We thank Drs Gordon and Shoesmith for their interest in our editorial, their complimentary remarks and their considered responses to what we said. Dr Gordon repeats our call to avoid prematurely abandoning categories or dimensions, and highlights the lack of known diagnostic biomarkers for psychosis, either as a whole or for current subtypes. Tandon et al ^{Reference Tandon, Keshavan and Nasrallah1} did not really consider this, quite reasonably, as their review focuses on what is known about the aetiology and pathogenesis of schizophrenia. As we have clarified in a forthcoming review, ^{Reference Lawrie, Olabi, Hall and McIntosh2} the lack of known biomarkers for psychosis (whether as categories or continua) is at least partly because the right sort of studies to find them have only rarely been done and reported in this light. The relevant populations need to be studied and then the results analysed according to the principles of clinical epidemiology (or evidence-based medicine), to extract the potential clinical significance for individuals of statistically significant abnormalities evident in groups of patients. Thus, for example, if one wished to identify specific diagnostic markers of schizophrenia that have clinical utility, a (preferably large) representative population of people in their first episode would need to be assembled, and predictive values and/or likelihood ratios calculated for the value of potential markers of schizophrenia as opposed to, say, bipolar disorder. Despite the paucity of studies, there are already a few well-replicated large differences between people with schizophrenia and healthy controls, which may also distinguish them from those with bipolar disorder. ^{Reference Lawrie, Olabi, Hall and McIntosh2} Not all of these require high-tech investigations. Simple clinical measures of neurodevelopmental aberration such as neurological soft signs, and even historical measures such as early social difficulties, are common in people who go on to develop schizophrenia but may not be in those with bipolar disorder. These already influence clinical decision-making but in an informal and rather haphazard fashion. The optimal method of eliciting and using such information needs further investigation, as outlined above and in our review. ^{Reference Lawrie, Olabi, Hall and McIntosh2}

Dr Shoesmith is absolutely right to remind us that any resource-intensive diagnostic procedure is going to be much less practical in less well-developed health services. This is of course an immediate and quite possibly fatal problem for any system requiring multiple ratings on continua and could be even more so if, for example, magnetic resonance imaging of the brain/mind turns out to be diagnostically valuable – as we suspect it might. ^{Reference Lawrie, Olabi, Hall and McIntosh2} In the long run, whatever turns out to be the best conceptual approach to psychosis for the maximal benefit of patients, and whether or not this has to be pioneered in leading clinical research centres, the process of formalising our diagnostic and therapeutic judgements will bring a much-needed and long-overdue re-engagement of psychiatry with the rest of medicine.

We are also grateful for the opportunity to respond to the letter from Professors Claridge and Barrantes-Vidal, especially those of us who after more than four decades still remember Professor Claridge's excellent and provocative teaching on, and seminal contributions to, the field of schizotypal cognitions, beginning as they did more than 30 years before this area became fashionable. We cite Paul Meehl as he is one of the very few commentators on diagnosis in psychiatry, whether psychologists or psychiatrists, to have offered a testable hypothesis that would allow one to make an informed decision about whether a categorical or continuous approach might be more valid. We recognise that there have been several alternative proposals to handling the complexity of psychosis, but very few of these have been tested in practice. To clarify our position, we are not opposed to continuous measures, be they psychological trait or cognitive test scores or brain imaging variables, nor are we particularly in favour of the status quo or hybrid models. We are simply arguing that any proposals to change our diagnostic approach to psychosis, which has survived to this day for some quite good reasons, should be based on data and therefore built on evidence rather than fashion or because something looks good on paper. We would very enthusiastically support, for example, a trial that tested the efficacy of one or more treatments on one or more continua of psychosis severity. Having said that, however, even if that trial generated informative results for clinical practice, any resulting practical system would of necessity have to include thresholds for treatment and would thereby create categories. As we said, continua may or may not be more valid than categories of psychosis, but clinical decisions require choices between alternative courses of action.