Silvia Maria Lavezzi (1), Maurizio Rocchetti (2), Paolo Bettica (3), Stefania Petrini (4), Giuseppe De Nicolao (1)
(1) Dipartimento di Ingegneria Industriale e dell'Informazione, Università degli Studi di Pavia, Pavia, Italy, (2) Independent consultant, Milan, Italy, (3) Italfarmaco S.p.A., Cinisello Balsamo (MI), Italy, (4) Confocal Microscopy Core Facility, Research Center, Bambino Gesù Children's Hospital, Rome, Italy
Objectives: Limited treatment options are currently available for Duchenne Muscular Dystrophy (DMD), the most common muscular dystrophy in childhood [1]. To objectively assess the efficacy of possible new treatments, muscle fibers cross sectional areas (CSAs) can be measured via skeletal muscle biopsies, at baseline and at least once after treatment. Being biopsy a destructive procedure, the measured tissue specimen changes on each occasion. As pre- and post-treatment fiber CSAs cannot be paired, their empirical distributions in the two occasions are compared. Here, it is shown how distributional data can be analysed within a population approach framework. In particular, the histological effects of the HDAC1 inhibitor Givinostat are assessed, based on data collected in a DMD clinical trial [2].
Methods: In the phase II DMD clinical study (sponsored by Italfarmaco S.p.A; study identifier: NCT01761292), 20 boys aged from 7 to For each patient, pre- and post-treatment CSA empirical distributions are compared in order to assess Givinostat-induced changes from baseline. A hierarchical statistical approach is proposed to identify CSA distributions and estimate drug effect on CSA, both (i) for each patient (lower level of the hierarchy) and (ii) in the population (higher level).
Lower level.
For the single patient, each fiber is seen as a sample drawn from a population of fibers. The pre- and post-treatment distributions of log-transformed CSAs (logCSAs) are modelled as two-component Gaussian mixtures. Drug effect on logCSAs is described by a shift, corresponding to a multiplicative factor on CSAs, intended as a drug potency parameter. Mixtures and shift are estimated in NONMEM 7.3 (FOCE). Muscle composition data are used for model validation: post-treatment tissue fractions (muscle fiber/fibrotic/fat/necrotic) are model-derived and compared to observed ones.
Higher level.
Typical values for the patient population and inter-patient variabilities are computed via Global Two Stage (GTS) and their uncertainty evaluated via boostrap.
Results:
Lower level.
The model describes well the distributional data. For all patients the drug potency parameter is >1 (shift>0) i.e. the CSAs increase following Givinostat treatment. All parameters are reliably estimated (CV%<~30% for all patients). The model-derived post-treatment muscle composition values are comparable to the observed ones.
Higher level.
Due to dose reductions in study Part 2, computation of typical parameters and inter-patient variabilities via GTS was performed separately on Group A (dose=25 mg BID) and Group B (37.5 mg BID). Drug effect was found to be stronger in Group B (drug potency parameter=2.05, vs 1.57 in Group A). Typical mixture parameters are comparable between the two groups. Inter-patient variability (CV%) in Group B is at most around 30%, while in Group A it exceeds 50% for some parameters. According to bootstrap, the uncertainty in typical values estimates has maximum CV%~5% for both groups. Uncertainty in inter-patient variabilities is comparable between Group A and B (CV%~20%), except for two terms, whose uncertainty is higher in Group B (CV%>80%).
Conclusions: The analysis of muscular distributional data has been addressed via a hierarchical statistical approach that has been applied to assess Givinostat therapeutic effects on DMD patients. Both the distribution of fiber CSA and the drug potency (summarized by a multiplicative factor on fiber sizes) have been characterized. At the lower level, pre- and post-treatment logCSA distributions were described via two-component Gaussian mixtures, identical apart from a shift. Drug potency was estimated as >1 for all subjects, indicating that Givinostat has positive histological effects; for Group B (higher dose), drug potency was stronger with respect to Group A. The predictive performance of the model was confirmed by muscle composition data. The validity of this novel modeling framework may extend to several other contexts (other diseases/drugs/measures) where outcomes are obtained as distributional data.
References:
[1] Mah JK, Korngut L, Dykeman J, Day L, Pringsheim T, Jette N (2014) A systematic review and meta-analysis on the epidemiology of Duchenne and Becker muscular dystrophy. Neuromuscul Disord 24:482-491
[2] Bettica P, Petrini S, D’Oria V, D’amico A, Catteruccia M, Pane M, Sivo S, Magri F, Brajkovic S, Messina S, Vita GL, Gatti B, Moggio M, Puri PL, Rocchetti M, De Nicolao G, Vita G, Comi GP, Bertini E, Mercuri E (2016) Histological effects of givinostat in boys with Duchenne muscular dystrophy. Neuromuscul Disord 26:643-649
Reference: PAGE 27 (2018) Abstr 8768 [www.page-meeting.org/?abstract=8768]
Poster: Methodology - Other topics