Computational Optimisation of Chemotherapy Regimens Accounting for Heterogeneity of Myelotoxicity - a Worked Example for Lymphoma.

In the last years, the development of dose and time intensified chemotherapeutic regimens led to considerable improvement of treatment outcome in several diseases such as Hodgkin’s lymphoma (Diehl et al, NEJM 2003), aggressive Non-Hodgkin’s lymphoma (Pfreundschuh et al, Blood 2004) or breast cancer. Although the granulopoietic growth factor G-CSF is widely used, leukopenia is still a major limiting factor for such therapy intensifications. The degree of leukopenia is very inhomogeneously distributed in the population of patients even if similar dose levels are applied. Thus, further intensification is usually limited by a small subgroup of patients who respond with particularly severe toxicity.

It was our objective to develop a method for estimating the potential for a differential dose-escalation with regard to expected myelotoxicity. For this purpose we used a biomathematical model of human granulopoiesis, which allows simulation of the dynamics of peripheral neutrophil concentration under a variety of chemotherapy regimens with or without G-CSF support (Engel et al, Blood 2004). This model is adapted to describe the observed heterogeneity of myelotoxic drug response.

We analyzed clinical data obtained from several large phase III-trials of the German Study Group for Aggressive Non-Hodgkin’s Lymphoma using CHOP-like chemotherapies. Patients were stratified into 3 risk groups (low, medium, high myelotoxicity) of approximately equal size. Stratification was performed using individual prognostic factors such as age, gender, ECOG performance status and LDH serum concentration.

We were able to estimate the drug induced bone marrow damages as functions of the given dose and the related stratum. These estimates could be used to modify drug dosing and scheduling of chemotherapy in order to achieve a homogenization of myelotoxicity between strata. The aim was to intensify therapy for the low and medium risk group without exceeding the expected toxicity of the high risk group.

The model simulations suggest that CHOP treatment intervals (G-CSF day 4-13) can be shortened in elderly patients from 14 to 11 and 12 days in the low and medium risk group, respectively. Alternatively, one may increase the dose of cyclophosphamide by 280mg/m2 and 100mg/m2 or apply additional etoposide (300mg/m2 and 150mg/m2) in these groups. Similar results can be obtained by a different approach using the observed degree of leukopenia during the first therapy cycle as a predictor for leukopenia in subsequent cycles. Using this risk factor for stratification into two groups, our model simulations show that cyclophosphamide doses can be increased by 360mg/m2 in the low risk group to yield a similar degree of leukopenia expected for the high risk group.

We conclude that our computational model can help to design more effective chemotherapy regimens for future dose intensified clinical trials by increasing the total dose delivered in the population. Our approach may be also applicable for myelotoxic chemotherapies of other malignancies.