USE OF ELECTRONIC MEDICAL RECORDS OF THE EMERGENCY DEPARTMENT FOR AN AUTOMATED EPIDEMIOLOGICAL SURVEILLANCE OF ATTEMPTED SUICIDE : PILOT STUDY IN A FRENCH UNIVERSITY HOSPITAL
The aim of our study was to assess whether an extraction and automated processing of the computerized emergency record would improve the estimate of the annual rate of emergency department visits for attempted suicide, compared with the annual rate produced in the framework of a national surveillance currently carried out by manual coding of emergency physicians and underestimating it.
METHODS:
A feasibility study was conducted in the emergency department of the Lyon University hospital (France) on the population of patients admitted in 2011 and 2012 in the ward. After automatic extraction and data preprocessing, including automatic normalization of textual medical data through Unique Concept Identifiers of the Unified Medical Language System, predictive association rules to classify the reason of the visit into « suicide attempts » vs. « other reasons » were developed. The performance of these rules was evaluated by comparison with a gold standard (reading the medical documents by medical practitioners).
RESULTS:
In a test sample of 339 patients admitted to the emergency in 2012 (99 admitted for attempted suicide, 14 with suicidal ideation and 226 without any of these non fatal suicidal behavior), the sensitivity for automatic detection varied from 94.9% [95% CI : 88.6%-98.3%] to 95.9% [95% CI : 90%-98.9%] and the specificity between 96.5% [95% CI : 93.1%-98.5%] and 97.8% [95% CI : 94.9%- 99.3%].
CONCLUSIONS:
This study demonstrates the usefulness of developing semantic extraction and data mining methods to improve the quality of epidemiological indicators produced as part of national surveillance of suicide attempts.