Automated coding of consultation data in Sentinel general practice
back to Durham meeting retour atelier Nantes 99
Automated coding of consultation data
in Sentinel general practice
L Letrilliart, C Vivoud, M Guiguet, A Flahault. INSERM U444, Paris.
Background- Beyond the perspective of mandatory coding of procedures and pathologies
to be imposed in the future (such as the France perspective), classification and coding of
medical data from consultation are to be used more and more for epidemiological
research in primary care, especially within practice-based networks. Because of the poor
acceptability of manual coding by the practitioners in some countries, automated entry of
free-text data should be considered as a possible alternative.
Methods- We developed a short computer program, written in SQL, aimed at encoding
medical data according to ICPC-1 in its French translation, via a self-learning procedure
derived from previous manually coded data. Its performance and reliability were
assessed using data on referrals to hospital in general practice.
Results- Using a database with 2300 reasons for referral to hospital to encode 800 new
reasons for referral, performance in textual matching was estimated to 85% and
agreement between codes to 80%. Epidemiological results derived from automatically
coded data are extrapolated in routine to the whole sample of data in the Sentinel
network in France.
Conclusion- Automated coding of consultation data in general practice seems to be a
reliable alternative to manual coding. Because it is a simple system, it may be also
implemented at the office level through integration into the medical software.
