2004D-0377 - International Conference on Harmonisation; Draft Guidance on E14 Clinical Evaluation of QT/QTc Interval Prolongation and Proarrhythmic Potential for Non-Antiarrhythmic Drugs and S7B Nonclinical Evaluation of the Potential for Delayed Ventricular Repolarization (QT Interval Prolongation) by Human PharmacInternational Conference on Harmonisation; Draft Guidance on E14 Clinical Evaluation of QT/QTc Interval Prolongation and Proarrhythmic Potential for Non-Antiarrhythmic Drugs and S7B Nonclinical Evaluation of the Potential for Delayed Ventricular Repolarization (QT Interval Prolongation) by Human Pharmaceuticals; Reopening of Comment Periodseuticals; Reopening of Comment Periods
FDA Comment Number : EC9
Submitter : Dr. Alfred Balch Date & Time: 02/23/2005 10:02:10
Organization : Novartis Pharmaceuticals Biostatistics
Category : Drug Industry
Issue Areas/Comments
GENERAL
GENERAL
Comments from Novartis Biostatistics

(Section 2.1.2) After 'However, drugs that prolong the mean QT/QTc interval by around 5 ms or less do not appear to cause TdP. On that basis, the positive control (whether pharmacological or non-pharmacological) should be well-characterized and consistently produce an effect corresponding to the largest change in the QT/QTc interval that is currently viewed as clinically not important to detect (a mean change of around 5 ms or less)...'

This statement appears to require that for all (or most) timepoints, that average prolongations for these timepoints are less than 5ms. This restriction is too stringent. It is also in contradiction with the 8ms rule that appears later in the guidance. Because of large variability of the primary endpoint by chance we will expect to see some timepoints where the mean change from baseline exceeds 5 msec. If means of enough groups of subjects (samples) at many different timepoints are measured from a population with true population mean close to 5ms it is virtually certain that some of the sample timepoints will have sample means in excess of 5 ms. By contrast, in this situation, a sufficiently large (number of subjects) experiment will still exclude (via confidence interval) a population average over 8 ms which has here been defined as the statistically- based safety threshold. Additionally, even in cases where there is no prolongation whatsoever, the number of subjects, required so that each timepoint's sample mean does not to exceed 5 ms is quite large.

The criterion to claim no meaningful drug-induced QT-prolongation may be too stringent.The primary endpoint for central tendency analysis,QT/QTc change-from-baseline at timepoint showing the 'largest time-matched mean difference' may not be fully informative as a primary endpoint. Selection bias of the time-points de-links the exposure-response relationship because the largest difference in QTc change from baseline may be the result from the low placebo value. The variability at a selected timepoint may be quite high, requiring a large sample size to maintain a sufficient statistical power. In addition, this endpoint is associated with multiplicity issue that discourages a design with more ECG time-points. A more useful criterion could be based on a maximum of a small and defined number of timepoints. The associated threshold for the upper limit of the two-sided 90%CI for the difference of primary endpoint from placebo of 8 ms may be too low, considering the high variability of the endpoint.

The commonly used positive control, for example, Avelox, normally shows a QTc prolongation, compared to placebo, of greater than 12 msec. This effect size is much larger than the 5ms in the guidance it is further exacerbated by taking the maximum value. The criterion or the endpoint should be changed to be more consistent with the prolongation in this very safe drug.
2.1.2 The term 'baseline' should be clarified and and it should be made clear what baseline value should be used for change from baseline. Research from Wang, et al. (2004) suggests that a time-averaged baseline should be appropriate.
2.1.2, 8th paragraph: Latin square and incomplete block designs should be mentioned as special cases of a crossover design that can overcome some of the problems. We suggest the text

'Crossover studies, including Latin square and incomplete block designs have some potential advantages:'

We also suggest the following be added: 'Alternatively, the use of a crossover design may remove the need for separate baselines if direct comparisons between test and reference are calculated'. At end of this paragraph add 'The ways to reduce intrinsic variability should include replicates at each time point, e.g., multiple ECGs extraction within a few minutes and measurement of multiple heartbeats. Rationale- Replicates should be clearly distinguished from multiple measurements separated by longer time intervals.