Prediction of readmission to acute psychiatric units

Many factors are known to influence readmission to psychiatric wards, and readmission rates have been suggested as proxy outcome indicators of quality. Korner returns were used to ascertain readmission rates for all psychiatric admissions to acute wards in North Staffordshire, 1987-1993. Predictor variables were derived from Korner returns or obtained from the 1991 Census data. Survival analysis techniques were used to examine which variables predicted readmission. A predictive model was derived using Cox regression, which followed the observed data at greater than chance probability (chi2=48.5, df=4, P < 0.001). A psychotic diagnosis was the most influential predictor of readmission. Length of stay is not predictive in the Cox regression model, which suggests patients are not being prematurely discharged. The derived models may have value in service planning, audit and resource allocation.