Alessandro De Carlo (1), Mirjam Crul(2,3), Tim Schutte (2), Lia van Zuylen (2,3), Idris Bahce (2), Chi Fong Loo (2), Daan van Valkengoed (2,4), Harmen Huls (2), Elena Maria Tosca (1), Paolo Magni (1)*, Imke Bartelink* (2,3)
(1) Laboratory of Bioinformatics, Mathematical Modelling and Synthetic Biology (BMS Lab), Department of Electrical Computer and Biomedical Engineering, University of Pavia, Pavia, Italy, (2) Amsterdam UMC, Location VUmc, The Netherlands, (3) Amsterdam Cancer Center, The Netherlands, (4) Division of System Pharmacology and Pharmacy, Leiden Academic Center for Drug Research, Leiden University, The Netherlands , (*) Shared authorship
Background: Carboplatin is a cytotoxic anticancer drug used in monotherapy or in combination with other chemotherapeutics and/or immunotherapies to treat different types of cancers [1]. As an alkylator, carboplatin affects all dividing cells, including healthy cells, thus provoking adverse events (AE) in patients [2]. Severe myelotoxicity is the most frequent dose limiting carboplatin-induced AE, and both therapeutic drug monitoring (TDM) and dose individualization, beyond the Calvert formula based on Cockgroft & Gault equation [3], have been proposed to circumvent its onset and preserve treatment efficacy [2,4].
Objective: A population PK-PD model (popPK-PD) of carboplatin effect on platelets (PLT) and neutrophils (NT) was developed and validated to inform TDM and dose individualization in cancer patients of the Amsterdam University Medical Center (UMC). To this aim, real world data from Amsterdam UMC Electronic Health Records (EHR) were leveraged. Target population was focused on the most frequent tumor diagnoses in which carboplatin is a standard treatment (i.e., lung, gynecological, and esophageal/gastric neoplasm).
Methods: NT, PLT and individual covariates (i.e. age, serum albumin, eGFR (based on CKD-EPI[5]), body surface area (BSA), body mass index BMI, concomitant immuno and/or chemotherapies) collected in the EHR of 580 targeted cancer patients treated at Amsterdam UMC with carboplatin between January 2019 and June 2022 were used to develop the PK-PD model of myelosuppression (training set). Given the absence of PK samples in clinical routine, individual carboplatin PK parameters were extrapolated using a literature popPK model developed on a Dutch patient population [6]. Prediction corrected VPC, AIC and BIC scores were used in the development stage to identify the final model. The predictive performances of the final popPK-PD model were validated on a test population of target cancer patients (N=211) treated at Amsterdam UMC with Carboplatin between June 2022 and January 2024. All the analyses were performed by using Monolix suite and R.
Results: Two joint Friberg models [7] best described carboplatin myelosuppressive effects on NT and PLT in the training set. Correlations between baseline PLT and NT parameters (=0.44) and NT and slopeNT (=0.72, NT rebound effect and carboplatin cytotoxic effect on NT, respectively) were found significant and were included in the final model to describe inter-individual variability (IIV). Covariate analysis led to include the following patient characteristics on model parameters to explain IIV:
- baseline PLT: stomach cancer diagnosis (β=-0.16), serum albumin/44 g/L (β=-0.50)
- PLT mean maturation time: Paclitaxel co-medication (β=0.31)
- slopePLT (Carboplatin toxicity effect on PLT): serum albumin/44 g/L (β=-0.26), Paclitaxel co-medication (β=-0.56), linear effect of treatment cycle (β=0.76)
- baseline NT: Pemetrexed co-medication (β=0.36), serum albumin/44 g/L (β=-0.47)
- slopeNT: eGFR/100 ml/min (β=-0.60)
These findings indicate a chemotherapy-related decrease of renal functions [8]. Furthermore, the estimated effect of albumin on baseline and drug effect may be related to patient disease or inflammation status [9,10]. All parameters of the final model were well estimated with RSE<30%. The final popPK-PD model was retrospectively evaluated on the test set to assess its suitability to inform carboplatin TDM through a full Bayesian approach. In particular, at each cycle i, data until cycle i-1 were used to estimate patient posterior parameters using the developed PK-PD model as prior information. Then, posterior parameters were used to predict next cycle observations. Results showed that, at each cycle (N=5), almost all observed NT (>85%) and PLT (>87%) fell within the 95% C.I. of posterior prediction.
Conclusions: The popPK-PD model developed on EHR data of 580 Amsterdam UMC patients showed good performances in both describing and predicting carboplatin-induced myelosuppression in a validation cohort of 211 patients. Results of the evaluation in the Bayesian framework suggest that model predictions can be used to inform next cycle dose adjustment in a TDM context. Moving forward, this model will be integrated into the Reinforcement Learning framework [11] to further optimize individualized dose suggestions [12,13] and refine the current Cockgroft & Gault based carboplatin dosing strategy.
References:
[1] Ho GY, Woodward N, Coward JIG. Cisplatin versus carboplatin: comparative review of therapeutic management in solid malignancies. Critical Reviews in Oncology/Hematology 2016;102:37–46.
[2] Gutierrez F, Gonzalez-de-la-Fuente GA, Nazco GJ et al. Hematological toxicity of carboplatin for gynecological cancer according to body mass index. European Journal of Clinical Pharmacology 2016;72:1083–9.
[3] Calvert AH, Newell DR, Gumbrell LA et al. Carboplatin dosage: prospective evaluation of a simple formula based on renal function. J Clin Oncol 1989;7:1748–56.
[4] Paci A, Veal G, Bardin C et al. Review of therapeutic drug monitoring of anticancer drugs part 1–cytotoxics. Eur J Cancer 2014;50:2010–9.
[5] Levey AS, Stevens LA, Schmid CH et al. A new equation to estimate glomerular filtration rate. Ann Intern Med 2009;150:604–12.
[6] Ekhart C, de Jonge ME, Huitema ADR et al. Flat dosing of carboplatin is justified in adult patients with normal renal function. Clin Cancer Res 2006;12:6502–8.
[7] Friberg LE, Henningsson A, Maas H et al. Model of chemotherapy-induced myelosuppression with parameter consistency across drugs. J Clin Oncol 2002;20:4713–21.
[8] Santos MLC, de Brito BB, da Silva FAF et al. Nephrotoxicity in cancer treatment: An overview. World J Clin Oncol 2020;11:190–204.
[9] Margraf A, Zarbock A. Platelets in Inflammation and Resolution. J Immunol 2019;203:2357–67.
[10] Soeters PB, Wolfe RR, Shenkin A. Hypoalbuminemia: Pathogenesis and Clinical Significance. JPEN J Parenter Enteral Nutr 2019;43:181–93.
[11] Sutton RS, Barto AG. Reinforcement Learning: An Introduction. MIT press, 2018.
[12] De Carlo A, Tosca EM, Fantozzi M et al. Reinforcement Learning and PK-PD Models Integration to Personalize the Adaptive Dosing Protocol of Erdafitinib in Patients with Metastatic Urothelial Carcinoma. Clinical Pharmacology & Therapeutics 2024; DOI: 10.1002/cpt.3176.
[13] De Carlo, Alessandro, Tosca, Elena Maria, Magni, Paolo. Integrating Reinforcement Learning and PK-PD modelling to enable precision dosing: a multi-objective optimization for the treatment of Polycithemia Vera patients with Givinostat. PAGE 31.
Reference: PAGE 32 (2024) Abstr 10981 [www.page-meeting.org/?abstract=10981]
Poster: Real-world data (RWD) in pharmacometrics