Using routine clinical and administrative data to produce a dataset of attendances at Emergency Departments following self-harm

C. Polling, A. Tulloch, S. Banerjee, S. Cross, R. Dutta, D.M. Wood, P.I. Dargan, M. Hotopf

Self-harm is a significant public health concern in the UK. This is reflected in the recent addition to the English Public Health Outcomes Framework of rates of attendance at Emergency Departments (EDs) following self-harm. However there is currently no source of data to measure this outcome. Routinely available data for inpatient admissions following self-harm miss the majority of cases presenting to services. We aimed to investigate (i) if a dataset of ED presentations could be produced using a combination of routinely collected clinical and administrative data and (ii) to validate this dataset against another one produced using methods similar to those used in previous studies. Using the Clinical Record Interactive Search system, the electronic health records (EHRs) used in four EDs were linked to Hospital Episode Statistics to create a dataset of attendances following self-harm. This dataset was compared with an audit dataset of ED attendances created by manual searching of ED records. The proportion of total cases detected by each dataset was compared. There were 1932 attendances detected by the EHR dataset and 1906 by the audit. The EHR and audit datasets detected 77 % and 76 % of all attendances respectively and both detected 82 % of individual patients. There were no differences in terms of age, sex, ethnicity or marital status between those detected and those missed using the EHR method. Both datasets revealed more than double the number of self-harm incidents than could be identified from inpatient admission records. It was possible to use routinely collected EHR data to create a dataset of attendances at EDs following self-harm. The dataset detected the same proportion of attendances and individuals as the audit dataset, proved more comprehensive than the use of inpatient admission records, and did not show a systematic bias in those cases it missed.