrave and Killackey4} had a follow-up of 24 months (n = 292). Again, these are different studies, we are not sure whether there was overlapping of samples in these two but we don't see how this would affect how good the test is at ruling in or out the risk of developing schizophrenia. The same can be said with studies by Cannon et al ^{Reference Cannon, Cadenhead, Cornblatt, Woods, Addington and Walker5} and Woods et al, ^{Reference Woods, Addington, Cadenhead, Cannon, Cornblatt and Heinssen6} the diagnostic attributes of the Cannon study were not identical to Woods’ study.

Dr Mitchell raises important points regarding the predictive validity of prodromal criteria. In particular, Dr Mitchell is right to suggest that the positive predictive value and negative predictive value statistics are more intuitively informative than sensitivity and specificity, and so their reporting would have been beneficial. We also agree that assessing the clinical usefulness of prodromal criteria requires further consideration. We plan to further examine this important question in a subsequent paper. We welcome Dr Mitchell's proposal for a randomised study where high-risk patients are randomised to predicting psychosis with or without formal tests for prodromal criteria.