Background. The number of registries is increasing, but few of them perform reliability audits by comparing the data contained in the database with data contained in hospital charts.
Methods. The European Liver Transplant Registry (ELTR) cocoordinating committee appointed an independent team to check the reliability of data contained in ELTR. Centers were selected at random. Ten percent of each center's files were selected at random, and 25 items per file were checked during the site visits. The rates of completeness and inconsistencies and the agreement between ELTR and charts were established. We also assessed the correlation between the quality of data and the visited centers' activity.
Results. Seven hundred thirty-four files from 21 centers have been audited between June 1998 and June 2001. The rate of ELTR completeness was 95%, and the rate of consistency between charts and ELTR was 98%. The agreement between the ELTR and charts review was very good for all conditions (kappa value ≤0.81). However, comparisons of rates between items indicated that specific items, mostly cause of death or graft failure and patient outcome, should be targeted for improvement. No significant correlation was found between the quality of data and the experience of visited centers. The mean (min-max) and median cost per audited file were EUR 60 (8-150) and EUR 44, respectively.
Conclusion. The results of audit visits indicate that ELTR data are reliable, and the scientific results of ELTR can be considered credible and representative of liver transplantation in Europe. The method could serve as a model for auditing a registry.
Epidemiologic findings are often no better than the quality of analyzed data (1). The completeness and reliability of cumulated data affect the validity of risk-adjustment models and the accuracy of comparisons between patient cohorts (2,3). High reliability is particularly important for independent variables in statistical analysis. Low reliability in such determinants can bias the data analysis in unpredictable ways, causing both underestimation and overestimation of effects (4). In recent years, dramatic improvements have been achieved in all fields of data management, including a wider choice of database software, the use of secure internet web sites for rapid data exchange, education of users regarding data capture and security techniques, and the development of logical data control programs (5,6). Despite these advances, errors are still possible, and reliability audits by comparing the data contained in the database with data contained in hospital charts (considered to be the gold standard for accurate information) have become essential.