Accuracy of Administrative Data for Detection and Categorization of Adult Congenital Heart Disease Patients from an Electronic Medical Record

Diagnostic codes used in healthcare administration have been employed extensively in clinical research to identify target patient populations, including demonstration of important clinical outcomes among adults with congenital heart disease. However, little is known about the reliability of code-derived data in this context. We sought to determine the accuracy of International Classification of Disease-9th Revision (ICD-9) diagnoses and the reliability of retrieval algorithms in adults with congenital heart disease (ACHD). Pilot testing of a hierarchical algorithm to identify ACHD patients and determine their principle congenital diagnosis was performed. A revised algorithm was then applied retrospectively to a sample of all outpatients seen by providers who see general cardiology and ACHD patients. Using all ICD-9 codes available from any encounter, accuracy for detection and categorization of sub-types were compared to physician chart review. After initial testing on 334 patients, the revised algorithm was applied to 740 patients. The sensitivity and specificity for ACHD patient identification from this specialty clinic population were 99 and 88 %, respectively. Of 411 (56 %) non-ACHD patients, 49 were incorrectly categorized as ACHD by the algorithm. Of ACHD patients, 326 of 329 were correctly identified by diagnostic codes and categorization of ACHD defect sub-type was correct in 263 (80 %). Administrative data can be used for identification of ACHD patients based on ICD-9 codes with excellent sensitivity and reasonable specificity. Accurate categorization that would be utilized for quality indicators by ACHD defect type is less robust. Additional testing should be done using non-referral populations.