A multi-model strategy in R to address cytopenia

Objectives: Hematological toxicity is a frequent adverse event of chemotherapeutic regimens and other oncology drugs [ 1] and also happens for biologics (see e.g. [ 2]). The most frequent application of modeling and simulation optimizes treatment regimen for better control of the hematological toxicity while maximizing efficacy [ 3]. Different software employing simple empirical to complex mechanistic models are being used currently (mostly in parallel) and better toolchain integration and cross-platform operability is sought for maximizing the use of the totality of evidence coherently. We therefore explore an R-based multi-model ecosystem for the assessment of cytopenia in preclinical safety. Methods: We have set up a regimen-myelosuppression and population workflow using a) a semi-mechanistic single-lineage model [ 4] and b) a more complex multi-lineage model [ 5] that has been integrated with 1) a simple phenobarbital PK time course prediction using a simple PK model in R or 2) a phenobarbital PBPK model in dogs implemented in the Open Systems Pharmacology suite [ 6]. We then compare experimental PK and hematology data representing typical findings [ 7] and case studies of dogs with (pan) cytopenia representing high-dose or idiosyncratic bone marrow [ 8] for all four setups. Results: The overall cell count kinetics under phenobarbital exposure and idiosyncratic bone marrow damage leading to cytopenia can be recapitulated with all models, given that parameter distributions are sampled efficiently at the extremities. We highlight the feasibility of harmonizing between different dog populations and performing mechanistic exploratory investigations towards the mechanism of bone marrow toxicity. Only a longer-term damage mechanism (as opposed to myeloablative chemotherapy) can explain the observed kinetics of blood cell counts, pointing to an indirect effect on bone marrow cells while at the same time experimental mechanistic insight into this toxicity is very limited. Conclusions: While simple PK/PD models can recapitulate preclinical safety (and veterinary safety) data on dose-dependent and idiosyncratic bone marrow toxicity in dogs, mechanistic explorations, and predictive dose-exposure tasks require a mechanistic framework. Unifying all approaches under the umbrella of an R ecosystem offers maximizing re-use of evidence and reducing animal testing in preclinical safety.