Detection of QT prolongation using a novel electrocardiographic analysis algorithm applying intelligent automation: Prospective blinded evaluation using the Cardiac Safety Research Consortium electrocardiographic database

Abstract

Background The Cardiac Safety Research Consortium (CSRC) provides both “learning” and blinded “testing” digital ECG datasets from thorough QT (TQT) studies annotated for submission to the US Food and Drug Administration (FDA) to developers of ECG analysis technologies. This manuscript reports the first results from a blinded “testing” dataset that examines Developer re-analysis of original Sponsor-reported core laboratory data. Methods 11,925 anonymized ECGs including both moxifloxacin and placebo arms of a parallel-group TQT in 191 subjects were blindly analyzed using a novel ECG analysis algorithm applying intelligent automation. Developer measured ECG intervals were submitted to CSRC for unblinding, temporal reconstruction of the TQT exposures, and statistical comparison to core laboratory findings previously submitted to FDA by the pharmaceutical sponsor. Primary comparisons included baseline-adjusted interval measurements, baseline- and placebo-adjusted moxifloxacin QTcF changes (ddQTcF), and associated variability measures. Results Developer and Sponsor-reported baseline-adjusted data were similar with average differences less than 1 millisecond (ms) for all intervals. Both Developer and Sponsor-reported data demonstrated assay sensitivity with similar ddQTcF changes. Average within-subject standard deviation for triplicate QTcF measurements was significantly lower for Developer than Sponsor-reported data (5.4 ms and 7.2 ms, respectively; p<0.001). Conclusion The virtually automated ECG algorithm used for this analysis produced similar yet less variable TQT results compared to the Sponsor-reported study, without the use of a manual core laboratory. These findings indicate CSRC ECG datasets can be useful for evaluating novel methods and algorithms for determining QT/QTc prolongation by drugs. While the results should not constitute endorsement of specific algorithms by either CSRC or FDA, the value of a public domain digital ECG warehouse to provide prospective, blinded comparisons of ECG technologies applied for QT/QTc measurement is illustrated.

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

“…As anticipated based on the comparison of the single-deltas there are almost no discernable differences in the time-profiles, with 63% of the time-points having less than 1 ms difference (86% within 2 ms) and all the confidence intervals overlap. These results are similar to numbers reported by Green et al (7 out of 11 time-points) for comparison of QT algorithms to detect effects of moxifloxacin [26]. The exposure-response relationships obtained from automated and semi-automated measurements for dofetilide and quinidine (Fig 5) and ranolazine and verapamil exhibited overlapping population predictions (Fig 6).…”

“…Moreover, the majority of the outliers are from two subjects near maximum concentration of dofetilide and quinidine that have extreme QTc prolongation (~80 ms) far beyond the normal level of regulatory concern (~10 ms). These results are comparable to the difference between two QT algorithms for detecting moxifloxacin effects, where a mean difference of <1 ms was observed with a SD of 9 ms [26]. Furthermore, in FDA study 2, which had less substantial QTc prolongation compared to FDA study 1, but still prolongation of ~40 ms, the average single-deltas difference were within 1 ms for all intervals with a SD of <3 ms.…”

“…The differences between the automated and semi-automated single-delta measurements are <1 ms for all intervals with a standard deviation of the difference in most cases being <3 ms, comparable to differences between two QT algorithms to detect moxifloxacin [26]. In addition, the intra-time point variability was not different between the two algorithms.…”

Sources of QTc variability: Implications for effective ECG monitoring in clinical practice

“…Similar to the necessity of having serial ECGs interpreted by the same observer, ECG measurements by different equipment may lead to substantial variability (Kligfield et al, , ). However, since the algorithms used in the commercial equipment are steadily advancing (Green et al, ) their clinical reliability is now probably at least equivalent to fully manual measurements especially if combined with visual verifications to eliminate occasional outliers (Hnatkova et al, ).…”

The Role of Public–Private Partnerships in Catalyzing the Critical Path

“…Currently, this database includes ECGs derived from moxifloxacin (the positive control in TQT studies), placebo data sets, and the congenital long QT syndrome data submitted to CSRC . Thus far, proof‐of‐principle studies have demonstrated the value of the ECG Warehouse for developing algorithms that assess drug‐induced QT prolongation potential …”