Separating Signal from Noise: PK/PD modelling of QT-interval prolongation

1
Separating Signal from NoisePK/PD modelling of
QT-interval prolongation Anne Chain1, Lutz
Harnisch2 Oscar Della Pasqua1
1 CPK MS Research Group , Clinical
Pharmacology Discovery Medicine,
GlaxoSmithKline, Greenford, UK. 2 Clinical
Pharmacokinetics MS, Clinical Pharmacology
Discovery Medicine, GlaxoSmithKline, Greenford,
UK.

• BACKGROUND
• The presence of a prolonged QT-interval has
  become an identifier for the risk of a unique
  from of polymorphic ventricular tachycardia,
  Torsade de Pointes (TdP). Since this finding can
  be a serious safety issue, policies and
  guidelines have been proposed to ensure that the
  effects of non-cardiovascular drugs on
  QT-interval are accurately characterised. Such
  policies have assumed that ECG measurements are
  highly reproducible. However, there is
  convincing evidence from clinical research that
  QT-interval assessments can show high variability
  if considered over a widely spanned time course.
  Therefore, any meaningful attempt to characterise
  drug-induced changes in ECG parameters requires
  identification of variability sources, as they
  will have major impact on clinical study design
  and sample size.
• OBJECTIVE
• The primary objective of this investigation was
  to develop a pharmacokinetic / pharmacodynamic
  (PK/PD) model to describe the time course and
  variability of QT-interval in healthy subjects.
  In addition, it was our aim to establish the
  relevance of external factors on the accuracy and
  reproducibility of ECG measurements.
• CLINICAL STUDY
• 30 healthy subjects were given a single oral dose
  of 160 mg d,l-sotalol (SOTACORTM), a beta-blocker
  well known to produce clinically significant
  QT-interval prolongation, according to a
  double-blind, randomised, placebo-controlled,
  crossover study design. Pharmacokinetic sampling
  was performed at various times up to 24h after
  dosing. 12-lead ECG was monitored continuously
  throughout the study and recording were made at
  different time points before and after dosing.
  QT-intervals were read from automated recordings
  and manual assessments, as defined by a
  cardiographer. The following factors have been
  specifically controlled during the study to
  minimise the effect of variability of QT-interval
  assessments
• - Wake-up time
• - Meal time
• - Blood sampling time relative to ECG
  assessments
• - Temperature of all meals and beverages (room
  temperature 20 - 25)
• - Room temperature
• - Assignment of nurse
• - Skin preparation
• - Location of ECG lead placement on chest
• - Supine position (angle 35)

RESULTS We have derived a population-based
correction factor to estimate the QT/RR relation
and subsequently quantify drug effect on
corrected QT-interval. Initially, we found
clusters derived from automated measurements
which cause major discrepancies in the
reproducibility of recordings. An iterative
mixture model was implemented to account for data
clustering and estimate a QT/RR relation for each
subject. A direct effect model using an Emax
function was sufficient to characterise the QT
response on d,l-sotalol exposure. Pharmacokinetics

PK/PD
Posterior distribution presented as background
density Red (10 around mean), green (95 CI)
Parameters CL L/hr VSS L KA hr-1
Median 9.39 147 1.27
CV 8.79 7.58 143
95 CI 6.33, 13.9 69, 313 0.647, 2.47
IIV 17.15 12.73 57.36
95 CI 3.16, 24.04 0, 38.39 0, 85.15
IIV Inter-individual variability. CV
refers to the accuracy of the parameter estimates.
Fig. 2 PK modelling results.
Pharmacodynamics
Parameters SLP INCPT ms EC50 ng/mL EMAX HILL
Median 0.306 389 3460 40 fix 1.66
CV 3.02 0.517 24.5 0 17.7
95 CI 0.288, 0.324 385, 393 1760, 5160 - 1.07, 2.25
IIV 19.75 3.332 114.9 180.6 -
IOV 7.823 1.233 - - -
Fig. 6 Final PK/PD modelling results.
Final Evaluation Effect size
Using the final model estimates and the following
equation, the effect size for sotalol in this
study is calculated to be 31 ms (95 CI 13 - 54)
at a concentration of 1500 ng/mL (max conc.) and
it is 6 ms (95 CI 1.2 - 18) at 500 ng/mL (mean
conc.).
Fig. 3 Raw QT data showing clustering in
automated readings (black) and its inconsistency
compared to manual readings (red).
Iterative Model Placebo automated ECG
Type I and Type II errors
Parameters SLP INCPT ms Delta ms
Median 0.447 401 15.9
IIV 4.405 2.347 27.11
IOV 19.47 1.992 35.64
PK MODEL
IOV Inter-occasion variability.
Fig. 4 Results from the iterative model with
placebo automated ECG.
Iterative Model Placebo manual ECG
PK/PD MODEL
q?2(p0.99, df3) 11.3
Fig. 7 Results from Type II error for assessing
power of a given trial.

• CONCLUSIONS
• This PK/PD modelling effort allowed us to
  describe the QT/RR and its relationship to
  d,l-sotalol exposure in an integrated manner.
• The iterative model used to describe the
  individual clustering effect in the automated ECG
  readings resulted in a similar level of precision
  that is obtained from manual readings.
• The precision of the correction factors in
  estimating the correlation between QT and RR was
  best described in the following order individual
  gt Fridericia gt Bazett gt no transformation.
• Power simulations showed the importance of PK/PD
  modelling to optimise trial size and mitigate a
  potential QTc liability for new chemical entities.

Parameters SLP INCPT ms
Median 0.337 392
IIV 20.45 3.45
IOV 15.43 1.811
Fig. 1 A schematic diagram of the stepwise
modelling approach.
Fig. 5 Results from the iterative model with
placebo manual ECG.