Evan B. Wang (1), Matthew M. Abernathy (1), Derek J. Leishman (1)
(1) Eli Lilly and Company, Indianapolis, IN, USA
Objectives: ICH S7B and E14 guidances require that all compounds without intended effects on cardiac electrophysiology undergo ventricular repolarization safety assessment through measurement of the QT interval of the electrocardiogram. Given the interdependency of the duration of the QT interval with heart rate, a multitude of correction techniques are used in both clinical and nonclinical trials. Utilizing real beat-to-beat heart rate data and a human ventricular action potential model, the objectives of this work were twofold: 1) demonstrate performance of standard fixed factor and linear regression QT correction methods, and 2) evaluate which method is optimal for correcting the QT interval with drug effects on heart rate and/or ventricular repolarization.
Methods: A simulation-based approach allows known repolarization and heart rate changes to be applied. Different correction methods can then be compared to the known change. Continuous 24-hour beat-to-beat heart rate data from a 64-year-old female was used to calculate QT intervals for a variety of theoretical cardiovascular drug effects (human ether-a-go-go or hERG channel inhibition at 10, 20, 30%, heart rate increase at 10, 20, 30%, and a combination of the two effects). Heart rate and percent hERG inhibition were used as inputs into the O’Hara-Rudy human ventricular action potential model [1], further modified by Dutta et al [2], outputting the corresponding APD90 value (time to repolarize 90% between the peak and resting potential), which is then converted to a QT interval using a human reference value. For all cases (control, drug effects), the QT intervals were then corrected to 60 beats/min using four methods (Bazett’s, Fridericia’s, linear regression using control slope, linear regression using treatment slope). This exercise was repeated for 1-minute averages of the beat-to-beat data to mimic standard laboratory practices.
Results: Correction methods were assessed based on accuracy (deviation from the “true” modeled interval) and precision (standard deviation of corrected QT intervals). The reported results [lower, upper] reflect the range for the 9 theoretical cardiovascular drug effect scenarios listed above. For both beat-to-beat and 1-minute averaged heart rate data, linear regression using the treatment slope is shown to be overall the most accurate (beat-to-beat: [-11.4, -3.0] ms, 1-min-avg: [-21.6, -6.7] ms) and precise method (beat-to-beat: [3.2, 5.3] ms, 1-min-avg: [3.8, 8.0] ms). Bazett’s method was the least accurate producing the highest standard deviation (beat-to-beat: [12.6, 26.5] ms, 1-min-avg: [15.8, 31.3] ms) and consistently over-predicted the QT interval (beat-to-beat: [32.3, 57.9] ms, 1-min-avg: [28.0, 56.6] ms). Fridericia’s method performed better than Bazett’s method but still produced higher standard deviation (beat-to-beat: [4.0, 9.6] ms, 1-min-avg: [7.32, 14.48] ms) and over-predicted the QT interval (beat-to-beat: [9.4, 15.7] ms, 1-min-avg: [8.6, 15.4] ms). Linear regression using the control slope was similar in accuracy (beat-to-beat: [-17.5, -3.4] ms, 1-min-avg: [-25.4, -8.3] ms) and precision (beat-to-beat: [3.3, 7.7] ms, 1-min-avg: [4.38, 9.45] ms) compared with treatment slope. Unlike Bazett’s and Fridericia’s methods, linear regression using control or treatment slopes under-predicted the modeled interval, although both methods were closer in magnitude. In cases where there is hERG inhibition with and without heart rate increase (slope of the treatment changes relative to the control), correcting based on the treatment slope showed the smallest standard deviation and was consistently closer to the true simulated data. The relative accuracy increases with increasing hERG inhibition. Only when the QT interval was modified solely by heart rate increase did linear regression using control slope show greater accuracy than using the treatment slope.
Conclusions: Combining real beat-to-beat heart rate data with a human ventricular action potential model convincingly shows that using the treatment slope is the optimal method to correct QT interval measurements. Additionally, this work provides a means to compare heart rate correction methods and to understand the limitations of applying certain methods when they are the only methods feasible.
References:
[1] O’Hara T, Virag L, Varro A, and Rudy Y (2011) Simulation of the Undiseased Human Cardiac Ventricular Action Potential: Model Formulation and Experimental Validation. PLoS Comput Biol. 7(5) e1002061.
[2] Dutta S et al. (2017) Optimization of an In silico Cardiac Cell Model for Proarrhythmia Risk Assessment. Front Physiol. 8(616).
Reference: PAGE 27 (2018) Abstr 8454 [www.page-meeting.org/?abstract=8454]
Poster: Drug/Disease Modelling - Safety