Population analysis of MG-ADL total score for efgartigimod Phase 3 study in myasthenia gravis patients: application of a bounded-integer model
Marostica E (1), Ahsman M (1*), Van Bragt T (3), Noukens J (3), Vis P (1), T’joen C (2), Ulrichts P (2), Guglietta A (2), Rossenu S (2), van Steeg T (1)
(1) LAP&P Consultants B.V., (2) argenx BV, (3) Curare Consulting BV, * current affiliation: MSD LLC
Objectives: Efgartigimod (ARGX-113) is an antibody fragment and a neonatal Fc receptor (FcRn) antagonist that has been developed for the treatment of patients with severe autoimmune diseases mediated by pathogenic immunoglobulin G (IgG) autoantibodies. In patients with generalized myasthenia gravis (gMG), efgartigimod lowered antibodies against the acetylcholine receptor (AChRAb). To support the development of efgartigimod for gMG, models were built to describe PK, total IgG and AChRAb reduction, which are linked to clinical outcome: Myasthenia Gravis Activities of Daily Living (MG-ADL) total score. This is a standardized 8-item patient-reported scale used to assess MG symptoms and effects on daily activities. The objective of the analysis was to support further development of efgartigimod by means of a model which was able to predict the responder rates after placebo or efgartigimod treatment in the ADAPT Phase 3 study ARGX-113-1704 [1].
Methods: Data from the Phase 3 study (ARGX-113-1704 [1]) in patients with gMG receiving multiple cycles of four once-weekly 10 mg/kg efgartigimod IV infusions were analyzed. Efgartigimod PK was described through a three-compartmental model with linear clearance. Total IgG model consisted of an indirect response turnover model, in which efgartigimod stimulated the degradation rate of total IgG (kout). An Emax model was used to describe the efgartigimod effect on total IgG, which was directly linked to the reduction of AChRAb. In turn, AChRAb reduction was linked to MG-ADL score using a continuous approach. Further, a placebo model was incorporated to capture the reduction in the score over time due to placebo. Inter-individual variability (IIV) was identified for the baseline score and the drug effect parameter. In addition, a categorical approach was also investigated to describe MG-ADL score. A bounded integer model was used [2], in which the baseline reflects the probability of MG-ADL score being in a specific category. As compared to the continuous approach, the AChRAb concentration, instead of the delayed one, was used as a driver for the score. A Markovian component to account for serial correlation between observations was also estimated.
Results: PK, total IgG, and binding AChRAb models adequately described observations from study ARGX-113-1704 and parameters were precisely estimated [3]. Both the continuous and categorical models adequately described MG-ADL score data and their IIV across treatment cycles in ARGX-113-1704. For both models, parameters were precisely estimated and they were used to simulate 1000 individual MG-ADL score profiles, based on which the median and the 5th and 95th percentiles of the responder rate in cycle 1 were derived. With the continuous model, the median responder rates were predicted to be lower than the observed ones in the Phase 3 study, i.e. 21.7% (95% PI: 13.3%, 31.4%) vs 37.3% [1] and 54.8% (95% PI: 44.0%, 65.5%) vs 67.9% [1] for placebo and efgartigimod, respectively, whereas the responder rates predicted with the bounded integer model were in line with the observed ones, i.e. 38.6% (95% PI: 27.7%, 48.2%) vs 37.7% [1] and 65.5% (95% PI: 54.8%, 76.2%) vs 67.9% [1] for placebo and efgartigimod, respectively. Since MG-ADL score data equal to zero were not adequately captured, the responder rates were not adequately predicted with the continuous model. The presence of zero values makes the distribution of MG-ADL score data censored and thus a continuous approach, which assumes normality of the data, can lead to biased results. The bounded integer model solved this issue. Further, the Markovian component, providing a higher probability that an observation has the same value as the previous one in time [2], provided a better description of the data, as well as match between predicted and observed responder rates.
Conclusions: The bounded integer model was able to adequately describe the MG-ADL score data and predict the responder rate in patient with gMG receiving either placebo or efgartigimod. As such, the developed bounded integer model was suitable to perform simulations to support further development of efgartigimod in patients with gMG. In addition, the model supports the link between autoantibody reduction and clinical improvement.
References:
[1] argenx BV. Clinical Study Report ARGX-113-1704: A Randomized, Double-Blind, Placebo-Controlled, Multicenter Phase 3 Trial to Evaluate the Efficacy, Safety and Tolerability of ARGX-113 in Patients with Myasthenia Gravis Having Generalized Muscle Weakness (ADAPT), 18 August 2020.
[2] Wellhagen GJ, et al. A Bounded Integer Model for Rating and Composite Scale Data. The AAPS Journal. 2019; 21(74):1-8.
[3] Marostica E, et al. Population PK/PD analysis for efgartigimod Phase 3 study in myasthenia gravis patients, PAGE Meeting 2023.
