Study data

We use de-identified data from two FDA-sponsored studies that are publicly available from the PhysioNet website [19]. The first study (Study A) [13, 20] contains 5,232 ECGs of 22 healthy subjects (11♀, age: 26.9 ± 5.5 y, BMI 23.1 ± 2.6 kg/m²) partaking in a randomized, double-blind, 5-period crossover clinical trial. Its aim was to investigate whether multichannel blocking drugs with different potentials for blocking potassium, late sodium, and calcium currents can be differentiated by their effect on the ECG. In the morning of each 24h treatment period, all subjects received a single dose of one of the drugs dofetilide (500 μg), quinidine sulfate (400 mg), ranolazine (1,500 mg), verapamil hydrochloride (120 mg), or placebo. ECGs were extracted at 16 pre-defined time-points (pre-dose and 0.5, 1, 1.5, 2, 2.5, 3, 3.5, 4, 5, 6, 7, 8, 12, 14, 24 h post-dose. The subjects rested in a supine position for 10 min during ECG extraction, and serum PK samples were taken at the end of the resting phase. The washout period between treatments was 7 days. For further details see [13].

The second study (Study B) [15, 21] contains 4,211 ECGs from a randomized, double-blinded, 5-period crossover clinical trial in 22 healthy subjects who differ from those in Study A (9♀, age: 26.1 ± 4.9 y, 69.9 ± 9.0 kg). It addresses the electrophysiological responses to hERG/iKR current blocking drugs with and without the addition of blockade of either late sodium or calcium current blocking drugs. The 5 treatment periods include dofetilide alone, mexiletine without and with dofetilide, lidocaine without and with dofetilide, moxifloxacin without and with diltiazem, and placebo. In each period, subjects were dosed 3 times per day in the morning (hour 0), afternoon (hour 4) and evening (hour 9.5) as described in Table 1. For details see [15].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Dosing schema for Study B (modified from [15]).

https://doi.org/10.1371/journal.pone.0204712.t001

Studies A and B, registered at ClinicalTrals.gov as NCT02308748 and NCT01873950, were performed at a phase I clinical research unit (Spaulding Clinical, West Bend, WI). Both studies were approved by the US Food and Drug Administration Research Involving Human Subjects Committee (RIHSC 13-011D and 14-022D) and the local institutional review board (Chesapeake IRB). All subjects gave written informed consent [13, 15].

In both studies, continuous 12 lead ECGs (Surveyor, Mortara Instrument, Milwaukee, WI) were recorded at a sampling rate of 500 Hz and an amplitude resolution of 2.5 uV. Triplicate 10-second ECGs were extracted at each time point using the Antares Software (AMPS LLC). The study data contain J-T_peak measurements, corrected for heart rate according to [13]. We used these values for comparison with our TVV-based parameters.

ECG processing

All ECG files were analyzed with eECG/ABBIOS (AbbVie, Inc.’s proprietary, validated, ECG analysis system) in a semi-automated manner. ECGs were reviewed to identify artifacts, abnormal heartbeats, and unreliable automated annotations. ECGs of concern were manually reviewed to identify and annotate a minimum of 3 heartbeats per ECG with unaffected T-waves and consistent T annotations. The average RR intervals and the Fridericia corrected QT (QTcF) intervals [22] were extracted and used for further analyses.

For each ECG, the 12 lead signals were adjusted to the isoelectric lines (defined by the median amplitude of the PQ interval and interpolated between the heart beats by a cubic spline, see [23]), low-pass filtered (bidirectional Bessel filter with 36 Hz), and exported as an annotated ECG file in HL7 format (aECG) with 500 Hz sampling frequency, including the P, Q, J, and T_end annotations. The aECG files were loaded into, and further processed using the R system [24]. The supporting information S3 Text compares the resulting QTcF interval data from our analyses with the published QTcF data [13].

20 ECGs of Study A contained isolated leads that were strongly corrupted by noise. Those individual leads were set to zero. Since our approach aims at characterizing spatiotemporal properties of the cardiac repolarization process, we reconstructed the vectorcardiogram (VCG) by means of the inverse Dower transformation [25], and extracted the T-loop by limiting the VCG to the repolarization time interval [J + 20ms; T_end]. The rationale for choosing the Dower matrix is described in the supporting information S1 Text. Only normal sinus beats were used in subsequent analyses.

Calculation of TVV and T vector trajectory quantiles TrX

Fig 1 illustrates the derivation of the TVV as the temporal derivative of the spatial T vector trajectory. The sequence of blue arrows represents a stroboscopic view in steps of 20 ms of the heart vector’s temporal evolution during repolarization. In continuous time, the tip of the blue vectors inscribes the black trajectory, which corresponds to the 3-dimensional T vector loop. The red vectors indicate the instantaneous velocity of the blue arrows’ (i.e. heart vector’s) movement. We refer to this quantity as TVV, and it is fundamental to the suggested biomarkers.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Derivation of the TVV from the T vector loop.

Displayed is the time course of the heart vector in 20 ms steps (blue arrows), forming the 3-dimensional T vector loop (black curve). The red arrows indicate the TVV, the velocity at which the heart vector is moving at the given times.

https://doi.org/10.1371/journal.pone.0204712.g001

We estimated the TVV by separately calculating the temporal derivatives , and for the three VCG components and calculated its magnitude as . We obtained the temporal derivatives by means of a differentiating Savitzky-Golay polynomial smoothing filter (R method sgolayfilt, package signal, version 0.7–6) of order 3 and filter length of 31 samples, equivalent to a time window length of 60 msec. The rationale for our choice of filter settings is described in the supporting information S2 Text.

Integration of the TVV over a time interval corresponds to a cumulative summation of all spatio-temporal changes of the T vector trajectory, and yields the length of the T vector trajectory in the given time span. Normalizing the total T vector trajectory length to 1 allows interpreting the time course of the integrated TVV as a distribution function for the trajectory’s time course. The X% quantile of this distribution corresponds to the time elapsed until X% of the total T vector trajectory length have been reached. To assess drug effects throughout the entire repolarization phase, we used the 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90%, and 100% trajectory quantiles as parameters, and named them Tr10, Tr20, …, Tr100. For each ECG, we calculated the T vector trajectory quantile TrX as the average of the corresponding quantiles of the normal sinus beats.

Since repolarization is known to be strongly affected by heart rate, the trajectory quantile parameters were corrected for heart rate by fitting a linear mixed effects model (R package lme4, version 1.1–13) of the form TrX~1 + RR + (1 + RR|Subject) to all drug-free data in Studies A and B. The heart rate corrected trajectory quantiles were calculated as TrXc = TrX − β * (RR − 1000), where β represents the model’s fixed effect slope estimate, and RR the ECG’s average beat interval measured in milliseconds. The rationale for our choice of heart rate correction method is described in the supporting information S4 Text.

Fig 2 illustrates derivation of the T vector trajectory quantiles with the ECGs of subject 1017 from Study A under placebo and under quinidine treatment where colors indicate the relative time points. The upper row displays the T vector loops in 3-dimensional space, indicating normal diurnal variation in the placebo column, and quinidine induced T loop morphology changes in addition to circadian effects. With respect to the body orientation, the front direction corresponds to x, left direction to y, and foot direction to the negative z VCG component. Small dots indicate the start of the T vector trajectories, which are 20ms after the J-point. The second row displays the time courses of the T vector strength (TVS), calculated as , which would be the basis signals for T_peak measurements as applied in the FDA studies A and B (note the flattening and trend toward notching for quinidine). The third row displays the time course of the TVV, and the fourth row the cumulated and normalized TVV over time, which is identical to the relative vector trajectory length at a given time. The horizontal lines denote the 30% and 70% trajectory quantiles. Note that the cumulated TVV curves are nicely aligned with stable relative positions under placebo, but intersect under quinidine. Notably, in the first hours after quinidine dosing corresponding to high quinidine plasma concentrations, the 30% quantiles are decreased, while the 70% quantiles are increased. The time axes in Fig 2 are corrected for heart rate according to Fridericia’s formula.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Illustration of the derivation of the TVV and the T vector trajectory quantiles.

Displayed are ECGs of subject 1017 in Study A under placebo (left column) and under quinidine treatment (right column). Colors indicate the relative time with respect to dosing. 1^st row: T vector loops in the 3-dimensional space. Each loop is the average loop of the normal sinus heartbeats in the first replicate at the given time point. The coordinate system is located in the center of the heart with the axes pointing to the body’s front, left side and to the foot. 2^nd row: Time courses of the T vector strength. 3^rd row: Time courses of the TVV. 4^th row: Cumulated and normalized TVV over time, that is the relative loop length reached at a given time, with 30% and 70% trajectory quantiles.

https://doi.org/10.1371/journal.pone.0204712.g002

Evaluation of drug effects

The T vector trajectory quantiles TrXc described above are the basis for our evaluation of drug effects on ventricular repolarization. In order to quantify a drug effect, single-delta parameter values were calculated as ΔP(t, TR) = P(t, TR) − P(t₀, TR), with P(t, TR) as the average parameter value of the replicate ECGs at time t under treatment TR for a given subject, and t₀ as the baseline time point. Subsequently, double-delta parameter values were derived as ΔΔP(t, Drug) = ΔP(t, Drug) − ΔP(t, Placebo).

We modeled the dependencies of the T vector trajectory quantile parameters on the drug concentrations by means of the mixed effects models ΔΔP~0 + C + (0 + C|Subject) (for single drugs in Study A) respectively ΔΔP~0 + C1 + C2 + C1:C2 + (0 + (C1 + C2)|Subject) (for two drugs in Study B), where ΔΔP is the placebo corrected change from baseline of the parameter P, and C, C1, C2 are the drug concentrations. The zero terms in both models enforce the regression line or plane to intersect with the origin.

The linear mixed effects models were fitted using the R package lme4, version 1.1–13. In order to obtain a drug effect quantification, we evaluated the model predictions at fixed drug concentrations which we considered representative for the range of values actually observed. For Study A, we chose the following drug concentrations: dofetilide: 2500 pg/ml, quinidine: 1500 ng/mL, ranolazine: 2000 ng/mL, verapamil: 100 ng/mL. Fig 3 illustrates these values as vertical grey lines in the dose-response diagrams.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Exposure response in Study A.

Effects of dofetilide, quinidine, ranolazine, and verapamil determined as exposure-response relation of the plasma drug concentrations on the double-delta parameters Tr30c, Tr50c, Tr70c, and Tr100c. The blue lines indicate the estimated slope with its 95% confidence interval. The colors of the data points denote individual subjects. The grey vertical lines denote the drug concentrations used for calculating the drug effect profiles in Fig 4.

https://doi.org/10.1371/journal.pone.0204712.g003

For Study B, which included dosing of two compounds, the representative drug concentrations were determined as the geometric mean drug concentration from the afternoon and evening dosage time points where both drugs had been administered (see Table 2). The model prediction’s 95% confidence intervals were estimated through a bootstrap simulation with 2000 replicates (R package pROC, version 1.10.0).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Representative plasma drug concentrations in Study B.

https://doi.org/10.1371/journal.pone.0204712.t002

For assessing the combined effects of mexiletine with dofetilide, we fitted the model to the pooled data from the dofetilide alone treatment period, and the period with combined mexiletine plus dofetilide treatment. The same approach was used for assessing the effects of lidocaine combined with dofetilide. In the assessment of the effect of diltiazem combined with moxifloxacin, we omitted the interaction term C1:C2 because no data exist for pure diltiazem treatment, and because the regression surfaces for the model including the interaction term revealed implausible distortions. Thus, the predictions of the pure diltiazem effects are model extrapolations and should be interpreted with an appropriate level of caution.

The effects of a drug onto the 10 trajectory quantiles Tr10c to Tr100c (together with the corresponding 95% confidence intervals) constitute what we call a drug effect profile (of TrXc). A drug effect profile reveals during which phase, and to what extent, a drug delays or accelerates the repolarization process. We derived effect profiles for all drugs in Study A and the various drug combinations in Study B, and examined the effect profile characteristics for their consistency across both studies, and for their plausibility from a physiological perspective.

Classification of pure hERG/iKR current block versus multichannel block

In order to verify the utility of the T vector trajectory quantile parameters in differentiating pure hERG/iKR current block (group I) from multichannel block (group II), we used the data from Study B. To facilitate direct comparisons with the literature, we followed the methods as described in [16]. ECGs from all time points at which selective hERG/iKr current blockers had been dosed (dofetilide and moxifloxacin alone arms in Study B) were considered for group I. ECGs from all time points with additional administration of late sodium blockers (mexiletine with dofetilide and lidocaine with dofetilide arms in Study B) were assigned to group II.

For J-T_peakc, QTcF and for each of the 10 trajectory quantiles, we fitted logistic regression classification models B~ΔΔP where B denotes the block type and ΔΔP the double-delta values of the given parameter (Method glm, R package stats, version 3.4.1). Furthermore, we compared the models for the individual parameters with a model consisting of all TrXc parameters B~ΔΔTr10c + ⋯ + ΔΔTr100c. We assessed the performance in separating group I ECGs from group II ECGs by calculating the area under the receiver-operating curve (AUC). The 95% AUC confidence interval was estimated using a stratified bootstrap technique with 2000 replicates (R package pROC, version 1.10.0).

Relation of T vector trajectory quantiles to J-T_peak and QT

Since the 100% trajectory quantile Tr100 represents the time interval from J + 20ms to the end of the T wave, we expect that the drug effects as captured by this parameter are similar to those obtained from the QT interval (respectively QTcF). To verify this, we calculated the drug effects of both parameters for Study A and Study B, and compared them visually. (Supporting information S5 Text describes an approach for statistically comparing the drug effects between QTcF and Tr100c.) Keeping in mind that T_peak is typically located roughly in the middle of the J-T_end interval, we likewise expect a relation between the J-T_peak interval and the central T vector trajectory quantiles. This was addressed by visually comparing the drug effects of published J-T_peak data to Tr50 and Tr60 values.

Simulations of ion channel blocks

In order to provide evidence for a physiological interpretation of our results, we calculated action potential waveforms using the O’Hara-Rudy model [26] for human endocardial myocytes with its default parameters. In addition, we simulated a 20% block of the hERG/IKr-Channel, a 90% block of the late sodium (INaL) channel, and a combination of both blocks, by reducing the corresponding model variables of maximum channel conductivity accordingly.

Interpretation of T vector trajectory quantiles

The X% T vector trajectory quantile TrX denotes the time when X% of the total T vector trajectory length has been reached. Arriving earlier at a particular point on the T vector trajectory (as compared to placebo) means that the repolarization process is accelerated in the corresponding phase of repolarization. Hence, the dose-response curve of this placebo-corrected change from baseline quantile shows a negative slope, and its effect profile value is negative. In contrast, an increase in the time required to reach a particular point on the T vector trajectory indicates a delay of the repolarization process, and is associated with a positive slope in the dose-response curve and a positive value in the effect profile. The magnitude of an effect profile value reflects the effect size, i.e. the extent of acceleration or delay, with respect to a given drug concentration.

We propose a link between cellular repolarization changes resulting from blockade of single and multiple ionic currents and the observed TrX effect profiles as illustrated in Fig 10: Inward sodium and calcium ion currents are predominantly active during earlier repolarization, and both contribute to the maintenance of the action potential’s plateau phase. In our simulation based on the O’Hara-Rudy model [26], blocking the late sodium ion channels causes a steeper descent of the AP in phase 2 (Fig 10 emerald curve vs. red curve). This represents an acceleration of the repolarization process and earlier restitution of the resting potential. Phase 3 of the AP appears only marginally affected by pure sodium ion channel block (Fig 10) as the red and emerald curve run almost in parallel in phase 3, i.e. their downslopes have comparable magnitude.

We expect that faster intracellular action potential changes are reflected in higher speed of progression along the T vector trajectory. If sodium-block related acceleration predominantly happens in phase 2 of the AP, it should increase the TVV in the initial part of the T vector trajectory. Hence, the corresponding lower T vector trajectory quantiles will be reached earlier, resulting in negative effect profile values for the lower quantiles. This phenomenon is observable in all effect profiles of the late sodium current blockers mexiletine (Fig 5A blue line) and lidocaine (Fig 5B blue line), where the lower quantiles Tr10c to Tr40c are continuously decreased. The accumulated level of time lead is maintained over the second part of repolarization with Tr50c to Tr100c remaining at a negative effect profile level comparable to that of Tr40c. This is in accordance with the limited effect of pure sodium ion channel block onto phase 3 of the AP in our simulation (Fig 10). Overall, this indicates that the slight QTcF shortening effect of these two drugs is mainly caused by an acceleration of early repolarization, which then is largely maintained.

Fig 10 also shows that pure hERG/iKr current block causes a reduction of the downslope in the AP in both phase 2 and phase 3 (green curve vs. red curve), corresponding to a delay which affects all parts of the repolarization process. The effect profiles of both pure hERG/iKr current blockers dofetilide (red curves in Figs 4, 5A and 5B) and moxifloxacin (red curve in Fig 5C), consistently reflect this. They show a continuously increasing delay, indicating that the QT prolongation caused by these drugs are effective over the entire course of repolarization.

A combined block of sodium and hERG/iKr currents results in a superposition of the single effects in the AP waveform (Fig 10 violet curve). The reduced downslope of pure hERG/iKr current block (green curve) is largely compensated for by the additional sodium channel block in phase 2 (Fig 10 violet curve vs. red curve), but is still evident in phase 3 (note the reduced downslope of the violet curve versus the red one). The effect profiles of the corresponding drug combinations dofetilide + mexiletine (Fig 5A, green curve) and dofetilide + lidocaine (Fig 5B, green curve) are well in line with this. Their sigmoid shape indicates a slight, but largely balanced, acceleration of the repolarization process in the early T vector trajectory quantiles Tr10c to Tr40c (associated with the effect of the sodium ion channel block in AP phase 2 and the counteracting hERG/iKr current block), which then becomes increasingly delayed by the preponderance of the hERG/iKr current block effect in AP phase 3.

In the case of mexiletine, the combined effect profile (Fig 5A, green curve) is an approximately linear combination of the pure drug effect profiles (Fig 5A, red curve and blue curve). For lidocaine, the effect profile of the drug combination (Fig 5B, green curve) is slightly more depressed compared to the pure lidocaine effect alone (Fig 5B, blue curve). This is owing to a significant contribution of the interaction term C1 * C2 in the mixed effects model to all T vector trajectory quantile parameters.

The presented interpretation is further supported by the effect profiles of drugs, which simultaneously affect multiple ion channels. Ranolazine blocks both the hERG/iKr and the late sodium currents, and shows the same sigmoid-shaped effect profile (Fig 4 blue curve) observed for the combinations of the individual channel blockers (Fig 5A and 5B green curves).

Verapamil predominantly blocks the L-type calcium ion channel, mainly affecting AP phase 2, i.e. early repolarization, and to a lesser extent the hERG/iKr current. In line with this, its effect profile (Fig 4, green curve) indicates a slight but significant acceleration of the heart’s electrical activity in the early repolarization, which is compensated for during late repolarization but does not result in significant prolongation of the overall repolarization duration, measured by the QTcF interval.

The effect profile of quinidine (Fig 4, orange curve) as a strong hERG/iKr current blocking drug as well as calcium and sodium current blocking agent indicates significant acceleration in the early repolarization phase and a strong delay in the mid and late repolarization phases. On the other hand, the effect profile of dofetilide (Fig 4, red curve) as a pure hERG/iKr current blocker indicates (with exception of Tr10c) continuously increasing delay throughout repolarization.

As reported in [15], the combination of diltiazem with moxifloxacin in the evening dose did unexpectedly slightly increase QTcF, despite a slightly reduced moxifloxacin plasma concentration compared to the afternoon (pure moxifloxacin). Our analysis shows that diltiazem slightly accelerates repolarization in the early phase (Fig 5C, green curve for Tr10c to Tr40c), which may be due to its calcium ion channel block. However, diltiazem delays the later repolarization phase, finally increasing the repolarization delay induced by moxifloxacin. Note that the pure diltiazem effect profile (Fig 5C, blue curve) was calculated as an extrapolation of the mixed effects model and should be interpreted with care.

The body surface potentials observed through the T vector loop are results of a complex integration of cellular action potentials from different myocardial cell types (endocardial, epicardial, M cells) with different activation times, modulated among others by spatial gradients in ion channel expression. Thus, the proposed link between cellular repolarization and TrX effect profiles should be taken as a reasonable and plausible attempt to explain our clinical findings rather than a rigorous mechanistic proof. Moreover, the available number of drugs was limited and should certainly be expanded in future. But in summary, our current results support the presented interpretation and suggest that the concept of the TrX effect profile constitutes an electrocardiographic fingerprint of a drug’s impact on repolarization. It reveals the presence, extent and relative phase of drug-induced alterations in the timing (delays or accelerations) of the ventricular repolarization process. TrXc effect profiles extend repolarization assessment beyond QTc and the T_peak fiducial point to a quasi-continuous view onto the entire repolarization process.

Differentiation of pure hERG/iKr current block versus multichannel block

Our results confirm that T vector trajectory quantiles enhance differentiation of multichannel block from pure hERG/iKr current block. In our data, pure hERG/iKr current block consistently manifests as an increasing delay of repolarization with a slightly higher slope between Tr10c to Tr60c compared to Tr60c to Tr100c (Figs 4 and 5, red lines). The presence of additional effective ion channel blocks is revealed by deviations from this profile. According to the previous discussion, the effect of late sodium ion channel block should be particularly evident during phase 2 of the AP. The finding that the “early” T vector trajectory quantiles Tr20c, Tr30c and Tr40c demonstrate the best separation (Fig 8), and that Tr40c is most effective among all T vector trajectory quantiles to identify additional late sodium block (Fig 8) is well in line with this assumption. Note that the performance of the 40% quantile of the T vector trajectoryTr40c, reaching an AUC of 0.9 (Fig 9), is significantly higher than that provided by J-T_peak (0.81) representing the best previously published ECG biomarker for that purpose [15, 16]. The potential of the effect profile in identifying multi-ion channel effects is highlighted by a further increase in performance observed when the entire profile is used in the classification process (Fig 9, AUC 0.94). Well aware of the danger of overfitting a 10-parameter-model to a data set of limited size, we take this improvement only as an additional hint that requires prospective validation.

Relation of TrX to QT and J-T_peak

Since QTcF and Tr100c just differ in the starting point (Q versus J+20ms) and in a slightly different correction formula for heart rate, it is not surprising that their effects are largely comparable in magnitude and precision. The slight differences may arise from drug effects on QRS duration, from the different heart rate correction methods for QTcF and Tr100c, or from systematic differences of our algorithms’ robustness with respect to identification of the J point and the onset of Q.

ECG effects described by the TrX parameters are in line with effects described by parameters, which are derived from the T_peak fiducial point: Pure hERG/iKr current blocking drugs delay repolarization through the entire repolarization process, indicated by continuously increasing parameters Tr10 to Tr100. Drugs or drug combinations that additionally block late sodium or calcium ion currents shorten the parameters Tr10 to Tr40. Accordingly, pure hERG/iKr current blocking drugs prolong both the J-T_peak and the T_peak-T_end intervals, while J-T_peak tends to be shortened by multichannel blocking drugs [27]. However, declaring T_peak as cutting point between early and late repolarization is questionable [28], and there is no generally accepted physiologic reason for choosing the T_peak to subdivide the repolarization interval for assessment of pro-arrhythmic effects [29]. It has recently been challenged whether a lack of J-T_peak prolongation can reliably differentiate between safe and proarrhythmic QT_end prolongation [30].

Using a simplified model of VCG generation, we would expect T_peak to represent the instant in time when the largest spatial gradient of intracellular potentials exists, i.e. a moment of maximum heterogeneity of the action potentials in all heart cells. From our new parameters, the Tr50c and Tr60c quantiles correspond most closely with J-T_peakc. For the majority of drugs, the effect size of J-T_peakc is located between that of Tr50c and Tr60c (Figs 6 and 7, lines on left side).

The clear trend for narrower confidence intervals of Tr50c and Tr60c, compared to J-T_peakc (Figs 6 and 7, lines on left hand side), suggests that the determination of the T vector trajectory quantiles is more robust than that of the T_peak position. This is not surprising in view of the demonstrable changes of T wave morphology induced by drugs, including severe flattening or notching. In such a setting, small alterations of the intracellular potential distribution may cause significant dislocation of the T_peak position, increasing J-T_peak parameter variability. We attribute the robustness of the T vector trajectory quantiles TrX to their unique identifiability even in the presence of significant changes in T wave / T loop morphology and consider this an advantage compared to J-T_peak.

Another conceptual difference between J-T_peak and the TrX parameters is that the J-T_peak determination is purely based on the T vector’s magnitude, while the TrX parameters capture changes of both the magnitude and the direction of the T vector progression. Since directional changes of the T vector trajectory reflect changes of the distribution of electrical activities in the whole heart, this information may contribute to the performance improvement in differentiating multi-channel blockade.