Automated Algorithm for J-Tpeak and Tpeak-Tend Assessment of Drug-Induced Proarrhythmia Risk

Abstract

Background: Prolongation of the heart rate corrected QT (QTc) interval is a sensitive marker of torsade de pointes risk; however it is not specific as QTc prolonging drugs that block inward currents are often not associated with torsade. Recent work demonstrated that separate analysis of the heart rate corrected J-Tpeakc (J-Tpeakc) and Tpeak-Tend intervals can identify QTc prolonging drugs with inward current block and is being proposed as a part of a new cardiac safety paradigm for new drugs (the "CiPA" initiative).

Methods: In this work, we describe an automated measurement methodology for assessment of the J-Tpeakc and Tpeak-Tend intervals using the vector magnitude lead. The automated measurement methodology was developed using data from one clinical trial and was evaluated using independent data from a second clinical trial.

Results: Comparison between the automated and the prior semi-automated measurements shows that the automated algorithm reproduces the semi-automated measurements with a mean difference of single-deltas <1 ms and no difference in intra-time point variability (p for all > 0.39). In addition, the time-profile of the baseline and placebo-adjusted changes are within 1 ms for 63% of the time-points (86% within 2 ms). Importantly, the automated results lead to the same conclusions about the electrophysiological mechanisms of the studied drugs.

Conclusions: We have developed an automated algorithm for assessment of J-Tpeakc and Tpeak-Tend intervals that can be applied in clinical drug trials. Under the CiPA initiative this ECG assessment would determine if there are unexpected ion channel effects in humans compared to preclinical studies. The algorithm is being released as open-source software.

Trial registration: NCT02308748 and NCT01873950.

Conflict of interest statement

The authors have declared that no competing interests exist.

Figures

Fig 1. Graphical illustration of the T-wave delineation method Illustration of two steps in the T-wave delineation method described in the text.

Panel A shows the definition of the empirical window based on heart rate where peak candidates will be detected. Panel B shows the smoothed derivative in solid black (scaled) together with the T-wave (red), and the four identified candidates: SR (slur–rising), peak and SF (slur–falling).

Fig 2. ECG examples.

A set of ECG examples showing the different type of ECG morphologies observed in the two prospective clinical trials. From left to right the panels represent a normal T-wave (A), flattened T-wave (B) and a T-wave with a prominent notch (C), the latter is only observed in 0.8% of the ECGs in FDA study 1 and never observed in FDA study 2. Lastly, the vertical lines correspond to the QRS onset (red), QRS offset (green), primary T-wave (green), secondary T-wave (blue) and T-wave offset (purple) all as identified semi-automatically.

Fig 3. Bland-Altman plots.

Bland-Altman plot for the J-Tpeak, Tpeak-Tend and QT intervals comparing semi-automated and fully automated measurements for FDA Study 1 [9] (top) and FDA Study 2 [10] (bottom). The solid horizontal lines represent the average difference and the dashed lines 2 times the standard deviation (SD) of the difference. In the bottom of each panel is the average difference (SD) of the single-deltas. Of note, only 8 and 9 measurements for J-Tpeakc and Tpeak-Tend, respectively for FDA study 1, were associated with differences >40 ms due to borderline notches (10 single-deltas in total). Removal of these measurements results in a reduction of the SD to <3 ms for all intervals. A representative ECG from one of the eight single-delta outliers identified in Fig 3 is shown in Fig 4B. For cases like this, the blinded adjudicated interpretation was that the initial hump was not large enough to be considered a peak, however, the algorithm identified it a peak thus resulting in a large outlier (>40 ms). However, even in the presence of significant QTc prolongation (~80 ms) outliers of this type were rare (eight single-deltas). In addition, for FDA Study 1 there were a few single-deltas with a difference in QTc of ~30 ms, an example ECG is shown in Fig 4C, where the T-wave barely met the criteria for amplitude.

Fig 4. Example ECGs.

An ECG with a small difference and two ECGs with outlier single-deltas (Fig 3). The left ECG (A) represents a normal ECG with small differences, middle ECG (B) representing an example with a large difference in the J-Tpeak and Tpeak-Tend intervals and the right ECG (C) representing an example of a moderate difference (~30 ms) in QTc. The solid vertical lines refer to semi-automatic measurements and the dashed lines to fully automatic.

Fig 5. Time profiles and exposure-response relationship for dofetilide and quinidine.

The plots in the top row show the time-profile for dofetilide (left) and quinidine (right) for J-Tpeakc (orange), Tpeak-Tend (blue) and QTc (gray) for semi-automated (solid lines) and automated (dashed lines). The plots in the bottom row show the corresponding exposure-response relationship.

Fig 6. Time profiles and exposure-response relationship for ranolazine and verapamil.

The plots in the top row show the time-profile for ranolazine (left) and verapamil (right) for J-Tpeakc (orange), Tpeak-Tend (blue) and QTc (gray) for semi-automated (solid lines) and automated (dashed lines). The plots in the bottom row show the corresponding exposure-response relationship. See. Fig 7 for time-profiles with some time points removed to enhance visualization due to crowding of early time-points.

Fig 7. Time profile for ranolazine and verapamil with some time-points removed.

The plots show the time-profile for ranolazine (left) and verapamil (right) for J-Tpeakc (orange), Tpeak-Tend (blue) and QTc (gray) for semi-automated (solid lines) and automated (dashed lines). This is the dame data as Fig 6 with some time points removed to enhance visualization due to crowding of early time-points.

Fig 8. Time profiles for drug-combinations for FDA study 2 (primary).

Time-profiles for the drug combinations studied in Johannesen et al. [10] are shown for QTc for dofetilide (A), dofetilide + mexiletine (B) and dofetilide + lidocaine (C) with corresponding time-profiles for the pharmacokinetic samples below. Panels D and E show the differences for QTc (gray), J-Tpeakc (orange) and Tpeak-Tend (blue) for dofetilide–mexiletine and dofetilide–lidocaine. In all the panels dashed are from automated measurements and solid are from semi-automated measurements.

Fig 9. Time profiles for drug-combinations for FDA study 2 (secondary).

Time-profiles for moxifloxacin with or without diltiazem for QTc (gray), J-Tpeakc (orange) and Tpeak-Tend (blue) from Johannesen et al. [10]. Dashed lines are from automated measurements and solid are from semi-automated measurements.