One example of a battery of nonclinical tests was proposed by Bass and colleagues in a 2008 publication discussing an initiative from the Health and Environmental Sciences Institute (HESI), a division of the International Life Sciences Institute (Bass et al. 2008). The authors noted that “the critical challenge in the pharmaceutical industry today is to identify experimental models, composite strategies, or biomarkers of cardiac risk that can distinguish a drug which prolongs cardiac ventricular polarization but is not proarrhythmic, from one that prolongs the QT interval and leads to TdP.” They observed that a problematic feature of the biomarkers of proarrhythmic risk discussed in both ICH S7B and ICH E14 is their low specificity and hence “the potential for promising new test agents to generate false-positive results with a high frequency,” an issue discussed in more detail in due course.

The paper by Bass and colleagues summarized the objectives of a workshop entitled “Moving towards better predictors of drug-induced Torsade de Pointes” that brought together experts in the field to work collaboratively towards a better fundamental understanding of the emerging science, trends, and methodologies relating to the prediction of drug-induced TdP. These objectives included the following (Bass et al. 2008):

The authors concluded that there was a need for both in vitro and in vivo TdP proarrhythmia models to help increase knowledge of arrhythmogenic mechanisms and substantially improve the predictive value of a nonclinical battery of tests for the clinical proarrhythmic liability of new drugs. They proposed a multifaceted strategy employing test systems of increasing complexity (i.e., concurrently assessing several parameters at a time), including single cells, isolated cardiac tissue, isolated hearts, intact animals, and diseased animals.

Acquisition of high-fidelity digital ECGs is achieved in one of two ways: employment of a 12-lead resting ECG device that records 10-s strips of ECG signals or use of a 12-lead Holter device that records continuous ECG signals in ambulatory participants for 24 h or longer. At the time of drafting ICH E14, ambulatory ECG monitoring was not considered to be sufficiently well validated to be used as the primary methodology employed in the assessment of drug-induced effects on the QTc interval. Therefore, a significant number of early TQT studies used 12-lead resting ECG recordings. Given the advances in Holter ECG recording technology, it has been demonstrated that Holter ECGs can be as accurate as 12-lead digital ECGs in the assessment of drug-induced changes in QT and RR intervals. Consequently, use of 12-lead Holter recorders in TQT studies steadily increased. TQT studies therefore now often employ Holter devices for collection of ECG waveforms, since this approach enables continuous digital waveforms to be recorded and short sections of 12-lead ECG strips to be extracted at prespecified study time points. This approach incorporates the advantage of selecting ECGs that are relatively free of artifacts and that are attained at a stable heart rate. Moreover, these 24-h recordings have a 50–100 times higher likelihood of detecting transient cardiac arrhythmias than conventional 10-s ECGs (Min et al. 2010).

The digital ECG waveform can be considered as a series of dots, with the precision of measurement afforded corresponding to the sampling frequency of the ECGs. Digital ECG recorders with a sampling frequency of 500 Hz (samples that are 2 msec apart) to 1000 Hz (samples are 1 msec apart) are optimal. In early TQT studies employing 12-lead Holter ECGs, the sampling rate employed was considerably lower, around 180 Hz, because of technological limitations at that time. However, Holter recorders used more recently have sampling rates of 500 Hz and 1000 Hz, equal to those of 12-lead resting ECG devices (Panicker et al. 2010).

Consistency in operator techniques relating to skin preparation, ECG lead placement, and the study participants’ position are important components of the collection of good-quality ECGs, since inconsistencies in recording methodology add to variability in ECG parameter values. Site training to provide clear instructions that ensure participants rest in a supine position for several minutes prior to the ECG acquisition time points to ensure a stable resting heart rate is essential: otherwise, activity-related heart rate changes could be falsely attributed to the study drug. Additionally, ECG recording must precede the drawing of blood for pharmacokinetic (PK) samples for exposure–response analysis, a topic discussed in the following chapter.

In studies using 12-lead resting ECG devices, the digital ECG files recorded at the site are transmitted via an in-built modem to the ECG receiving system on the server of the lab where they will be analyzed. After the confirmation of demographics, ECG files are converted into FDA-compliant Extensible Markup Language (XML) files that are conveyed to the ECG analysis software at the lab. When employing 12-lead Holter devices, the digital ECG data are recorded on a removable memory card that may be sent via a courier to the lab. The digital Holter file is downloaded and used for the extraction of discrete 10-s ECG files that are analyzed by a process similar to that used for 12-lead resting ECG. Alternatively, the digital Holter recording from the removable memory card may be transferred to the investigational site computer and transmitted to the lab’s server via a secure web upload.

The labs that analyze ECGs collected during a TQT study are highly specialized for this purpose and are referred to as core ECG labs. This nomenclature is similar in intent to the term central lab: central labs are used to analyze different types of biological samples (e.g., blood and urine) taken during many large clinical trials. The name central lab refers to testing labs to which samples from all of the investigational sites in a clinical trial are shipped. This strategy has two important advantages compared with using a large collection of local labs, i.e., laboratories close to each investigational site:

cGCP guidances require that all labs have data-audit trails, standard procedures, trained staff, archives of samples and data, and routine quality assurance inspections (Prokscha 2007). If multiple labs were used, assurance would be required that cGCP requirements were met at every one. Use of a single central lab makes assurance of cGCP much simpler.

Analysis of data that have been collected from many sources and then pooled into one data set can lead to considerable statistical problems. Since the majority of lab measurements are surrogates and reference ranges can vary from analytical method to method, many different types of errors can occur during lab testing. These can be due to variation in many aspects of data collection, including the technician, an instrument, the environment, and the reagents used. While statistical approaches to standardize values from several local laboratories, each with their own reference ranges, have been described (Chuang-Stein 1992), central labs provide a real statistical advantage: all samples are analyzed in the same manner in the same laboratory, thereby providing a much greater degree of fidelity. Thus, the role of the ECG core lab is not restricted just to ECG analysis as discussed in the next section but also encompasses aspects of ECG acquisition and processing as well as submission of the annotated digital ECG data to regulatory agencies.

ICH E14 recognizes that the threshold of regulatory concern in a TQT study is small, i.e., a mean QT prolongation of about 5 msec. Given that the length of a “typical” QT interval is in the order of 400 msec, the increase of interest is in the order of 1 %. Detecting such a small change requires a high degree of accuracy and precision in ECG measurement. ICH E14 recommends skilled readers without specifying the specific training needed. A technician performing the initial read followed by a cardiologist overread is considered adequate. It is the responsibility of the core lab to have appropriate criteria for reader selection and training and benchmarking their competence (Panicker et al. 2009). ICH E14 further states that the ECGs must be analyzed by a small number of readers and that all ECGs from the same participant be read by the same reader (blinded to treatment arm) to maintain consistency in QT measurement. A specific assessment of reader variability is also required, with a subset of the ECGs being reread to quantify inter- and intra-reader variability. Each core lab should have a well-defined approach to quantifying reader variability (Salvi et al. 2014). Without appropriate care, a QT interval measured in the same ECG by the same reader blinded to his or her previous measurement may vary up to 25 msec (Malik and Camm 2001).

The method of QT interval measurement is also of great importance (Salvi et al. 2010). ICH E14 recognizes that there are several different methodologies for the measurement of ECG intervals which differ in “what to measure” in terms of items and characteristics including conventions for lead selection, waveform presentations, definition of T-wave offset, and handling of U-waves. It is not prescriptive about the method chosen but emphasizes the need to use a consistent approach. In this context, the use of a positive control such as moxifloxacin, a drug whose effects on ECG waveforms are very well known, is an internal validation of the QT interval measurement methodology and the accuracy of the ECG readers.

A core lab’s ECG analysis system also enables various techniques of measurement. ICH E14 recommends one of the following approaches for TQT studies:

Regardless of which technique is used, readers should be trained in, and follow, standard operating procedures with prospectively defined criteria on how to place or correct fiducials for interval measurement. Although fully automated methods are available and offer the advantage of being consistent and reproducible, they have not yet been sufficiently validated and can yield misleading results, especially for less-than-good-quality ECGs and in the presence of morphologic or rhythm abnormalities. However, with improvements in computer algorithms, there are now validated computer-assisted techniques, often described as highly automated reading, wherein intervals are measured by a computer algorithm that also provides confidence scores based on the quality, morphology, and measured values. Only select ECGs exceeding prespecified quality cutoffs then undergo human overread. The ECG readers also manually assess the shape of the T wave for any drug-induced changes and for the presence of other diagnostic features such as cardiac arrhythmias and conduction abnormalities. A combination of validated technology solutions, trained readers, and scientifically rigorous methods is therefore essential to ensure high-quality ECG data.

QTc values are used in analysis of potential drug-induced QT interval prolongation, where the “c” stands for corrected for heart rate. Many correction factors have been developed to address this issue (see Camm et al. 2004). The more commonly used correction factors, QTcB (Bazett’s correction) and QTcF (Fridericia’s correction) employed in this manner date back almost 100 years in the context of clinical practice.

When discussing the history of the reasons behind the choice of the preferred formula for adjusting QT interval values, Camm and colleagues commented as follows (Camm et al. 2004, p 44):

Of all the formulas used in the past, the most commonly used are Bazett’s square-root formula (QTc = QT/RR^1/2) and Fridericia’s cube-root formula (QTc = QT/RR^1/3). Between the two, Bazett’s formula is more commonly used [in clinical practice] and most reported values are given using Bazett’s formula because of its simplicity (most simple calculators have a function for a square root but not for a cube-root computation, which gives a practical advantage to Bazett’s over Fridericia’s correction).

Bazett’s correction is still the most commonly used correction factor in clinical practice: most ECG machines used in clinical practice settings automatically provide this correction.

Bazett’s and Fridericia’s corrections essentially aim to normalize the QT interval to a heart rate of 60 bpm. The nomenclature (and notation) used for these two corrections is as follows, where RR is the interval between the R wave of one ECG and the R wave of the subsequent ECG:
and
ICH E14 proposed that several values should be reported when presenting the results from a TQT study. These include the following:

However, QTcB is now widely accepted to be an inferior method of correcting for heart rate, and using QTcF is considered appropriate in most situations.

An alternative correction employs linear regression techniques to normalize the QT interval to a heart rate of 60 bpm. ICH E14 discusses a correction calculated in this manner in the Framingham study (see Kannel and Sorlie 1975; Tisdale, et al. 2007). Regression approaches can also be used from other large databases in a similar manner: employing pretreatment data from a study, QT interval measurements are regressed on (1-RR) to obtain estimates of the slope for the following equation:
Another correction methodology employs linear regressions that are calculated for each individual participant in a TQT study: this methodology yields the individual corrected QT or QTcI. Pretreatment QT and heart rate data are used to fit a separate regression for each participant. This means that a slope coefficient is applied to each participant on an individual basis, rather than using the same coefficient in a population-level regression approach. The employment of QTcI has been advocated as the primary endpoint in TQT studies where the drug being tested is known to exert notable influence on heart rate. However, when QTcI is employed as the primary endpoint, it is good practice to report QTcB (for historical reasons) and QTcF as well (Litwin et al. 2008).