Improved complexity stratification in congenital heart disease; the impact of including procedural data on accuracy and reliability.

Background: In order to manage a class of diseases as broad as congenital heart disease (CHD), multiple "manually generated" classification systems defining CHDs as mild, moderate and severe have been developed and used to good effect. As databases have grown, however, such "manual" complexity scoring has become infeasible. Though past attempts have been made to determine CHD complexity algorithmically using a list of diagnoses alone, missing data and lack of procedural information have been significant limitations. Methods: We built an algorithm that can stratify the complexity of patients with CHD by integrating their diagnoses with a list of their previous procedures. Specific procedures which address a missing diagnosis or imply a certain operative status were used to supplement the diagnosis list. To verify this algorithm, CHD specialists manually checked the classification of 100 children and 100 adults across four hospitals in Australia. Results: Our algorithm was 99.5% accurate in the manually checked cohort (100% in children and 99% in adults) and was able to automatically classify more than 90% of a cohort of over 24,000 CHD patients, including 92.5% of children (vs 84.4% without procedures, p < 0.0001) and 91.1% of adults (vs 70.4% without procedures; p < 0.0001). Conclusions: CHD complexity scoring is significantly improved by access to procedural history and can be automatically calculated with high accuracy.