Covariate selection in pharmacometric analyses: a review of methods

Full random effects models (FREM): A practical usage guide

The full random-effects model (FREM) is an innovative and relatively novel covariate modeling technique. It differs from other covariate modeling approaches in that it treats covariates as observations and captures their impact on model parameters using their covariances. These unique characteristics mean that FREM is insensitive to correlations between covariates and implicitly handles missing covariate data. In practice, this implies that covariates are less likely to be excluded from the modeling scope in light of the observed data. FREM has been shown to be a useful modeling method for small datasets, but its pre-specification properties make it a very compelling modeling choice for late-stage phases of drug development. The present tutorial aims to explain what FREM models are and how they can be used in practice.

Machine-Learning Assisted Screening of Correlated Covariates: Application to Clinical Data of Desipramine

Stepwise covariate modeling (SCM) has a high computational burden and can select the wrong covariates. Machine learning (ML) has been proposed as a screening tool to improve the efficiency of covariate selection, but little is known about how to apply ML on actual clinical data. First, we simulated datasets based on clinical data to compare the performance of various ML and traditional pharmacometrics (PMX) techniques with and without accounting for highly-correlated covariates. This simulation step identified the ML algorithm and the number of top covariates to select when using the actual clinical data. A previously developed desipramine population-pharmacokinetic model was used to simulate virtual subjects. Fifteen covariates were considered with four having an effect included. Based on the F1 score (an accuracy measure), ridge regression was the most accurate ML technique on 200 simulated datasets (F1 score = 0.475 ± 0.231), a performance which almost doubled when highly-correlated covariates were accounted for (F1 score = 0.860 ± 0.158). These performances were better than forwards selection with SCM (F1 score = 0.251 ± 0.274 and 0.499 ± 0.381 without/with correlations respectively). In terms of computational cost, ridge regression (0.42 ± 0.07 seconds/simulated dataset, 1 thread) was ~20,000 times faster than SCM (2.30 ± 2.29 hours, 15 threads). On the clinical dataset, prescreening with the selected ML algorithm reduced SCM runtime by 42.86% (from 1.75 to 1.00 days) and produced the same final model as SCM only. In conclusion, we have demonstrated that accounting for highly-correlated covariates improves ML prescreening accuracy. The choice of ML method and the proportion of important covariates (unknown a priori) can be guided by simulations.

Serum bile acid change correlates with improvement in pruritus in patients with primary biliary cholangitis receiving linerixibat

BACKGROUND & AIMS

Total serum bile acid (TSBA) levels are elevated in patients with primary biliary cholangitis (PBC) and may mediate cholestatic pruritus. Linerixibat, an ileal bile acid transporter inhibitor, improved pruritus in patients with PBC. We explored the relationship between linerixibat dose, TSBA concentration, and pruritus.

METHODS

Data from Phase 1/2 trials were used to develop a population kinetic-pharmacodynamic model to characterize the linerixibat dose-TSBA relationship. Individual Bayesian parameter estimates for participants in the GLIMMER study were used to derive the area under the TSBA concentration curve over 24 h (AUC0-24). Time-matched post hoc estimates of AUC0-24 were correlated with pruritus reported on a 0-10 numerical rating scale. Baseline TSBA concentration was correlated with change from baseline (ΔBL) in monthly itch score (MIS). ΔBL in model-estimated TSBA AUC0-24 was correlated with time-matched ΔBL in weekly itch score (WIS) or MIS.

RESULTS

Linerixibat dose dependently reduced TSBA AUC0-24, reaching steady state after 5 days. Baseline TSBA levels in GLIMMER did not correlate with ΔBL in MIS. ΔBL in TSBA AUC0-24 correlated with improved WIS over 12 weeks of treatment (r = 0.52, p < 0.0001). Of participants with a ≥30% decrease in TSBA AUC0-24, 60% were pruritus responders (≥2-point improvement in WIS from baseline).

CONCLUSIONS

Linerixibat treatment leads to rapid, dose-dependent TSBA reductions. Baseline TSBA levels do not correlate with on-treatment pruritus change, suggesting they do not predict linerixibat response. Change in TSBA AUC0-24 correlates significantly with, and can be predictive of, pruritus improvement in patients with PBC.

Covariate modeling in pharmacometrics: General points for consideration

Modeling the relationships between covariates and pharmacometric model parameters is a central feature of pharmacometric analyses. The information obtained from covariate modeling may be used for dose selection, dose individualization, or the planning of clinical studies in different population subgroups. The pharmacometric literature has amassed a diverse, complex, and evolving collection of methodologies and interpretive guidance related to covariate modeling. With the number and complexity of technologies increasing, a need for an overview of the state of the art has emerged. In this article the International Society of Pharmacometrics (ISoP) Standards and Best Practices Committee presents perspectives on best practices for planning, executing, reporting, and interpreting covariate analyses to guide pharmacometrics decision making in academic, industry, and regulatory settings.

Using forest plots to interpret covariate effects in pharmacometric models

The inclusion of covariates in pharmacometric models is important due to their ability to explain variability in drug exposure and response. Clear communication of the impact of covariates is needed to support informed decision making in clinical practice and in drug development. However, effectively conveying these effects to key stakeholders and decision makers can be challenging. Forest plots have been proposed to meet these communication needs. However, forest plots for the illustration of covariate effects in pharmacometrics are complex combinations of model predictions, uncertainty estimates, tabulated results, and reference lines and intervals. The purpose of this tutorial is to outline the aspects that influence the interpretation of forest plots, recommend best practices, and offer specific guidance for a clear and transparent communication of covariate effects.

Impact of covariate model building methods on their clinical relevance evaluation in population pharmacokinetic analyses: comparison of the full model, stepwise covariate model (SCM) and SCM+ approaches

Covariate analysis in population pharmacokinetics is key for adjusting doses for patients. The main objective of this work was to compare the adequacy of various modeling approaches on covariate clinical relevance decision-making. The full model, stepwise covariate model (SCM) and SCM+ PsN algorithms were compared in a clinical trial simulation of a 383-patient population pharmacokinetic study mixing rich and sparse designs. A one-compartment model with first-order absorption was used. A base model including a body weight effect on CL/F and V/F and a covariate model including 4 additional covariates-parameters relationships were simulated. As for forest plots, ratios between covariates at a specific value and that of a typical individual were calculated with their 90% confidence interval (CI90) using standard errors. Covariates on CL, V and KA were considered relevant if their CI90 fell completely outside the reference area [0.8-1.2]. All approaches provided unbiased covariate ratio estimates. For covariates with a simulated effect, the 3 approaches correctly identify their clinical relevance. However, significant covariates were missed in up to 15% of cases with SCM/SCM+. For covariate with no simulated effects, the full model mainly identified them as non-relevant or with insufficient information while SCM/SCM+ mainly did not select them. SCM/SCM+ assume that non-selected covariates are non-relevant when it could be due to insufficient information, whereas the full model does not make this assumption and is faster. This study must be extended to other methods and completed by a more complex high-dimensional simulation framework.

Go beyond the limits of genetic algorithm in daily covariate selection practice

Covariate identification is an important step in the development of a population pharmacokinetic/pharmacodynamic model. Among the different available approaches, the stepwise covariate model (SCM) is the most used. However, SCM is based on a local search strategy, in which the model-building process iteratively tests the addition or elimination of a single covariate at a time given all the others. This introduces a heuristic to limit the searching space and then the computational complexity, but, at the same time, can lead to a suboptimal solution. The application of genetic algorithms (GAs) for covariate selection has been proposed as a possible solution to overcome these limitations. However, their actual use during model building is limited by the extremely high computational costs and convergence issues, both related to the number of models being tested. In this paper, we proposed a new GA for covariate selection to address these challenges. The GA was first developed on a simulated case study where the heuristics introduced to overcome the limitations affecting currently available GA approaches resulted able to limit the selection of redundant covariates, increase replicability of results and reduce convergence times. Then, we tested the proposed GA on a real-world problem related to remifentanil. It obtained good results both in terms of selected covariates and fitness optimization, outperforming the SCM.

Predicting hypovitaminosis C with LASSO algorithm in adult critically ill patients in surgical intensive care units: a bi-center prospective cohort study

Vitamin C played pleiotropic roles in critical illness and vitamin C insufficiency was predictive of the development of multiple organ failure. Currently, the prevalence of vitamin C insufficiency in Chinese critically ill patients is rarely determined and there are no established bedside tools to predict hypovitaminosis C. To develop a nomogram to identify patients with high risk of hypovitaminosis C, we performed a bi-center prospective cohort study at two ICUs of the first and sixth medical center in PLA General Hospital, Beijing, China from May 6th to July 31st, 2021 We identified 322 eligible patients. 62.4% patients were hypovitaminosis C. 7 features, including source of infection, the level of serum albumin, age, male gender, sepsis, vascular disease, and wasting of vitamin C by the kidney, were selected using LASSO algorithm and therefore included in the nomogram. In the testing set, our model showed moderate discrimination ability with areas under the curve of 0.75 [0.64-0.84]. Variable importance evaluated by SHAP value highlighted two novel important predictors, i.e., abdominal infection and the level of serum albumin. In conclusion, we first reported a high burden of vitamin C insufficiency in Chinese adult patient in the ICU. We also constructed a prediction model to timely identify patients with high risk of hypovitaminosis C, which allows the clinicians to choose appropriate candidates for Vitamin C repletion in clinical practice or clinical trials.

Properties of the full random-effect modeling approach with missing covariate data

During drug development, a key step is the identification of relevant covariates predicting between-subject variations in drug response. The full random effects model (FREM) is one of the full-covariate approaches used to identify relevant covariates in nonlinear mixed effects models. Here we explore the ability of FREM to handle missing (both missing completely at random (MCAR) and missing at random (MAR)) covariate data and compare it to the full fixed-effects model (FFEM) approach, applied either with complete case analysis or mean imputation. A global health dataset (20 421 children) was used to develop a FREM describing the changes of height for age Z-score (HAZ) over time. Simulated datasets (n = 1000) were generated with variable rates of missing (MCAR) covariate data (0%-90%) and different proportions of missing (MAR) data condition on either observed covariates or predicted HAZ. The three methods were used to re-estimate model and compared in terms of bias and precision which showed that FREM had only minor increases in bias and minor loss of precision at increasing percentages of missing (MCAR) covariate data and performed similarly in the MAR scenarios. Conversely, the FFEM approaches either collapsed at≥ $$ \ge $$ 70% of missing (MCAR) covariate data (FFEM complete case analysis) or had large bias increases and loss of precision (FFEM with mean imputation). Our results suggest that FREM is an appropriate approach to covariate modeling for datasets with missing (MCAR and MAR) covariate data, such as in global health studies.

Covariates in population pharmacokinetic studies of critically ill adults receiving β-lactam antimicrobials: a systematic review and narrative synthesis

Introduction

Population pharmacokinetic studies of β-lactam antimicrobials in critically ill patients derive models that inform their dosing. In non-linear mixed-effects modelling, covariates are often used to improve model fit and explain variability. We aimed to investigate which covariates are most commonly assessed and which are found to be significant, along with global patterns of publication.

Methods

We conducted a systematic review, searching MEDLINE, Embase, CENTRAL and Web of Science on 01 March 2023, including studies of critically ill adults receiving β-lactam antimicrobials who underwent blood sampling for population pharmacokinetic studies. We extracted and categorized all reported covariates and assessed reporting quality using the ClinPK checklist.

Results

Our search identified 151 studies with 6018 participants. Most studies reported observational cohorts (120 studies, 80%), with the majority conducted in high-income settings (136 studies, 90%). Of the 1083 identified covariate instances, 237 were unique; the most common categories were patient characteristics (n = 404), biomarkers (n = 206) and physiological parameters (n = 163). Only seven distinct commonly reported covariates (CLCR, weight, glomerular filtration rate, diuresis, need for renal replacement, serum albumin and C-reactive protein) were significant more than 20% of the time.

Conclusions

Covariates are most commonly chosen based on biological plausibility, with patient characteristics and biomarkers the most frequently investigated. We developed an openly accessible database of reported covariates to aid investigators with covariate selection when designing population pharmacokinetic studies. Novel covariates, such as sepsis subphenotypes, have not been explored yet, leaving a research gap for future work.

Population pharmacokinetic-pharmacodynamic modeling of clopidogrel for dose regimen optimization based on CYP2C19 phenotypes: A proof of concept study

Clopidogrel is an antiplatelet drug used to reduce the risk of acute coronary syndrome and stroke. It is converted by CYP2C19 to its active metabolite; therefore, poor metabolizers (PMs) of CYP2C19 exhibit diminished antiplatelet effects. Herein, we conducted a proof-of-concept study for using population pharmacokinetic-pharmacodynamic (PK-PD) modeling to recommend a personalized clopidogrel dosing regimen for individuals with varying CYP2C19 phenotypes and baseline P2Y12 reaction unit (PRU) levels. Data from a prospective phase I clinical trial involving 36 healthy male participants were used to develop the population PK-PD model predicting the concentrations of clopidogrel, clopidogrel H4, and clopidogrel carboxylic acid, and linking clopidogrel H4 concentrations to changes in PRU levels. A two-compartment model effectively described the PKs of both clopidogrel and clopidogrel carboxylic acid, and a one-compartment model of those of clopidogrel H4. The CYP2C19 phenotype was identified as a significant covariate influencing the metabolic conversion of the parent drug to its metabolites. A PD submodel of clopidogrel H4 that stimulated the fractional turnover rate of PRU levels showed the best performance. Monte Carlo simulations suggested that PMs require three to four times higher doses than extensive metabolizers to reach the target PRU level. Individuals within the top 20th percentile of baseline PRU levels were shown to require 2.5-3 times higher doses than those in the bottom 20th percentile. We successfully developed a population PK-PD model for clopidogrel considering the impact of CYP2C19 phenotypes and baseline PRU levels. Further studies are necessary to confirm actual dosing recommendations for clopidogrel.

Comparison of two-stage and joint TGI-OS modeling using data from six atezolizumab clinical studies in patients with metastatic non-small cell lung cancer

Two-stage and joint modeling approaches are the two main approaches to investigate the link between longitudinal tumor size data and overall survival (OS) and anticipate clinical trial outcome. We here used a large database composed of one phase II and five phase III clinical trials evaluating atezolizumab (an immunotherapy) in monotherapy or in combination with chemotherapies in 3699 patients with non-small cell lung cancer to evaluate the differences between both approaches in terms of parameter estimates, magnitude of covariate effects, and ability to predict OS. Although the two-stage approach may underestimate the magnitude of the impact of tumor growth rate (KG ) on OS compared to joint modeling approach (hazard ratios [HRs] of 0.42-2.52 vs. 0.25-2.85, respectively, for individual KG varying from the 5th and 95th percentiles), this difference did not lead into poorer performance of the two-stage approach to describe the OS distribution in the six clinical studies. Overall, two-stage and joint modeling approaches accurately predicted OS HR with a median (range) difference with the observed OS HR of 0.02 (0.01-0.18) and 0.03 (0.00-0.19), in all cases considered, respectively (e.g., for IMpower150: 0.80 [0.66-0.95] vs. 0.82 [0.70-0.95], respectively, whereas the observed OS HR was 0.80). In our setting, the two-stage approach accurately predicted the benefit of atezolizumab on OS. Further work is needed to verify if similar results are achieved using phase Ib or phase II clinical trials where the number of patients and measurements is limited as well as in other cancer indications.

Operational characteristics of full random effects modelling ('frem') compared to stepwise covariate modelling ('scm')

An adequate covariate selection is a key step in population pharmacokinetic modelling. In this study, the automated stepwise covariate modelling technique ('scm') was compared to full random effects modelling ('frem'). We evaluated the power to identify a 'true' covariate (covariate with highest correlation to the pharmacokinetic parameter), precision, and accuracy of the parameter-covariate estimates. Furthermore, the predictive performance of the final models was assessed. The scenarios varied in covariate effect sizes, number of individuals (n = 20-500) and covariate correlations (0-90% cov-corr). The PsN 'frem' routine provides a 90% confidence intervals around the covariate effects. This was used to evaluate its operational characteristics for a statistical backward elimination procedure, defined as 'frem_posthoc' and to facilitate the comparison to 'scm'. 'Frem_posthoc' had a higher power to detect the true covariate with lower bias in small n studies compared to 'scm', applied with commonly used settings (forward p < 0.05, backward p < 0.01). This finding was vice versa in a statistically similar setting. For 'frem_posthoc', power, precision and accuracy of the covariate coefficient increased with higher number of individuals and covariate effect magnitudes. Without a backward elimination step 'frem' models provided unbiased coefficients with highly imprecise coefficients in small n datasets. Yet, precision was superior to final 'scm' model precision obtained using common settings. We conclude that 'frem_posthoc' is also a suitable method to guide covariate selection, although intended to serve as a full model approach. However, a deliberated selection of automated methods is essential for the modeller and using those methods in small datasets needs to be taken with caution.

Population pharmacokinetics of nadroparin for thromboprophylaxis in COVID-19 intensive care unit patients

AIMS

Nadroparin is administered to COVID-19 intensive care unit (ICU) patients as thromboprophylaxis. Despite existing population pharmacokinetic (PK) models for nadroparin in literature, the population PK of nadroparin in COVID-19 ICU patients is unknown. Moreover, optimal dosing regimens achieving anti-Xa target levels (0.3-0.7 IU/mL) are unknown. Therefore, a population PK analysis was conducted to investigate different dosing regimens of nadroparin in COVID-19 ICU patients.

METHODS

Anti-Xa levels (n = 280) from COVID-19 ICU patients (n = 65) receiving twice daily (BID) 5700 IU of subcutaneous nadroparin were collected to perform a population PK analysis with NONMEM v7.4.1. Using Monte Carlo simulations (n = 1000), predefined dosing regimens were evaluated.

RESULTS

A 1-compartment model with an absorption compartment adequately described the measured anti-Xa levels with interindividual variability estimated for clearance (CL). Inflammation parameters C-reactive protein, D-dimer and estimated glomerular filtration rate based on the Chronic Kidney Disease Epidemiology Collaboration equation allowed to explain the interindividual variability of CL. Moreover, CL was decreased in patients receiving corticosteroids (22.5%) and vasopressors (25.1%). Monte Carlo simulations demonstrated that 5700 IU BID was the most optimal dosing regimen of the simulated regimens for achieving prespecified steady-state t = 4 h anti-Xa levels with 56.7% on target (0.3-0.7 IU/mL).

CONCLUSION

In our study, clearance of nadroparin is associated with an increase in inflammation parameters, use of corticosteroids, vasopression and renal clearance in critically ill patients. Furthermore, of the simulated regimens, targeted anti-Xa levels were most adequately achieved with a dosing regimen of 5700 IU BID. Future studies are needed to elucidate the underlying mechanisms of found covariate relationships.

Machine learning-guided covariate selection for time-to-event models developed from a small sample of real-world patients receiving bevacizumab treatment

Therapeutic outcomes in patients with metastatic colorectal cancer (mCRC) receiving bevacizumab treatment are highly variable, and a reliable predictive factor is not available. Progression-free survival (PFS) and overall survival (OS) were recorded from an observational, prospective study after 5 years of follow-up, including 46 patients with mCRC receiving bevacizumab treatment. Three vascular endothelial growth factor (VEGF)-A and two intercellular adhesion molecule-1 genes polymorphisms, age, gender, weight, dosing scheme, and co-treatments were collected. Given the relatively small number of events (37 [80%] for the PFS and 26 [57%] for the OS), to study the effect of these covariates on PFS and OS, a covariate analysis was performed using statistical and supervised machine learning techniques, including Cox regression, penalized Cox regression techniques (least absolute shrinkage and selection operator [LASSO], ridge regression, and elastic net), survival trees, and survival forest. The predictive performance of each method was evaluated in bootstrapped samples, using prediction error curves and the area under the curve of the receiver operating characteristic. The LASSO penalized Cox-regression model showed the best overall performance. Nonlinear mixed effects (NLME) models were developed, and a conventional stepwise covariate search was performed. Then, covariates identified as important by the LASSO model were included in the base NLME models developed for PFS and OS, resulting in improved models as compared to those obtained with the stepwise covariate search. It was shown that having gene polymorphisms in VEGFA (rs699947 and rs1570360) and ICAM1 (rs1799969) are associated with a favorable clinical outcome in patients with mCRC receiving bevacizumab treatment.

Efficient and relevant stepwise covariate model building for pharmacometrics

Covariate modeling is an important opportunity for pharmacometrics to influence decision making in drug development. The stepwise covariate model (SCM) building procedure is the most common method for covariate model development. Despite its advantages, the traditional SCM method is known to have long runtimes and the suboptimal ability to select relevant covariates, especially in more complex phase III settings. In this work, two alternative approaches are presented: SCM+, which introduces the “adaptive scope reduction” and changes to general estimation settings, and “stage‐wise filtering,” which groups covariates into categories based on their importance (mechanistic, structural, and exploratory). The three methods (SCM, SCM+, and SCM+ with stage‐wise filtering) are applied to data from a simulated phase III population pharmacokinetic study and are compared in terms of efficiency and relevance. The two SCM+ methods were considerably more efficient than the traditional SCM: the number of function evaluations was reduced by 70% for SCM+ and by 76% for SCM+ with stage‐wise filtering compared to SCM; the corresponding number of executed models was reduced by 44% for SCM+ and 70% for SCM+ with stage‐wise filtering. In addition, among the three methods, SCM+ with stage‐wise filtering selected the highest number of relevant covariates. Given the improved efficiency and ability to select relevant covariates shown in this work, the use of SCM+ and stage‐wise filtering can greatly increase the efficiency of covariate modeling in drug development, which will ultimately facilitate more timely support for decision making.

Contribution of machine learning to tumor growth inhibition modeling for hepatocellular carcinoma patients under Roblitinib (FGF401) drug treatment

Machine learning (ML) opens new perspectives in identifying predictive factors of efficacy among a large number of patients’ characteristics in oncology studies. The objective of this work was to combine ML with population pharmacokinetic/pharmacodynamic (PK/PD) modeling of tumor growth inhibition to understand the sources of variability between patients and therefore improve model predictions to support drug development decisions. Data from 127 patients with hepatocellular carcinoma enrolled in a phase I/II study evaluating once‐daily oral doses of the fibroblast growth factor receptor FGFR4 kinase inhibitor, Roblitinib (FGF401), were used. Roblitinib  PKs was best described by a two‐compartment model with a delayed zero‐order absorption and linear elimination. Clinical efficacy using the longitudinal sum of the longest lesion diameter data was described with a population PK/PD model of tumor growth inhibition including resistance to treatment. ML, applying elastic net modeling of time to progression data, was associated with cross‐validation, and allowed to derive a composite predictive risk score from a set of 75 patients’ baseline characteristics. The two approaches were combined by testing the inclusion of the continuous risk score as a covariate on PD model parameters. The score was found as a significant covariate on the resistance parameter and resulted in 19% reduction of its variability, and 32% variability reduction on the average dose for stasis. The final PK/PD model was used to simulate effect of patients’ characteristics on tumor growth inhibition profiles. The proposed methodology can be used to support drug development decisions, especially when large interpatient variability is observed.

Machine Learning and Pharmacometrics for Prediction of Pharmacokinetic Data: Differences, Similarities and Challenges Illustrated with Rifampicin

Pharmacometrics (PM) and machine learning (ML) are both valuable for drug development to characterize pharmacokinetics (PK) and pharmacodynamics (PD). Pharmacokinetic/pharmacodynamic (PKPD) analysis using PM provides mechanistic insight into biological processes but is time- and labor-intensive. In contrast, ML models are much quicker trained, but offer less mechanistic insights. The opportunity of using ML predictions of drug PK as input for a PKPD model could strongly accelerate analysis efforts. Here exemplified by rifampicin, a widely used antibiotic, we explore the ability of different ML algorithms to predict drug PK. Based on simulated data, we trained linear regressions (LASSO), Gradient Boosting Machines, XGBoost and Random Forest to predict the plasma concentration-time series and rifampicin area under the concentration-versus-time curve from 0–24 h (AUC_0–24h) after repeated dosing. XGBoost performed best for prediction of the entire PK series (R²: 0.84, root mean square error (RMSE): 6.9 mg/L, mean absolute error (MAE): 4.0 mg/L) for the scenario with the largest data size. For AUC_0–24h prediction, LASSO showed the highest performance (R²: 0.97, RMSE: 29.1 h·mg/L, MAE: 18.8 h·mg/L). Increasing the number of plasma concentrations per patient (0, 2 or 6 concentrations per occasion) improved model performance. For example, for AUC_0–24h prediction using LASSO, the R² was 0.41, 0.69 and 0.97 when using predictors only (no plasma concentrations), 2 or 6 plasma concentrations per occasion as input, respectively. Run times for the ML models ranged from 1.0 s to 8 min, while the run time for the PM model was more than 3 h. Furthermore, building a PM model is more time- and labor-intensive compared with ML. ML predictions of drug PK could thus be used as input into a PKPD model, enabling time-efficient analysis.

Population Pharmacokinetic Modelling of Intravenous Immunoglobulin Treatment in Patients with Guillain-Barré Syndrome

Background and Objective

Intravenous immunoglobulin (IVIg) at a standard dosage is the treatment of choice for Guillain–Barré syndrome. The pharmacokinetics, however, is highly variable between patients, and a rapid clearance of IVIg is associated with poor recovery. We aimed to develop a model to predict the pharmacokinetics of a standard 5-day IVIg course (0.4 g/kg/day) in patients with Guillain–Barré syndrome.

Methods

Non-linear mixed-effects modelling software (NONMEM^®) was used to construct a pharmacokinetic model based on a model-building cohort of 177 patients with Guillain–Barré syndrome, with a total of 589 sequential serum samples tested for total immunoglobulin G (IgG) levels, and evaluated on an independent validation cohort that consisted of 177 patients with Guillain–Barré syndrome with 689 sequential serum samples.

Results

The final two-compartment model accurately described the daily increment in serum IgG levels during a standard IVIg course; the initial rapid fall and then a gradual decline to steady-state levels thereafter. The covariates that increased IgG clearance were a more severe disease (as indicated by the Guillain–Barré syndrome disability score) and concomitant methylprednisolone treatment. When the current dosing regimen was simulated, the percentage of patients who reached a target ∆IgG > 7.3 g/L at 2 weeks decreased from 74% in mildly affected patients to only 33% in the most severely affected and mechanically ventilated patients (Guillain–Barré syndrome disability score of 5).

Conclusions

This is the first population-pharmacokinetic model for standard IVIg treatment in Guillain–Barré syndrome. The model provides a new tool to predict the pharmacokinetics of alternative regimens of IVIg in Guillain–Barré syndrome to design future trials and personalise treatment.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40262-022-01136-z.

Population Pharmacokinetic Modeling of Lenvatinib in Chinese Patients with Advanced Hepatocellular Carcinoma using Real-World Data

Lenvatinib is a novel oral angiogenesis inhibitor approved in China for the treatment of unresectable hepatocellular carcinoma(HCC) without prior systemic treatment. We described the population pharmacokinetics of lenvatinib in Chinese patients with advanced HCC and explore the potential patient characteristics associated with lenvatinib pharmacokinetics using real-world data. A total of 266 samples, provided by 127 Chinese patients with advanced HCC, were analyzed by nonlinear mixed-effects modeling. Monte Carlo simulation was conducted to assess impact of covariates on the exposure to lenvatinib. The clearance of lenvatinib in Chinese patients with advanced HCC was 5.3L/h, and alkaline phosphatase(ALP), total bilirubin(TB) and sex were identified as important covariate sassociated with it. The clearance of Child-Pugh class B patients(4.82L/h) was significantly lower than that of Child-Pugh class A patients (5.53L/h), and the systemic exposure increased with the increase of ALP and TB. There were sex differences in the pharmacokinetic characteristics of lenvatinib. The clearance of women was significantly lower than that of men (4.61L/h vs. 5.6L/h, P< 0.001), and the area under the plasma concentration-time curve of women was about 20% higher than that of men. In this study, a population pharmacokinetic model of lenvatinib was established, which can be used to simulate clinical trials or various dosing scenarios. Our findings provide important new insights for optimizing the use of lenvatinib in patients with advanced HCC. This article is protected by copyright. All rights reserved.

High-Dosage Fosfomycin Results in Adequate Plasma and Target-Site Exposure in Morbidly Obese and Nonobese Nonhyperfiltration Patients

The objectives of this study were the identification in (morbidly) obese and nonobese patients of (i) the most appropriate body size descriptor for fosfomycin dose adjustments and (ii) adequacy of the currently employed dosing regimens. Plasma and target site (interstitial fluid of subcutaneous adipose tissue) concentrations after fosfomycin administration (8 g) to 30 surgery patients (15 obese/15 nonobese) were obtained from a prospective clinical trial. After characterization of plasma and microdialysis-derived target site pharmacokinetics via population analysis, short-term infusions of fosfomycin 3 to 4 times daily were simulated. The adequacy of therapy was assessed by probability of pharmacokinetic/pharmacodynamic target attainment (PTA) analysis based on the unbound drug-related targets of an %fT_>MIC (the fraction of time that unbound fosfomycin concentrations exceed the MIC during 24 h) of 70 and an fAUC_0-24h/MIC (the area under the concentration-time curve from 0 to 24 h for the unbound fraction of fosfomycin relative to the MIC) of 40.8 to 83.3. Lean body weight, fat mass, and creatinine clearance calculated via adjusted body weight (ABW) (CLCR_{CG_ABW}) of all patients (body mass index [BMI] = 20.1 to 52.0 kg/m²) explained a considerable proportion of between-patient pharmacokinetic variability (up to 31.0% relative reduction). The steady-state unbound target site/plasma concentration ratio was 26.3% lower in (morbidly) obese than nonobese patients. For infections with fosfomycin-susceptible pathogens (MIC ≤ 16 mg/L), intermittent "high-dosage" intravenous (i.v.) fosfomycin (8 g, three times daily) was sufficient to treat patients with a CLCR_{CG_ABW} of <130 mL/min, irrespective of the pharmacokinetic/pharmacodynamic indices considered. For infections by Pseudomonas aeruginosa with a MIC of 32 mg/L, when the index fAUC_0-24h/MIC is applied, fosfomycin might represent a promising treatment option in obese and nonobese patients, especially in combination therapy to complement β-lactams, in which carbapenem-resistant P. aeruginosa is critical. In conclusion, fosfomycin showed excellent target site penetration in obese and nonobese patients. Dosing should be guided by renal function rather than obesity status. (This study has been registered in the EU Clinical Trials Register under EudraCT no. 2012-004383-22.).

Development of a new score for early mortality prediction in trauma ICU patients: RETRASCORE

Background

Severity scores are commonly used for outcome adjustment and benchmarking of trauma care provided. No specific models performed only with critically ill patients are available. Our objective was to develop a new score for early mortality prediction in trauma ICU patients.

Methods

This is a retrospective study using the Spanish Trauma ICU registry (RETRAUCI) 2015–2019. Patients were divided and analysed into the derivation (2015–2017) and validation sets (2018–2019). We used as candidate variables to be associated with mortality those available in RETRAUCI that could be collected in the first 24 h after ICU admission. Using logistic regression methodology, a simple score (RETRASCORE) was created with points assigned to each selected variable. The performance of the model was carried out according to global measures, discrimination and calibration.

Results

The analysis included 9465 patients: derivation set 5976 and validation set 3489. Thirty-day mortality was 12.2%. The predicted probability of 30-day mortality was determined by the following equation: 1/(1 + exp (− y)), where y = 0.598 (Age 50–65) + 1.239 (Age 66–75) + 2.198 (Age > 75) + 0.349 (PRECOAG) + 0.336 (Pre-hospital intubation) + 0.662 (High-risk mechanism) + 0.950 (unilateral mydriasis) + 3.217 (bilateral mydriasis) + 0.841 (Glasgow ≤ 8) + 0.495 (MAIS-Head) − 0.271 (MAIS-Thorax) + 1.148 (Haemodynamic failure) + 0.708 (Respiratory failure) + 0.567 (Coagulopathy) + 0.580 (Mechanical ventilation) + 0.452 (Massive haemorrhage) − 5.432. The AUROC was 0.913 (0.903–0.923) in the derivation set and 0.929 (0.918–0.940) in the validation set.

Conclusions

The newly developed RETRASCORE is an early, easy-to-calculate and specific score to predict in-hospital mortality in trauma ICU patients. Although it has achieved adequate internal validation, it must be externally validated.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13054-021-03845-6.

Postoperative breakthrough pain in paediatric cardiac surgery not reduced by increased morphine concentrations

BACKGROUND

Morphine is commonly used for postoperative analgesia in children. Here we studied the pharmacodynamics of morphine in children after cardiac surgery receiving protocolized morphine.

METHODS

Data on morphine rescue requirements guided by validated pain scores in children (n = 35, 3-36 months) after cardiac surgery receiving morphine as loading dose (100 μg kg^-1) with continuous infusion (40 μg kg^-1 h^-1) from a previous study on morphine pharmacokinetics were analysed using repeated time-to-event (RTTE) modelling.

RESULTS

During the postoperative period (38 h (IQR 23-46)), 130 morphine rescue events (4 (IQR 1-5) per patient) mainly occurred in the first 24 h (107/130) at a median morphine concentration of 29.5 ng ml^-1 (range 7-180 ng ml^-1). In the RTTE model, the hazard of rescue morphine decreased over time (half-life 18 h; P < 0.001), while the hazard for rescue morphine (21.9% at 29.5 ng ml^-1) increased at higher morphine concentrations (P < 0.001).

CONCLUSIONS

In this study on protocolized morphine analgesia in children, rescue morphine was required at a wide range of morphine concentrations and further increase of the morphine concentration did not lead to a decrease in hazard. Future studies should focus on a multimodal approach using other opioids or other analgesics to treat breakthrough pain in children.

IMPACT

In children receiving continuous morphine infusion, administration of rescue morphine is an indicator for insufficient effect or an event. Morphine rescue events were identified at a wide range of morphine concentrations upon a standardized pain protocol consisting of continuous morphine infusion and morphine as rescue boluses. The expected number of rescue morphine events was found to increase at higher morphine concentrations. Instead of exploring more aggressive morphine dosing, future research should focus on a multimodal approach to treat breakthrough pain in children.

Optimizing gentamicin dosing in different pediatric age groups using population pharmacokinetics and Monte Carlo simulation

Introduction

The use of once daily dosing of aminoglycosides in pediatrics is increasing but studies on dose optimization targeting the pediatric population are limited. This study aimed to derive a population pharmacokinetic model of gentamicin and apply it to design optimal dosing regimens in pediatrics.

Methods

Population pharmacokinetics of gentamicin in pediatrics was described from a retrospective chart review of plasma gentamicin concentration data (peak/ trough levels) of pediatric patients (1 month − 12 years), admitted to non-critically ill pediatrics. Monte Carlo simulations were performed on the resulting pharmacokinetic model to assess the probability of achieving a C_max/MIC target of 10 mg/L over a range of gentamicin MICs of 0.5–2 mg/L and once daily gentamicin dosing regimens. Results: A two-compartment model with additive residual error best described the model with weight incorporated as a significant covariate for both clearance and volume of distribution. Monte Carlo simulations demonstrated a good probability of target attainment even at a MIC of 2 mg/L, where neonates required doses of 6-7 mg/kg/day and older pediatrics required lower daily doses of 4–5 mg/kg/day while maintaining trough gentamicin concentration below the toxicity limit of 1 mg/L. Conclusion: Once daily dosing is a reasonable option in pediatrics that allows target attainment while maintaining trough gentamicin level below the limits of toxicity.

Nonparametric goodness-of-fit testing for parametric covariate models in pharmacometric analyses

The characterization of covariate effects on model parameters is a crucial step during pharmacokinetic/pharmacodynamic analyses. Although covariate selection criteria have been studied extensively, the choice of the functional relationship between covariates and parameters, however, has received much less attention. Often, a simple particular class of covariate‐to‐parameter relationships (linear, exponential, etc.) is chosen ad hoc or based on domain knowledge, and a statistical evaluation is limited to the comparison of a small number of such classes. Goodness‐of‐fit testing against a nonparametric alternative provides a more rigorous approach to covariate model evaluation, but no such test has been proposed so far. In this manuscript, we derive and evaluate nonparametric goodness‐of‐fit tests for parametric covariate models, the null hypothesis, against a kernelized Tikhonov regularized alternative, transferring concepts from statistical learning to the pharmacological setting. The approach is evaluated in a simulation study on the estimation of the age‐dependent maturation effect on the clearance of a monoclonal antibody. Scenarios of varying data sparsity and residual error are considered. The goodness‐of‐fit test correctly identified misspecified parametric models with high power for relevant scenarios. The case study provides proof‐of‐concept of the feasibility of the proposed approach, which is envisioned to be beneficial for applications that lack well‐founded covariate models.

Evaluation of the pharmacokinetics of prednisolone in paediatric patients with acute lymphoblastic leukaemia treated according to Dutch Childhood Oncology Group protocols and its relation to treatment response

Glucocorticoids form the backbone of paediatric acute lymphoblastic leukaemia (ALL) treatment. Many studies have been performed on steroid resistance; however, few studies have addressed the relationship between dose, concentration and clinical response. The aim of the present study was to evaluate the pharmacokinetics of prednisolone in the treatment of paediatric ALL and the correlation with clinical parameters. A total of 1028 bound and unbound prednisolone plasma concentrations were available from 124 children (aged 0–18 years) with newly diagnosed ALL enrolled in the Dutch Childhood Oncology Group studies. A population pharmacokinetic model was developed and post hoc area under the curve (AUC) was tested against treatment outcome parameters. The pharmacokinetics of unbound prednisolone in plasma was best described with allometric scaling and saturable binding to proteins. Plasma protein binding decreased with age. The AUC of unbound prednisolone was not associated with any of the disease parameters or treatment outcomes. Unbound prednisolone plasma concentrations correlated with age. No effect of exposure on clinical treatment outcome parameters was observed and does not substantiate individualised dosing. Poor responders, high‐risk and relapsed patients showed a trend towards lower exposure compared to good responders. However, the group of poor responders was small and requires further research.

A Large Impact of Obesity on the Disposition of Ivermectin, Moxidectin and Eprinomectin in a Canine Model: Relevance for COVID-19 Patients

Ivermectin (IVM) and moxidectin (MOX) are used extensively as parasiticides in veterinary medicine. Based on in vitro data, IVM has recently been proposed for the prevention and treatment of COVID-19 infection, a condition for which obesity is a major risk factor. In patients, IVM dosage is based on total body weight and there are no recommendations to adjust dosage in obese patients. The objective of this study was to establish, in a canine model, the influence of obesity on the clearance and steady-state volume of distribution of IVM, MOX, and a third analog, eprinomectin (EPR). An experimental model of obesity in dogs was based on a high calorie diet. IVM, MOX, and EPR were administered intravenously, in combination, to a single group of dogs in two circumstances, during a control period and when body weight had been increased by 50%. In obese dogs, clearance, expressed in absolute values (L/day), was not modified for MOX but was reduced for IVM and EPR, compared to the initial control state. However, when scaled by body weight (L/day/kg), plasma clearance was reduced by 55, 42, and 63%, for IVM, MOX and EPR, respectively. In contrast, the steady-state volume of distribution was markedly increased, in absolute values (L), by obesity. For IVM and MOX, this obese dog model suggests that the maintenance doses in the obese subject should be based on lean body weight rather than total weight. On the other hand, the loading dose, when required, should be based on the total body weight of the obese subject.

An Integrated Paediatric Population PK/PD Analysis of dDAVP: How do PK Differences Translate to Clinical Outcomes?

INTRODUCTION

The bioequivalence of two formulations of desmopressin (dDAVP), a vasopressin analogue prescribed for nocturnal enuresis treatment in children, has been previously confirmed in adults but not in children. In this study, we aimed to study the pharmacokinetics (PK) and pharmacodynamics (PD) of these two formulations, in both fasted and fed children, including patients younger than 6 years of age.

METHODS

Previously published data from one PK study and one PK/PD study in children aged between 6 and 16 years were combined with a new PK/PD study in children aged between 6 months and 8 years, and analysed using population PK/PD modelling. Simulations were performed to further explore the relative bioavailability of both formulations and evaluate current dosing strategies.

RESULTS

The complex absorption behaviour of the lyophilizate was modelled using a double input, linked to a one-compartmental model with linear elimination and an indirect response model linking dDAVP concentration to produced urine volume and osmolality. The final model described the observed data well and elucidated the complexity of bioequivalence and therapeutic equivalence of the two formulations. Simulations showed that current dosing regimens using a fixed dose of lyophilizate 120 μg is not adequate for children, assuming children to be in the fed state when taking dDAVP. A new age- and weight-based dosing regimen was suggested and was shown to lead to improved, better tailored effects.

CONCLUSIONS

Bioequivalence and therapeutic equivalence data of two formulations of the same drug in adults cannot be readily extrapolated to children. This study shows the importance of well-designed paediatric clinical trials and how they can be analysed using mixed-effects modelling to make clinically relevant inferences. A follow-up clinical trial testing the proposed dDAVP dosing regimen should be performed.

CLINICAL TRIAL REGISTRATION

This trial has been registered at www.clinicaltrials.gov (identifier NCT02584231; EudraCT 2014-005200-13).

A novel method based on unbiased correlations tests for covariate selection in nonlinear mixed effects models: The COSSAC approach

Building a covariate model is a crucial task in population pharmacokinetics and pharmacodynamics in order to understand the determinants of the interindividual variability. Identifying a good covariate model usually requires many runs. Several procedures have been proposed in the past to automatize this task. The most commonly used is Stepwise Covariate Modeling (SCM). Here, we present a novel stepwise method based on statistical tests between individual parameters sampled from their conditional distribution and the covariates. This strategy, called the COnditional Sampling use for Stepwise Approach based on Correlation tests (COSSAC), makes use of the information contained in the current model to choose which parameter‐covariate relationship to try next. This strategy greatly reduces the number of covariate models tested, while retaining on its search path the models improving the log‐likelihood (LL). In this article, we detail the COSSAC method and its implementation in Monolix, and evaluate its performance. The performance was assessed by comparing COSSAC to the traditional SCM method on 17 representative data sets. For the large majority of cases (15 out of 17), the final covariate model is identical (11 cases) or very similar (4 cases with LL differences less than 3.84) with both procedures. Yet, COSSAC requires between 2 to 20 times fewer runs than SCM. This represents a decisive speed up, especially for models that take long to run and would not be tractable using the SCM method.

Population Pharmacokinetics and Pharmacodynamics of Ciprofloxacin Prophylaxis in Pediatric Acute Lymphoblastic Leukemia Patients

BACKGROUND

Ciprofloxacin is used as antimicrobial prophylaxis in pediatric acute lymphoblastic leukemia (ALL) to decrease infections with gram-negative bacteria. However, there are no clear guidelines concerning prophylactic dose.

AIMS

To determine the pharmacokinetics and pharmacodynamics (PKPD) of ciprofloxacin prophylaxis in a pediatric ALL population. The effect of patient characteristics and antileukemic treatment on ciprofloxacin exposure, the area under the concentration time curve over minimal inhibitory concentration (AUC24/MIC) ratios, and emergence of resistance were studied.

METHODS

A total of 615 samples from 129 children (0-18 years) with ALL were collected in a multicenter prospective study. A population pharmacokinetic model was developed. Microbiological cultures were collected prior to and during prophylaxis. An AUC24/MIC of ≥125 was defined as target ratio.

RESULTS

A 1-compartment model with zero-order absorption and allometric scaling best described the data. No significant (P < .01) covariates remained after backward elimination and no effect of asparaginase or azoles were found. Ciprofloxacin AUC24 was 16.9 mg*h/L in the prednisone prophase versus 29.3 mg*h/L with concomitant chemotherapy. Overall, 100%, 81%, and 18% of patients at, respectively, MIC of 0.063, 0.125, and 0.25 mg/L achieved AUC24/MIC ≥ 125. In 13% of the patients, resistant bacteria were found during prophylactic treatment.

CONCLUSION

Ciprofloxacin exposure shows an almost 2-fold change throughout the treatment of pediatric ALL. Depending on the appropriateness of 125 as target ratio, therapeutic drug monitoring or dose adjustments might be indicated for less susceptible bacteria starting from ≥ 0.125 mg/L to prevent the emergence of resistance and reach required targets for efficacy.

Predicting Unacceptable Pain in Cardiac Surgery Patients Receiving Morphine Maintenance and Rescue Doses: A Model-Based Pharmacokinetic-Pharmacodynamic Analysis

BACKGROUND

Optimal analgesic treatment following cardiac surgery is crucial for both patient comfort and successful postoperative recovery. While knowledge of both the pharmacokinetics and pharmacodynamics of analgesics is required to predict optimal drug dosing, models quantifying the pharmacodynamics are scarce. Here, we quantify the pharmacodynamics of morphine by modeling the need for rescue morphine to treat unacceptable pain in 118 patients after cardiac surgery.

METHODS

The rescue morphine event data were analyzed with repeated time-to-event (RTTE) modeling using NONMEM. Postoperative pain titration protocol consisted of continuous morphine infusions (median duration 20.5 hours) with paracetamol 4 times daily and rescue morphine in case of unacceptable pain (numerical rating scale ≥4).

RESULTS

Patients had a median age of 73 years (interquartile range [IQR]: 63-77) and median bodyweight of 80 kg (IQR: 72-90 kg). Most patients (55%) required at least 1 rescue morphine dose. The hazard for rescue morphine following cardiac surgery was found to be significantly influenced by time after surgery, a day/night cycle with a peak at 23:00 (95% confidence interval [CI], 19:35-02:03) each day, and an effect of morphine concentration with 50% hazard reduction at 9.3 ng·mL-1 (95% CI, 6.7-16).

CONCLUSIONS

The pharmacodynamics of morphine after cardiac surgery was successfully quantified using RTTE modeling. Future studies can be used to expand the model to better predict morphine's pharmacodynamics on the individual level and to include the pharmacodynamics of other analgesics so that improved postoperative pain treatment protocols can be developed.

A Modified Hybrid Wald's Approximation Method for Efficient Covariate Selection in Population Pharmacokinetic Analysis

One important objective of population pharmacokinetic (PPK) analyses is to identify and quantify relationships between covariates and model parameters such as clearance and volume. To improve upon existing covariate model development methods including stepwise procedures and Wald's approximation method (WAM), this paper introduces an innovative method named the hybrid first-order conditional estimation (FOCE)/Monte-Carlo parametric expectation maximization (MCPEM)-based Wald's approximation method with backward elimination (BE), or H-WAM-BE. Compared with WAM, this new method uses MCPEM to obtain full covariance matrix after running FOCE to obtain full model parameter estimates, followed by BE to select the final covariate model. Two groups of datasets (simulation datasets and rituximab datasets) were used to compare the performance of H-WAM-BE with two other methods, likelihood ratio test (LRT)-based stepwise covariate method (SCM) and H-WAM with full subset approach (H-WAM-F) in NONMEM. Different scenarios with different sample sizes and sampling schemes were used for simulating datasets. The nominal model was used as the reference to evaluate the three methods for their ability to accurately identify parameter-covariate relationships. The methods were compared using the number of true and false positive covariates identified, number of times that they identified the reference model, computation times, and predictive performance. Best-performing H-WAM-BE methods (M2 and M4) showed comparable results with LRT-based SCM. H-WAM-BE required shorter or comparable computation times than LRT-based SCM and H-WAM-F regardless of the model structure, sample size, or sampling design used in this study.

Machine learning in pharmacometrics: Opportunities and challenges

The explosive growth in medical devices, imaging and diagnostics, computing, and communication and information technologies in drug development and healthcare has created an ever-expanding data landscape that the pharmacometrics (PMX) research community must now traverse. The tools of machine learning (ML) have emerged as a powerful computational approach in other data-rich disciplines but its effective utilization in the pharmaceutical sciences and PMX modelling is in its infancy. ML-based methods can complement PMX modelling by enabling the information in diverse sources of big data, e.g. population-based public databases and disease-specific clinical registries, to be harnessed because they are capable of efficiently identifying salient variables associated with outcomes and delineating their interdependencies. ML algorithms are computationally efficient, have strong predictive capabilities and can enable learning in the big data setting. ML algorithms can be viewed as providing a computational bridge from big data to complement PMX modelling. This review provides an overview of the strengths and weaknesses of ML approaches vis-à-vis population methods, assesses current research into ML applications in the pharmaceutical sciences and provides perspective for potential opportunities and strategies for the successful integration and utilization of ML in PMX.

Evaluation of clinical and genetic factors in the population pharmacokinetics of carbamazepine

Aims

Carbamazepine can cause hypersensitivity reactions in ~10% of patients. An immunogenic effect can be produced by the electrophilic 10,11‐epoxide metabolite but not by carbamazepine. Hypothetically, certain single nucleotide polymorphisms might increase the formation of immunogenic metabolites, leading ultimately to hypersensitivity reactions. This study explores the role of clinical and genetic factors in the pharmacokinetics (PK) of carbamazepine and 3 metabolites known to be chemically reactive or formed through reactive intermediates.

Methods

A combination of rich and sparse PK samples were collected from healthy volunteers and epilepsy patients. All subjects were genotyped for 20 single nucleotide polymorphisms in 11 genes known to be involved in the metabolism or transport of carbamazepine and carbamazepine 10,11‐epoxide. Nonlinear mixed effects modelling was used to build a population‐PK model.

Results

In total, 248 observations were collected from 80 subjects. A 1‐compartment PK model with first‐order absorption and elimination best described the parent carbamazepine data, with a total clearance of 1.96 L/h, central distribution volume of 164 L and absorption rate constant of 0.45 h⁻¹. Total daily dose and coadministration of phenytoin were significant covariates for total clearance of carbamazepine. EPHX1‐416G/G genotype was a significant covariate for the clearance of carbamazepine 10,11‐epoxide.

Conclusion

Our data indicate that carbamazepine clearance was affected by total dose and phenytoin coadministration, but not by genetic factors, while carbamazepine 10,11‐epoxide clearance was affected by a variant in the microsomal epoxide hydrolase gene. A much larger sample size would be required to fully evaluate the role of genetic variation in carbamazepine pharmacokinetics, and thereby predisposition to carbamazepine hypersensitivity.

Application of Machine Learning for Tumor Growth Inhibition - Overall Survival Modeling Platform

Machine learning (ML) was used to leverage tumor growth inhibition (TGI) metrics to characterize the relationship with overall survival (OS) as a novel approach and to compare with traditional TGI‐OS modeling methods. Historical dataset from a phase III non‐small cell lung cancer study (OAK, atezolizumab vs. docetaxel, N = 668) was used. ML methods support the validity of TGI metrics in predicting OS. With lasso, the best model with TGI metrics outperforms the best model without TGI metrics. Boosting was the best linear ML method for this dataset with reduced estimation bias and lowest Brier score, suggesting better prediction accuracy. Random forest did not outperform linear ML methods despite hyperparameter optimization. Kernel machine was marginally the best nonlinear ML method for this dataset and uncovered nonlinear and interaction effects. Nonlinear ML may improve prediction by capturing nonlinear effects and covariate interactions, but its predictive performance and value need further evaluation with larger datasets.

Quantification of Scheduling Impact on Safety and Efficacy Outcomes of Brain Metastasis Radio- and Immuno-Therapies: A Systematic Review and Meta-Analysis

Objectives: The goal of this quantitative research was to evaluate the impact of various factors (e.g., scheduling or radiotherapy (RT) type) on outcomes for RT vs. RT in combination with immune checkpoint inhibitors (ICI), in the treatment of brain metastases, via a meta-analysis. Methods: Clinical studies with at least one ICI+RT treatment combination arm with brain metastasis patients were identified via a systematic literature search. Data on 1-year overall survival (OS), 1-year local control (LC) and radionecrosis rate (RNR) were extracted; for combination studies which included an RT monotherapy arm, odds ratios (OR) for the aforementioned endpoints were additionally calculated and analyzed. Mixed-effects meta-analysis models were tested to evaluate impact on outcome, for different factors such as combination treatment scheduling and the type of ICI or RT used. Results: 40 studies representing a total of 4,359 patients were identified. Higher 1-year OS was observed in ICI and RT combination vs. RT alone, with corresponding incidence rates of 59% [95% CI: 54-63%] vs. 32% [95% CI: 25-39%] (P < 0.001). Concurrent ICI and RT treatment was associated with significantly higher 1-year OS vs. sequential combinations: 68% [95% CI: 60-75%] vs. 54% [95% CI: 47-61%]. No statistically significant differences were observed in 1-year LC and RNR, when comparing combinations vs. RT monotherapies, with 1-year LC rates of 68% [95% CI: 40-90%] vs. 72% [95% CI: 63-80%] (P = 0.73) and RNR rates of 6% [95% CI: 2-13%] vs. 9% [95% CI: 5-14%] (P = 0.37). Conclusions: A comprehensive, study-level meta-analysis of brain metastasis disease treatments suggest that combinations of RT and ICI result in higher OS, yet comparable neurotoxicity profiles vs. RT alone, with a superiority of concurrent vs. sequential combination regimens. A similar meta-analysis using patient-level data from past trials, as well as future prospective randomized trials would help confirming these findings.

Comparison of covariate selection methods with correlated covariates: prior information versus data information, or a mixture of both?

The inclusion of covariates in population models during drug development is a key step to understanding drug variability and support dosage regimen proposal, but high correlation among covariates often complicates the identification of the true covariate. We compared three covariate selection methods balancing data information and prior knowledge: (1) full fixed effect modelling (FFEM), with covariate selection prior to data analysis, (2) simplified stepwise covariate modelling (sSCM), data driven selection only, and (3) Prior-Adjusted Covariate Selection (PACS) mixing both. PACS penalizes the a priori less likely covariate model by adding to its objective function value (OFV) a prior probability-derived constant: [Formula: see text], Pr(X) being the probability of the more likely covariate. Simulations were performed to compare their external performance (average OFV in a validation dataset of 10,000 subjects) in selecting the true covariate between two highly correlated covariates: 0.5, 0.7, or 0.9, after a training step on datasets of 12, 25 or 100 subjects (increasing power). With low power data no method was superior, except FFEM when associated with highly correlated covariates ([Formula: see text]), sSCM and PACS suffering both from selection bias. For high power data, PACS and sSCM performed similarly, both superior to FFEM. PACS is an alternative for covariate selection considering both the expected power to identify an anticipated covariate relation and the probability of prior information being correct. A proposed strategy is to use FFEM whenever the expected power to distinguish between contending models is < 80%, PACS when > 80% but < 100%, and SCM when the expected power is 100%.

A phase 1 pooled PK/PD analysis of bone resorption biomarkers for odanacatib, a Cathepsin K inhibitor

To develop a framework for evaluating the resorption effects of Cathepsin K (CatK) inhibitors and to inform dose regimen selection, a pharmacokinetic/pharmacodynamic (PK/PD) model for odanacatib (ODN) was developed based upon data from Phase 1 studies. Pooled PK/PD data from 11 studies (N = 249) were fit reasonably to a population inhibitory sigmoid E_max model. Body weight on E₀ (baseline uNTx/Cr, urinary N-terminal telopeptide normalized by creatinine) and age on E_max (fractional inhibition of the biomarker response) were significant covariates for biomarker response. Simulations of typical osteoporosis patients (by age, sex and weight) indicated minimal differences between sexes in concentration-uNTx/Cr relationship. There was no evidence that regimen (daily vs. weekly dosing) influenced the PK/PD relationship of resorption inhibition for odanacatib. PK/PD models based on data from odanacatib (ODN) Phase 1 studies demonstrated that uNTx/Cr was an appropriate bone resorption biomarker for assessment of the effects of a CatK inhibitor. The models also identified the determinants of response in the PK/PD relationship for ODN (body weight on E₀ and age on E_max).

A Novel Pathological Scoring System for Hepatic Cirrhosis with Hepatocellular Carcinoma

Purpose

This study aimed to propose an effective quantitative pathological scoring system and to establish nomogram to assess the stage of cirrhosis and predict postoperative survival of hepatocellular carcinoma (HCC) with cirrhosis patients after hepatectomy.

Methods

The scoring system was based on a retrospective study on 163 patients who underwent partial hepatectomy for HCC with cirrhosis. The clinicopathological and follow-up data of 163 HCC with cirrhosis patients who underwent hepatectomy in our hospital from 2010 to 2014 were retrospectively reviewed. A scoring system was established based on the total value of independent predictive factors of cirrhosis. The results were validated using 97 patients operated on from 2011 to 2015 at the same institution. Nomogram was then formulated using a multivariate Cox proportional hazards model to analyze.

Results

The scoring system was ultimately composed of 4 independent predictive factors and was divided into 3 levels. The new cirrhosis system score strongly correlated with Child–Pugh score (r=0.8058, P<0.0001) 3 months after surgery; higher cirrhosis system scores predicted poorer liver function and stronger liver damage 3 months after surgery. Then, a four-factor nomogram for survival prediction was established. The concordance indices were 0.79 for the survival-prediction nomogram. The calibration curves showed good agreement between predictions by the nomogram and actual survival outcomes.

Conclusion

This new scoring system of cirrhosis can help us predict the liver function and liver injury 3 months after surgery, and the nomogram enabled accurate predictions of risk of overall survival in patients of HCC with cirrhosis after hepatectomy.

Population Pharmacokinetics of Imipenem in Critically Ill Patients: A Parametric and Nonparametric Model Converge on CKD-EPI Estimated Glomerular Filtration Rate as an Impactful Covariate

BACKGROUND

Population pharmacokinetic (popPK) models for antibiotics are used to improve dosing strategies and individualize dosing by therapeutic drug monitoring. Little is known about the differences in results of parametric versus nonparametric popPK models and their potential consequences in clinical practice. We developed both parametric and nonparametric models of imipenem using data from critically ill patients and compared their results.

METHODS

Twenty-six critically ill patients treated with intravenous imipenem/cilastatin were included in this study. Median estimated glomerular filtration rate (eGFR) measured by the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation was 116 mL/min/1.73 m2 (interquartile range 104-124) at inclusion. The usual dosing regimen was 500 mg/500 mg four times daily. On average, five imipenem levels per patient (138 levels in total) were drawn as peak, intermediate, and trough levels. Imipenem concentration-time profiles were analyzed using parametric (NONMEM 7.2) and nonparametric (Pmetrics 1.5.2) popPK software.

RESULTS

For both methods, data were best described by a model with two distribution compartments and the CKD-EPI eGFR equation unadjusted for body surface area as a covariate on the elimination rate constant (Ke). The parametric population parameter estimates were Ke 0.637 h-1 (between-subject variability [BSV]: 19.0% coefficient of variation [CV]) and central distribution volume (Vc) 29.6 L (without BSV). The nonparametric values were Ke 0.681 h-1 (34.0% CV) and Vc 31.1 L (42.6% CV).

CONCLUSIONS

Both models described imipenem popPK well; the parameter estimates were comparable and the included covariate was identical. However, estimated BSV was higher in the nonparametric model. This may have consequences for estimated exposure during dosing simulations and should be further investigated in simulation studies.

Pharmacokinetics and population pharmacokinetics in pediatric oncology

Pharmacokinetic research has become increasingly important in pediatric oncology as it can have direct clinical implications and is a crucial component in individualized medicine. Population pharmacokinetics has become a popular method especially in children, due to the potential for sparse sampling, flexible sampling times, computing of heterogeneous data, and identification of variability sources. However, population pharmacokinetic reports can be complex and difficult to interpret. The aim of this article is to provide a basic explanation of population pharmacokinetics, using clinical examples from the field of pediatric oncology, to facilitate the translation of pharmacokinetic research into the daily clinic.

Population pharmacokinetics of orally administrated bromopride: Focus on the absorption process

Bromopride is a prokinetic and antiemetic drug used to treat nausea and vomiting. Although its prescription is common in Brazil, there is a lack of studies about bromopride pharmacokinetics. Therefore, the aims of this study were to investigate the population pharmacokinetics of bromopride and to evaluate the influence of covariates on its absorption. This study is a retrospective analysis of data collected from bioequivalence studies. The data was modeled using MONOLIX 2018R2. Assuming one-compartment and linear elimination, the absorption phase was evaluated with different structural models. The model of sequential first- and zero-order with combined error and exponential inter-individual variability in all parameters best described the atypical absorption profile of bromopride. Population estimates were first-order absorption rate (ka) of 0.08 h ^- 1, fraction of dose absorbed by first-order (Fr) of 32.60%, duration of the zero-order absorption (Tk0) of 0.88 h with latency time (Tlag) of 0.47 h, volume of distribution of 230 l and clearance of 46.80 l h ^- 1. Bodyweight affects Tk0, dosage form was found to correlate with Tk0 and Tlag, while gender affects Tlag. However, simulations evaluating the clinical importance of these covariates on steady-state indicated minimal changes on bromopride exposure. The mixed absorption model was reasonable to describe the absorption process of bromopride because it had the flexibility to fit multiple-peaks profile and shows good agreement with physicochemical properties of drug.

Population pharmacokinetics and covariate analysis of Sym004, an antibody mixture against the epidermal growth factor receptor, in subjects with metastatic colorectal cancer and other solid tumors

Sym004 is an equimolar mixture of two monoclonal antibodies, futuximab and modotuximab, which non-competitively block the epidermal growth factor receptor (EGFR). Sym004 has been clinically tested for treatment of solid tumors. The present work characterizes the non-linear pharmacokinetics (PK) of Sym004 and its constituent antibodies and investigates two types of covariate models for interpreting the interindividual variability of Sym004 exposure. Sym004 serum concentration data from 330 cancer patients participating in four Phase 1 and 2 trials (n = 247 metastatic colorectal cancer, n = 87 various types advanced solid tumors) were pooled for non-linear mixed effects modeling. Dose regimens of 0.4-18 mg/kg Sym004 dosed by i.v. infusion weekly or every 2nd week were explored. The PK profiles for futuximab and modotuximab were parallel, and the parameter values for their population PK models were similar. The PK of Sym004 using the sum of the serum concentrations of futuximab and modotuximab was well captured by a 2-compartment model with parallel linear and saturable, Michaelis-Menten-type elimination. The full covariate model including all plausible covariates included in a single step showed no impact on Sym004 exposure of age, Asian race, renal and hepatic function, tumor type and previous anti-EGFR treatments. The reduced covariate model contained statistically and potentially clinically significant influences of body weight, albumin, sex and baseline tumor size. Population PK modeling and covariate analysis of Sym004 were feasible using the sum of the serum concentrations of the two constituent antibodies. Full and reduced covariate models provided insights into which covariates may be clinically relevant for dose modifications and thus may need further exploration.

Population Pharmacokinetic/Pharmacodynamic Modeling of O-Desmethyltramadol in Young and Elderly Healthy Volunteers

BACKGROUND

Age-related changes in the concentration-effect relationship of (+)-O-desmethyl-tramadol [(+)-ODM], tramadol's active metabolite, are not documented in the elderly.

OBJECTIVE

The objective of this study was to characterize, in elderly and young subjects, the (+)-ODM pharmacokinetic and pharmacodynamic relationship to examine the effect of age after single-dose administration of tramadol 200 mg extended-release tablets.

METHODS

A population analysis of a double-blind, randomized, placebo-controlled, two-period cross-over study including 13 elderly (aged ≥75 years) subjects with mild renal insufficiency and 16 young (aged 18-40 years) subjects was conducted. For 48 h post-dose, blood samples were collected and pain tolerance thresholds measured using an electrically stimulated pain model. A pharmacokinetic/pharmacodynamic model incorporating a one-compartment pharmacokinetic model for (+)-ODM parameterized with first-order formation rate, clearance (CL/f_m), volume of distribution (V/f_m) and a sigmoid maximum effect (E_max) model incorporating baseline (E₀) and placebo effect was used.

RESULTS

Maximum plasma concentrations of (+)-ODM occurred later and plasma concentrations declined more slowly in the elderly than in young subjects. In the elderly, V/f_m was 76% larger and CL/f_m 16% slower. Baseline (E₀) and sensitivity (C₅₀) for pain tolerance were similar between young and elderly subjects. However, the E_max parameter was 2.5 times higher in the elderly and maximum possible treatment-related effect was 169 (135-221) in the young and 194 (149-252) in the elderly; that is, 15% higher in the elderly.

CONCLUSIONS

This exploratory analysis suggests that age-related differences exist in the distribution and elimination of (+)-ODM, including a 76% larger distribution outside the central compartment and 16% slower clearance of (+)-ODM. These pharmacokinetic changes are associated with a 15% higher maximum possible treatment-related effect and carry the potential for greater efficacy but also the potential for increased side effects at the same dose in elderly subjects. Clinicaltrials.gov identifier: NCT02329561.

Nicotine Population Pharmacokinetics in Healthy Adult Smokers: A Retrospective Analysis

Background and Objective

Characterizing nicotine pharmacokinetics is challenging in the presence of background exposure. We performed a combined retrospective population pharmacokinetic analysis of 8 trials, including exposure to Tobacco Heating System and cigarettes (both inhaled), nicotine nasal spray and oral nicotine gum.

Method

Data from 4 single product use trials were used to develop a population pharmacokinetic model with Phoenix^® NLME™ and to derive exposure parameters. Data from 4 separate ad libitum use studies were used for external validation. A total of 702 healthy adult smokers (54% males; 21–66 years of age; smoking ≥10 cigarettes/day; from US, Europe and Japan) were eligible for participation.

Results

Two-compartment linear disposition combined with zero-order absorption model was adequate to describe nicotine pharmacokinetics, and a mono-exponentially decreasing background component was utilized to account for nicotine carry-over effects. Apparent nicotine clearance was typically 0.407 L/min in males and 26% higher in females (68% inter-individual variability). Bioavailability was product-specific, decreased with increasing nicotine ISO yield, and increased with increasing body weight. Absorption duration was apparently prolonged with nicotine gum. The typical initial and terminal half-lives were 1.35 and 17 h, respectively. The presence of menthol did not impact the determinants of the area under the curve. The model adequately described the external validation data.

Conclusions

The population model was able to describe in different populations the nicotine pharmacokinetics after single product use and after 4 days of ad libitum use of Tobacco Heating System, cigarettes, and of different nicotine replacement therapies with various routes of administration.

Electronic supplementary material

The online version of this article (doi:10.1007/s13318-017-0405-2) contains supplementary material, which is available to authorized users.

A population pharmacokinetic model of cabozantinib in healthy volunteers and patients with various cancer types

Purpose

An integrated population pharmacokinetic (popPK) model was developed to describe the pharmacokinetics (PK) of tyrosine kinase inhibitor cabozantinib in healthy volunteers (HVs) and patients with various cancer types and to identify any differences in cabozantinib PK across these populations.

Methods

Plasma concentration data used to develop the popPK model were obtained from nine clinical trials (8072 concentrations from 1534 HVs or patients) of cabozantinib in HVs and patients with renal cell carcinoma (RCC), medullary thyroid carcinoma (MTC), glioblastoma multiforme, castration-resistant prostate cancer, or other advanced malignancies.

Results

PK data across studies were adequately characterized by a two-compartment disposition model with dual first- and zero-order absorption processes and first-order elimination. Baseline demographic covariates (age, weight, gender, race, and cancer type) were generally predicted to have a small-to-moderate impact on apparent clearance (CL/F). However, MTC cancer type did show an approximately 93% higher CL/F relative to HVs following chronic dosing, resulting in approximately 40–50% lower predicted steady-state cabozantinib plasma concentrations.

Conclusion

This popPK analysis showed cabozantinib CL/F values to be higher for patients with MTC and may account for the higher dosage required in this patient population (140-mg) to achieve plasma exposures comparable to those in patients with RCC and other tumor types administered a 60-mg cabozantinib tablet dose. Possible factors that may underlie the higher cabozantinib clearance observed in MTC patients are discussed.

Electronic supplementary material

The online version of this article (10.1007/s00280-018-3581-0) contains supplementary material, which is available to authorized users.