Reliability of report coding of hospital referrals in primary care versus practice-based coding.

Coding of medical data according to a suitable classification is useful to epidemiological research in primary care but its implementation at practice-based level may be considered as extra work by participating practitioners. Secondary coding from reported data can be considered as a possible alternative to practice-based coding. The purpose of this study was to assess the inter-rater reliability of report coding versus practice-based coding of morbidity data. Via teleinformatics, 300 French general practitioners from the French Sentinel epidemiological network transmitted in free text, on a continuous real-time basis, the health problems generating each hospital referral they made since August 1997. All these reports were centrally coded according to the International Classification of Primary Care (ICPC). A subsample of 120 reports were coded in local practices for comparison. Codes resulting from blind centralized free-text coding were compared with practice-based codes. For the 120 referrals reported, the K measure of agreement for the number of codes was 0.65 (95% confidence interval [CI], CI: 0.52-0.77), and for the chapters selected, 0.84 (95% CI: 0.78-0.91). Discrepancies attributable to the centralized coding only occurred for 7.5% of the referrals, and were due to the lack of specificity of the information transmitted as free text. A thesaurus of correspondences between problem(s) generating referrals and ICPC codes was built from 5000 referrals, and has been used routinely for the automated report coding of an additional sample of 1691 referrals. We conclude that centralized coding is a reliable alternative to practice-based coding in primary care, provided that physicians give sufficiently specific information.