The aim of the study was to identify the best predictors of the course of burden during a 2-year follow-up period. The study is part of the Munich 5-year follow-up study in relatives of first admitted patients with schizophrenia or depression.

A cohort of 60 key relatives was assessed based on a transactional stress model concerning objective and subjective burden, well-being, self-rated symptoms and global satisfaction with life. The stressors were defined as patients' illness variables, and the potential predictors included different dispositions and resources of the relatives. Effects were analyzed by regression models. In a first step, the main predictors of burden were identified at each assessment. In a second step, the resulting predictors were included in a Generalized Linear Modeling procedure.

Caregivers' burden improved significantly, but well-being and self-rated symptoms remained elevated. In the final regression model, expressed emotion, neuroticism, generalized negative stress response and life stressors resulted as the best predictors of burden. The effects were rather time invariant than time dependent.

In order to effectively work on long-standing unfavourable patterns of stress response, family interventions should be long-term and targeted to vulnerable caregivers who could be identified by virtue of their personality traits.