Model-informed precision dosing of carboplatin in cancer patients by leveraging myelosuppression data from electronic health records

Abstract

Aims: This study aimed to develop and validate a population pharmacokinetic-pharmacodynamic (pop-PK-PD) model to describe carboplatin-induced myelosuppression in cancer patients and support dose individualization. Methods: Data from 580 cancer patients treated with carboplatin at Amsterdam UMC between 2019 and 2022 were used for model development, focusing on lung, gynaecological and gastric/oesophageal cancers. Platelet (PLT) and neutrophil (NT) counts, along with patient-specific covariates (e.g., age, serum albumin, eGFR), were extracted from Electronic Health Records and used in the analysis. Given the absence of pharmacokinetic (PK) samples, PK parameters were derived from a literature carboplatin pop-PK model. Model applicability to inform personalized carboplatin dosing was evaluated on a separate cohort of 210 patients treated between 2022 and 2024 in the same centre. Results: Two joint Friberg models effectively described carboplatin-induced myelosuppression. Serum albumin, eGFR and paclitaxel and pemetrexed co-medications were included in the final model. On the test cohort, >85% of NT and >87% of PLT observations fell within the 90% confidence interval of Bayesian model predictions, confirming that the model can support dose adjustments for subsequent treatment cycles. An example of model-based dose adjustments is also presented with a simulation study. Conclusions: The pop-PK-PD model demonstrated strong performance in describing and predicting carboplatin-induced myelosuppression, thus providing a valuable strategy for dose personalization. Further refinements and validation steps are needed before integrating such an approach into clinical workflows.