Use of a Cystatin C-Based GFR Equation in a Population Pharmacokinetic Model of Methotrexate Clearance in Adult Patients with Lymphoma

Abstract

Background

High-dose methotrexate (HDMTX) is a key treatment for lymphoma with central nervous system involvement. Whether incorporating cystatin C into glomerular filtration rate estimation improves methotrexate (MTX) clearance prediction remains unclear.

Objectives

We aimed to evaluate whether cystatin C-inclusive glomerular filtration rate equations improve MTX clearance prediction and to explore the relationship between MTX exposure and acute kidney injury (AKI) in adult patients with lymphoma receiving HDMTX.

Methods

This was a prospective single-center study performed on 80 adult patients with lymphoma receiving HDMTX (1.5–8 g/m²) over a 4-h infusion. A population pharmacokinetic model was constructed using data from 80 administrations of HDMTX and 427 serum MTX concentrations. The population pharmacokinetic model estimated MTX concentrations were included in a logistic regression to assess the relationship between MTX exposure and AKI.

Results

A two-compartment model best described the pharmacokinetic data, with baseline albumin and CKD-EPI creatinine-cystatin C (eGFRCr-CysC) as significant covariates on clearance. Seventeen patients (21%) developed any-stage AKI. Among those receiving ≤ 3.5 g/m², model-estimated 4-h MTX concentrations were associated with AKI (odds ratio: 1.02 per µmol/L; p = 0.0038), with an optimal threshold of 160 µmol/L (area under the concentration–time curve: 0.818). Patients above this threshold were 22 times more likely to experience AKI (p = 0.0005). This association was not observed in patients treated with 8 g/m². Despite a lower dose and exposure, patients receiving ≤ 3.5 g/m² demonstrated a stronger concentration–toxicity relationship.

Conclusions

Our results support the use of cystatin C-inclusive glomerular filtration rate estimates in MTX pharmacokinetic modeling and suggest early MTX concentration sampling may identify AKI risk, enabling proactive, AKI-mitigating clinical interventions during HDMTX therapy.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40262-026-01618-4.

Key Points

Cystatin C is a kidney function marker that was more strongly associated with methotrexate clearance than serum creatinine in adults with lymphoma. Including cystatin C in kidney function estimates improved the prediction of methotrexate elimination and supports its use in monitoring during therapy.

Measuring an early methotrexate concentration (4 h after the start of the infusion) helped predict acute kidney injury in patients receiving 3.5 g/m2 of methotrexate (but not in those receiving 8 g/m2), highlighting the value of an early sample as a tool to identify at-risk patients and support proactive care to prevent kidney injury and other complications.

Introduction

High-dose methotrexate (HDMTX, ≥ 1 g/m²) is a cornerstone of treatment for primary central nervous system (CNS) lymphoma and patients with systemic disease at risk for relapse in the CNS because it is one of the few agents that reliably crosses the blood–brain barrier [1–3]. Though the preferred dosage remains unclear, methotrexate (MTX) appears to achieve therapeutic concentrations in the CNS when more than 3 g/m² is administered intravenously over 4 h [4, 5]. Current HDMTX regimens are selected based on a patient’s disease etiology, with the CNS as the sole area of disease or as a secondary spread of systemic disease, most commonly resulting in the use of either 3.5 g/m² or 8 g/m² of HDMTX [6–8]. Rapid control of CNS involvement for primary diffuse large B-cell lymphoma is necessary to prevent significant neurologic morbidity and is often treated with HDMTX intravenously at 8 g/m² over 4 h [9]. Conversely, patients may receive 3.5 g/m² to treat heterogeneous presentations of systemic disease, including systemic diffuse large B-cell lymphoma, primary CNS lymphoma with systemic involvement, or cases where HDMTX is administered prophylactically to prevent CNS relapse [10]. While previous studies have not stratified HDMTX-associated toxicities by treatment regimen (3.5 g/m² vs 8 g/m²), existing data suggest that 3.5 g/m² may be associated with higher rates of MTX-related adverse events despite its lower dosing [10, 11].

Optimizing absolute MTX exposure through appropriate dose selection and supportive care during post-infusion clearance (CL) is necessary to maximize anti-tumor effects while minimizing treatment-related toxicity [12–14]. Supportive care adjustments (e.g., fluids, sodium bicarbonate, folinic acid) are reactions to laboratory values (e.g., serum MTX concentration, serum creatinine) collected during routine therapeutic drug monitoring after HDMTX administration. Standard therapeutic drug monitoring for adults receiving HDMTX over a 4-h infusion includes initial sample collections at 24 and 48 h after the start of the MTX infusion, with subsequent collections occurring every 6–24 h until the patient achieves discharge status [15].

Methotrexate is primarily eliminated through the kidneys via glomerular filtration, making timely monitoring of renal function and accurate glomerular filtration rate (GFR) estimation imperative for safe and efficacious HDMTX delivery [16, 17]. The primary method for GFR estimation in pharmacokinetic (PK) models for MTX and the most common approach to HDMTX adjustments in clinical practice relies on serum creatinine and estimated creatinine CL based on the Cockcroft–Gault equation [7, 12, 18–21]. As the terminal byproduct of skeletal muscle catabolism, non-renal determinants, including altered muscle mass, deconditioning, and malnutrition, all decrease the accuracy of serum creatinine-based GFR estimates in patients with cancer [22]. Furthermore, serum creatinine is a delayed and insensitive indicator for AKI, which has a variable incidence for patients receiving HDMTX, with a range from 1.5 to 30% [23, 24]. These limitations contribute to the poor performance of serum creatinine and estimated creatinine CLas markers of MTX elimination [18, 25, 26].

Cystatin C is an endogenous low-molecular-weight protein that has emerged as a practical adjunct or alternative to serum creatinine for GFR estimation [27, 28]. Cystatin C is less dependent on age, sex, race, or muscle mass than serum creatinine, but may be affected by non-GFR determinants such as systemic inflammation, thyroid dysfunction, and obesity [29–31]. Serum cystatin C has been used to assess kidney function in patients treated for CNS lymphoma when serum creatinine was thought to be a suboptimal GFR surrogate [32]. A recent non-compartmental analysis demonstrated that cystatin C-inclusive GFR estimates, whether expressed in mL/min or mL/min/1.73 m², predicted MTX CL better than estimates based on serum creatinine alone [33]. These GFR estimating (eGFR) equations have not been evaluated for associations with MTX CL in patients with lymphoma receiving HDMTX therapy. The primary objectives of this study were: (1) to perform a population PK modeling analysis evaluating the roles of serum creatinine, cystatin C, and eGFR in predicting MTX CL and (2) to explore the association between serum MTX concentrations and the occurrence of AKI in adult patients receiving HDMTX therapy.

Methods

Setting and Patient Population

This was a prospective single-center study that included adult (aged ≥ 18 years) patients with histologically confirmed lymphoma admitted for intravenous HDMTX between January 2018 and December 2019 at Mayo Clinic in Rochester, MN, USA. Patients with a new diagnosis or relapse of disease were eligible and could be enrolled regardless of prior MTX exposure. Patients were excluded if receiving a MTX infusion extending beyond 4 h as dictated by the treatment protocol, if they presented with AKI of any stage according to Kidney Disease Improving Global Outcomes (KDIGO) guidelines prior to MTX therapy, or currently receiving renal replacement therapy [34]. Written informed consent was provided before all data collection and sampling. The protocol was approved by the Mayo Clinic Institutional Review Board and followed the ethical standards of the 1964 Declaration of Helsinki with adherence to all relevant regulations of the US Health Insurance Portability and Accountability Act.

High-dose MTX (1.5 g/m², 3.5 g/m², or 8 g/m²) as monotherapy every 21-28 days, HDMTX (3.5 g/m²) in combination with rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone every 21 days with HDMTX administered on day 14 of the cycle (MR-CHOP), or HDMTX (8 g/m²) in combination with rituximab and temozolomide (MRT) was prescribed per published treatment protocols at the discretion of the primary hematologist or the hospital-based care team [7, 8, 35]. Table S1 of the Electronic Supplementary Material (ESM) provides additional information about the treatment regimens, HDMTX administration, and HDMTX-based supportive management.

Study procedures and the diagnostic performance of the liquid chromatography-mass spectrometry MTX assay with a lower limit of detection of 10 ng/mL (~0.02 µmol/L) have previously been described [33]. Briefly, samples of blood and urine were collected by the Mayo Clinic Clinical Research and Trials Unit at specific, predetermined timepoints (e.g., 4, 12, 24, 48, 72, and 96 h after the start of the HDMTX infusion) to evaluate MTX concentrations, serum creatinine, and serum cystatin C. In addition to the prospectively collected specimens, data were abstracted from existing information available in the electronic health records, including patient demographics, height, weight, comorbid conditions, and other laboratory data relevant to the study.

Population PK Modeling

Base PK Model

A population PK analysis using non-linear mixed-effects modeling was performed using NONMEM (Version 7.5; ICON Software Development, Ellicott City, MD, USA). Model development and analysis were facilitated by the Pirana Modeling Workbench (Version 3.0.0) and Perl-speaks-NONMEM (Version 5.2.6). Both two- (ADVAN3 TRANS4) and three-compartment (ADVAN11 TRANS4) structural base models were evaluated. The analyses were performed using log-transformation of serum MTX concentrations with the first-order conditional estimation with interaction method. A log-normal distribution was assumed for each PK parameter, with inter-individual variability described using an exponential model. Covariances among inter-individual variability terms were systematically evaluated by adding off-diagonal elements to the OMEGA matrix and retained only if supported by significant reductions in the objective function and stable parameter estimation. Residual unexplained variability was modeled using a proportional error structure implemented via the log-transformed both sides approach to account for heteroscedasticity across the concentration–time profile. The final base structure used a two-compartment model parameterized as CL (L/h), volume of distribution of the central compartment (V1; L), inter-compartmental CL (Q; L/h), and volume of distribution of the peripheral compartment (V2; L). A review of the concentration–time profiles (Fig. S1 of the ESM) visually supports the use of a two-compartment structural model.

Covariate Analysis

The relationships between pre-defined individual patient-level data and treatment details on empirical Bayes estimates of CL and V1 were examined during an exploratory covariate analysis. Covariates demonstrating biologically plausible or statistically meaningful trends were advanced to stepwise covariate modeling (Sect. 2.2.3). Correlations among covariates were evaluated using scatterplots and correlation matrices, and highly correlated variables were not simultaneously tested to avoid redundancy.

Evaluated covariates included baseline demographic data (i.e., age, body surface area, body mass index, sex, and weight), baseline treatment details (i.e., the purpose of treatment and drug administration number), time-dependent renal function biomarkers (i.e., serum creatinine and serum cystatin C), and baseline laboratory values (albumin, alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, blood urea nitrogen, hemoglobin, platelets, and total bilirubin). Missing baseline laboratory data were replaced using population median values or regression-based estimates, and sensitivity analyses confirmed that parameter estimates remained consistent across imputation methods.

Continuous covariates, such as baseline albumin, hemoglobin, and platelets, were also evaluated as categorical variables using clinically relevant thresholds to improve interpretability and facilitate clinical translation. Hypoalbuminemia was defined as an albumin ≤ 3.5 g/dL. Anemia was defined as a hemoglobin ≤ 13.5 g/dL for men, and < 12.0 g/dL for women. Thrombocytopenia was defined as platelets ≤ 150 × 10⁹/L. Baseline body mass index was evaluated as both a continuous variable (kg/m²) and as a categorical variable using the Centers for Disease Control and Prevention definitions of underweight, healthy, overweight, and obese for adults [36]. All serum creatinine and serum cystatin C were modeled as longitudinal (baseline and time-dependent) covariates using absolute laboratory values and transformed into eGFR (mL/min/1.73 m²) using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) serum creatinine 2021 equation (CKD-EPI_Cr), the CKD-EPI serum cystatin C 2012 equation (CKD-EPI_CysC), or the CKD-EPI creatinine-cystatin 2021 equation (CKD-EPI_Cr-CysC) [27, 28]. Longitudinal changes to serum creatinine, cystatin C, and eGFR are displayed in Fig. S2 of the ESM.

Stepwise Covariate Modeling

Forward stepwise model-building evaluated continuous covariates using linear, exponential, and power models structures based on the results of the exploratory covariate analysis, while categorical variables were implemented as fractional changes using indicator variables. Variable inclusion was driven by several diagnostic metrics, which included using the objective function value (OFV; a decrease in OFV by at least 6.635 [p < 0.01]), Akaike information criterion (AIC), and Bayesian information criterion for quantitative model comparisons, reductions in the inter-individual variability, residual standard error, and model shrinkage. Model bias was determined by mean prediction error, and model precision was determined by root mean squared error. Visual evaluations of the individual predictions versus observed, population predictions versus observed, conditional weighted residuals versus time, and conditional weighted residuals versus population predictions goodness-of-fits plots were also considered.

Model Evaluation

A prediction-corrected visual predictive check was used to assess the performance and predictive accuracy of the final population PK model. Using the prediction-corrected visual predictive check command in Pirana, simulated concentration–time profiles were generated based on the model parameter estimates and individual covariates, then binned using the automatic binning algorithm implemented in Pirana. Prediction-corrected percentiles were then calculated by normalizing concentrations to the median of the dosing and covariate distribution for each simulated timepoint. The observed and simulated prediction-corrected percentiles were overlaid to visually assess the model’s ability to describe the observed data. A bootstrap resampling analysis evaluated the robustness of parameter estimates and quantified their uncertainty. A total of 1000 bootstrap datasets were generated by sampling with replacement from the original dataset. The model was refitted to each bootstrap dataset, and the distributions of the parameter estimates were used to calculate each parameter’s 95% confidence intervals (CIs). The final model parameters were compared to the median values obtained from the bootstrap estimates to confirm the stability and reliability of the model.

Exposure-Response Analysis

Fisher’s exact test assesses the independence of any stage AKI based on a patient’s protocol-defined MTX dose (≤ 3.5 g/m² vs 8 g/m²). The final population PK model generated patient-specific post hoc estimates of MTX concentrations at the end of the 4-h infusion. Model-generated concentrations were summarized using the median and interquartile range (IQR). Pairwise comparisons between groups were performed using the Wilcoxon rank-sum test. A binomial logistic regression model with a logit link function was fit in R (Version 4.5.0) using the glm function to evaluate the association between 4-h MTX concentration and the occurrence of any stage AKI. Patients were stratified by protocol-defined MTX dose (≤ 3.5 g/m² vs 8 g/m²) for logistic regression analysis, irrespective of chemotherapy regimen (monotherapy or combination). Univariate logistic regression models were fit separately for each protocol-defined MTX dose (≤ 3.5 g/m² vs 8 g/m²), and a multi-variate model including protocol-defined MTX dose as a covariate was also performed. Predicted probabilities from these models were plotted to visually illustrate the exposure–response relationship for each MTX dose group. These models were not intended for individual risk prediction. Receiver operating characteristic (ROC) curve analyses using Youden’s Index were performed to evaluate model discrimination and identify a concentration threshold with optimal sensitivity and specificity, intended as a candidate value for future prospective evaluation. The identified threshold was further assessed for association with any stage AKI using Fisher’s exact test and quantified by odds ratio (OR). An unadjusted p value < 0.05 was considered statistically significant.

Results

Patient Baseline Demographics and Clinical Characteristics

Eighty patients satisfied the inclusion criteria and were enrolled in the study. The median age was 69 (IQR: 59, 76) years, 54 (68%) were male, and 74 (93%) were White. The median body weight was 81 kg (IQR: 70, 92) and the median body surface area was 1.97 m² (IQR: 1.8, 2.14) as calculated by the Du Bois method. Baseline eGFR ranged between 83 mL/min and 99 mL/min, depending on the equation used. There were five patients with a history of chronic kidney disease at baseline. Additional baseline demographics, clinical characteristics, laboratory values, and HDMTX-based chemotherapy treatment information are provided in Table 1. Baseline albumin was the only variable with missingness; values were replaced for 17 (21%) patients using population median or regression-based estimates as described in the Methods. A total of 43 (54%) patients received a MTX dose ≤ 3.5 g/m² for prophylaxis of lymphoma relapse in the CNS, and 37 (46%) patients received a MTX dose of 8 g/m² for the treatment of lymphoma with CNS involvement.

Table 1.

Baseline demographics, clinical characteristics, and chemotherapy information for patients with lymphoma at the Mayo Clinic prescribed HDMTX

Characteristic	Patients (N = 80)
Age (years), median (IQR)	68.6 (59.3–75.9)
Patients aged 65 years or older, n (%)	51 (64)
Male, n (%)	54 (68)
Caucasian, n (%)	74 (93)
Weight (kg), median (IQR)	80.5 (69.7–92.3)
BSA (m2), median (IQR)	1.97 (1.80–2.14)
Male	2.04 (1.93–2.15)
Female	1.77 (1.68–1.93)
Diagnosis, n (%)
DLBCL	44 (57)
Primary DLBCL of the CNS	23 (30)
EBV positive DLBCL of the elderly	6 (8)
Other	7 (9)
Bone marrow involvement, n (%)	10 (12.5)
Percent involvement, median (IQR)	15 (9–35)
Kidney parameters at baseline
History of chronic kidney disease, n (%)	5 (6)
Serum creatinine (mg/dL)	0.8 ± 0.4
Cystatin C (mg/dL)	1.1 ± 0.5
Estimated kidney function (mL/min), median (IQR)
Cockcroft-Gault eCrCl	99 ± 46
CKD-EPI eGFRcreatinine	93 ± 27
CKD-EPI eGFRcystatin C	83 ± 26
CKD-EPI eGFRcreatinine-cystatin C	88 ± 24
Albumin at baseline (g/dL), median (IQR)	3.8 (3.5–4.2)
Chemotherapy regimen distribution, n (%)
HDMTX 8 g/m2 with rituximab and temozolomide	37 (46)
HDMTX 3.5 g/m2 in combination with R-CHOP	29 (36)
HDMTX monotherapy	14 (18)
Protocol-defined dose (g/m2), n (%)
8	37 (46)
3.5	34 (43)
1.5	9 (11)
Delivered dose (g), median (IQR)	7.55 (4.83–11.25)
Prevention of CNS involvement	4.85 (3.4–6.85)
Treatment of active CNS disease	10.85 (7.8–13.5)
Disease burden, n (%)
Systemic disease	37 (46)
CNS disease	30 (38)
Both	13 (16)
MTX dose history
First MTX dose, n (%)	38 (48)
Second MTX dose, n (%)	26 (32)
Beyond second MTX dose, n (%)	16 (20)
Number of doses beyond second dose, median (IQR)	4 (3, 9)

BSA body surface area, CNS central nervous system, CKD EPI Chronic Kidney Disease Epidemiology Collaboration, DLBCL diffuse large B-cell lymphoma, EBV Epstein–Barr virus, eCrCl estimated creatinine clearance, eGFR estimated glomerular filtration rate, HDMTX high-dose methotrexate, IQR interquartile range, MTX methotrexate, R-CHOP rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone

Population PK Modeling

Base Model

The serum MTX concentration–time data were best described using a two-compartment PK model. Initial model building attempted to parameterize the interindividual variability for CL, V1, Q, and V2; however, parameterizations of the inter-individual variability for V1 and Q were not supported (parameterization of OMEGAs approached 0). A two-parameter OMEGA matrix (CL-V1) enabled the parameterization of V1 with a moderate correlation (R = 0.51) and ∆OFV = − 4.63, but resulted in 82% relative standard error and 37% shrinkage. A full OMEGA block was not supported by these data. Sensitivity testing led to a parsimonious model with parameterization of CL and V2 only. The base model parameterization for the four PK parameters (CL, V1, Q, V2) resulted in stable estimates with RSEs of 5.3%, 7.7%, 13.5%, and 12.7% for CL, V1, Q, and V2, respectively, and a base OFV of 55.9 (Table 2).

Table 2.

Final population pharmacokinetic model parameters for high-dose methotrexate in adult patients with lymphoma

Parameters	Final model			Bootstrap
	Estimate	RSE (%)	Shrinkage (%)	Median	95% CI
CL (L/h)	11.8	5.1		11.8	10.8–12.8
ΘeGFR	0.80	11.5		0.79	0.64–0.95
ΘAlbumin	0.69	21.3		0.67	0.42–0.97
V1 (L)	42.2	6.4		41.9	37.8–46.5
Q (L/h)	0.35	11.6		0.35	0.29–0.43
V2 (L)	6.67	10.4		6.63	5.56–7.88
IIV CL	0.02	24.3	13.3	0.02	0.01–0.03
IIV V2	0.03	23.9	31	0.03	0.02–0.05
Residual error	0.20	11.6	12.2	0.20	0.16–0.23

IIV and residual error are reported as the variance of the selected parameter

CI confidence interval, CL clearance, eGFR estimated glomerular filtration rate, IIV inter-individual variability, Q intercompartmental clearance, RSE relative standard error, V1 volume of distribution of the central compartment, V2 volume of distribution of the peripheral compartment

Covariate Model

A preliminary univariate analysis demonstrated that MTX CL (L/h) was statistically associated with age, albumin, serum cystatin C, serum creatinine, and eGFR. Demographics (i.e., body surface area, body mass index, and weight), baseline treatment details (i.e., the purpose of treatment and drug administration number), and baseline laboratory values (i.e., alkaline phosphatase, alanine aminotransferase, aspartate aminotransferase, blood urea nitrogen, hemoglobin, platelets, and total bilirubin) were not selected for stepwise covariate modeling. Improvements to the base model with the inclusion of these variables are described in Table S2 of the ESM. Serum cystatin C outperformed serum creatinine, producing a lower OFV (OFV = − 25.5 vs 13.5, respectively) and explained more inter-individual variability on CL, reducing inter-individual variability by 10.4% compared with 3.7%. However, the addition of the CKD-EPI_Cr-CysC eGFR resulted in the largest decreases in OFV (ΔOFV = 117.1, p < 0.0001), AIC, and Bayesian information criterion (Table S2 of the ESM) compared with other eGFR equations and recorded renal function values, and reduced the inter-individual variability on CL from 28.5 to 17%, corresponding to approximately 11.5% of total variability explained by eGFR. The individual- and population-level bias and precision were similar for all renal models (Table S3 of the ESM). The CKD-EPI_Cr-CysC eGFR variable was added to the model equation using a power function (Eq. 1) with a reference value set to the median GFR value, which enabled parameter estimation that reflects the kidney function in this adult population. The addition of albumin to the CKD-EPI_Cr-CysC model resulted in an additional improvement in OFV (ΔOFV = 21.1, p < 0.0001), AIC, and Bayesian information criterion (Table S2 of the ESM), and explained 2.9% of the inter-individual variability. Therefore, the final model includes the CKD-EPI_Cr-CysC eGFR and baseline albumin (Eq. 1).

$${\text{CL}}_i={\text{CL}}_p\ast {\left(\frac{{\text{eGFR}}_i}{76}\right)}^{{\theta }_{\text{eGFR}}}\ast {\left(\frac{{\text{Alb}}_i}{4}\right)}^{{\theta }_{\text{Albumin}}}$$ 1

Goodness-of-fit plots show that the final model did not display model bias nor misspecification (Fig. S3 of the ESM).

Model Evaluation

The median and 95% CI for each PK parameter are reported for the bootstrap analysis of 1000 simulated patients (Table 2). The bootstrap analysis was in agreement with the model parameters derived from the population PK model. In addition to the bootstrap analysis, a prediction-corrected visual predictive check is presented in Fig. 1. The median prediction interval displays continuity with the median observed concentrations from the simulated patient data.

Fig. 1.

Prediction-corrected visual predictive check of the final population pharmacokinetic model for high-dose methotrexate in adult patients with lymphoma. The solid black line represents the median observed concentration for the simulation (N = 1000). The gray shaded area is the median prediction interval. The dashed black lines represent the 10th and 90th percentiles of the observed concentrations. The shaded blue areas represent the 10th and 90th percentiles of the prediction intervals. The ‘•’ are observed patient data. The analysis was completed in Pirana using the automatic binning algorithm. The figure was created using NONMEM2R (CRAN, Version 0.2.5)

HDMTX Treatment, Pharmacokinetics, and AKI

A total of 17 (21%) patients experienced MTX-related AKI, with the first recorded event occurring 9 h after the start of the infusion (Fig. S4 of the ESM). Of those 17 patients, 13 (16% of total) experienced any stage AKI after receiving ≤ 3.5 g/m² of HDMTX, and 11 of the 17 patients (14% of total) experienced either stage 2 or stage 3 AKI. Patient demographics, baseline laboratory values used in the covariate model, and estimated PK parameters were similar between the 43 patients (54%) who received ≤ 3.5 g/m² of MTX and the 37 patients (46%) who received 8 g/m² of MTX (Table 3). However, the frequency of any-stage AKI appeared to differ by MTX dose (p = 0.054), with patients receiving ≤ 3.5 g/m² being 3.58 times more likely to experience any-stage AKI than those with 8 g/m² [95% CI: 1.1–10.8] (Table 3). Patients receiving ≤  3.5 g/m² HDMTX were also 2.59 times more likely to experience Stage 2 and Stage 3 AKI compared with those with 8 g/m² [95% CI 0.68–9.53; p = 0.208]

Table 3.

Impact of dose on MTX pharmacokinetics and AKI

Variables	MTX dose: ≤3.5 g/m2 N = 43	MTX dose: 8 g/m2 N = 37	P value
Age (years)	68.4 (59.2–76.6)	68.8 (61.4–73.7)	0.835
Sex (male)	33 (76)	21 (56)	0.0962
Protocol dose (g/m2)	2.6 (1.75–3.48)	6.0 (4.7–7.7)	<0.0001
Total dose (g)	5.1 (3.4–6.9)	11.5 (9.1–13.75)	<0.0001
Kidney parameters at baseline
Serum creatinine (mg/dL)	0.90 (0.79–1.08)	0.89 (0.66–1.09)	0.365
Cystatin C (mg/dL)	1.03 (0.86–1.17)	1.08 (0.90–1.23)	0.352
Estimated kidney function (mL/min)
CKD-EPI eGFRcreatinine	89.3 (71.4–96.3)	90.1 (68.2–102.8)	0.647
CKD-EPI eGFRcystatin C	74.0 (60.7–89.7)	65.4 (56.0–85.1)	0.249
CKD-EPI eGFRcreatinine-cystatin C	82.5 (69.3–96.5)	80.2 (63.8–95.4)	0.661
Albumin at baseline (mg/L)	3.9 (3.3–4.2)	3.8 (3.5–4.1)	0.850
Clearance of methotrexate (L/h)	11.9 (10.6–13.1)	11.4 (10.6–12.1)	0.307
Occurrence of any stage of AKI (N = 17)	13 (30)a	4 (11)a	0.054

^aPercent of column total

AKI acute kidney injury, CKD-EPI Chronic Kidney Disease Epidemiology Collaboration, eGFR estimated glomerular filtration rate, MTX methotrexate

Despite receiving a lower protocol dosage (3.5 g/m²), lower total dose (g), and lower model-estimated 4-h concentration (µmol/L), patients receiving ≤ 3.5 g/m² exhibited a stronger concentration–toxicity association, as shown by the plotted predicted probabilities from a multi-variate logistic regression (Fig. 2) of any-stage AKI by MTX dose group (≤ 3.5 g/m², 8 g/m²) using the 4-h serum MTX concentration estimated from the final model with eGFR (CKD-EPI_Cr-CysC) and albumin on CL. The impact of MTX dose and serum concentration on the presence of AKI was further explored in post hoc univariate logistic regressions in the full cohort, where the HDMTX protocol dose (OR: 0.97, 95% CI 0.84–1.10; p = 0.66), total dose (OR: 0.89, 95% CI 0.66–1.16; p = 0.40), and 4-h serum MTX (OR: 1.00, 95% CI 0.99–1.01; p = 0.83) showed no positive association with AKI.

Fig. 2.

Multi-variate logistic regression comparing the serum methotrexate concentration at 4 h post-infusion and the occurrence of any acute kidney injury for adult patients with lymphoma receiving a high dose. The solid blue line represents the plotted probability for patients on a ≤ 3.5-g/m² dose of high-dose methotrexate. The solid black line represents the plotted probability for patients on an 8-g/m² dose of high-dose methotrexate. The shaded regions represent the 95% confidence interval. The points represent patient data. The dashed black lines represent the 50% probability for any acute kidney injury for each protocol dose and are provided for visual reference only

The median model-estimated 4-h concentrations were higher in patients who experienced AKI within that administration compared with patients who did not experience AKI, which was observed in each respective disease type (Table 4). Univariate modeling on patients receiving ≤ 3.5 g/m² revealed that the estimated 4-h MTX concentration (per µmol/L) was significantly associated with higher odds of any stage AKI (OR: 1.02, 95% CI 1.01–1.05; p = 0.0038) compared with the null model (AIC: 42.85 vs 54.6). The ROC curve for patients receiving ≤ 3.5 g/m² identified the optimal discriminatory point as 160 µmol/L, with an ROC area under the concentration–time curve of 0.818 (95% CI 0.650–0.986), a sensitivity of 0.77, and a specificity of 0.87. The presence of any stage AKI was dependent upon a patient’s estimated 4-h concentration > 160 µmol/L (p = 0.0005), with patients above this threshold being 22 times more likely to experience any stage [95% CI 3.7–89.4]. Conversely, the 4-h concentration (per µmol/L) was not associated with the presence of any stage AKI for patients receiving 8 g/m² (OR: 1.00, 95% CI 0.99–1.01; p = 0.47) compared to the null model (AIC: 28.81 vs 27.34, respectively).

Table 4.

Model estimated 4-h serum MTX concentration by dose and AKI status

Treatment groups	AKI (N = 17)	No AKI (N = 63)	Row total
MTX dose: ≤ 3.5 g/m2; N = 43	182.8 (153.4–241.8)	116.2 (78.1–150.9)	129.2 (84.1–178.8)
MTX dose: 8 g/m2; N = 37	335.6 (265.4–479.6)	313.3 (231.5–385.1)	315.7 (240.2–534.5)

Model estimated MTX concentrations (µmol/L) are reported median (interquartile range)

AKI acute kidney injury, MTX methotrexate

Discussion

This study of 80 adult patients with lymphoma receiving HDMTX therapy investigated the roles of serum creatinine, cystatin C, and eGFR on MTX CL and sought to determine the association between serum MTX concentrations and the occurrence of any stage AKI. The results demonstrate that (1) adding the CKD-EPI_Cr-CysC eGFR equation to the base two-compartment model markedly improved the estimation of MTX CL compared with the CKD-EPI_Cr and CKD-EPI_CysC eGFR equations in adult patients with lymphoma; (2) that serum MTX concentrations were only associated with the occurrence of any stage AKI for patients receiving HDMTX at ≤ 3.5 g/m², but not for patients receiving HDMTX at 8 g/m²; and (3) the collection of the 4-h post-infusion serum MTX sample provided insights that could enable proactive supportive care modifications that mitigate the risk for AKI in adults receiving HDMTX for the management of lymphoma.

The population PK modeling of MTX CL in adult patients with lymphoma after HDMTX therapy was best described using a two-compartment structural model that included the CKD-EPI_Cr-CysC eGFR equation, and albumin as clinical covariates (Table S2 of the ESM). The addition of the CKD-EPI_Cr-CysC eGFR resulted in a ΔOFV = 117.1, and explained approximately 11.5% of the total inter-individual variability. Baseline albumin resulted in a ΔOFV = 21.1 and explained 2.9% of the inter-individual variability. Our study is in accordance with prior population PK models of HDMTX that demonstrate a two-compartment model as best for short infusion protocols, and our estimated parameters for CL and volume of distribution are in line with other publicly available works in adult patients [37–40]. Additionally, stepwise covariate modeling identified other variables, including albumin, that were significant predictors of MTX pharmacokinetics and of MTX CL, which adds to the growing body of evidence on albumin’s role in MTX CL [39–42]. Albumin has also been associated with increased risk of HDMTX-induced AKI, supporting its role as a baseline biomarker before initiating therapy [43, 44].

Including the CKD-EPI_Cr-CysC eGFR equation as a longitudinal covariate significantly improved model performance compared with other eGFR formulas or individual biomarkers (Table S2 of the ESM), and accounted for approximately 11.5% of total inter-individual variability on CL. This is consistent with prior findings that eGFR reflects baseline filtration capacity but not dynamic processes such as renal tubular secretion, urine alkalinization, or transient treatment-related renal effects that also influence MTX elimination. Cystatin C has demonstrated utility in monitoring kidney function of children with acute lymphoblastic leukemia receiving HDMTX treatment [45]. Additionally, cystatin C can be incorporated into GFR estimation in patients with characteristics that predispose serum creatinine-based eGFR to overestimate true kidney function such as cachexia or reduced skeletal muscle mass [29, 32]. The inclusion of the CKD-EPI_Cr-CysC eGFR equation on MTX CL has only recently been described in adult patients with lymphoma using a non-compartmental analysis [33]. Cystatin C as a laboratory value and CKD-EPI_CysC eGFR were sufficient to describe population HDMTX pharmacokinetics better than serum creatinine (Table S3 of the ESM). These data support that cystatin C, in substitution of or in addition to serum creatinine, offers invaluable insights into HDMTX pharmacokinetics and could greatly benefit proactive supportive care changes to mitigate excessive MTX exposure associated with the occurrence of AKI. These results also confirm findings from the non-compartmental analysis and strengthen the evidence for cystatin C being an integral marker to assess kidney function in patients requiring HDMTX therapy [11, 46]. While cystatin C is not yet universally adopted, it is increasingly accessible because of its compatibility with existing automated laboratory platforms. The primary cost to institutions is the reagent itself, which is estimated at $5–$10 per test, approximately 8–20 times the cost of creatinine reagents (~ $0.50). However, this cost is comparable to, or slightly lower than, other commonly used biomarkers such as C-reactive protein, troponin T, and B-type natriuretic peptide [47, 48]. Furthermore, the 2025 Medicare Clinical Diagnostic Laboratory Fee Schedule lists reimbursement for cystatin C at $18.52 per test, versus $5.12–$9.46 for creatinine (Centers for Medicare & Medicaid Services, 2025). Although formal cost-effectiveness analyses specific to drug dosing decisions have not yet been published, the potential for improved dosing precision and toxicity risk stratification may justify its use in select high-risk populations.

Supportive care practices during HDMTX therapy require routine collection of serum creatinine to monitor for changes in kidney function and prompt intervention to mitigate AKI [7, 33, 49]. Prior works have documented the impact of demographic, clinical, and PK variables on the occurrence of AKI following HDMTX therapy [39, 49–51]. Uniquely, this exposure–response analysis determined that the 4-h post-infusion serum MTX concentration was associated with any stage AKI, but its effect was dependent upon the patient’s disease type (Table 3). The foundational pharmacologic principle that higher doses produce higher exposures and greater response was not observed. In our cohort, adult patients receiving the largest dose of HDMTX (8 g/m²) with the highest 4-h serum MTX concentrations had a lower frequency of reported AKI compared with patients receiving a lower dose of HDTMX (≤ 3.5 g/m²) with lower 4-h serum MTX concentrations (Tables 3, 4). This violation of our underlying pharmacological assumptions ascertains that a combination of factors, rather than dose alone, could predispose patients to a heightened risk of HDMTX-associated toxicities. Stratifying the analysis allowed us to isolate the exposure–response relationship within each clinically distinct subgroup, reducing confounding from the protocol-defined dose and disease type. This approach revealed that the association between 4-h MTX concentrations and any stage AKI was significant in the ≤ 3.5-g/m² group but not in the 8-g/m² group, suggesting a patient- or disease-specific susceptibility to toxicity. The ≤3.5 g/m² patients were also 2.59 times more likely to experience Stage 2 and Stage 3 AKI compared to those with 8 g/m2. This is supported by our previous report that showed patients with bone marrow involvement receiving HDMTX had significantly greater odds of developing AKI compared with patients without bone marrow involvement (OR: 5.09, 95% CI 1.26–20.60, p = 0.023). Additionally, recent literature demonstrates the unexpectedly substantial toxicity (delayed CL and treatment interruptions) when 3.5 g/m² of MTX was given concurrently with R-CHOP for CNS prophylaxis [10, 52]. Ultimately, understanding a patient’s basal metabolic capacity before an HDMTX infusion could proactively guide supportive care and optimize MTX exposure [53].

Importantly, standard supportive care protocols for adult patients receiving HDMTX do not typically include the collection of a 4-h post-infusion sample. A key strength of this study was the collection of early MTX samples at 4 and 12 h post-infusion, enabling detection of concentration–AKI relationships not typically observed in standard protocols [7, 8]. The 4-h sample also represents a feasible and clinically meaningful timepoint collected at the end of the infusion, preceding the earliest observed AKI events in this cohort. Although formal optimal sampling analyses were not performed, this timepoint offers an actionable balance between model informativeness and clinical feasibility. Further, recent findings from an evaluation of juvenile pigs suggest that serum MTX and serum creatinine concentrations collected very early after an HDMTX infusion can be used to predict impending AKI [54]. This 4-h post-HDMTX sample could be a powerful addition to future treatment protocols, though it requires confirmation in a larger sample population. Additionally, expanding protocols to include the collection of serum cystatin C at 4-h could further aid in a kidney function assessment and our understanding of a patient’s MTX CL.

This study has several limitations. First, our study population had a limited sample size from a single center; however, our cohort is one of the larger prospective PK studies in adults with lymphoma to incorporate pre-defined early serum MTX concentration monitoring and serum cystatin C concentrations as an additional biomarker of kidney function. Additionally, the consecutive enrollment over the study period permitted consistent supportive management strategies across our population. Second, there is significant known heterogeneity in the MTX protocol dose amount, therapy indication, chemotherapeutic protocol, and disease burden, and unknown heterogeneity such as smoking status, chronic inflammation, or thyroid disease that can affect cystatin C. Some patients were enrolled for their first MTX dose versus those receiving a subsequent dose, and the MTX indication of lymphoma treatment versus HDMTX to prevent lymphoma relapse in the CNS. Furthermore, the analysis did not include the effect of concurrent medications on HDMTX elimination; however, current treatment protocols hold known perpetrators before and during HDMTX therapy, resulting in a low likelihood that adverse medications would have been concomitantly administered during MTX therapy. This population-level heterogeneity, limited sample size, and lack of data for repeated MTX administrations could explain the higher residual variability observed in our final model parameters. However, this population represents a real-world practice and facilitated comparisons among adult patients with lymphoma, all of whom received HDMTX as a 4-h infusion for lymphoma treatment or prevention of lymphoma relapse in the CNS. Third, while patients were permitted to refuse sampling at any point, over 90% of the requested samples were fulfilled, making investigators confident in the modeling results. Fourthly, 20% of patients experienced AKI, which is a limited number of patients with AKI to explore the association between exposure and response. Additional urine data (i.e., urine volume, urine pH) were collected per treatment protocol but not included in the analysis, given protocolization of hyperhydration and urine alkalinization. Last, the PK model and logistic regression did not undergo external validation. While the logistic regression and ROC curve offer invaluable insights into MTX pharmacokinetics and any stage AKI, this evidence would need to be explored further in a future prospective study. Importantly, AKI is a time-dependent outcome, with the earliest event in this cohort occurring 9 h after infusion initiation. The analysis described herein used all available MTX PK data available through 4 h after the start of the infusion to remove the confounding feedback loop of the presence of AKI slowing MTX CL, and slowed MTX CL increasing the risk of AKI. The exposure–response analysis was fit for purpose; however, time-to-event modeling is a needed next step to better describe the occurrence of MTX-associated AKI.

Conclusions

Our findings suggest that (1) cystatin C improves MTX CL estimation; (2) disease burden may increase susceptibility to MTX-related toxicities; and (3) early 4-h MTX concentrations may inform supportive care adjustments to mitigate AKI risk in adult patients with lymphoma receiving HDMTX. These observations support our study objective to refine early identification of patients at an elevated toxicity risk, and further demonstrate the potential for cystatin C to enhance creatinine-based GFR estimation and improve characterization of MTX exposure.

A prospective multi-center study is warranted to validate these results and confirm the clinical utility of early MTX concentration monitoring and cystatin C-inclusive GFR estimation in lymphoma care. Additional work should explore whether these strategies can be integrated into model-informed precision dosing frameworks to further improve safety outcomes.

Supplementary Information

Below is the link to the electronic supplementary material.

Funding

This project was supported in part by the Hematology Oncology Pharmacy Association (HOPA) Foundation Research Grant Program, the National Institutes of Health Center for Translational Science Activities Grant Number TL1 TR002380 from the National Center for Advancing Translational Science (NCATS), the National Center Institute Grant Number P30 CA015083, and K23AI143882 from the National Institute of Allergy and Infectious Diseases (principal investigator, Erin F. Barreto). Its contents are solely the opinion of the authors and do not necessarily represent the official views of the National Institutes of Health.

Declarations

Conflicts of interest/competing interests

This work was conducted in the absence of a competing interest or financial relationship; however, the authors report the following: Zachary L. Taylor and Laura B. Ramsey have received research funding from SERB Pharmaceuticals. Erin F. Barreto serves in a consulting or advisory role for Wolters-Kluwer and Baxter-Health. Erin F. Barreto has received research funding from Numares AG. Andrew D. Rule receives royalties from UpToDate. Nelson Leung has stock or ownership interests in Senseonics, AbbVie, Verrica Pharmaceuticals, and Checkpoint Therapeutics. Nelson Leung has received research funding from Omeros. Thomas E. Witzig has received honoraria, from Curio Science, serves in a consulting or advisory rolefor Tornado Therapeutics and Salarius Pharmaceuticals, and has received research funding fromCelgene, Acerta Pharma, Kura Oncology, Acrotech Biopharma, and Karyopharm Therapeutics. Jason N. Barreto serves in a consulting or advisory role for Optum, and has received research funding from Asellas Pharma and Numares AG. Kristin C. Cole, Kianoush B. Kashani, and Carrie A. Thompson have no conflicts of interest that are directly relevant to the content of this article.

Ethics approval

This prospective single-center study was conducted in accordance with Good Clinical Practice guidelines and the ethical principles of the Declaration of Helsinki. The protocol was reviewed and approved by the Mayo Clinic Institutional Review Board (ID: 17-007837). Patient confidentiality was maintained in compliance with the US Health Insurance Portability and Accountability Act.

Consent to participate

All participants provided written informed consent prior to enrollment in accordance with the protocol that was approved by the Mayo Clinic Institutional Review Board.

Consent for publication

The authors affirm that human research participants provided informed consent for publication of their data at the time of study enrollment, as part of the informed consent process.

Availability of data and material

The datasets that support the findings of this study are not publicly available because of institutional privacy policies but may be made available from the corresponding author upon reasonable request and with appropriate data use agreements.

Code availability

Population PK modeling was performed using NONMEM (Version 7.5) with Pirana and Perl-speaks-NONMEM (PsN) workbench tools. No custom code was developed for this study beyond standard model scripting within these platforms. Model scripts may be made available upon reasonable request to the corresponding author.

Authors’ contributions

ZLT, EFB, KMC, ADR, KBK, NL, CAT, TEW, LBR, and JNB contributed to the conception and design of the research. ZLT, EFB, KMC, and JNB performed the research. ZLT, EFB, KMC, ADR, KBK, CAT, TEW, LBR, and JNB analyzed the data. All authors contributed to the drafting and critical revision of the manuscript. All authors have read and approved the final version of the manuscript and agree to be accountable for all aspects of the work.