Higher abatacept exposure after transplant decreases acute GVHD risk without increasing adverse events

Abstract

In the ABA2 study, the T-cell costimulation blockade agent, abatacept, was safe and effective in preventing acute graft-versus-host disease (aGVHD) after unrelated-donor hematopoietic cell transplant (HCT), leading to US Food and Drug Administration approval. Here, we performed a determination of abatacept pharmacokinetics (PK), which enabled an examination of how abatacept exposure-response relationships affected clinical outcomes. We performed a population PK analysis of IV abatacept using nonlinear mixed-effect modeling and assessed the association between abatacept exposure and key transplant outcomes. We tested the association between the trough after dose 1 (Ctrough_1) and grade (GR) 2 or 4 aGVHD (GR2-4 aGVHD) through day +100. An optimal Ctrough_1 threshold was identified via recursive partitioning and classification tree analysis. This demonstrated that abatacept PK was characterized by a 2-compartment model with first-order elimination. The ABA2 dosing regimen was based on previous work targeting a steady-state abatacept trough of 10 μg/mL. However, a higher Ctrough_1 (≥39 μg/mL, attained in ∼60% of patients on ABA2) was associated with a favorable GR2-4 aGVHD risk (hazard ratio, 0.35; 95% confidence interval, 0.19-0.65; P < .001), with a Ctrough_1 <39 μg/mL associated with GR2-4 aGVHD risk indistinguishable from placebo (P = .37). Importantly, no significant association was found between Ctrough_1 and key safety indicators, including relapse, and cytomegalovirus or Epstein-Barr virus viremia. These data demonstrate that a higher abatacept Ctrough_1 (≥39 μg/mL) was associated with a favorable GR2-4 aGVHD risk, without any observed exposure-toxicity relationships. This trial was registered at www.clinicaltrials.gov as #NCT01743131.

Graphical abstract

Figure 1.

Classification tree analysis with bootstrap for aGVHD GR2-4 at day +100. Classification tree analysis of the developmental cohort is shown (n = 128) (the validation cohort [n = 56] is not included). This method uses recursive partitioning based on the Gini purity index and enables the identification of variables that best predict GR2-4 aGVHD. Each box represents a node (second row in a box: the percentage of GR2-4 aGVHD within the node; third row in a box: sample size of the node and fraction of the initial node size). The initial tree was pruned per the crossvalidated error to avoid overfitting. The final model predicts the risk of aGVHD GR2-4 at day +100 by C_{trough_1} ≥39 vs <39 μg/mL (31% vs 62%). C_{trough_1} was identified as the most predictive classifier in 73.6% of the 1000 bootstrap data sets. Of these, the median C_{trough_1} threshold value was 39.0 μg/mL (95% CI, 24.1-50.9). BuCy, busulfan/cyclophosphamide; BuFlu, busulfan/fludarabine; CyTBI, cyclophosphamide/total body irradiation; FluMel, fludarabine/melphalan.

Figure 2.

E-R relationships, using local polynomial regression (developmental cohort). Shown in the blue line is the local polynomial regression (ie, Loess smooth regression) for E-R relationships for 3 exposure variables: (A) C_{trough_1}; (B) C_{max_1}; and (C) AUC₁. The shaded area represents the 95% CI. C_{trough_1} showed the lowest P value (logistic regression). Each dot represents a single patient and associated C_{trough_1} and the probability of the event (0 [−], event absent; 1 [+], event present). AUC₁, AUC of abatacept concentration-time; C_{max_1}, maximum abatacept concentration after the first dose.

Figure 3.

Exposure-outcome relationships based on the C_{trough_1} of key secondary outcomes in the developmental cohort. Shown are graphical assessments of E-R relationships, using scatter plots with local polynomial regression lines (ie, Loess smooth regression; blue line). Each dot represents a single patient and associated C_{trough_1} and the probability of the event (0 [−], event absent; 1 [+], event present). This analysis demonstrates no associations between C_{trough_1} and any of the tested outcomes. The shaded area represents 95% CI for the typical value. CMV- and EBV-related outcomes are assessed as approximation because the censoring event (ie, relapse) is treated as absence of the outcome of interest. All of the other outcomes do not have censoring events. P values are calculated using logistic regression.

Figure 4.

Cumulative incidence of GR2-4 aGVHD divided by C_{trough_1}. (A) Cumulative incidence of GR2-4 aGVHD for patients receiving placebo and those with C_{trough_1} <39 μg/mL (pink) vs those with a C_{trough_1} ≥39 μg/mL (green) (developmental cohort). (B) Cumulative incidence of GR2-4 aGVHD for patients receiving placebo and those with C_{trough_1} <39 μg/mL (pink) vs those with a C_{trough_1} ≥39 μg/mL (green) (validation cohort). (C) Cumulative incidence of GR2-4 aGVHD for patients receiving placebo (red), those with C_{trough_1} <39 μg/mL (blue), and those with a C_{trough_1} ≥39 μg/mL (green) (developmental cohort). (D) Cumulative incidence of GR2-4 aGVHD for patients receiving placebo (red), those with C_{trough_1} <39 μg/mL (blue), and those with a C_{trough_1} ≥39 μg/mL (green) (validation cohort). The number of patients at risk are listed below the graph. Cumulative incidence plots represent unadjusted estimates, whereas the HR estimates are adjusted for the graft type.

Figure 5.

Exposure simulation for C_{trough_1}. Shown is the simulation of C_{trough_1} for patients in the 7/8 URD HCT cohort (left) and the 8/8 URD HCT cohort (right). The top graphs depict simulations that include a maximum dose of 1000 mg. The bottom graphs depict simulations that have no maximum dose. The black line represents the 50th percentile and the gray range represents the 90th percentile interval (bottom: 5th, top: 95th) of 1000 simulated C_{trough_1} values. The range above the red dashed lines represents the optimal target for GR 2-4 aGVHD (C_{trough_1} ≥39 μg/mL).