Elena Maria Tosca (1), Roberta Bartolucci (1), Stefano Castellano (1), Francesco Fiorentini (2), Manzoni Sara (3), Silvia Di Tollo (3), Caserini Maurizio (3), Maurizio Rocchetti (4), Paolo Bettica (3), Paolo Magni (1)
1) Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia (Italy); 2) Accellera S.r.l., Nerviano (Italy) 3) Drug Development, Italfarmaco Group, Cinisello Balsamo (Italy) 4) Consultant, Rho (Italy)
Introduction: Polycythemia Vera (PV) is a chronic myeloproliferative neoplasm characterized by an abnormal production of red and white blood cells (WBC) and platelets (PLT), for which no worldwide recognized cures exist [1,2]. A new histone-deacetylase inhibitor (Drug G) is currently under investigation as treatment option for PV. A pivotal Phase III trial for the comparison of the efficacy/safety profile of Drug G in PV patients with the first line treatment, is being planned. Study protocol states a fixed starting dose and a subsequent individual dose adaptation: at each treatment cycle (28 or 14 days), dose is increased for low efficacy and decreased or temporarily stopped for moderate and severe toxicities, respectively (grade I or II thrombocytopenia and/or neutropenia); in addition, phlebotomies could be performed if HCT≥45%.
Objective: A clinical trial simulation approach was here proposed to assess the adaptive dosing regimen and to predict the outcomes of the planned Phase III trial, i.e. the expected fraction of respondent patients at the End Of Treatment (EOT, 8th month).
Methods: A pop PK model for Drug G was already available. A PK/PD model for PLT, WBC and HCT dynamics was developed based on the myelosuppression Friberg model [3] where phlebotomies were included as a 10% reduction of the HCT level. The model was identified in NONMEM on data from Phase II studies adopting a non-linear mixed effect approach. In addition, the probability of performing a phlebotomy when HCT≥45% was modeled based on phlebotomy administered during Phase II trials. The PK/PD model and the probabilistic phlebotomy model was, then, used to develop a clinical trial simulator in MATLAB. It was composed by i) a NONMEM-based virtual population generator for PV patients, ii) a control algorithm implementing the adaptive dosing strategy, iii) a NONMEM-based simulator predicting hematological responses for each given treatment period. To predict outcomes of the future Phase III trial, 10 populations of 120 virtual PV patients were generated extracting individual model parameters from their joint distributions. For each of the 1200 subjects, the first 28-day treatment cycle was simulated. Individual PLT, WBC and HCT levels at day 28th were evaluated and dose for the subsequent cycle adjusted accordingly by the control algorithm. If HCT≥45%, phlebotomies were performed based on the probabilistic model. The subsequent treatment cycle was simulated and dose adjusted again based on PLT, WBC and HCT levels at the end of the cycle. These steps were iteratively repeated until EOT. For each subject, the entire trial period was simulated 500 times to account for the residual unexplained variability (RUV). Finally, patient fraction with a complete hematological response (CHR), i.e. PLT≤400×109/L, WBC≤10×109/L and HCT<45% without phlebotomies in the last 3 months of treatment, was evaluated.
Results: The PK/PD model adequately described PLT, WBC and HCT dynamics in treated PV patients. Inter-individual variability was included on all PD parameters except γ for HCT. Correlations were included between the 3 drug potencies, the feedback processes of PLT and WBC, the mean transit times of PLT and WBC. Correlations were also included between the RUV parameters of PLT and WBC. The analysis of phlebotomy events showed that the probability of performing a phlebotomy when HCT≥45% increased with the increase of HCT level. A logistic regression was used to identify the probability of performing a phlebotomy as a function of the HCT value, PPhleb(HCT). Accordingly, if HCT≥45%, phlebotomies were modeled as a Bernoulli random variable with probability of success p=PPhleb(HCT). 10 virtual populations of 120 PV patients were generated with the simulator. For each subject of each population, PLT, WBC and HCT responses to treatment were iteratively simulated (500 times for each patient) following the adaptive dosing strategy designed for the study. Based on the 1200*500 simulations, the proportion of subjects with a CHR was computed.
Conclusions: A clinical trial simulator was developed to predict the outcomes of a future Phase III clinical trial on an investigated compound for PV, providing an invaluable support to study planning and decision-making process. Indeed, accordingly to the simulated results, the study protocol was revised, allowing to increase the expected percentage of virtual patients with a CHR.
References:
[1] Barbui T, Tefferi A, Vannucchi AM, Passamonti F, Silver RT, Hoffman R, et al. Philadelphia chromosome-negative classical myeloproliferative neoplasms: Revised management recommendations from European LeukemiaNet. Leukemia. 2018.
[2] Passamonti F, Maffioli M, Mora B. Therapy of polycythemia vera: Is it time to change? Oncotarget. 2017.
[3] Friberg LE, Henningsson A, Maas H, Nguyen L, Karlsson MO. Model of chemotherapy-induced myelosuppression with parameter consistency across drugs. J Clin Oncol. 2002;20(24):4713–21.
Reference: PAGE 30 (2022) Abstr 10054 [www.page-meeting.org/?abstract=10054]
Poster: Drug/Disease Modelling - Oncology