Readmissions - can they be predicted on admission?

Abstract

This paper looks at the development of logistic regression models to predict readmissions for medical patients on their initial admission to hospital. The design of our study was a retrospective analysis of a large dataset drawn from a range of secondary sources - medical, nursing, therapy and social care records. Three northern hospitals and related community health districts and social care organizations in the UK participated. Records of 1,192 patients discharged from medical wards during the period April 1992-March 1995 were analysed. Readmission within six weeks of discharge was the main outcome measure. Four logistic regression equations were produced. Three individual site equations were calculated and classification levels for readmission of 17-22 per cent were achieved. Component factors that differed in importance were age, GP contact, social services contact, marital status and living status. The weakest equation was the equation that encompassed patients from all three sites, which classified 7 per cent of readmissions. It is possible to develop equations that will explain readmission for a fifth of medical patients on admission to individual hospitals. Further exploratory work needs to be undertaken to explore reasons for differences between districts and develop more generalizable predictive equations.