Inter-rater reliability of a refined index of melancholia: the CORE system.

We report an inter-rater reliability study of the modified 18-item, sign-based CORE index of melancholia, undertaken on 205 ratings of 35 patients by five clinical research psychiatrists. Inter-rater agreement about the presence or absence of individual items was slight to moderate when examined by the kappa coefficient, but moderate to high when examined by the intraclass correlation statistic. For total CORE scores, perhaps the most important application of the index of melancholia, high levels of intraclass correlation coefficients (ranging 0.79 to 0.90 across the varying rater dyads) were established. When a single cut-off score (of 7/8) was used to allocate patients to either a 'melancholic' or 'non-melancholic' class, only moderate agreement was established between raters in such 'class' assignments--a limitation which can be redressed by imposing a 'probable/possible melancholia' band of scores.