The promises and potential pitfalls of pharmacokinetic model–based dosing in cellular therapy

TO THE EDITOR:

It was with great interest that I read the commentaries by my colleagues, Sánchez-Salinas et al and Wicha et al, regarding our analysis, “Identifying an optimal fludarabine exposure for improved outcomes after axi-cel therapy for aggressive B-cell non-Hodgkin lymphoma (B-NHL).”^1-3 In the former commentary, the authors present data from their analysis of measured fludarabine pharmacokinetic (PK) levels in patients with aggressive B-NHL receiving CD19 chimeric antigen receptor (CAR) T-cell therapy with axicabtagene ciloleucel or tisagenlecleucel. Using these levels, they calculated a measured exposure and then compared those values with the population PK (popPK) model-predicted exposure (what we used in our analysis) that relies on patient and treatment variables rather than measured PK levels.² They found that the model-predicted method often underestimated the fludarabine exposure compared with the measured method. The main question was clearly stated in their title: is estimated exposure an accurate surrogate for measured fludarabine levels in CAR T-cell patients? Wicha et al³ expressed similar concerns regarding the variability when evaluating fludarabine exposure using model-based estimates without the use of therapeutic drug monitoring (TDM), that is, PK levels.

I will start by agreeing with my colleagues’ list of the potential reasons why the model-predicted fludarabine exposures differed from measured-predicted exposures, including the following: the popPK model was developed in patients receiving fludarabine in the context of allogeneic hematopoietic cell transplantation (allo-HCT) and not in patients receiving CD19 CAR T cells, the fludarabine doses used in allo-HCT were different and fludarabine was combined with different chemotherapies other than cyclophosphamide, and the model-predicted exposure may have limited ability to capture variations in weight and renal function that may occur over the course of the 3-day lymphodepleting chemotherapy (LD chemo) regimen.^4,5 Both commentaries express their most well-taken point; that is, calculating the exposure based on measured levels is certainly a truer reflection of actual fludarabine exposure than model-predicted exposure.^2,3 

Our efforts with regard to fludarabine dosing should evoke the most successful example of TDM in the cellular therapy field, which is the use of PK-directed dosing of busulfan in patients undergoing allo-HCT.^6,7 Although implemented as a standard at many centers, the approach is resource intensive and may not be amenable to widespread use. As we consider the feasibility of individualized fludarabine dosing with CD19 CAR T-cell therapy, it is imperative that we make the methods highly adaptable given that complex and labor-intensive processes may limit their routine applicability. The use of a model-based exposure method is simpler than TDM, but as was echoed by Sánchez-Salinas et al,² the model estimates must be predictable and reliable to be used in practice. Of interest, what may be required would be iteration of the previously published popPK model by adding a large cohort of PK level data points from patients with aggressive B-NHL receiving CD19 CAR T cells or altogether building a new popPK model specific to these patients. Bolstering the existing model or building a CAR T-cell specific model may remove some of the confounding variables such as patient characteristics and concomitant chemotherapies given with fludarabine.

A critical concept in PK/pharmacodynamic analyses is that estimates of exposure are only as reliable as the data that were used to formulate the popPK model and, therefore, the PK model itself. Langenhorst et al’s⁴ popPK model used in our study (and in the Sánchez-Salinas et al’s² work) was derived using data from 258 adult and pediatric patients whose median age was 18 years. In contrast, most patients receiving CD19 CAR T cells for aggressive B-NHL are significantly older and are therefore more likely to have renal dysfunction. I agree with the commentators that measured PK levels each day along with use of daily creatinine levels would likely improve the accuracy of the estimated fludarabine exposures. Although changes in weight are possible over the 3-days of LD chemo, creatinine (and therefore creatinine clearance) rather than weight is more likely to dynamically change during the LD chemo period. Owing to data limitations in our multicenter analysis, we used baseline (day −5) weights and creatinine values to estimate cumulative fludarabine exposure. If a patient’s creatinine clearance changed considerably between day −5 and −3 of LD chemo, this would not have been captured by our PK estimates. Ultimately, this represents a limitation of our data.¹ Wicha et al³ commented on the limitations of the Cockcroft-Gault equation in estimating estimated glomerular filtration rate (eGFR); however, Langenhorst’s popPK model relies on the Cockcroft-Gault equation in adults (and Schwartz equation in younger patients). In addition, most standard of care chemotherapy dosing strategies, including those for fludarabine, use Cockcroft-Gault equation for eGFR estimation to guide dose reductions in cases of lower eGFR. Therefore, at least for now, we are bound to using the Cockcroft-Gault equation regardless of whether model-based or TDM methods are used to estimate fludarabine exposure. Using a different eGFR estimation method would likely lead to better precision, but this would warrant a revision of the fludarabine popPK model itself.

Using measured levels to calculate exposure will get us closer to the actual drug exposure rather than popPK model-predicted estimates. What remains clear is that with current standard body surface area–based fludarabine dosing, achieving an optimal exposure is a “shot in the dark.” The model-based dosing we used in our previous analysis may get us closer but given that fludarabine PK levels were not imputed into the model, there is considerable uncertainty regarding the exposure estimations. Therefore, the narrow potential optimal fludarabine exposure of 18 to 20 mg × h/L we found must be approached with caution and, as we discussed in our publication, would require prospective validation ideally in a randomized study before enacting in practice.¹ 

Despite many years of using busulfan TDM in patients undergoing allo-HCT, center-specific differences in exposure estimation methods led to considerable variability.⁸ However, the development, validation, and use of busulfan popPK models have led to more harmonization in practice.^7,8 

Whether we may eventually consider model-based dosing for its simplicity or rather use measured PK-directed fludarabine dosing for its accuracy, collaboration, and harmonization across multiple centers is key to bringing precision chemotherapy dosing to clinical practice. As such, I thank my colleagues for their critical, but fair assessments, and I look forward to working with them on optimizing LD chemo for improved outcomes after CD19 CAR T-cell therapy.^2,3 

Contribution: M.S. conceived of, wrote, and edited this response to the commentaries.

Conflict-of-interest disclosure: M.S. served as a paid consultant for McKinsey & Company; Angiocrine Bioscience, Inc; and Omeros Corporation; received research funding from Angiocrine Bioscience, Inc, Omeros Corporation, Amgen Inc, Bristol Myers Squibb, and Sanofi; served on ad hoc advisory boards for Kite, a Gilead company and Miltenyi Biotec; and received honoraria from i3Health, Medscape, and CancerNetwork for CME-related activity and honoraria from IDEOlogy.

Correspondence: Michael Scordo, Adult Bone Marrow Transplant Service, Cellular Therapy Service, Department of Medicine, Memorial Sloan Kettering Cancer Center, 530 E 74th St, New York, NY 10021; email: scordom@mskcc.org.