In the paper, titled “Evaluation of QT Liability for PF‐05251749 in the Presence of Potential Circadian Rhythm Modification,”
the authors presented results from assessment of QT‐liability based on (i) clinical data and (ii) simulations using models for a drug that affects QT circadian rhythm.Regarding part (i), we question the conclusion that the assumptions of the prespecified linear mixed effect (LME) model from the scientific whitepaper
were satisfied:It is inconclusive whether the heart rate (HR) effect on QT may have been corrected by the Fridericia method (QTcF). In Figure S1A,
ΔHR appears to be above the threshold of 10 bpm at a few observations. Because placebo and baseline adjusted HR plots are not provided, it is difficult to evaluate the appropriateness of the use of QTcF.The presence of hysteresis cannot be excluded, as shown in Figure S1C
at the two highest doses.The relationship between concentration and QTcF does not appear to be linear (Figure S1D
).The authors did not address the whitepaper requirement for pooling data from different studies (e.g., heterogeneity evaluation).For the reasons listed above, the prespecified LME model should not have been applied to the PF‐05251749 data. The authors adequately applied an alternative nonlinear mixed effect (NLME) model, which accounts for drug‐induced circadian rhythm change. We do not question the interpretation of the results from the NLME model. However, application of the prespecified LME model was not properly justified during the assessment. Therefore, one cannot compare the results from the NLME and whitepaper LME models.Regarding part (ii), the authors investigated via simulations if drug‐induced QTc circadian rhythm changes may affect the inference of C‐QTc slope estimate and the false‐negative rate in detecting QTc liability with the LME and NLME models. In this scenario, we argue that the assumptions of the LME model should be met, even for simulations studies. For example, the whitepaper indicates that the prespecified LME model should not be used for drugs that show “PK/PD hysteresis.”
If QT circadian rhythm change due to a drug is detected, one should develop a model that best describes such data. That model may be an NLME model that accounts for circadian rhythm, which may result in a different conclusion from the LME model.In summary, the “general principles for model development and evaluation”
described in the whitepaper does not support use of the LME model in cases when basic assumptions are not met. Therefore, the call for reconsideration of the LME model if circadian rhythm change is expected based on the pharmacology of a drug is unwarranted.