In this analysis, using a newly developed electronic assessment tool (OPCRIT+), we identified a six-component symptom structure underlying the psychopathology recorded in a large, mixed-diagnostic, inpatient cohort. Using component scores to indicate severity, we demonstrated distinct symptom profiles across different clinical diagnoses for five of the six components. Furthermore, these severity scores provided significant predictive value, which was more informative than diagnosis, for a range of clinical outcome measures.
The component structure we extracted is similar to those reported in studies using the original OPCRIT for this purpose 
. In fact, the five most commonly reported components (or factors) in those studies were also extracted in our PCA: mania, depression, negative symptoms, disorganization and positive symptoms (although the specific OPCRIT items associated with these components varies somewhat across studies). This similarity occurred despite the fact that over half of the patients in our study belonged to diagnostic categories outside the psychotic and affective spectrum, from where cohorts in the other studies were drawn. One notable difference in our component structure however, was the extraction of an ‘anxiety’ component. This occurred due to the additional items in OPCRIT+ allowing the diagnosis of anxiety spectrum disorders.
The extracted components explained 46% of the variance in the symptom data being recorded. This is at the lower end of the range seen in the studies cited above (mean: 52.2% range: 39–71%). There are a number of possible explanations for this. For example, it may be because our PCA contained ratings from a large number of doctors, whereas those in the cited studies typically contained far fewer raters. Alternatively, it could have resulted from the addition of patients whose primary diagnosis was outside the psychotic and affective spectrum and who may have presented with more heterogeneous symptom profiles. Despite this, the successful extraction of an underlying component structure is a vital first step in onward use of the data.
Following the PCA, we created component scores for all subjects to indicate severity levels on each of the six symptom dimensions. We then investigated the distributions of these scores as a function of clinical diagnosis. There were distinct distributions, by diagnosis, for five out of the six components, demonstrated by different median scores and proportions of ‘high-scorers’. Scores on the anxiety dimension did not differ in these respects, indicating that doctors were rating all in-patients as having similar levels of anxiety. Different distributions of symptoms between diagnoses would be expected and support the construct validity of measuring symptom severity in this way. It is notable though, from inspection of the median and inter-quartile range figures, that there was substantial symptom heterogeneity within diagnoses. This variability, in its most extreme form meant that, for example, there were patients with a diagnosis of F10 ‘Mental and behavioural disorders due to use of alcohol’ in the upper and lower 5% of scores on four out of the six dimensions (positive symptoms, mania, depression and anxiety).
We then investigated the predictive power of component scores by following an existing literature whose aim has been to establish the superiority of dimensional, categorical or combinatorial representations of psychopathology. There were five clinical outcome measures where dimensional representations of illness alone provided the best model, whereas there was only one measure where a categorical representation alone was best. There were no measures where a combined approach provided the best solution. The superiority of dimensional over categorical representations of psychopathology, as demonstrated here, is in agreement with other studies which have asked this question using the original OPCRIT 
; although one study concluded that combinatorial approaches were best 
. It is important to note however, in relation to the above observations, that we were using ICD diagnoses collapsed to the 2-digit level (due to variation in the way clinical diagnoses were documented). It may be, that at the three digit level or higher (e.g. F10.52), categorical representations of psychopathology would exhibit greater predictive power as well as less symptom heterogeneity.
Despite their overall superiority to diagnosis in this analysis, the predictive value of the component scores, for this set of clinical outcome variables, was only modest (indicated by low R2 values and eight measures having no association with the ‘symptoms only’ model). It is therefore important that the utility of this approach in other research realms (e.g. biomarker research) is explored further, particularly as one intended use of the data will be to characterize associated biological and neuroimaging information being gathered in a Bioresource (Biobank) operated by the trust and its partners. It may be that categorical or combinatorial representations of psychopathology are more appropriate for other research areas. Crucially though, via the adoption of OPCRIT+ by SLaM, researchers will now have access to both symptom and diagnosis data recorded in the clinic.
In summary, our analysis has demonstrated that using OPCRIT+, symptom data being routinely recorded across a broad diagnostic spectrum within inpatient settings can be reused to represent severity levels on psychopathological dimensions. This has been achieved despite the very different methodological circumstances between our study and the previous use of OPCRIT for this purpose. Symptom dimensions are applicable across a variety of research and clinical applications and have the potential to add significant explanatory power to many types of analyses.