Examining the relationship between length of stay and readmission rates for selected diagnoses in Scottish hospitals

Before a meaningful evaluation of the performance of a hospital over any output or outcome can be made, it is essential that the case mix of both that hospital and of its comparison group are standardized. This must extend beyond age and sex to include all pertinent available information relating to the individual which is available in routine data sets. One of the restrictions of analyses made upon such data sets tends to be the paucity of the social and demographic data collected. Census data may be used to provide some detail concerning the localities from which the individual patients are drawn by means of the postcode, and in this manner it may be seen how area characteristics are related to standardized outcomes. Similarly, variables relating to hospital practice or structure may affect the outcomes of individual patients. This forms the basis of a hierarchy, with individuals living within localities being discharged from hospitals; subsequent modelling of the data should take this structure into account. This process is illustrated using discharges from general surgery in Scotland. The data are subdivided into eight homogeneous groups and each is analysed separately using the multilevel modelling software ML3 for both length of stay and readmissions. Using crude figures, there is a slight positive correlation between the two measures at the hospital level. However, the use of hospital residuals from a length-of-stay model as explanatory variables in a model for readmission rates brings to light the negative association between the length of stay and readmissions, and suggests that pressures to reduce hospital lengths of stay may not be without consequences in terms of other outcomes.