Joint modeling of time-varying exposure data and progression-free survival in elotuzumab-treated patients with relapsed/refractory multiple myeloma
Jennifer Sheng (1), Manish Gupta (1), Jun Shen (1), Amit Roy (1), Chaitali Passey (1), Tai-Tsang Chen (1)
(1) Bristol-Myers Squibb, Lawrenceville, NJ, USA
Objectives: Exposure-efficacy (E-R) analyses with Cox proportional hazard (CPH) assume a constant hazard throughout treatment and often use a static value without capturing longitudinal data until the time-to-event [1,2]. A time-varying CPH model incorporates longitudinal data but due to large random errors, may lead to highly biased and inefficient estimates [1]. Here, we explore a joint model (JM) simultaneously integrating longitudinal data and time-to-event data, thereby improving assessment of longitudinal exposure of elotuzumab (plus lenalidomide/dexamethasone [ELd]) on PFS for the treatment of relapsed/refractory multiple myeloma.
Methods: Joint modeling of longitudinal exposure data and PFS was performed using R JM package [3], with evaluable elotuzumab exposure from ELd-treated patients (N = 310) from ELOQUENT-2 (NCT01239797). Elotuzumab was administered at 10 mg/kg once weekly for the first 2 cycles and then twice weekly onwards, resulting in exposure changes over time. The E-R analysis of elotuzumab applied Cave,ss and baseline covariates to the CPH model, predicting clinical outcomes [4]. The final JM used the piecewise proportional hazard model with Gauss-Hermite integration method, and was selected based on Bayesian information criteria using the clinically observed Cmin time profile. Sensitivity analysis was performed with Cmin1 using the CPH model and time-varying Cmin in the JM. Additional sensitivity analyses used the clinically observed evaluable Cmin profile and population pharmacokinetic (PPK) simulated Cmin profile [4]. Survival dynamic probabilities of individuals were predicted using the final JM.
Results: The JM suggested a weaker dependence of survival outcome (PFS) on longitudinal elotuzumab exposure, and stronger association for other covariates, such as lactate dehydrogenase vs the CPH model. JM results were comparable between clinically observed Cmin and PPK-simulated Cmin, with overlapping mean plus one standard error. Individual dynamic predictions provide dynamic assessment of survival as additional longitudinal clinical data become available.
Conclusions: Joint modeling of longitudinal data and PFS provides a new approach to predict clinical benefits using dynamic data. Ultimately, the goal of joint modeling is to utilize earlier clinical data (including progressive disease biomarkers) to predict individual clinical benefit and to reveal the continuous interplay between PK, PD and efficacy to support clinical decisions.
References:
[1] Ibrahim JG, Chu H, Chen LM. Basic concepts and methods for joint models of longitudinal and survival data. J Clin Oncology (2010) 28(16): 2796-801.
[2] Zhang D, Chen MH, Ibrahim JG, et al. Assessing model fit in joint models of longitudinal and survival data with applications to cancer clinical trials. Stat Med (2014) 33(27): 4715-33.
[3] Rizopoulos D. Joint models for longitudinal and time-to-event data with applications in R. 2012. CRC Press, Boca Raton, FL, USA.
[4] Gibiansky L, Passey C, Roy A, et al. Model-based pharmacokinetic analysis of elotuzumab in patients with relapsed/refractory multiple myeloma. J Pharmacokinet Pharmacodyn (2016) 43(3): 243-57.
