How disability progression is correlated with serum neurofilament light chain dynamics in relapsing remitting multiple sclerosis patients treated with Alemtuzumab: results from a mechanistic drug disease joint model
Tom Chebassier1,2, Hoai-Thu Thai2, Vincent Thuillier3, Sophie Fliscounakis-Huynh4, Julie Bertrand1
1Université Paris Cité and Université Paris Sorbonne Paris Nord, Inserm, IAME, F-75018 Paris, France, 2Sanofi, Translational Medicine Unit, Quantitative Pharmacology, Disease modeling, Sanofi, 94250 Gentilly, France, 3Clinical modeling & Evidence Integration, Sanofi, Gentilly, France, 4Sanofi, Translational Medicine Unit, Quantitative Pharmacology, Disease modeling, Sanofi, Gentilly, France on behalf of IT&M Stats, 94250 Gentilly, France
Introduction: Relapsing Remitting Multiple Sclerosis (RRMS) is an auto-immune disease characterized by demyelinating plaques in the central nervous system[1]. If decreasing relapse rate is the main target of drug development, delaying disability progression is an unmet medical need[2]. Currently, no blood biomarker is approved by regulatory agencies to predict disability progression. However, serum Neurofilament light chain (sNfL) has emerged as a promising candidate for early prediction of disease progression in neurology and is hereby explored [3]. Alemtuzumab induces a prolonged lymphodepletion by targeting CD52+ lymphocytes [4] which results in a long-lasting decrease in relapse rate[5] and a reduced disability progression[6], associated with sNfL decrease[7]. Objectives: The aim of this work was to develop a semi-mechanistic model for sNfL dynamics and explore its link with disability progression in RRMS patients treated by alemtuzumab. Methods: Data were extracted from alemtuzumab arm of the CARE MS-1 clinical trial, a 2-year-rater-masked randomized controlled phase III trial, and its first extension CARE MS-1-EXT that followed patients for 6 more years. Absolute lymphocyte count (ALC) was measured monthly while sNfL was measured every 6 months in CARE MS-1 and every 3 months in CARE MS-1-EXT. First disability progression was defined as at least 1.5-point increase on Expanded Disability Status Scale (EDSS) from baseline if baseline EDSS=0 or 1-point increase if baseline EDSS=5.5, confirmed 24 weeks after (FCDP24W). Demographic, disease-related and biochemistry covariates were collected at baseline[7]. We used a nonlinear mixed effects modeling approach in Monolix2024R1 to jointly model the ALC, sNfL and time to FCDP24W data. First, we built the sub-model of the longitudinal ALC and sNfL data. Second, we selected a parametric hazard model to describe the time to FCDP24W data. Then, we used a two-stage approach to assess the link between sNfL and FCDP24W. Model selection was based on Bayesian Information Criteria Corrected, residual standard errors being <50% and goodness of fits plots[8]. We used the COnditional Sampling for Stepwise Approach based on Correlation tests algorithm[9] to identify covariates in the longitudinal sub-model and stepwise covariate model building approach[9] in the survival sub-model. Results: A total of 351 patients who received 12 mg/day alemtuzumab through IV infusion for 5 consecutive days at baseline and for 3 consecutive days one year later in CARE MS-1 and as needed in CARE MS-1-EXT were analyzed. We selected a model where lymphocytes are produced at a 1st order rate kin from a proliferating compartment constantly at steady state and cleared linearly at a rate kout stimulated by Alemtuzumab. A Kinetic-Pharmacodynamic model was used with 1st-order elimination rate for alemtuzumab derived from previous analyses[10]. In our model, sNfL is produced and cleared via 0-order and 1st-order rates kG and kS respectively and lymphocytes stimulate its synthesis in a linear manner through a parameter ßALC->kg with inter occasion variability to capture sNfL peaks. The release of sNfL in the blood significantly increased with age at baseline (Wald p-value <2.10^-12) and with the burden of the disease (Wald p-value <4.10^-12). Also, the elimination of sNfl from the blood decreased with EDSS at baseline (Wald p-value <3.10^-4) and time since last relapse (Wald p-value <7.10^-3). Time to FCDP24W was best described by an exponential model. Older patients showed worse disability progression (Wald p-value <6.10^-5). Change from baseline in sNfL, predicted using for each patient the mean of their ßALC->kg estimates at each occasion, was associated with time to FCDP24W (Wald p-value < 0.002). This link function reflects how disability progression is associated with the underlying dynamics of sNfL and not the occasional peak. Conclusion: We developed a semi-mechanistic joint model that well captured the ALC and sNfL dynamics and disability progression in patients with RRMS treated with Alemtuzumab. Further studies are warranted to quantify i) the strength of the link with sNfL via individual dynamic predictions and ii) how other RRMS endpoints, such as imaging data can improve the prediction of disability progression.