Pattern recognition analysis of digital ECGs: Decreased QT measurement error and improved precision compared to semi-automated methods

doi:10.1016/j.jelectrocard.2012.11.012

Journal of Electrocardiology

Volume 46, Issue 2, March–April 2013, Pages 118-125

https://doi.org/10.1016/j.jelectrocard.2012.11.012 Get rights and content

Abstract

Background and Purpose

Machine-read QT measurements employing T-wave detection algorithms (ALG) are not accepted by regulatory agencies for the primary analysis of thorough QT (TQT) studies. Newly developed pattern recognition software (PRO) which matches ECG waveforms to user-defined templates may improve this situation.

Methods

We compared RR, QT, QTc, QT variability, T-end measurement errors, and individual QT rate correction factors and their associated coefficients of determination (R²) following ALG and PRO analysis. Machine-read QTc values were compared with core laboratory semi-automated (SA) values for verification.

Results

Compared to ALG, PRO reduced the frequency of T-end measurement errors (5.6% vs. 0.1%), reduced the intra-individual QT variability (12.6 ± 5.9 vs. 4.9 ± 1.1 ms) and allowed the recovery of 3/58 subjects that exhibited an unacceptable (< 0.9) R².

Conclusions

PRO adjusted for ALG-based T-end measurement errors and provided an accurate and precise automated method for continuous QT analysis, thus offering an alternative to resource-intensive semi-automated analyses currently performed by ECG core laboratories.

Introduction

Currently, the accurate clinical assessment of potential drug-induced QT interval prolongation remains a major concern for regulatory agencies. Rarely, prolongation of ventricular repolarization may lead to torsade de pointes (TdP), a life-threatening polymorphic ventricular tachycardia. Despite several attempts to develop new and more reliable biomarkers predictive of TdP,1, 2, 3 the assessment of cardiac repolarization during a thorough QT (TQT) study remains the gold standard where virtually all new drug candidates are required to demonstrate relative safety for the risk of inducing TdP, indirectly assessed as an estimation of their potential to prolong the QT interval.

Pharmaceutical companies generally outsource TQT data analysis to ECG core laboratories which specialize in semi-automated (SA) ECG annotations performed by well-trained cardiologists. The analysis performed by such laboratories typically consists of semi-automatically adjudicated annotations of 3–9 beats obtained during periods of heart rate stability (± 2 bpm) extracted at predefined time points prospectively defined in the study protocol.⁴ Although this methodology provides reliable QT interval measurements during the specified intervals, it has many well recognized and, to date, unresolved limitations including the exclusion of the majority of the recorded data, insufficient temporal resolution, substantial time and resource (manpower, cost) requirements, and inter-observer variability.5, 6, 7, 8 Importantly, the sparse QT–RR data derived from fully manual or SA ECG analysis precludes the derivation and use of individual QT rate corrections, now generally accepted as the most accurate and reliable method to control for the effects of heart rate on the QT interval.⁹

Most of the fully automated T-end detection algorithms employ traditional analytical approaches (e.g. first-derivative adaptive threshold, tangent-based return to baseline) which usually perform well in noise-free, stable ECG recordings.10, 11 However, electrical noise, baseline drift, and T-wave morphological changes will unavoidably lower the accuracy of algorithm-based ECG interval measurements, particularly affecting the reproducible determination of the T-end offset.12, 13 Recently developed pattern recognition software enables reliable machine-read analysis of continuous ECGs by matching individual ECG waveforms to user-selected, noise-free, well-inscribed, and manually adjudicated ECG complexes which serve as reference waveform analysis templates.

In this study, we considered a well-characterized automated ECG analysis algorithm,¹⁴ and assessed potential improvements in raw QT interval measurement accuracy and precision with the addition of an ECG pattern recognition module (Ponemah Physiology Platform, ver. 5.0, Data Sciences International, St. Paul, MN). The algorithm (ALG) and pattern recognition (PRO) modules were compared by performing parallel analyses of continuous ECG data collected from all periods of a TQT study which included placebo, two doses of saquinavir, a QT-prolonging drug that induces changes in T-wave morphology,¹⁵ and the standard clinical reference agent, moxifloxacin. The current machine-read methods were validated by direct comparison of the reference moxifloxacin-induced QT-prolonging effect with parallel time-matched values obtained by an ECG core laboratory during the primary TQT analysis. Confirmation of accurate and precise fully automated QT analytical method, particularly in the setting of variable T-wave morphology,¹⁶ would provide a powerful new method for continuous beat-to-beat ECG analysis, possibly supplanting the current labor intensive manual adjudication of sparse ECG samples. Accordingly, the goals of the current study were twofold: (1) to directly compare the accuracy and precision of ALG and PRO T-end determinations in the presence of drug-induced QT prolongation and/or T-wave morphological changes and (2) to characterize and compare the ability of both machine-read analyses to detect benchmark moxifloxacin-induced QTc changes previously adjudicated by an ECG core laboratory employing SA analysis of beats extracted from a common data set.

Section snippets

Study design

Data used in this study were extracted from a TQT study previously evaluated by an ECG core laboratory. Sixty healthy subjects were enrolled and 24-h continuous 12-lead digital ECGs (1000 Hz, Mortara H12 + Holter devices, Mortara Instruments, Milwaukee, WI) were acquired on day − 1 (pretreatment baseline) and day 3 (on-drug) for each period. The study was performed at a single center, employing a double-blind, randomized, four-period, four-way crossover design. All subjects received each of four

Raw QT and RR interval analysis

For all subjects and periods, PRO analysis matched 91% ± 10% of all beats employing 4.2 ± 1.3 templates (range 2–7), yielding ~ 72-K valid beats for each 24-h period. Generally, unmatched complexes were due to either electrical noise or environmental disturbances.

Visual characterization of T-end adjudication with both ALG and PRO yielded 671 and 10 measurement errors, corresponding to incidence rates of 5.6% and 0.1%, respectively (p < 0.001, chi-squared test). ALG measurement errors usually consisted

Discussion

While the feasibility of reliable machine-read QT interval measurements has been previously demonstrated,5, 6, 7, 8, 10, 13 this methodology is not yet routinely accepted by the regulatory agencies as a standalone clinical ECG evaluation method. Natural and drug-induced changes in T-wave morphology remain a major concern when employing machine-read analysis of continuously digitized ECGs. Currently, there is no clinical evidence to support the notion that a machine-read analysis, which performs

Conclusions

Machine-read ECG methodologies are rapidly evolving and, as demonstrated for the current TQT study, advanced measurement algorithms can precisely replicate the results of semi-automated visual interpretations currently performed by core laboratory cardiologists, but with reduced QT variability and improved temporal resolution. The current results support the use of pattern recognition methods when evaluating continuously recorded digital ECGs, especially when the study involves the analysis of

Limitations

The current study assessed moxifloxacin-induced QTca changes based on machine-read values derived from continuous QT measurements from lead V₄ which were then compared to moxifloxacin-induced QTcS changes based on semi-automated lead II QT measurements performed by an ECG core laboratory. Comparison of repolarization results obtained in different leads and employing different methodologies should be interpreted with caution. In the current study, lead V₄ offered optimal waveform amplitudes for

References (23)

C. Antzelevitch
Role of transmural dispersion of repolarization in the genesis of drug-induced torsades de pointes
Heart Rhythm
(2005)
R.R. Shah
Drug-induced QT, dispersion: does it predict the risk of torsade de pointes?
J Electrocardiol
(2005)
N. Sarapa et al.
Comparison of QTinno, a fully automated electrocardiographic analysis program, to semiautomated electrocardiographic analysis methods in a drug safety study in healthy subjects
J Electrocardiol
(2009)
B. Tyl et al.
Improvement and limitation of the reliability of automated QT measurement by recent algorithms
J Electrocardiol
(2011)
X. Zhou et al.
A multidifferentiator-based approach to the reliable determination of T-wave offset in electrocardiograms
J Electrocardiol
(2011)
R.W. Chui et al.
A comparison of three software platforms for automated ECG analysis
J Pharmacol Toxicol Methods
(2009)
V. Salvi et al.
Comparison of 5 methods of QT interval measurements on electrocardiograms from a thorough QT/QTc study: effect on assay sensitivity and categorical outliers☆
J Electrocardiol
(2011)
P. Smetana et al.
Individual patterns of dynamic QT/RR relationship in survivors of acute myocardial infarction and their relationship to antiarrhythmic efficacy of amiodarone
J Cardiovasc Electrophysiol
(2004)
M. Malik et al.
Study as part of a first-in-man study
J Clin Pharmacol
(2008)
A.A. Fossa
Dynamic beat-to-beat modeling of the QT–RR interval relationship: analysis of QT prolongation during alterations of autonomic state versus human ether a-go-go-related gene inhibition
J Pharmacol Exp Ther
(2004)

F. Extramiana et al.

Contrasting time- and rate-based approaches for the assessment of drug-induced QT changes

J Clin Pharmacol

(2007)

Cited by (7)

QT interval correction assessment in the anesthetized guinea pig
2015, Journal of Pharmacological and Toxicological Methods
The anesthetized guinea pig (ANES GP) has proven to be an effective small animal model to evaluate cardiac electrophysiologic effects of drug-candidate molecules during lead optimization. While heart rate (HR) corrected QT interval (QTc) is a key variable to determine test article-dependent repolarization effects, ideal correction methods are an area of constant debate given the potential influence of anesthesia, autonomic tone, species, strain and gender on the QT/HR relationship. The aim of this study was to characterize the ability of common correction formulas to normalize rate-dependent effects on the QT interval in the ketamine/xylazine ANES GP.
Atrial pacing (n = 10), ivabradine or ephedrine (n = 6/group) infusions were used, respectively to evaluate the effects of a wide range of HRs on the QT/HR relationship. Correction formulas (Bazett [QTcb], Fridericia [QTcf] and Van de Water [QTcVdW]) were applied and the best fit formula was determined with the aid of the slope of their QT–HR linear relationship.
From 100 to 220 bpm, QTcb underestimated the change in QT interval duration (QT/HR slope = 0.35 to 0.67). However, QTcVdW was more appropriate in this HR range (QT/HR slope = − 0.07 and 0.09). At higher HRs (> 220 bpm), QTcb performed better (QT/HR slope = − 0.02 and 0.07) as compared to QTcf (QT/HR slope = − 0.18 to − 0.1) and QTcVdW (QT/HR slope = − 0.2 to − 0.17) (p < 0.01). All the correction formulas identified dofetilide- and sotalol-dependent repolarization delay (n = 6/group) but QTcb and QTcf demonstrated reduced sensitivity as compared to fixed cardiac pacing (p < 0.01). In contrast, QTcVdW resulted in an apparent underestimation of the QT interval duration at HR levels above the basal ketamine/xylazine ANES GP HRs (> 220 bpm) with ephedrine (n = 6).
The best fit correction formula in the ANES GP was highly dependent on the HR range. In the ketamine/xylazine model, QTcVdW performed best with HR < 220 bpm and QTcb performed best with HR > 220 bpm. The QTcVdW correction formula was thus selected in the ketamine/xylazine ANES GP since HRs in this model are generally within the optimal range for this correction formula.
Preclinical QT safety assessment: Cross-species comparisons and human translation from an industry consortium
2014, Journal of Pharmacological and Toxicological Methods
Citation Excerpt :
Employing a common analytical scheme, the current data present a parallel QT–RR comparison across all species and clearly demonstrate that preclinical variability can be reduced to the same extent as in man when the species-specific QT–RR domains are considered in the rate-correction model. Assays conducted across diverse platforms must be referenced to a common gold standard to accurately assess translation (Meyer et al., 2013). For repolarization studies, moxifloxacin is the most common clinical reference agent and was thus used to benchmark all current preclinical results.
In vivo models have been required to demonstrate relative cardiac safety, but model sensitivity has not been systematically investigated. Cross-species and human translation of repolarization delay, assessed as QT/QTc prolongation, has not been compared employing common methodologies across multiple species and sites. Therefore, the accurate translation of repolarization results within and between preclinical species, and to man, remains problematic.
Six pharmaceutical companies entered into an informal consortium designed to collect high-resolution telemetered data in multiple species (dog; n = 34, cynomolgus; n = 37, minipig; n = 12, marmoset; n = 14, guinea pig; n = 5, and man; n = 57). All animals received vehicle and varying doses of moxifloxacin (3–100 mg/kg, p.o.) with telemetered ECGs (≥ 500 Hz) obtained for 20–24 h post-dose. Individual probabilistic QT–RR relationships were derived for each subject. The rate-correction efficacies of the individual (QTca) and generic correction formulae (Bazett, Fridericia, and Van de Water) were objectively assessed as the mean squared slopes of the QTc–RR relationships. Normalized moxifloxacin QTca responses (Veh Δ%/μM) were derived for 1 h centered on the moxifloxacin Tmax.
All QT–RR ranges demonstrated probabilistic uncertainty; slopes varied distinctly by species where dog and human exhibited the lowest QT rate-dependence, which was much steeper in the cynomolgus and guinea pig. Incorporating probabilistic uncertainty, the normalized QTca-moxifloxacin responses were similarly conserved across all species, including man.
The current results provide the first unambiguous evidence that all preclinical in vivo repolarization assays, when accurately modeled and evaluated, yield results that are consistent with the conservation of moxifloxacin-induced QT prolongation across all common preclinical species. Furthermore, these outcomes are directly transferable across all species including man. The consortium results indicate that the implementation of standardized QTc data presentation, QTc reference cycle lengths, and rate-correction coefficients can markedly improve the concordance of preclinical and clinical outcomes in most preclinical species.
Comparative TQT analysis with three fully-automated platforms: Comparison to core laboratory semi-automated results
2013, Journal of Electrocardiology
Citation Excerpt :
Briefly, data used in this study were extracted from a TQT study previously evaluated by an ECG core laboratory. Both the current study design and semi-automated methodologies have been described in detail elsewhere.4 It is well-recognized that selection of the lead to be analyzed with machine-read methods is critical as electrical noise, baseline drift, and T-wave morphological changes will impact the accurate and reproducible determination of T-end offset.2,15
Technological advances in machine-read QT measurement now enable detailed and precise cardiac repolarization assessments. This study assessed the applicability of three state-of-art ECG measurement applications to provide reliable continuous analyses from data obtained in a positive thorough QT study previously characterized with sparse semi-automated measurements performed by an ECG core laboratory.
Continuous RR, QT, QTc measurements, and individual QT/RR relationships and their associated intra- and inter-subject variability were derived in parallel with BioQT, Ponemah PRO, and WinAtrec analysis software.
Despite slight vendor-specific differences in measurement variability and QTc, all machine-read methods demonstrated requisite assay sensitivity and yielded similar conclusions in accordance with SA analysis.
Three commercially available ECG analytical software applications reliably detected the drug-induced QT prolonging effects and replicated the SA core-laboratory conclusions, with greatly improved temporal resolution and reduced analytical costs. With broader experience, these data suggest that current SA methodologies could be effectively replaced by fully automated ECG analysis.
The "thorough QT study": A valid paradigm to test new algorithms for QT interval measurements?
2013, Journal of Electrocardiology
Neonatal ECG screening andQT correction: Themarch towards consistency and accuracy
2018, European Heart Journal
Electrocardiographic data quality in thorough QT/QTc studies
2014, Drug Safety

View all citing articles on Scopus

^☆: No declared financial support or conflicts of interest.

View full text

Pattern recognition analysis of digital ECGs: Decreased QT measurement error and improved precision compared to semi-automated methods☆

Abstract

Background and Purpose

Methods

Results

Conclusions

Introduction

Section snippets

Study design

Raw QT and RR interval analysis

Discussion

Conclusions

Limitations

Heart Rhythm

J Electrocardiol

J Electrocardiol

J Electrocardiol

J Electrocardiol

J Pharmacol Toxicol Methods

J Electrocardiol

Individual patterns of dynamic QT/RR relationship in survivors of acute myocardial infarction and their relationship to antiarrhythmic efficacy of amiodarone

J Cardiovasc Electrophysiol

Study as part of a first-in-man study

J Clin Pharmacol

Dynamic beat-to-beat modeling of the QT–RR interval relationship: analysis of QT prolongation during alterations of autonomic state versus human ether a-go-go-related gene inhibition

J Pharmacol Exp Ther

Contrasting time- and rate-based approaches for the assessment of drug-induced QT changes

J Clin Pharmacol