Box 1.

Covariates used to predict high impact user status in descending order of importance in explaining the variation

Covariate	Comments
Number of emergencies in 365 days before index spell	Modelled as a continuous variable
Number of emergencies between 366 days and 36 months before index spell	Modelled as a continuous variable
Charlson index of comorbidity	Was devised originally to predict death. Capped at an upper limit of 6 (few patients had scores above this)
Age	Five-year age bands up to 90+
Ambulatory care sensitive condition	Conditions considered most amenable to case management. 19 groups as per Table 1, plus a `group 0' for all other conditions
Ethnicity	Six groups: white, black, Indian sub-continent, Chinese, `unknown' and `other'
Standardized admission ratio (SAR) (log-transformed)	SAR adjusted for age and sex for all emergencies for the ward of residence for 2000/2001 to 2002/2003 combined
Area-level lifestyle group	Postcodes were allocated to one of 61 groups based on various lifestyle factors using the MOSAIC classification
Source of admission	Home, nursing home, other hospital, etc.
Area-level deprivation	2004 Index of Multiple Deprivation for the patient's area of residence, divided into fifths
Number of consultant episodes in index spell	Modelled as a continuous variable
Sex	Male or female