OBJECTIVE: Rates of hospital presentation for self-harm in England were compared using different national and local data sources.
DESIGN: The study was descriptive and compared bespoke data collection methods for recording self-harm presentations to hospital with routinely collected hospital data.
SETTING: Local area data on self-harm from the 3 centres of the Multicentre Study of Self-harm in England (Oxford, Manchester and Derby) were used along with national and local routinely collected data on self-harm admissions and emergency department attendances from Hospital Episode Statistics (HES).
PRIMARY OUTCOME: Rate ratios were calculated to compare rates of self-harm generated using different data sources nationally and locally (between 2010 and 2012) and rates of hospital presentations for self-harm were plotted over time (between 2003 and 2012), based on different data sources.
RESULTS: The total number of self-harm episodes between 2010 and 2012 was 13 547 based on Multicentre Study data, 9600 based on HES emergency department data and 8096 based on HES admission data. Nationally, routine HES data underestimated overall rates of self-harm by approximately 60% compared with rates based on Multicentre Study data (rate ratio for HES emergency department data, 0.41 (95% CI 0.35 to 0.49); rate ratio for HES admission data, 0.42 (95% CI 0.36 to 0.49)). Direct local area comparisons confirmed an overall underascertainment in the HES data, although the difference varied between centres. There was a general increase in self-harm over time according to HES data which contrasted with a fall and then a rise in the Multicentre Study data.
CONCLUSIONS: There was a consistent underestimation of presentations for self-harm recorded by HES emergency department data, and fluctuations in year-on-year figures. HES admission data appeared more reliable but missed non-admitted episodes. Routinely collected data may miss important trends in self-harm and cannot be used in isolation as the basis for a robust national indicator of self-harm.