Purpose: To examine the feasibility of escalating carboplatin area under the concentration-time curve (AUC), using dose predictions based on individual estimates of drug clearance, in a phase I trial of multicycle carboplatin, paclitaxel, and cyclophosphamide chemotherapy with peripheral-blood stem-cell (PBSC) replacement.
Patients and methods: Forty-four patients (37 breast, seven ovarian) received 165 courses. Initial target carboplatin AUC was 10 mg/ml x min, with interpatient escalation in increments of 25%. Initial carboplatin dose estimates used creatinine clearance (CrCl) to estimate carboplatin clearance. Subsequent clearance and dose estimates were determined using a model incorporating Bayesian estimation and two measured carboplatin plasma ultrafiltrate concentrations.
Results: Median clearance was 80.5 mL/min/m2 (range, 41.6 to 131.8). Carboplatin doses up to 2,440 mg/m2 per course were administered without major extramedullary toxicity. Doses varied 2.6-fold at each exposure level. Using the Bayesian model, AUC was predicted with a mean accuracy of 101.2% (83% using CrCl). Ninety-six of 117 courses were within 25% of the target AUC. This model was less biased (0.15 v -2.35 mg/mL x min) and more precise (2.76 v 3.52) in predicting AUC compared with one using CrCl. Hematologic recovery was not prolonged with increasing exposure. The carboplatin maximum-tolerated systemic exposure (MTSE) was 13.3 mg/mL x min (level five). The dose-limiting toxicity was cardiac toxicity, which occurred at dose levels six and seven.
Conclusion: Results demonstrate that (1) CrCl is a poor estimator of carboplatin clearance in this population, and (2) the use of a model incorporating limited sampling and Bayesian estimation improves the precision of carboplatin clearance estimation and is suitable for targeting carboplatin exposure in an ambulatory setting.