Modelling late-life depression

To determine if we could find support for a three-class depression sub-typing model (and identify differentiating constituent clinical features) in a sample of elderly depressed patients.

Depression is currently modelled dimensionally, with little concession to descriptive psychopathology and distinction of meaningful clinical depressive sub-types. We have proposed a three-class hierarchical specificity model for sub-typing the depressive disorders (comprising psychotic, melancholic and non-melancholic depression), with specificity referring to two clinical features (psychotic symptoms and psychomotor disturbance or PMD) separating the first two classes from a residual non-melancholic class.

Subjects were aged 65 years or more, non-demented and being treated for depression. Extensive clinical assessment was undertaken, while several standardised measures were administered. 'Bottom up' analyses were data driven, while 'top down' analyses respected DSM-III-R decision rules. Dimensional and categorical multivariate analyses sought to identify features differentiating psychotic depression (PD), melancholic depression (MEL) and a residual non-melancholic (NON-MEL) class.

Of the 123 referred patients (having a mean age of 75.6 years), 46 had DSM-defined PD, 46 had MEL and 31 were assigned as NON-MEL. Mean total CORE scores (measuring PMD) more clearly distinguished the groups than scores on two depression severity measures. Psychotic depression was best distinguished from melancholic depression by psychotic features, as well as more severe PMD and anhedonia. Melancholic depression was best distinguished from non-melancholic depression by PMD, terminal insomnia and pathological guilt.

The specificity of PMD to the definition of the psychotic and melancholic depression was confirmed in our elderly depressed sample. Clinical features identified as distinguishing psychotic, melancholic and non-melancholic depression were broadly consistent with findings from our previous studies involving younger subjects and with our three-class hierarchical model.