Stressor, vulnerability and depression: a question of replication.

One of the most fertile areas in social psychiatry in the last two decades has been the study of the aetiology of depressive illness. The number of reports appearing requires periodic stocktaking to keep abreast of developments, but the criteria by which findings should be evaluated are not always self-evident and merit some consideration. This theme is particularly topical in view of a recent editorial in Psychological Medicine which, in discussing one theoretical model of depressive disorder in women, concludes decisively: 'In summary Brown & Harris' vulnerability model has not been replicated' (Tennant, 1985). In commenting on and disagreeing with this conclusion, we hope to illuminate a number of more general issues surrounding replication studies and the criteria for their interpretation. There are various approaches to the study of life events, vulnerability and psychiatric disorder. Some 90% of all studies have employed questionnaire-based measures that have proved not only to be burdened with problems of low reliability and poor validity, but also to have a predilection for reflecting general rather than specific phenomena (e.g. cumulative life-event scores for a period of time rather than the characteristics such as threatfulness of particular life events). The problems arising from the use of such measures have been compounded rather than eased by the use of more complex research designs. For example, longitudinal studies have typically failed to utilize life events occurring in the follow-up period, presumably because of the problem with such measures of establishing time order in relation to any onset (e.g. Henderson et al. 1981). Since events of aetiological importance tend to occur within a few weeks of onset, many relevant stressors are bound to be missed. An alternative tradition relying on intensive interviews and rating scales completed by investigators has, in contrast, developed reliable and apparently valid measures and established the presence of large associations between certain types of stressor and the onset of depression. Results on the whole have been so clear that the main focus of recent interest has been the possibility that certain ongoing vulnerability factors increase risk, but only in the presence of a stressor. It is important to distinguish here the replicability of the finding that a high proportion of depressive onsets are preceded by stressors of these types from the replicability, at a different level, of a vulnerability model. If conceived broadly, this model implies some form of synergy between stressor and vulnerability factor in which the risk of onset when a stressor is experienced without a vulnerability factor is considerably higher than when a vulnerability factor is experienced without a stressor, although this may exceed the risk when neither factor is present. In narrower terms, the vulnerability model implies that when a vulnerability factor is present without a stressor, the risk is no higher than in the absence of both. The Brown-Harris model has typically been defined in the narrower of these traditions, although the authors have stated that they do not see any difference of any particular theoretical significance between the two (Brown & Harris, 1978 ft), especially as measurement error will serve to reduce the chances of confirming the narrower version (Brown & Harris, 1986). It is therefore worth stating at the outset in discussing replication that too absolute a distinction between the broader and narrower versions at this stage of research may lose the wood for the trees. The message which both versions seek to convey is that the onset of depression is basically in response to a stressor, specifically characterized as a provoking agent (for definition, see Brown & Harris, 1978 a), and that vulnerable subjects are considerably more likely to respond in this way. Either version has implications for the