OBJECTIVES: To compare the proportions of emergency department (ED) attendances following self-harm that result in admission between hospitals, examine whether differences are explained by severity of harm and examine the impact on spatial variation in self-harm rates of using ED attendance data versus admissions data. SETTING: A dataset of ED attendances and admissions with self-harm to four hospitals in South East London, 2009-2016 was created using linked electronic patient record data and administrative Hospital Episode Statistics.
DESIGN: Proportions admitted following ED attendance and length of stay were compared. Variation and spatial patterning of age and sex standardised, spatially smoothed, self-harm rates by small area using attendance and admission data were compared and the association with distance travelled to hospital tested.
RESULTS: There were 20 750 ED attendances with self-harm, 7614 (37%) resulted in admission. Proportion admitted varied substantially between hospitals with a risk ratio of 2.45 (95% CI 2.30 to 2.61) comparing most and least likely to admit. This was not altered by adjustment for patient demographics, deprivation and type of self-harm. Hospitals which admitted more had a higher proportion of admissions lasting less than 24 hours (54% of all admissions at highest admitting hospital vs 35% at lowest). A previously demonstrated pattern of lower rates of self-harm admission closer to the city centre was reduced when ED attendance rates were used to represent self-harm. This was not altered when distance travelled to hospital was adjusted for.
CONCLUSIONS: Hospitals vary substantially in likelihood of admission after ED presentation with self-harm and this is likely due to the differences in hospital practices rather than in the patient population or severity of self-harm seen. Public health policy that directs resources based on self-harm admissions data could exacerbate existing health inequalities in inner-city areas where these data may underestimate rates relative to other areas.