Fragment-Based Prediction of the Clinical Occurrence of Long QT Syndrome and Torsade de Pointes

We present a study directly linking clinically observed adverse events to molecular structure. The method is applied to predict the long QT syndrome (LQS) and the resulting condition torsade de pointes (TdP) that can lead to sudden death. The predictive models are created by correlating biochemically significant chemical substructures, derived from a database of marketed drugs, to reports of adverse events in the FDA adverse event reporting system, which contains all events reported to the FDA since 1997. We compute the reporting ratio for each drug/event combination and perform a chi(2) test to determine whether there is a statistically significant association of each drug to reports of LQS and TdP. Linear models are then used to identify chemical substructures that are most consistently associated with the adverse event. The results for LQS and TdP are compared to models for LQS based on human ether-a-go-go-related gene binding and tested for statistical significance by comparing to models created with a randomized dependent variable. The ability to identify compounds associated with LQS and TdP is approximately five times improved in comparison to models based on randomized data, suggesting that there is a significant relationship between specific chemical structures and these adverse events.