Algorithmic complexity stratification for congenital heart disease patients

Background: Congenital Heart Disease (CHD) encompasses a huge variety of rare diagnoses that range in complexity and comorbidity. To help build clinical guidelines, plan health services and conduct statistically powerful research on such a disparate set of diseases there have been various attempts to group pathologies into mild, moderate, or severe disease. So far, however, these complexity scores have required manual specialist input for every case, and are therefore missing in large databases where this is impractical, or quickly outdated when guidelines are revised. Methods: We used the up-to-date European Society of Cardiology guidelines to create an algorithm to assign complexity scores to CHD patients using only their diagnosis list. Two CHD specialists then independently assigned complexity scores to a random sample of patients. Results: Our algorithm was 96% accurate where both specialists agreed on a complexity score; this occurred 68% of the time overall, and 79% of the time in moderate or complex CHD. The algorithm “failed” mainly when diagnoses were insufficiently specific, usually for septal defects (where size was unspecified), or where complexity depends on the procedure performed (e.g. atrial/arterial switch for transposition of the great arteries). Conclusions: We were able to algorithmically determine the complexity scores of a majority of patients with CHD based on their diagnosis list alone. This could allow for automatic complexity scoring of most patients in large CHD databases, for example our own Registry of the Congenital Heart Alliance of Australia and New Zealand. This will facilitate targeted research into the management, outcomes and burden of CHD.