Lia Liefaard (1)
(1) GlaxoSmithKline, Stevenage, United Kingdom
Objectives: To analyse dose-response count data from two cough challenging compounds, citric acid (CA) and capsaicin (Caps), and to quantify the effect of GSK2339345 on the dose-response relationships, using mixed effects analysis.
Methods: The cough challenge test consisted of subjects receiving increasing doses of CA (2-fold increments; 0.03-4M) or Caps (2-fold increments; 0.49-1000uM), and counting of coughs for 30 seconds following each dose. The effect of GSK2339345 was tested by performing the tests after administration of GSK2339345 or placebo (A/P) in each subject in randomised order [1, 2]. The resulting dose-response datasets had data from 9/9 (A/P) subjects for CA, and from 11/10 (A/P) subjects for Caps.
The dose-response profiles of CA and Caps were analysed by NLME modelling using NONMEM 7.2.0 [3]. The relationship between CA or Caps dose and number of coughs was modelled using an Emax function (gamma fixed to 1). The effect of treatment with GSK2339345 was investigated by implementing proportional differences between A/P in Emax and ED50. Between-subject variability (BSV) was tested as exponential error on model parameters. Poisson and Negative Binomial (NB) distributions were investigated. Model development was guided by log likelihood ratio test (-2LL) and exploratory and diagnostic plots as presented by Zamuner et al [4].
Results: For all models tested (with/without BSV/treatment effect) the NB distr models performed better than Poisson for both CA and Caps. Interestingly, implementing a treatment difference between GSK2339345 and placebo in ED50 for CA or in EMAX for Caps significantly improved the fit for Poisson, but not for NB distr models. The NB distr models without treatment effect had lower Obj F than the Poisson distr models with treatment effect, suggesting there was no real treatment effect, but rather, in the Poisson distr models it described variability within a subject. The exploratory and diagnostic plots showed little difference in the dose-response curves for CA and Caps between A/P, confirming that any treatment effect will be small.
Conclusions: Even if only data from 9-11 subjects were available, the dose-response count data with CA and Caps could be described by an Emax model with Poisson or NB distributions. A small treatment effect of GSK2339345 was identified in the Poisson, but not in the NB distr models; this was likely because of variability within subjects.
Sponsored by GSK (study 117720)
References:
[1] Hilton EC, Baverel PG, Woodcock A, Van Der Graaf PH, Smith JA. Pharmacodynamic modelling of cough responses to capsaicin inhalation calls into question the utility of the C5 end point. J Allergy Clin Immunol (2013) 132:847-55.
[2] Badri H, Satia I, McGarvey L, MarksKonczalik J, Murdoch RD, Cheesbrough A, Warren F, Siederer S, Smith JA. A randomised, double-blind (sponsorunblind), placebo controlled, cross-over study to investigate the efficacy, effect on cough reflex sensitivity, safety, tolerability and pharmacokinetics of inhaled GSK2339345 in patients with chronic idiopathic cough using an aqueous droplet inhaler. British Thoracic Society Winter Meeting (2015) P238.
[3] Beal SL, Sheiner LB, Boeckmann AJ & Bauer RJ (Eds.) NONMEM Users Guides. 1989-2011. Icon Development Solutions, Ellicott City, Maryland, USA.
[4] Zamuner S, Sahota T, Liefaard L. Modelling development for count data: NONMEM vs R. PAGE 24 (2015) Abstr 3629 [www.page-meeting.org/?abstract=3629].
Reference: PAGE 25 () Abstr 5748 [www.page-meeting.org/?abstract=5748]
Poster: Drug/Disease modeling - Other topics