Predictive performance of population pharmacokinetic models of imatinib in chronic myeloid leukemia patients

Background and aim Chronic myeloid leukemia is a myeloproliferative neoplasm associated with the specific chromosomal translocation known as the Philadelphia chromosome. Imatinib is a potent BCR-ABL tyrosine kinase inhibitor, which is approved as the first line therapy for CML patients. There are various population pharmacokinetic studies available in the literature for this population. However, their use in other populations outside of their cohort for the model development has not been evaluated. This study was aimed to perform the predictive performance of the published population pharmacokinetic models for imatinib in CML population and propose a dosing nomogram. Methods A systematic review was conducted through PubMed, and WoS databases to identify PopPK models. Clinical data collected in adult CML patients treated with imatinib was used for evaluation of these models. Various prediction-based metrics were used for assessing the bias and precision of PopPK models using individual predictions. Results Eight imatinib PopPK model were selected for evaluating the model performance. A total of 145 plasma imatinib samples were collected from 43 adult patients diagnosed with CML and treated with imatinib. The PopPK model reported by Menon et al. had better performance than all other PopPK models. Conclusion Menon et al. model was able to predict well for our clinical data where it had the relative mean prediction error percentage ≤ 20%, relative median absolute prediction error ≤ 30% and relative root mean square error close to zero. Based on this final model, we proposed a dosing nomogram for various weight groups, which could potentially help to maintain the trough concentrations in the therapeutic range. Supplementary Information The online version contains supplementary material available at 10.1007/s00280-024-04644-w.


Introduction
Chronic myeloid leukemia (CML) is a myeloproliferative neoplasm associated with the specific chromosomal translocation in chromosomes 9 and 22, which is known as the Philadelphia chromosome (Ph).Around 95% of the patients with CML are detected with Ph chromosome [1].In the general community, the diagnosis of CML is found to be 1-2 instances per 100 000 population.Adults with newly diagnosed leukemia account for about 15% of all leukemia cases [2].Imatinib mesylate (IM) belongs to the potent class of anticancer drugs known as BCR-ABL tyrosine kinase selective inhibitors.It was introduced in the year 2000 and thereafter, it has become the gold standard for treatment of CML.The death rate due to CML has come down from the range of 10-20% to 1-2% after the introduction of imatinib [2,3].
In patients with CML, imatinib is given at a standard dose of 400 mg in chronic phase and 600 mg in accelerated and blast phase [4], orally once daily, which is well tolerated.It has been noted that imatinib has high intra-individual variability (IIV) and between subject variability (BSV) in the pharmacokinetic parameters [5][6][7].Despite the excellent response imatinib produces, some CML patients show resistance to drug or therapy failure after a primary response.Variations in plasma trough concentration (C trough ) of Imatinib could affect clinical responses in CML, and it is also related to the occurrence of prominent adverse events (AEs) such as edema or fluid retention, neutropenia, grade 3 myalgia and others [6,[8][9][10][11][12].Standard doses are used for treating different phases of CML; however, attaining clinical outcome in all the treated patients is a challenge.Trough concentrations greater than 1 µg/mL have been shown to help achieve higher efficacy [13][14][15].Few studies have shown that if the trough concentration are kept between 1and 1.5 µg/ mL, there is a higher chance of therapeutic success without drug resistance and a lower chances of developing AEs [8,16].Population pharmacokinetics (PopPKs) is a branch of pharmacometrics that comes from quantitative clinical pharmacology, which explores the sources and correlates to variability in drug concentrations [17].
Pharmacometrics in precision medicine considers variability and recommends dosage regimen according to the sub-populations and has broad applications in clinical oncotherapeutics [18].This study was aimed to evaluate the predictive performance of the published population pharmacokinetics model for Imatinib in CML and propose a dosing nomogram based on the model which had a better prediction ability.

Systematic review for identification of PopPK models
A systematic review was conducted by two reviewers independently to identify relevant literatures.If there was any difference in opinion during this process, then a third reviewer was consulted to resolve the issue and come to an agreement.The systematic search was conducted by following PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-analyses) criteria in the PubMed and Web of Science databases.The articles published till December 31, 2022, were included in the review.The keywords used in the databases were "Imatinib", "Population Pharmacokinetic", "PopPK", "Chronic myeloid leukemia" and, "CML".Imatinib population pharmacokinetics in CML patients published in English language were included in the review.Conference abstracts, editorials and book chapters were excluded.

Data extraction and model evaluation
Details of study like title and author details, study design, treatment and drug quantification details and pharmacokinetic model related data which includes model structure, parameter values, BSV, RUV, and covariates were collected.

PopPK models replication
All selected PopPK models were replicated using the structural model and typical parameters obtained from the published values of each candidate model.Empirical Bayesian estimate was used to derive individual-based predictions for imatinib plasma concentration for each time point.

Clinical data for external evaluation of PopPK models
The clinical data for 49 subjects was collected in a prospective observational study from CML patients, who were on Imatinib therapy through routine therapeutic drug monitoring.This study was conducted in ACTREC, Mumbai a tertiary care oncology hospital after approval from institutional ethics committee.The following demographics and clinical patient information were collected: gender, age, body-weight, height, liver function test, blood cells count, concomitant medication, single nucleotide polymorphisms (SNPs) which included ABCB1, ABCG2, CYP3A4 genotypes, and imatinib treatment information (Dose, sampling time, plasma concentration, stage of CML).

Drug quantification
A high-performance liquid chromatography technique developed by Peng et al. [1] was used to ascertain the plasma drug concentration.Erlotinib was used as the internal standard and the method was validated over a linear range of 100 to 10,000 ng/ml [19].

External evaluation of PopPK model predictive ability
The predictive performance of the selected PopPK model was performed with the external dataset using the PUMAS AI package.Graphical representations of the results were performed with R version 4.2.1.The PUMAS package version 2.3.0 with Julia computing language was the tool used to run all the pharmacometric simulations [20].

Model evaluation using prediction-based metrics
Relative mean prediction error percentage [rMPE (%)], relative median absolute prediction error percentage [rMAPE (%)] and relative root mean squared error (rRMSE) are the most commonly used metrics for evaluating model bias and precision [21].In this study, we used rMPE (%) to predict the bias of the PopPK models.The rMPE (%) was calculated by comparing the individual predicted concentration (C ipred ) estimated by using the parameters in the structural and stochastic models to the final estimates of each PopPK model to the observed concentration (C obs ) for each subject.

Literature search and study selection
A total of 78 research publications were identified through the PubMed and Web of Science database searches.After the title and abstract screening, a total of 15 articles were left for review.Out of 15 articles, seven were excluded as they were therapeutic drug monitoring (TDM) studies and the PopPK model of Imatinib for gastrointestinal stromal tumors.Eight population pharmacokinetic studies of Imatinib in CML [22][23][24][25][26][27][28][29] met the inclusion criteria and were included in this study after the scrutiny of the literature.Supplementary Table 1 summarizes the demographic data of studies such as weight, dose, study population and covariates tested in the models.Supplementary Table 2 shows the search strategy and Fig. 1 depicts the study selection methodology.In most of the studies, the modeling tool used for the development of the PopPK models was NONMEM software package.The relation between different covariates and pharmacokinetic parameters and the values of BSV were reported in these studies.Visual predictive check and bootstrap analysis were the common reported approaches for model evaluation.
PK parameters such as Clearance (CL), absorption rate constant (Ka), and volume of distribution (Vd), IIV, RUV, and inferences from various studies are presented in Table 1.All the selected studies reported one-compartment model for imatinib with either first-order or zero-order absorption with or without lag time.Proportional error model was the most common error model used in these studies for explaining residual error.

Clinical data
From a total of 49 subjects, six were dropped from the analysis because of the unusually high-drug concentration found during the elimination phase in the data, which did not make any pharmacokinetics sense.External validation was performed using the remaining data from 43 adult CML patients with 145 imatinib plasma concentrations.Blood samples were collected between 0.5 and 24 h after administration of imatinib doses, based on the convenience of patients when they were able to provide samples.The distribution of sample collection time after dosing is depicted in Supplementary Fig. 1.All imatinib concentrations were measured accurately.Most of the patients were in chronic phase (74.4%) of CML.Table 2 displays the clinical and demographic characteristics of the external validation dataset.Imatinib was given orally, with doses ranging from 400 to 800 mg, at a median dose of 400 mg once daily.Samples taken during and after 20 h of administration have been regarded as representing the trough concentration.43 of the 145 imatinib plasma concentrations were trough concentrations.Only 17 (39.5%)imatinib trough concentrations fell within the therapeutic range of 1.0-1.5 µg/mL [8]; the remaining samples were either above or below this level.

External validation
The findings of the model performance based on predictionbased diagnostics using the 8 PopPK models selected are shown in Table 3.The plot of observed vs. individual predicted concentrations for each model is depicted in Fig. 2 ), rMAPE of 5.59% (between 30%), and the lowest rRMSE of 0.16 (close to zero).Figure 3 demonstrates the precision of the imatinib PopPK models that were examined [29].The Menon et al.'s model is a one-compartment with first-order elimination and zero-order absorption.Body weight was included allometrically in the model to explain CL and Vd.

Simulation approach for designing dosing nomogram
The Menon et al. model was used to develop the dosing nomogram in the weight range of 45-120 kg (Table 4).Virtual subjects were first created for the range of covariates with narrow intervals to create a dosing nomogram.Body weights ranging from 40 to 120 kg were divided into groups with 3 kg intervals.The PK profile for different weight groups was simulated with 1000 subjects in each group using imatinib doses ranging from 400 to 800 mg with a 100 mg interval.All weight groups were tested with each dosage regimen to determine the best-dosage regimen for patients.The target trough concentration for the simulation was set at 1.0-1.5 µg/mL, which falls within the acceptable therapeutic range.More than 95% of the simulations in each weight group fell within the therapeutic range.The clinician may use the recommended dosing nomogram to maintain the imatinib concentration in the therapeutic range.

Discussion
In the last decade, a number of population pharmacokinetic studies were published about many drugs in different populations.Predictability and applicability of these models outside their study settings have not been evaluated.There is a need for evaluating these models using either external data from other published studies or from prospective clinical studies.Such evaluations will lend these models amenable for application in clinical situation that needs precision dosing [16,30].All the chosen PopPK models were externally evaluated with our clinical data.In these evaluations, the model reported by Golabchifar et al. [23], Menon et al. [29], and Widmer et al. [27] performed well with the acceptable Fig. 1   count by the factor of two, then both CL and Vd increased by 8% and 5% respectively [25].Takahashi et al. stated that therapeutic drug monitoring along with ABCG2 421 C > A genotyping might be helpful in improved imatinib therapy management as patients with ABCG2 421C/A or A/A genotype were associated with a higher trough concentration than ABCG2 421C/C genotype [14] but when we tried to predict with ABCG2 as a covariate we found no significant impact on the trough concentration among these genotypes.Nevertheless, therapeutic drug monitoring might be important to understand whether the patient's trough concentrations is above or below the required threshold value (1-1.5 µg/mL).
Several studies have reported that the trough concentration for imatinib is higher in patients with major molecular response (MMR) than in patients without MMR.The same pattern was recognized for complete cytogenetic response (CCyR).The patients had significantly higher MMR (minimum of 3 log reduction in the level of BCR-ABL transcript from a standard baseline after initiation of therapy), when the trough concentration was found to be higher than 1 µg/ mL [13][14][15]33].There is still a debate on the therapeutic range of imatinib in CML therapy [34].Main reason attributed to the treatment resistance for imatinib in CML patients   was trough concentrations lower than 1 µg/mL [35].Higher imatinib trough concentrations of 1.5 µg/mL & 3 µg/mL have been reported to be associated with toxicities such as neutropenia, grade 3 myalgia, fluid retention and eyelid edema [5,9,10].
In Tyrosine Kinase Inhibitor Optimization and Selectivity (TOPS) trial, it has been reported that there is an association between the imatinib C trough and the frequency of AEs.Imatinib C trough above 1.165 µg/mL had high MMR and CCyR but C trough above 3.18 µg/mL has resulted in higher frequency of grade 3 or 4 AEs.However, the upper limit for the therapeutic window have not been reported [36].Recent studies have suggested that the imatinib C trough can be maintained between 1 and 1.5 µg/mL because of higher probability of therapeutic success without drug resistance and lower probability of developing AEs [7,8,37].Based on these reports, we considered the window between 1 and 1.5 µg/mL for proposing dosage regimen for various bodyweight ranges.
Pharmacometric simulations using the identified generalizable model for the weight range of 45-120 kg with standard regimen resulted in C trough values less than 1 µg/mL in certain weight groups.In certain body-weight groups, trough concentrations were well over 1.5 µg/mL.These simulations showed the need for varied doses in different body-weight groups as opposed to the present fixed dosage regimen to all the body-weight ranges to attain the therapeutic range.
The dosage regimen proposed is based on the simulation for the typical subjects with specific body weight considering variability.The current recommendation remains a broad guideline for initial doses in patients with specific body weights and the doses may have to be adjusted again based on the trough levels of imatinib, if they are outside the desired ranges.
This study has several limitations, including a small number of patients (n = 43) and imatinib plasma samples (n = 145), which may limit the robustness of the results, differences between structural and stochastic PopPK models due to the high BSV in imatinib PK previously described in the literature, and the impact of ethnicity on model performance.The findings of this study also suggest that additional imatinib TDM samples are required when the patient did not show signs of improvement in MMR and CCyR.

Fig. 2
Fig. 2 Individual predicted concentrations by each model vs. observed concentrations from the present clinical study

Fig. 3 Table 4
Fig. 3 Relative median aboslute prediction error distribution of the imatinib PopPK models evaluated using external clinical dataset.Dashed black line, rMAPE equal to ± 30% (acceptable bias) . The PopPK model developed by Golabchifer et al., Menon et al., and Widmer et al. had good prediction ability.Other PopPK models that were tested underpredicted imatinib concentrations in our population.However, the Menon et al. model outperformed the Golabchifer and Widmer model in terms of prediction ability, with rMPE of 0.98% (between 20% PRISMA flow chart a covariate on CL and Vd.In contrast, Golabchifar et al.'s model included no covariates on the PK parameters, whereas Widmer et al.'s model included age, gender, stage of diagnosis, and weight.The clinical dataset available for external white blood cell (WBCs) count and hemoglobin (Hb) value.Patients with higher body-weight and Hb values reported to have increased CL by 12% and 86% and increase in the Vd by 32% and 60% respectively.If a patient had reduced WBC

Table 1
Study parameters and Inference CL Clearance (Liter/hr), Vd Volume of distribution (Liter), Ka Absorption rate constant (per hour), D1 Duration for absorption (hour), IIV Interindividual variability, RUV Residual unexplained variability, PE Proportional error, AE Additive error, %RSE Relative Standard Error Above table displays the pharmacokinetic parameters and inferences of available/collected studies

Table 2
Patient Clinical and Demographic characteristics (N = 43)

Table 3
An overview of prediction-based diagnostics