Two probabilistic methods to characterize and link drug related ECG changes to diagnoses from the PTB database: Results with Moxifloxacin

The QT interval is the lead biomarker for predicting risk of severe ventricular pro-arrhythmia in drug safety assessment. QT is a weak predictor, with no direct link to clinical endpoints.We propose two methods that might bridge the gap. Method A is based on the study level probabilistic intra-subject clustering of 12-lead based ECG waveform similarities, comparing pre-treatment with placebo and treatment within and between study periods. Method B: population related approach where all 12-lead ECG waveforms from a study are compared with the fully diagnosed 30.000 ECGs in the PTB database. ECGs are assigned a probability for normality and a limited number of cardiac main diagnoses contained therein. In a crossover study with moxifloxacin 200 and 400 mg iv, QT-prolongation and method A revealed significant treatment effects, while method B showed no relevant treatment related change to non-normal diagnosis.