386069 A Generalizable Optimization Algorithm for Clinically-Relevant Patient-Specific Cancer Chemotherapy Scheduling
Chemotherapy is the most commonly employed method for systemic cancer treatment of solid tumors and their metastases. The balance between successful killing of cancer cells and minimizing the toxicity to the host remains a challenge for clinicians in deploying chemotherapy treatments. With toxicity as the primary reason a patient is either dose-reduced or dose-delayed in the clinic, our approach explicitly incorporates treatment-induced toxicities into the schedule design. As a case study, we synthesize administration schedules for docetaxel, a widely used chemotherapeutic ($765 million in sales in 2012) employed as a monoagent or in combination for the treatment of breast, non-small-cell lung, prostate, and head and neck cancers. The primary adverse effect of docetaxel treatment is myelosuppression, characterized by neutropenia, a low plasma absolute neutrophil count (ANC). Through the use of model-based systems engineering tools, we aim to provide treatment schedules for docetaxel and its combination therapies that reduce toxic side effects and improve patient treatment outcome.
Mirroring our past work, the current algorithm employs models of tumor growth, drug pharmacokinetics, and pharmacodynamics – both anticancer effect and toxicity, as characterized by ANC. Also included is a toxicity-rescue therapy, in granulocyte colony stimulating factor (G-CSF), that serves to elevate ANC. The single-agent docetaxel chemotherapy schedule minimizes tumor volume (the performance objective), over a multi-cycle horizon, subject to toxicity and logistical constraints imposed by clinical practice. In docetaxel treatment, the solving horizon length is 8 cycles(24 treatment weeks) where the tumor size is minimized every week along the treatment horizon. The numbers of variables and constraints in the docetaxel scheduling problem are 408,000 and 408,200 respectively. This problem is solved through a dynamic optimization step following an enumeration step. The enumeration step is carried to solve for all of the discrete variables (when to dose) and provide an initial guest for the dynamic optimization. Pyomo with Ipopt as the solver is then used to optimize the dose magnitude. The optimal solution provides a clinically-feasible treatment schedule for docetaxel chemotherapy within an order of 1-2 hours of computational time.
This single-agent chemotherapy scheduling problem formulation is extended to combination chemotherapy using docetaxel-cisplatin or docetaxel-carboplatin drug pairs. The two platinum agents display different toxicities, with cisplatin exhibiting kidney function damage (dynamically modelled through creatinine clearance changes) and carboplatin demonstrating the same myelosuppression effects as docetaxel. These case studies provide two different challenges to the algorithm: (i) cisplatin scheduling significantly increases in the number of variables and constraints (585,000 and 885,000 respectively with the same treatment horizon of 8 cycles and same objective function of minimizing tumor size weekly), thereby challenging the computational engine and formulation; (ii) carboplatin’s overlapping toxicity tests the ability of the algorithm to schedule drugs with different mechanisms of action (they act in different phases of the cellular growth cycle) that have the same toxic side effect. Oftentimes in the clinic, the use of agents with overlapping toxicities is avoided, for fear of severe toxic side effects. The simulated results demonstrate the algorithm’s flexibility, with docetaxel and cisplatin or carboplatin scheduled to provide tumor kill and toxicity within clinically acceptable limits. With an increase in number of variables and constraints in the combination chemotherapy, the algorithm is able to provide an optimal treatment schedule within 24 hours of computational time.
Overall, a clinically-relevant chemotherapy scheduling optimization algorithm is provided for designing single agent and combination chemotherapy, when toxicity and pharmacokinetic/pharmacodynamic information for the agents in question is available. Furthermore, the algorithm can be extended to patient-specific treatment via updating of the pharmacokinetic/pharmacodynamic models as data (e.g., tumor volume, toxicity measurements) is collected during treatment.
See more of this Group/Topical: Food, Pharmaceutical & Bioengineering Division