Data from outcomes registry databases are being increasingly used for peer review and public reporting. However, administrative and clinical databases are mostly unaudited; thus, their accuracy has not been verified.Methods
Outcomes data from all coronary artery bypass operations from a single cardiac surgery practice were entered into The Society of Thoracic Surgeons (STS) National Cardiac Database. From our practice of 18 surgeons, we audited 247 (10%) of the clinical records of patients undergoing surgery in 2001 and correlated them with all 315 elements of the STS National Cardiac Database for verification of accuracy. Inaccuracies were defined as a disagreement with a nominal or categorical variable or, for continuous variables, as the value not being within a predetermined window. When discrepancies existed, the hospital clinical record was assumed to be accurate. Outcomes discrepancies were then analyzed by four major categories: components of the preoperative risk algorithm, operative mortality, major complications, and other outcomes.Results
Discrepancies were noted in 5% (16) or fewer of the audited fields for 98.8% of the records. Of the 32 variables in the mortality risk algorithms, discrepancies were present in fewer than 10% of the audits on 30 of the 32 variables. More than 95% of the audited charts had zero or one discrepancy in the seven most important variables in the mortality risk models. Operative mortality was determined to be completely accurate with no discrepancies between the database and the audited clinical record. Among major complications, the error rate was less than 1% for all complications except prolonged ventilation (4.0%). A higher rate of discrepancies did exist in some of the other variables, including discharge medications (14.1%) and ventilator time (36.4%).Conclusions
A detailed audit of a clinical outcomes registry database demonstrated that the major fields within this specific database including operative mortality, major complications, and the significant factors in the risk algorithm were highly accurate. Process improvement factors were identified to further increase the accuracy of data collection.