International trends have seen a policy shift from chronic psychiatric institutions and long hospital admissions towards acute, short hospital stays and community-based care. Following this process a dramatic increase in relapse rates has been noted among a particular subgroup of psychiatric patients. The general decline in number of hospital beds and mental hospital population size seems to have directly paralleled the increasing rate of readmissions for certain patients. Design. A retrospective cohort of 180 admissions was followed up for 12 months, after an index discharge, by means of multiple hospital and community-based record reviews.Each readmission was analysed as an event using a survival analysis model. Setting. Chris Hani Baragwanath Hospital, Gauteng.Subjects. A random sample of patients admitted during a 6-month period in 1996.Outcome measures. Time to readmission.Results. Two hundred and eighty-four admissions were analysed. The only factor that provided a significant protective effect was being married or cohabiting (P = 0.015). Clinic attendance showed a slight protective effect early on but conferred a significantly higher risk of readmission on those who had been out of hospital for a long period (P = 0.001). Only 21% of discharged patients ever attended a clinic. The overall risk of readmission was significantly higher in the first 90 days post discharge. Conclusions.The lack of impact of length of hospital stay and use of depot neuroleptics on time to readmission may indicate that patients are being kept for appropriate duration and that the most ill patients are receiving depot medication.Several sampling and statistical artefacts may explain some of our findings. These results confirm the worldwide difficulty in finding consistent and accurate predictors of readmission. Low rates of successful referral to community aftercare need to be addressed before their effectiveness can be reasonably assessed. The inherent instability of the post-discharge period is a potential area for further investigation and intensive management. The influence of length of hospital stay on relapse rates remains a disputed area. S AfrWith regard to aftercare variables, poor treatment compliance was found to predict relapse, 9,13,17 but one study 9 found that use of outpatient services, number of clinic visits and access to carewere not significant factors.Quality of life indicators seem poorly predictive of the 'revolving door' syndrome. 9 In this category family criticism of the patient 9 and unsatisfactory family relations 11,16 are the only variables that have been associated with high relapse rates.Finally, it has been suggested that the inconsistent and contradictory findings in much of the research may be accounted for by variables that are significant only when they interact with other variables, for example employment and age with living status. 10The above studies 1 17 were done in developed countries and a review of the literature in Medline over the last 10 years reveals a paucity of research on risk fa...