Case study of Exposure-Safety analysis in Phase 1 oncology - lessons learned
Anne Keunecke (1), Sebastiaan C. Goulooze (1), Teun M. Post (1), Murad Melhem (2), Herbert Struemper (3), Tianli Wang (2)
(1) LAP&P Consultants, Leiden, the Netherlands (2) GSK, Waltham, MA, USA (3) GSK, Durham,NC, USA
Objectives:
Exposure-safety (E-S) analyses in Phase 1 oncology studies can support future study design elements, particularly selection of dosing regimens. To better inform such decisions, the right choice of pharmacokinetic (PK) exposure metric (e.g., AUC or Cmax) as a driver for safety is essential.
Here, we discuss learnings from a case study on the Phase 1 E-S analysis of a cancer therapeutic with immune modulation potential. The objectives of the analysis were to quantify the exposure-response relationship and to investigate, via simulations, the impact of dose frequency and infusion duration on the frequency of adverse events (AEs). Additionally, it was explored whether the inclusion of covariates, such as baseline C-reactive protein (CRP), into the E-S model influenced the choice of PK driver of safety.
Methods: Previously, a population PK (popPK) model (3-compartment, linear elimination) was developed based on dense PK data from the Phase 1 study analysed in this study. Baseline CRP was the strongest covariate for CL, and therefore exposure. This model was used to derive individual PK metrics for each dosing interval (AUC and Cmax).
As the dependent variable, the counts of AEs with a grade ≥2 were modelled per dosing interval (i.e., one or three weeks). The count data were modelled using a truncated generalised Poisson model, maximizing the information compared to logistic regression. Exposure metrics, demographic variables, and other relevant variables were examined as potential drivers of the AE response. Baseline CRP was considered a mechanistic candidate covariate.The model was evaluated using visual predictive checks.
Results:
Baseline CRP and AE counts were found to be positively correlated in the E-S model, independent of the inclusion of exposure metrics (p<0.001, OFV drop>10.8). Because CRP is also strongly associated with drug CL, and therefore with AUC, baseline CRP could be a confounder in the E-S relationship. It was investigated whether the choice of the most predictive PK metric was dependent on the inclusion of the potential confounder baseline CRP in the E-S model. Without baseline CRP, AUC resulted in a OFV drop of 11.07 compared to 8.91 for Cmax. With baseline CRP, Cmax was found to be the superior driver compared to AUC (OFV drop of 10.22 vs 5.93 points, respectively). Findings regarding the impact of infusion duration on safety were strongly dependent on this choice of PK metric, as infusion duration would reduce Cmax, but not influence AUC.
The dosing interval (Q1W versus Q3W) was not found to be significant covariate in the model. Because counts of AE were modelled ‘per dosing interval’, this means that for a given dose level, each dose (regardless of whether the previous dose occurred one or three weeks ago) would result in a similar expected number of AE. However, for a given observation interval of three weeks, for Q1W dosing regimen three doses were administered, whereas for Q3W only one dose. Thus, the risk of experiencing at least one AE with a grade >= 2 within a particular time interval is higher for Q1W dosing compared to Q3W dosing at all dose levels.
Conclusions:
The choice to model counts of AE per dosing interval (as opposed to occurrence over the trial period) was motivated by the fact that most AE occurred during 1-2 days after each dose. This choice had to be considered to adequately simulate results on the relative safety profiles of Q1W versus Q3W dosing, and to properly communicate the outcome.
This case study is an instructive example of confounding in E-S analyses with important consequences on the analysis results. CRP was identified as a significant and mechanistically plausible covariate of clearance and safety events. Confounding was mitigated by using a multivariable analysis including the confounding variable instead of univariate exposure-response [1]. With this approach Cmax as opposed to AUC was identified as the most important exposure metric driving AEs, consistent with the acute nature of the safety events. The correct choice of exposure metric here makes a crucial difference when e.g. assessing the impact of infusion duration on safety. In summary, our case study underlines the importance of considering the potential for confounding not just for exposure-efficacy but also for E-S analyses.
References:
[1] Dai et al, Clinical Pharmacology & Therapeutics. 2020 Dec;108(6):1156-70