Zhiyuan Tan 1,2, Elise Smolders 3, Elianne de Boer 4, Helle-Brit Fiebrich-Westra 4, Erik Metscher 2, Yuri van der Burgt 2, Amy Rieborn 2, Jan Willem de Groot 4, Peter Plomp 5, Aymara Sancho-Araiz 1, Swantje Völler 1, Catherijne Knibbe 1,6, Jan Gerard Maring 3, Dirk Jan Moes 2
1 Department of Clinical Pharmacy & Toxicology, Leiden University Medical Center (Leiden, The Netherlands), 2 Division of Systems Pharmacology and Pharmacy, Leiden Academic Centre for Drug Research, Leiden University (Leiden, The Netherlands), 3 Department of Clinical Pharmacy, Isala Hospital Zwolle (Zwolle, The Netherlands), 4 Oncology Center Department, Isala Hospital, Zwolle (Zwolle, The Netherlands), 5 Pulmonary Disease, Isala Hospital Zwolle (Zwolle, The Netherlands), 6 Department of Clinical Pharmacy, St Antonius Hospital (Nieuwegein, The Netherlands)
Background: The approved dose of pembrolizumab was established in phase I-III clinical studies with 2 mg/kg every three weeks (Q3W) and later changed to a fixed dose of 200 mg Q3W or 400 mg every six weeks (Q6W) using in-silico studies[1]. A flat exposure-response relationship has been reported in the approved dose range[2], while a time-dependent decrease in clearance (CL) during treatment correlates with individual response[3].
Objectives: The first objective was to evaluate and refine the published pembrolizumab population pharmacokinetic (POPPK) models. The second objective was to determine the limited sampling strategy (LSS) to efficiently estimate the individual delta CL (ΔCL(%)). Lastly, the POPPK model with the LSS were used to evaluate whether (changes in) CL can be used as a descriptive biomarker for (non-)responders and to design model-informed dose interval individualization to reduce drug expenses.
Methods: Patients from the prospective non-randomized clinical study entitled MINUTE (Evaluating the Safety of Shortened Infusion Times for different Oncological Immunotherapie, NCT06031233) with at least one pembrolizumab PK measurement were included. Published pembrolizumab POPPK models[3-5] were externally evaluated and refined where needed. Two- and three-sample limited sampling strategies were evaluated to estimate ΔCL(%) using the final model. Subsequently, the ability to identify responders and non-responders based on individual ΔCL(%) values and treatment outcomes was assessed. Finally, model-informed dose interval optimization strategies were simulated, aiming to maintain trough concentrations (Cmin) ≥ 10 mg/L as the pharmacokinetic target for clinical recommendations.
Results: Data from 93 individuals with 267 observations (median 3 [range 1–5] per individual) were available. The two-compartment time-varying CL POPPK model outperformed the other published models and CL, volume of distribution, cancer diagnosis effect on CL were refined. A two‑sample limited sampling strategy at Day 21 (1st cycle Cmin) and Day 84 (4th cycle Cmin) accurately estimated ΔCL(%) (mean absolute prediction error (MAPE) = 0.42%), or at Day 21 and Day 63 (3rd cycle Cmin) with MAPE=0.63%. For individuals with decrease of ΔCL(%) > 30% by Day 84, the dosing interval could be extended from 200 mg Q3W to 200 mg Q6W while maintaining Cmin within a 90% prediction interval above the 10 mg/L target. A decrease of ΔCL(%) < 10% by 4th cycle indicated potentially non-response to pembrolizumab. These findings were translated into the following practical clinical algorithm for model-informed dose individualization and early non-responder identification: 1. Initiate pembrolizumab at the standard dose of 200 mg every 3 weeks (Q3W) for the first four cycles. 2. Collect trough concentrations at Day 21 (end of cycle 1) and Day 63 (end of cycle 3) using the validated two-sample limited sampling strategy. 3. Estimate individual ΔCL(%) at Day 84 via the refined time-varying clearance POPPK model using maximum a posteriori Bayesian estimation. 4. At Day 84 (prior to cycle 5), apply the following decision rules to personalize the dosing interval: • If clearance has decreased by >30% (ΔCL(%) < –30%), extend the regimen to 200 mg every 6 weeks (Q6W) from cycle 5 onward; If clearance has decreased by <10% (ΔCL(%) > –10%), flag the patient as a potential non-responder and prioritize early radiological assessment (RECIST/iRECIST) or treatment modification; For intermediate changes (–30% ≤ ΔCL(%) ≤ –10%), continue 200 mg Q3W with repeated monitoring.
This PK-guided algorithm uses patient-specific measured concentrations to estimate ΔCL(%) and directly calculates the individualized dosing interval while preserving target exposure and enabling a projected 45% reduction in drug expenses for qualifying responders.
Conclusions: In this study a refined time-varying CL POPPK model best described the real-world pembrolizumab concentration versus time data. Furthermore, a practical two-sample limited sampling schedule at Day 21 and Day 63 to estimate individual ΔCL(%) was identified. Early ΔCL(%) could potentially be a descriptive target for the identification of non‑responders to pembrolizumab. For patients showing ΔCL(%) decrease > 30% by Day 84, a model-informed dose interval extension from 200 mg Q3W to Q6W was proposed which potentially reduced drug expenses by 45%.
References:
[1] Patnaik, A. et al. Clinical Cancer Research 2015, 21 (19), 4286-4293.
[2] Chatterjee, M. S et al. CPT Pharmacometrics Syst Pharmacol 2017, 6 (1), 29-39.
[3] Li, H. et al. Journal of pharmacokinetics and pharmacodynamics 2017, 44 (5), 403-414.
[4] Hurkmans, D. P. et al. J Immunother Cancer 2021, 9 (6), e002344.
[5] Ahamadi, M. et al. CPT Pharmacometrics Syst Pharmacol 2017, 6 (1), 49-57.
Reference: PAGE 34 (2026) Abstr 12195 [www.page-meeting.org/?abstract=12195]
Poster: Clinical Applications