A Naïve Bayes classifier for differential diagnosis of Long QT Syndrome in children

This study examined disease models most indicative of risk of Long QT Syndrome (LQTS) in children. Data mined from electronic health records of children confirmed with (n=248) and without (n=101) a diagnosis of LQTS were used to develop a patient profile for LQTS. The profile consisted of 44 distinct features, 17 of which were enriched in LQTS patients. Notably, 66.9% of subjects with a diagnosis of LQTS fell into a category of “low” (22.6%) or “intermediate” (44.3%) risk using a current LQTS risk assessment standard. We developed and trained a machine learning process for LQTS classification by applying a Naïve Bayes model to our LQTS cohort. The model classified patients with a sensitivity of 91.1% and a specificity of 73.3%. These results suggest that data mining of clinical data in conjunction with a Bayesian modeling approach can lead to a diagnostic system for prediction of LQTS in children.