A reliable method to retrieve accident & emergency data stored on a free-text basis.

Accident & Emergency (A & E) data on asthma-related attendances are useful for studies on the effectiveness of asthma interventions, and to determine the relationship of environmental factors to asthma and asthma epidemics. The final diagnoses made in the A & E departments are not usually coded when entered into hospital databases in the U.K., although the "presenting complaint' can be retrieved from the computerized Hospital Information & Support Systems (HISS), from a free-text attendance diagnosis field entered by the reception clerk when the patient arrives at the A & E department. The validity of this as an indication of the final diagnosis is unevaluated. The aim of this study was to measure the validity of the string "asth' in the A & E attendance diagnosis field for identifying patients attending the A & E departments of two hospitals for asthma-related conditions. A list of patients who attended the A & E department of two hospitals was retrieved from the HISS along with the attendance diagnosis field. If the attendance diagnosis field contained the text string "asth', mentioned wheeze or breathing problems, or the patients were referred by their GP without any diagnostic information entered on HISS, the records were selected for evaluation. The remaining attendances, which were not evaluated further, were attributed to another diagnosis based on the evidence of the recorded attendance diagnosis. The results indicated that the string "asth' in the attendance diagnosis field had a sensitivity of 80.3% (95% CI 75.1-85.5%) and a specificity of 96.7% (95% CI 95.6-97.8%) for a final diagnosis of asthma. It is concluded that free-text attendance diagnosis fields in hospital databases can be searched with suitable strings to obtain reliable data on attendance with asthma. As part of another investigation, the present authors attempted to retrieve a list of the attendances with asthma at the same two A & E departments at a time that was reportedly associated with an epidemic of asthma following a thunderstorm. On this occasion, the string "asth' proved to be significantly less sensitive. The possible reasons for this and the implications for using this method for identifying cases are discussed.