Elena Maria Tosca 1, Paolo Magni 1
1 University Of Pavia (Pavia, Italia)
Introduction: Patient-derived xenograft (PDX) mouse models represent a key preclinical platform for evaluating anticancer drug efficacy, with large-scale PDX encyclopedias and public datasets supporting their translational value (1). However, converting tumor growth inhibition (TGI) observed in PDXs into quantitative predictions of clinical efficacy remains a critical bottleneck in drug development. Among RECIST-based endpoints (2), overall response rate (ORR) is one of the earliest most decision-relevant metrics in phase II/III oncology trials.
Objective: We present CliP (Clinical Predictor), a translational PK/PD modeling framework designed to predict RECIST-based efficacy from preclinical TGI data. In this work, CliP was evaluated across seven case studies spanning multiple solid tumor indications to assess its generalizability in predicting clinical ORR.
Data: The case studies included: Gemcitabine in pancreatic cancer (GEM-PA), Sorafenib in hepatocellular carcinoma (SOR-HCC), Everolimus in gastric cancer (EVE-GC), Cetuximab in colorectal cancer (CET-CC), Encorafenib and Binimetinib in cutaneous melanoma (ENCO-MEL and BIN-MEL), Imlunestrant in HER2+ER- breast cancer (IML-BC). For each case study, TGI data from a panel of PDX models were collected from Novartis PDX Encyclopedia (1), Hubase database and literature sources for IML-BC (3) (n=27, 24, 38, 43, 33, 32 and 10 for GEM-PA, SOR-HCC, EVE-GC, CET-CC, ENCO-MEL, BIN-MEL and IML-BC, respectively). Published phase III clinical ORR data were used for external validation (4–10). Genetic characterization was considered when available.
Methods: CliP integrates a preclinical TGI modeling platform with a virtual clinical trial simulation framework for the clinical setting. The workflow includes: i) characterization of tumor-related and drug-related parameter distributions in PDXs; ii) inter-species allometric scaling of model parameters with propagation of inter-PDX variability; iii) development of human TGI model, informed by the scaled parameters and linked to clinical PK; iv) generation of virtual cancer patient populations, defined by a set of individual TGI model parameters (sampled from scaled multivariate lognormal distribution), and covariates (sampled from clinical distributions); v) simulation of tumor diameter dynamics according to clinical protocol and study design; vi) classification of simulated tumor size trajectories into RECIST-categories and computation of ORR in the population. Uncertainty propagation was performed via Monte Carlo simulations (500 replicates). Median predicted ORR values were reported.
Results: Population TGI models were successfully identified for all case studies (11). For ENCO–MEL (BRAF-V600) and IML–BC (ESR1), mutation status significantly increased estimated drug potency, highlighting the framework’s ability to incorporate genetic covariates. Allometrically scaled parameters informed the human TGI model and enabled generation of virtual patient populations matching clinical study size. Across diverse tumor types and mechanisms of action, CliP accurately reproduced the observed ORR, with predicted medians within the 95%CI of the clinical outcomes. Observed vs predicted ORR(%) were: 11.11% vs 10.6% for GEM–PA, 6.7% vs 4.2% for SOR–HCC, 4.48% vs 1.45% for EVE–GA, 8% vs 11% for CET–CC, 84.02% vs 84% for ENCO–MEL, 15.24% vs 11% for BIN–MEL, and 14.3% vs 16% for IML–BC (ESR1-mutated subgroup), respectively. Agreement was outstanding: R² = 0.99; RMSE = 2.55 percentage points (pp). Absolute differences were consistently ≤3 pp across case studies, except for BIN–MEL (4.24 pp). Even where point accuracy was lower, predicted control disease rate (CDR) remained consistent with observations, supporting model robustness. (Observed vs Predicted CDR: 43.27% vs 40.1% for EVE–GA, 39.4% vs 41.3% for CET–CC, 58.36% vs 57.6% for BIN–MEL).
Conclusions: CliP represents a quantitative translational framework to systematically predict clinical ORR from preclinical PDX efficacy data routinely generated during drug development programs. Here, its predictive performance was demonstrated across heterogeneous tumor types and drug classes, including case studies from both successful programs and indications with limited efficacy (e.g., GRANITE-1), highlighting its robustness across diverse translational contexts. CliP is currently undergoing further refinement and evaluation within ongoing drug development programs (12). Validation against patient-level datasets will further strengthen its credibility and support its integration into translational pharmacometric workflows.
By enabling prospective efficacy forecasting from preclinical data, CliP supports model-informed decision making, candidate prioritization, and risk reduction in oncology drug discovery and development.
References:
1. Gao H, et al. High-throughput screening using patient-derived tumor xenografts to predict clinical trial drug response. Nat Med. 2015;21(11):1318–25. doi:10.1038/nm.3954
2. Eisenhauer EA, et al. New response evaluation criteria in solid tumours: revised RECIST guideline (version 1.1). Eur J Cancer. 2009;45(2):228–47. doi:10.1016/j.ejca.2008.10.026
3. Sherman S, et al. Imlunestrant a next-generation oral SERD overcomes ESR1 mutant resistance in estrogen receptor-positive breast cancer. JCI Insight. 2025 23;10(12):e188051. doi:10.1172/jci.insight.188051
4. Von Hoff DD, et al. Increased survival in pancreatic cancer with nab-paclitaxel plus gemcitabine. N Engl J Med. 2013 31;369(18):1691–703. doi:10.1056/NEJMoa1304369
5. Cheng AL, et al. Sunitinib versus sorafenib in advanced hepatocellular cancer: results of a randomized phase III trial. J Clin Oncol Off J Am Soc Clin Oncol. 2013 10;31(32):4067–75. doi:10.1200/JCO.2012.45.8372
6. Ohtsu A, et al. Everolimus for previously treated advanced gastric cancer: results of the randomized, double-blind, phase III GRANITE-1 study. J Clin Oncol Off J Am Soc Clin Oncol. 2013 1;31(31):3935–43. doi:10.1200/JCO.2012.48.3552
7. Jonker DJ, et al. Cetuximab for the treatment of colorectal cancer. N Engl J Med. 2007 15;357(20):2040–8. doi:10.1056/NEJMoa071834
8. Dummer R, et al. Encorafenib plus binimetinib versus vemurafenib or encorafenib in patients with BRAF-mutant melanoma (COLUMBUS): a multicentre, open-label, randomised phase 3 trial. Lancet Oncol. 2018 1;19(5):603–15. doi:10.1016/S1470-2045(18)30142-6
9. Dummer R, et al. Binimetinib versus dacarbazine in patients with advanced NRAS-mutant melanoma (NEMO): a multicentre, open-label, randomised, phase 3 trial. Lancet Oncol. 2017 1;18(4):435–45. doi:10.1016/S1470-2045(17)30180-8
10. US. Food and Drug Administration. Multi-Discipline Review and Evaluation: INLURIYO (Imlunestrant) NDA 218881. Report No.
11. Simeoni M, Magni P, et al. Predictive pharmacokinetic-pharmacodynamic modeling of tumor growth kinetics in xenograft models after administration of anticancer agents. Cancer Res. 2004 1;64(3):1094–101. doi:10.1158/0008-5472.can-03-2524
12. Flaherty C. ELEVATE Trial: Elacestrant Combos Boost PFS in ER+/HER2- Breast Cancer | Targeted Oncology – Immunotherapy, Biomarkers, and Cancer Pathways. 2026. Available from: https://www.targetedonc.com/view/elevate-trial-elacestrant-combos-boost-pfs-in-er-her2–breast-cancer
Reference: PAGE 34 (2026) Abstr 11912 [www.page-meeting.org/?abstract=11912]
Poster: Oral: Preclinical and Translational modelling to support drug discovery and development