AN INCIDENCE-SEVERITY MODEL FOR CYTOKINE RELEASE SYNDROME FOLLOWING DOSE-PRIMING REGIMENS OF ELRANATAMAB
Donald Irby1, Professor Lena E. Friberg2, Professor Mats O. Karlsson2, Jennifer Hibma1
1Pfizer Research and Development, Pfizer Inc, 2Department of Pharmacy, Uppsala University
Introduction: Cytokine Release Syndrome (CRS) is a common adverse event with bispecific antibody (BsAb) therapy. Dose-priming regimens are routinely devised to mitigate CRS, though there are few established models capable of adequately simulating their CRS-related properties. A two-part mixture modeling strategy has been implemented to describe the time course of any CRS grade (CRS yes/no) following various dose-priming regimens of elranatamab, a humanized BsAb that targets the B-cell maturation antigen (BCMA) on myeloma cells and CD3 on T-cells [1]. However, that model does not discriminate between the grades of CRS. We have therefore developed an extension of our previous work; an incidence-severity model utilizing a 3-state discrete-time Markov model (DTMM) to distinguish between grades 0, 1, and 2/3 CRS. Methods: CRS events data from 271 patients with relapsed/refractory multiple myeloma were analyzed from 4 phase I/II clinical trials that tested 3 dose-priming regimens, as described elsewhere [1]. A two-part mixture modeling approach was used, as in our previous model, where there is an incidence model in the first part (described as a mixture model) and now a 3-state DTMM in the second step characterizing the conditional severity of CRS over time. The transition probabilities between CRS grades 0 and 1, and 0 and 2/3, respectively, were adjusted with separate Bateman functions to describe response and tolerance effects [2]. Early (Cmax,D1) and event time-based (Cmax,event) elranatamab exposure metrics of free analyte were considered during Markov and incidence model development, respectively. Additional covariates evaluated in both models included premedication (dexamethasone, acetaminophen, and diphenhydramine) use and baseline soluble BCMA (free analyte; sBCMA). Time-varying IL-6 pathway inhibitors (tocilizumab or siltuximab; TOCI) use was also considered in the Markov model. All model building was performed in NONMEM version 7.5.0 using the FOCE method with the Laplace and Likelihood options [3]. Simulations were performed for each of the observed dose-priming regimens. First, CRS event times were simulated from the Markov model and the event-time information was used with the elranatamab PK model to calculate Cmax,event [4]. Then, the CRS incidences were obtained using the incidence model. Results: Peak elranatamab exposure on the day of the first CRS event (log(Cmax,event)) was retained as a linear term in the logit for the CRS mixture probability in the incidence model. The predicted odds ratio (OR; 6.0) suggests a 500% increase in the odds of any grade CRS occurring for every 3-fold increase in Cmax,event. As expected, increasing Cmax,D1 was associated with an increased “amount” term in the Bateman function (representing a “stimulation effect”) for both 0-1 and 0-2/3 state transitions (18.6% for ?01 and 45.0% for ?02/3, respectively, when comparing the 90th/10th percentiles of Cmax,D1) as well as an increase in the elimination rate constant of the Bateman function (representing a “tolerance effect”) for each of the same state transitions (estimated decreases of 51.9% for MET01 and 97.9% for MET02/3, where ke01 = 1/MET01 and ke02/3 = 1/MET02/3). Premedication was associated with a 35.5% reduction in the typical stimulation effect for a 0-2/3 state transition (?02/3), which is consistent with its intended use. High sBCMA was associated with a low logit probability for a 1-0 state transition (OR: 0.58 comparing the 90th/10th percentiles of sBCMA) and a longer duration of grade 1 CRS, which could reflect a greater potential for trimer formation. Finally, TOCI was associated with a higher logit probability for a 1-0 state transition (OR: 2.75) and a decreased duration of grade 1 CRS. The dosing simulations confirmed the model’s adequacy to recapitulate the observed incidence rates of grade 1 (observed: 48.9, 43.2, and 39.5% vs. simulated: 37.0, 47.5, and 38.4% for 4/20/76 mg, 12/32/76 mg, and 44/0/76 mg, respectively) and grade 2/3 CRS (observed: 11.1, 14.2, and 48.8% vs. simulated: 16.4, 20.0, and 49.6% for 4/20/76 mg, 12/32/76 mg, and 44/0/76 mg, respectively). Conclusions: We have successfully developed an incidence-severity model, incorporating a 3-state DTMM, to differentiate between predicted grades of CRS following dose priming regimens of elranatamab. This kind of model may allow for a more nuanced approach to the optimization of dose-priming regimens for BsAbs.