Perrine Courlet (a)*, Josiah Ryman (b)*, Andrew Santulli (c), Scott Van Wart (c), Karthik Venkatakrishnan (b), Wei Gao (b)
(a) Merck Institute for Pharmacometrics, Ares Trading S.A. (an affiliate of Merck KGaA, Darmstadt, Germany), Lausanne, Switzerland ; (b) EMD Serono Research and Development Institute, Inc., Billerica, Massachusetts, USA.; (c) Enhanced Pharmacodynamics, LLC, Buffalo, NY, USA ; *Co-first authors with equal contribution to the work
Objectives:
Dose optimization is a critical objective of early clinical development of oncology drugs. The value of longitudinal tumor size in Phase 1 dose escalation studies can be maximized using modeling and simulation to inform dose selection1. To this end, designing Phase 1 trials requires careful consideration of various factors such as dose and sample size.
We present a workflow using pharmacokinetic (PK) data, tumor growth inhibition (TGI) modeling, and clinical trial simulations (CTS) to aid in the identification of the most informative phase 1 trial design for efficient dose selection for later phases of development.
Methods:
A virtual population database was first generated using a QSP model, with modifications from a previously published model for antitumor activity of antibody-drug conjugates as a representative therapeutic modality with opportunities for model-informed dosage optimization2,3. Virtual PK and longitudinal tumor size data were generated for five dose levels (DL) of interest at predefined PK and tumor assessment timepoints assuming representative drug and system parameters for a population with colorectal carcinoma (CRC).
As the data is limited in a phase 1 trial, more conventional and identifiable PK-TGI models (including a two cell populations or Claret model4) were fit to the data (nlmixr5, step 1). The PK-TGI model with the best fit was selected.
In step 2, various Phase 1 trial designs, including diverse DLs and sample sizes were generated (referred to as simulation scenarios thereafter) by sampling from the virtual population. Tumor size profiles for each virtual subject reflected the planned clinical trial schedule of tumor size assessment. PK-TGI model parameters were estimated for each trial. These steps were repeated to generate 500 datasets per simulation scenario, which were subsequently fit using the step 1 PK-TGI model. This allows for the evaluation of the impact of study designs on dose selection.
Step 3 sampled from each estimated parameter distribution, which generated 1000 simulated clinical trial datasets with 100 subjects per DL (mrgsolve6). This enabled the evaluation of tumor shrinkage at various timepoints for each scenario and DL.
For each CTS, a target dose was identified, defined as the lowest dose producing a median tumor shrinkage ≥20% relative to baseline at week 87. Earlier and later tumor assessment timepoints were also considered and results were compared. This dose was compared to the “true” reference dose based upon the QSP simulated dataset using the aforementioned definition (step 4).
Results:
In step 1, a Claret model with Emax relationship between exposure and killing rate was the best model. It adequately captured the virtual population tumor size time-course data, which reflected variability in target expression, baseline tumor size, tumor growth, drug PK, and drug effect in participants with CRC.
Step 2 provided summary statistics of the final parameter estimates across the 500 clinical trial simulation-estimation runs to provide insight into the identifiability of a given trial design scenario. Largest coefficient of variation (CV, used as a measure of parameter variability across the estimations) ranged from 7% to 40% depending on the model parameters and simulation scenario. The largest CV was observed for the variability on the resistance parameter, particularly for scenarios with small sample sizes.
Steps 3 and 4 demonstrated that a design with balanced sample sizes (e.g. 12 patients at each DL) better informed the model fits, thereby increasing the probability of selecting an active dose (the QSP-predicted “true” dose). Unbalanced designs (e.g. small vs large sample sizes at low vs high doses) tended to underestimate the drug resistance, which increased the probability of selecting a non-optimal dose. Thus, the balanced trial design better informed the dose selection in this case study, lending support to the incorporation of dose-ranging backfill cohorts in Phase 1 studies to inform dose selection strategies in subsequent clinical development.
Conclusions:
Our workflow provides valuable proof-of-principle for leveraging longitudinal tumor size data to enhance dose selection decision-making in oncology drug development, aiding in early trial design optimization. This framework can be applicable to other therapies including combination regimens. Ongoing work involves evaluation of adaptive trial design, patient drop-out, and impact of tumor sampling times.
References:
[1] Bruno R, Bottino D, De Alwis DP, Fojo AT, Guedj J, Liu C, Swanson KR, Zheng J, Zheng Y, Jin JY. Progress and opportunities to advance clinical cancer therapeutics using tumor dynamic models. Clinical Cancer Research. 2020 Apr 15;26(8):1787-95.
[2] Liao MZ, Lu D, Kågedal M, Miles D, Samineni D, Liu SN, Li C. Model‐Informed Therapeutic Dose Optimization Strategies for Antibody–Drug Conjugates in Oncology: What Can We Learn From US Food and Drug Administration–Approved Antibody–Drug Conjugates?. Clinical Pharmacology & Therapeutics. 2021 Nov;110(5):1216-30
[3] Singh AP, Guo L, Verma A, Wong GG, Shah DK. A cell-level systems PK-PD model to characterize in vivo efficacy of ADCs. Pharmaceutics. 2019 Feb 25;11(2):98.
[4] Claret L, Girard P, Hoff PM, Van Cutsem E, Zuideveld KP, Jorga K, Fagerberg J, Bruno R. Model-based prediction of phase III overall survival in colorectal cancer on the basis of phase II tumor dynamics. Journal of Clinical Oncology. 2009 Sep 1;27(25):4103-8.
[5] Fidler M, Xiong Y, Schoemaker R, Wilkins J, Trame M, Hooijmaijers R, Post T, Wang W (2022). nlmixr: Nonlinear Mixed Effects Models in Population Pharmacokinetics and Pharmacodynamics. R package, https://CRAN.R-project.org/package=nlmixr.
[6] Baron K (2024). mrgsolve: Simulate from ODE-Based Models. R package, https://github.com/metrumresearchgroup/mrgsolve.
[7] Colloca GA, Venturino A, Guarneri D. Early tumor shrinkage after first-line medical treatment of metastatic colorectal cancer: a meta-analysis. International journal of clinical oncology. 2019 Mar 15;24:231-40.
Reference: PAGE 32 (2024) Abstr 10929 [www.page-meeting.org/?abstract=10929]
Poster: Methodology - Study Design