Background: The Arthroplasty Clinical Outcomes Registry NSW (ACORN) was initiated in 2012. ACORN is a registry piloting within NSW, Australia with several participating hospitals; it aims to monitor patient-centred outcomes and post-surgical complications after total hip and knee arthroplasty. Using retrospective audit methodology, we aimed to investigate the completeness and accuracy of data in ACORN.
Methods: We undertook a reabstracting audit of 100 clinical records of patients who underwent surgery in 2012/2013 (50 each from hospitals A and B). These records represented 27% (100/367) of patient entries in the ACORN registry, all of which were collected at either hospital A or hospital B. Firstly, data completeness was determined by identifying the proportion of missing data in the original data pro forma. Secondly, accuracy of the initial data extraction was determined by comparing these data to reabstracted data collated by an auditor blind to the outcomes of the initial extraction. Inaccuracies were ascertained to be a disagreement between categorical variables and for continuous data, a pre-determined window of error was established. Benchmarks for data completeness and accuracy were set at 95.0%; kappa and intraclass coefficient (ICC) calculations were also utilised to supplement this analysis. In addition, registry completeness (the percentage capture of eligible patients) was also determined as part of the data quality analysis.
Results: Completeness and accuracy of submitted datasets were evaluated to be 99.0% (1259/1272) and 94.0% (2159/2296) respectively for Hospital A, and 99.3% (1589/1600) and 96.1% (2444/2542) for Hospital B. The majority of accuracy discrepancies pertained to medical history data. For Hospital A, 57.1% (28/49) of variables met the accuracy benchmark of 95%; 74.5% (38/51) of variables in Hospital B met this benchmark. Of the number of patients eligible for inclusion in the registry, 93.5% (660/706) were found to be included.
Conclusion: Levels of data completeness and accuracy were found to be high in the submitted datasets for both hospitals. However, important deficits were identified in the accuracy of patient comorbidities. More specific and clear data definitions, and a more thorough examination of medical records would be possible methods to improve the accuracy of deficient areas.