Aims: It is usually more difficult to build a PK-PD model with discrete symptomatic disease endpoints than a PK-PD model with continuous biomarkers because a symptomatic endpoint has often discrete scores with unbalanced scales. A static Investigator’s Global Assessment (sIGA) is one of the primary efficacy endpoint of atopic dermatitis evaluation and a basis for FDA approval of atopic dermatitis evaluation in clinical study. It has a discrete score (0, 1, 2, 3, 4 or 5) and had a difficulty to apply the model for continuous values like indirect turnover model. A continuous-time Markov exposure-response model has been published by Lacroix et. al. in 2014 to model ACR20, 50 and 70 which are the discrete efficacy endpoints in rheumatoid arthritis area simultaneously (1). The model includes Markov element and it enables to consider individual time course of discrete scores. We hypothesized that the sIGA could be described with kind of continuous-time Markov exposure-response model and has developed the model for future clinical study design optimization based on simulated longitudinal sIGA scores.
Methods: A population PK-PD (sIGA) model was implemented to the data from the patients with atopic dermatitis in randomized double blind phase 2 study (ClinicalTraials.gov Identifier: NCT01986933) which included 5 arms (0.1, 0.5, 2 mg/kg Q4W, 2 mg/kg Q8W and placebo) up to 84 days after the first administration using a first-order conditional estimation method in NONMEM version 7.3.0. Modified continuous-time Markov exposure-response model was applied to sIGA with above 5 categories. Emax model was used to link the relationship between observed serum nemolizumab concentration and sIGA. Visual predictive check was performed to confirm a predictability of the model and bootstrapping method was utilized to estimate median and 90% confidence interval of the estimated parameters.
Results: The data of 1488 sIGA observations from 264 patients with severe to moderate atopic dermatitis was applied to the continuous-time Markov exposure-response model. The run has been converged successfully. The sIGA score showed improvement with the nemolizumab treatment and the rates of the patients who were in each category were calculated and compared to the simulated outputs as visual predictive check. The model described the observed data well. Population mean (90% bootstrapped range) of Emax on transfer rate constant of disease improvement, that of disease progression and EC50 were 121% (53.9% to 217%), -32.3% (-66.5% to -4.66%) and 1830 (749 to 5490) ng/mL, respectively. The mean serum nemolizumab trough concentration in 2 mg/kg arm was 8110 ng/mL and it was suggested that almost maximum effect was obtained between 0.5 and 2 mg/kg dose.
Conclusion: The sIGA scores were well described by the continuous-time Markov exposure-response model. The model had a longitudinal predictability of the sIGA scores with any dose regimen of nemolizumab. Continuous-time Markov exposure-response model is relatively accessible to implement and one of the recommended option to fit a symptomatic and categorical endpoint with discrete scores.
Lacroix BD, Karlsson MO, Friberg LE. Simultaneous exposure-response modeling of ACR20, ACR50, and ACR70 improveement scores in rheumatoid arthritis patients treated with certolizumab pegol. CPT Pharmacometrics Syst. Pharmacol. 2014 Oct 29;3:e143.