Abstracts

A Message from France Mentré, Editor‐in‐Chief; Lena Friberg, Deputy Editor‐in‐Chief; Paolo Denti and Colin Pillai, on behalf of the WCoP 2020 Organizing Committee

We know we speak for many when we say how disappointed we were that we were not able to gather in Cape Town this past April for the opportunity to share important and exciting research from all across the globe. CPT: Pharmacometrics & Systems Pharmacology (PSP) planned to publish the conference abstracts in a special supplement. When the conference was initially moved to 2021, PSP planned to wait and publish the abstracts then. As the global situation with COVID‐19 evolved, and WCoP was, of necessity, moved to 2022, we realized that many authors would not want to wait so long to share their research with the wider world. We then contacted the authors, asking if they would wish to have their abstracts published in a 2020 supplement of PSP. All of those who agreed are included in this special issue of PSP. We hope you will find their research of interest, as we do. It makes us wish even more that we could have had a chance to speak with all of these authors directly, and even more anxious to join everyone for WCoP March 29 to April 1, 2022, in Cape Town, South Africa.

01

Linezolid Population Pharmacokinetics in South African MDR/XDR TB Patients

M. T. Abdewahab¹, S. Wasserman², J. CM Brust³, K. Dheda¹, L. Wiesner¹, G. Maartens^1,2, P. Denti¹

¹University of Cape Town, Cape Town, South Africa; ²Wellcome Centre for Infectious Diseases Research in Africa, University of Cape Town, Cape Town, South Africa; ³Albert Einstein College of Medicine, New York, NY, USA

Background

The World Health Organization recently updated its treatment guidelines to include linezolid as a core agent for drug resistant tuberculosis (DR‐TB). Linezolid has a narrow therapeutic index and frequently causes dose‐related haematological toxicity and neuropathy. An understanding of linezolid pharmacokinetics (PK) is necessary to inform dose optimization; however, there are limited PK data for linezolid in DR‐TB patients with HIV‐co‐infection. We aimed to characterize the population PK of linezolid in South African patients with DR‐TB and explore effect of co‐medication and HIV‐co‐infection on drug exposure.

Methods

Data were obtained from a PK sub‐study in a randomized controlled trial, NExT (NCT02454205) and an observational study, PROBeX, which enrolled adult patients with pulmonary DR‐TB. Participants were sampled at pre‐ and 1, 2, 3, 4, 5, 6, 8 and 24 post observed dose, and plasma linezolid concentrations measured by LC‐MS/MS. PK data were analyzed using NONMEM 7.4 with FOCE‐I.

Results

Thirty participants underwent intensive PK sampling; 26 were on standard daily dose of 600 mg and 4 had dose reduction to 300 mg due to adverse events. Nineteen patients were male and 15 HIV‐positive. Median weight (WT) was 58.5 kg (IQR 49.8–67.6), median creatinine clearance was 116 ml/min (IQR 103–139) and median treatment duration was 59 days. A one compartment model with transit absorption fitted the data well; disposition parameters were allometrically scaled using WT. In the final model, the typical values for clearance (CL) and central volume were 3.15 L/h and 36.4 L and 30.4 % between subject variability in CL, respectively. HIV co‐infection and creatinine clearance had no significant effect on CL.

Conclusion

Our model characterized PK of linezolid in South African patients with DR‐TB. The exposure was not affected by HIV status.

02

Meropenem Population Pharmacokinetics in Adult Patients with Rifampicin‐Sensitive Pulmonary Tuberculosis

A. A. Abulfathi¹, E. Van Brakel², H. Feyt², N. Gupte³, N. Vanker², V. De Jager², G. L. Barnes³, E. Nuermberger³, S. E. Dorman⁴, A. Diacon^2,5, K. E. Dooley⁶, E. M. Svensson^7,8

¹Stellenbosch University, Cape Town, South Africa; ²Task Applied Science, Bellville, South Africa; ³Johns Hopkins University Baltimore, MD, USA; ⁴Medical University of South Carolina, Charleston, SC, USA; ⁵Stellenbosch University, Cape Town, South Africa; ⁶Johns Hopkins University Center for Tuberculosis Research, Baltimore, MD, USA; ⁷Uppsala University, Uppsala, Sweden; ⁸Radboud University Medical Center, Nijmegen, the Netherlands

Background

Meropenem is being investigated for repurposing as an anti‐tuberculosis drug. This study aimed to develop a meropenem population PK model and identify covariates improving predictive performance.

Methods

Approval to conduct the study was granted by Pharma‐Ethics. Pulmonary tuberculosis patients were randomized to one of four treatment groups: meropenem 2 g thrice daily plus oral rifampicin 20 mg/kg once daily, meropenem 2 g thrice daily, meropenem 1 g thrice daily, and meropenem 3 g once daily. Meropenem was administered by intravenous infusion over 0.5–1 h. All patients also received oral amoxicillin/clavulanate. Intensive plasma PK sampling over 8 h was done on study day 13. First‐order conditional estimation with interaction was used in NONMEM software. The best model was chosen based on likelihood metrics, goodness‐of‐fit plots, and parsimony. Covariates were tested stepwise.

Results

A total of 404 concentrations from 49 patients were analyzed. A 2‐compartment model parameterized with clearance (CL), inter‐compartmental clearance (Q), central (V1) and peripheral (V2) volumes of distribution fit the data well. Typical values of CL, Q, V1 and V2 were 11.6 L/h, 3.84 L/h, 13.6 L, and 3.82 L, respectively. The relative standard errors of the parameter estimates ranged from 4.5 to 29.8%. The covariates included in the final model were creatinine clearance on CL, and allometric scaling of body weight on all disposition parameters. An effect of age on CL as previously reported could not be identified.

Conclusion

A two‐compartment model describes meropenem population PK in tuberculosis patients with good precision in parameter estimates. The final model will be used for an integrated PK/PD analysis linking meropenem exposure to early bactericidal activity.

03

Drug Development for the Treatment, Control and Elimination of Onchocerciasis: A Pooled Population Pharmacokinetic Analysis of Emodepside (BAY 44‐4400) in Healthy Subjects

F. AssmUS; ^1,2, R. M. Hoglund^1,2, I. Scandale³, J. Tarning^1,2

¹Mahidol University, Bangkok, Thailand; ²University of Oxford, Oxford, UK; ³Drugs for Neglected Diseases Initiative, Geneva, Switzerland

Background

Emodepside is under clinical development for the treatment of onchocerciasis (river blindness), a neglected tropical disease caused by the parasitic worm Onchocerca volvulus. The aim of this study was to characterize the population pharmacokinetic properties of emodepside in healthy subjects and recommend a dosing regimen to evaluate in patients.

Methods

Pharmacokinetic data were pooled from 142 subjects enrolled in three phase 1 studies (single‐dose/multiple‐dose dose‐escalation study, relative bioavailability study). All subjects received an oral dose of emodepside, in fasted or fed state, with the majority of participants receiving a liquid formulation. The pharmacokinetic analysis also included data from two immediate‐release tablet formulations. Ethical approval was granted by the London – Brent Research Ethics Committee. All cohorts were pooled and analyzed using nonlinear mixed‐effects modeling (NONMEM v.7.4). The final population pharmacokinetic model was used to simulate clinical dosing scenarios.

Results

Emodepside pharmacokinetics was well‐described by a transit‐compartment absorption model, followed by a three‐compartment disposition model. Food intake and increased dose were associated with delayed absorption. Furthermore, the absorption of tablets was slower compared to the liquid formulation. Bioavailability was estimated as 69% (tablet A) and 80% (tablet B) relative to solution, and 76% (fed state) relative to fasted state. Simulations suggested that twice daily dosing of 15 mg emodepside (tablet B) administered without food for 10 days was required to reach target concentrations.

Conclusion

Pharmacokinetic modeling and simulation was used to derive an optimized dosing regimen for emodepside for a planned phase 2 clinical trial in Africa.

04

Evaluation of Who's Weight‐Band Pediatric Dosing Recommendations for Dolutegravir Using Physiologically‐Based Pharmacokinetic Modeling

S. Atoyebi, D. Bolanle, O. Bolaji, A. Olagunju

Obafemi Awolowo University, Ile Ife, Nigeria

Background

Dolutegravir‐based regimen is now the preferred first‐line option for people living with Human Immunodeficiency Virus (HIV) including children initiating antiretroviral therapy. In this study, a physiologically‐based pharmacokinetic (PBPK) model was developed, qualified, and used to evaluate the adequacy of the recent WHO pediatric dosing recommendation in achieving plasma exposure above EC90 in children weighing 3–5.9 kg.

Methods

A whole‐body PBPK model was developed using Simbiology (MATLAB 2017b). Processes governing drug disposition were described with ordinary differential equations parameterized by key drug‐specific and system properties. Predicted pharmacokinetic (PK) parameters in children (n = 100) weighing 6–9.9 kg were validated using clinical data from ODYSSEY and IMPAACT P1093 studies in children with 0.5–1.5 fold difference as acceptance threshold. Simulated virtual population of children (n = 100) weighing 3–5.9 kg received the 10 mg dose of dolutegravir once daily. Predicted plasma PK parameters at steady‐state were computed and number of children with Ctrough values below EC90 (0.32 mg/L) were observed.

Results

Model predictions were within the stated acceptance threshold. Median age (range) of the simulated pediatric population in the 3–5.9 kg weight‐band were 4.49 months (1.00–10.0). Predicted pharmacokinetics of dolutegravir dispersible tablets were: geometric mean (%CV) of C24, Cmax and AUC0‐24 were 0.585 mg/L (47.3), 3.99 mg/L (20.5) and 40.6 mg.h/L (27.7) respectively. Predicted plasma exposure was similar to adults’ and above the EC90 (0.32 mg/L) in 92% of the population.

Conclusion

Model predictions support the WHO weight‐band dosing recommendations for children weighing 3–5.9 kg. Confirmation of this in prospective safety and efficacy trials is now warranted.

05

Using Physiologically‐Based Pharmacokinetic Models to Predict Bictegravir Pharmacokinetics in Pregnant Women

S. Atoyebi, O. Bolaji, A. Olagunju

Obafemi Awolowo University, Ile Ife, Nigeria

Background

A bictegravir‐based regimen (BIKTARVY) has been approved for antiretroviral therapy in adults, but there are no available pharmacokinetic (PK) data in pregnant women to support its use in pregnancy. In this study, a physiologically‐based pharmacokinetic (PBPK) model was developed, qualified, and deployed to evaluate the PK of 50 mg bictegravir in pregnant adults.

Methods

A PBPK model was developed using Simbiology (MATLAB 2017b). Processes governing drug disposition were described with ordinary differential equations parameterised by key drug‐specific and system properties. Predicted PK parameters in virtual non‐pregnant adults (n = 100) were validated with clinical data from BIKTARVY label with 0.5–1.5‐fold difference as acceptance threshold. Pregnancy‐induced biological changes were added to the qualified model to give a pregnancy PBPK model. Simulated virtual pregnant women (n = 100) in second and third trimesters received 50 mg bictegravir once daily. Predicted plasma PK parameters at steady‐state during pregnancy were computed and compared with non‐pregnant adults.

Results

Mean of predicted vs. reported Cmin, 2.23 vs. 2.61 mg/L; Cmax, 4.07 vs. 6.15 mg/L; AUC0‐24, 71.4 vs. 102 mg.h/L. Geometric mean (% CV) of predicted PK parameters of bictegravir in pregnant women at second and third trimesters were: Cmin, 2.59 (32.1) and 1.65 (17.1) mg/L; Cmax, 3.27(26.8) and 2.28 mg/L (13.1); AUC0‐24, 68.1 (28.9) and 45.6 mg.h/L (14.5) respectively. Geometric mean ratios of predicted plasma exposure in pregnant to non‐pregnant adults were 1.0 and 0.68 in second and third trimesters, respectively.

Conclusion

Model predictions show that pregnancy does not lead to clinically relevant changes in bictegravir pharmacokinetics. Confirmation of this in prospective trials is now warranted.

06

Model‐Based Statistical Approaches for Pharmacokinetic Bioequivalence Studies with Sparse Sampling

J. Bertrand¹, F. Loingeville¹, T. T. Nguyen¹, F. Mentré¹, K. Mollenhoff², H. Dette², S. Sharan³, S. Guoying³, S. Grosser³, L. Zhao³, L. Fang³

¹Université de Paris, IAME, INSERM, Paris, France; ²Ruhr‐Universität Bochum, Germany; ³US Food and Drug Administration, Silver Spring, MD, USA

Background

In traditional bioequivalence (BE) analysis, two one‐sided tests (TOST) are performed on AUC and Cmax obtained by NCA from crossover or parallel studies. Rich sampling is not always feasible, so we proposed a model‐based TOST (MB‐TOST) for sparse sampling. However, the MB approach can lead to increased type I error due to underestimation of asymptotic standard error (SE) of the treatment effect.¹ Therefore, we investigated alternative approaches to correct this inflation. Further, for data with high variability, methods based on the TOST suffer from conservative type 1 error. Therefore, we proposed another test than the TOST, called Bioequivalence Optimal Testing (BOT).

Methods

The alternatives to the asymptotic approximation of the SE were i) parametric bootstrap, ii) a posteriori distribution sampled by HMC and iii) Gallant correction. The BOT compares the absolute value of the treatment effect to the fifth quantile of a folded normal. All approaches were evaluated by clinical trials simulation for crossover and parallel trials, rich and sparse designs, low and high variability, under H0 and H1.

Results

On crossover trial with sparse design, the type 1 error inflation observed for MB‐TOST when using an asymptotic SE is corrected using all three alternative calculations of the SE. On parallel trials with high between subject variability, both NCA‐TOST and MB‐TOST obtained very conservative type 1 errors. Using BOT, type 1 errors were close to the nominal level.

Conclusion

When variability is large compare to N, TOST can be problematic and BOT should be used whether one performs NCA or MB bioequivalence. For sparse design, three alternatives to asymptotic SE were evaluated and a posteriori distribution sampled by HMC proved to be the best.

Reference

1. Dubois et al. Model‐based analyses of bioequivalence crossover trials using the stochastic approximation expectation maximisation algorithm. Stat. Med.30, 2582–2600 (2011).

07

An Empirical Approach to Identifying Electrophysiological Correlates of Topiramate‐Related Working Memory Impairment Using Pharmacokinetic‐Pharmacodynamic Modeling

S. P. Callisto, C. M. Barkley, M. Fiecas, R. Brundage, A. K. Birnbaum, S. E. Marino

University of Minnesota, Minneapolis, MN, USA

Background

Topiramate (TPM) is a commonly prescribed anti‐seizure drug that frequently impairs working memory (WM) function. Here, in order to quantify TPM‐induced changes in neural activity, TPM was administered to subjects before they completed a modified Sternberg WM task while their electroencephalogram (EEG) was recorded. Although the effects of TPM on the EEG are not well‐described, we hypothesized that changes in neural activity would occur in the theta (4–8 Hz) and alpha (8–12 Hz) frequency bands, measures known to correlate with WM function.

Methods

Healthy volunteers (n = 27) received a blinded single dose of lorazepam, TPM, or placebo in random order on separate visits of a study approved by the University of Minnesota IRB. A population PK‐PD model was developed to quantify the relationship between TPM plasma concentration and theta band power observed during the retention phase of the WM task. Post hoc sparse principal component analysis (sPCA) was conducted to determine whether TPM‐induced changes would be better described by a composite of activity in the theta and alpha bands. Relationships between the resulting principal components and WM task performance measures were quantified using linear regression.

Results

Two groups were identified with differential sensitivity to TPM‐induced theta band power increases (0.66 µg/mL⁻¹ vs. 0.08 µg/mL⁻¹). Although alpha power explained >80% of variability in EEG data, reaction time on the WM task correlated most strongly with theta band power.

Conclusion

sPCA identified a composite EEG‐derived index of TPM‐related WM impairment which (i) could be used downstream in PK/PD modeling to quantify drug effects on multiple processes, and (ii) verified that the EEG index selected correlated with behavioral measures of drug effect.

08

Implementation and Evaluation of the Sampling Importance Resampling Algorithm in Saemix

M. Chanel¹, E. Comets^1,2

¹Université de Paris, IAME, INSERM, F‐75018 Paris, France; ²INSERM, CIC 1414, 35700 Rennes, France; Université Rennes‐1, 35700 Rennes, France

Background

Estimating the uncertainty on model parameters is a key component to using nonlinear mixed effect models in drug development and decision‐making. Alternative approaches to using the asymptotic Fisher Information Matrix (FIM) have been proposed to estimate uncertainty in small samples. In the present work, we implement the Sampling Importance Resampling (SIR) in the R package saemix and investigate its performance through simulations.

Methods

SIR consists in sampling parameter vectors from a proposal distribution and resampling them with a weight depending on their likelihood.¹ We implemented the algorithm in a development version of saemix adapted to handle discrete models. We first evaluated SIR in the setting of Dosne et al.¹, both with the original design and with a more challenging sparse design. We then applied the SIR algorithm to estimate the uncertainty of the parameters in a model representing the evolution of a binomial response with time, in two groups of subjects. We used as proposal distribution either the exact FIM computed using adaptive gaussian quadrature (AGQ) with ad hoc code, or a bootstrap estimate.

Results

SIR was able to recover the true parameter uncertainty with covariance matrix from saemix as proposal distribution for models with continuous outcome, and with covariance matrix by AGQ or bootstrap as proposal distribution for models with categorical outcomes. Estimations for very small datasets could not however be improved, reflecting the scant information available with binary outcomes.

Conclusion

SIR provided good estimates of SE for models with continuous or categorical outcomes. However the computational cost was high and the SE were generally unreliable in very small datasets.

Reference

1. Donne, A., Bergstrand, M., Harling, K., & Karlsson, M. An automated sampling importance resampling procedure for estimating parameter uncertainty. J. Pharmacokinet. Pharmacodyn. 44, 509–520 (2017).

09

Modeling of the Time to AUR or BPH‐Related Surgery After Placebo, Monotherapy or Combination Therapy with Tamsulosin and Dutasteride

S. D'Agate¹, C. Chavan², M. Manyak³, J. M. Palacios‐Moreno⁴, M. Oelke⁵, M. C. Michel⁶, C. G. Roehrborn⁷, O. Della Pasqua^1,8

¹University College London, UK; ²GlaxoSmithKline, India; ³GlaxoSmithKline, USA; ⁴GlaxoSmithKline, Spain; ⁵St. Antonius Hospital, Gronau, Germany; ⁶Johannes Gutenberg University, Mainz, Germany; ⁷University of Texas Southwestern Medical Center, Dallas, TX, USA; ⁸GlaxoSmithKline, UK

Background

Men with moderate‐to‐severe lower urinary tract symptoms (LUTS) attributed to benign prostatic hyperplasia (BPH) at risk for progression are frequently treated with tamsulosin (TAM) and dutasteride monotherapy (DUT) or combination therapy (CT). This study aimed to describe the hazard of acute urinary retention or BPH‐related surgery (AUR/S) in this patient population. Simulations were then performed to assess the impact of the covariate factors on the baseline incidence of the events, disentangling it from the drug effects.

Methods

Time‐to‐event modeling was performed using data from patients (N = 10238) enrolled into six clinical studies receiving placebo, TAM, DUT, or CT. A hazard function was used to describe the time to the first AUR/S. Multivariate analysis was performed to explore the influence of relevant clinical and demographic factors, as well as drug treatment, on the baseline hazard. Predictive performance of the final model was assessed by graphical and statistical methods.

Results

An exponential hazard model was found to best describe the time to first AUR/S in patients with moderate or severe LUTS/BPH symptoms. The baseline hazard rate (95%CI) was 7.78·10⁻⁵ (6.86·10⁻⁵–8.70·10⁻⁵) day⁻¹. Baseline values of IPSS, PSA, prostate volume and maximum urine flow were identified as covariates. DUT and CT were found to significantly decrease the risk of events with HRs (95%CI) of 0.432 (0.352–0.512) and 0.336 (0.249–0.423), respectively. TAM did not affect the baseline hazard with similar parameter estimates observed in patients receiving placebo.

Conclusion

Our model shows the effect of disease modifying properties of DUT and CT on the baseline hazard of AUR/S. The use of symptomatic treatment (TAM) has no impact on the individual baseline hazard.

Study sponsored by GSK.

10

Population Pharmacokinetics of Daunorubicin and Its Metabolite Daunorubicinol in Adult and Paediatric Patients with Acute Myeloid LEUKEMIA

A. Di Deo¹, S. Oosterholt¹, M. Locci de Oliveira², V. Lucia Lanchote², O. Della Pasqua¹

¹University College London, London, UK; ²University of São Paulo, Brazil

Background

Anthracyclines, such as daunorubicin (DNR), are important for the treatment of acute myeloid leukemia (AML). However, the optimal dosage in pediatric population is still unclear. Drug resistance and disease relapse have been major factors in limiting the success of AML therapy. Interindividual variability in the exposure to DNR could contribute to difference in treatment outcome in AML. Using a model‐based approach the present study aimed to assess the influence of covariate factors on the PK of DNR in patients with AML.

Methods

Data from adult patients (n = 12) were used to evaluate the disposition of DNR and its main metabolite, daunorubicinol (DOL) using nonlinear mixed‐effects modeling. Subsequently, the model was used to 1) define a sparse sampling scheme in pediatric patients and 2) optimize the dose to ensure comparable exposure to DNR and DOL in children, irrespective of their body weight. Prior distributions on model parameters were used to optimize sampling time and intervals required to characterize the time course of DNR and DOL concentrations in a pediatric population.

Results

A five‐compartment model was found to describe the PK of DNR. The model allowed the prediction of the PK profile of DNR in children and subsequent optimization of the sampling scheme, including three samples per patient. Simulations showed optimal exposure at a dose of 60 mg/m² in children.

Conclusion

The use of a model‐based approach reveals that PK of DNR and DOL can be described by allometric principles and further characterized using sparse sampling. In conjunction with optimality concepts, it is possible to establish optimal dosing regimens to be used in prospective clinical trials in pediatric patients with AML.

11

Pharmacokinetics/Pharmacodynamics of High‐Dose Isoniazid Against Multi Drug Resistant Tuberculosis

K. Gausi¹, S. Miyahara², E. IgnatiUS; ³, X. Sun², L. Moran⁴, R. Hafner⁵, S. Rosenkranz², S. Swindells⁶, A. H. Diacon⁷, K. E. Dooley³, P. Denti¹

¹University of Cape Town, Cape Town, South Africa; ²Harvard T.H. Chan School of Public Health, Boston, MA, USA; ³Johns Hopkins University School of Medicine, Baltimore, MD, USA; ⁴Social & Scientific Systems, Inc., Silver Spring, MD, USA; ⁵National Institute of Allergy and Infectious Diseases, Bethesda, MD, USA; ⁶University of Nebraska Medical Center, Omaha, NE, USA; ⁷Task Applied Science, Bellville, South Africa

Background

There is accumulating evidence that higher than standard doses of isoniazid (INH) are effective against low to intermediate level INH resistant strains. This study aimed to characterize the association between INH serum concentrations achieved at standard or high doses and its early bactericidal activity in INH‐resistant Tuberculosis with inhA mutation.

Methods

The clinical study ACTG A5312 is comparing the early bacterial activity of INH 5 mg/kg in drug‐sensitive patients with 5, 10, and 15 mg/kg in patients with isolates carrying an inhA mutation. Participants were on INH monotherapy for 7 days, with sputum collected overnight daily and intensive PK sampling on day 6. Nonlinear mixed‐effect modeling was used to jointly model drug‐induced bacterial killing in colony‐forming units (CFU) and time to culture positivity (TTP) to determine correlations with INH serum levels and genetic acetylator types.

Results

59 participants were recruited. A two‐compartment model with NAT2 genotype effect on clearance well‐described INH pharmacokinetics (PK). A mono‐exponential model described the decline in CFU, while an exponential growth model using CFU as MGIT inoculum described TTP. The L2 method was implemented to account for correlation of residual errors from the two models. An Emax model was used to relate INH AUC to kill rate of bacteria. Based on the PK model simulation, 33%, 52%, and 80% of patients on 5, 10, and 15 mg/kg, respectively, were predicted to attain at least EC80.

Conclusion

Increase in INH dose to 15 mg/kg in patients with inhA mutation has high probability of achieving early bactericidal activity similar to that of standard‐dose against drug‐sensitive bacteria.

12

Pharmacokinetic‐Pharmacodynamic Analysis to Evaluate the Effect of Moxifloxacin and Verapamil on Electrocardiographic Safety Markers, J‐Tpeak, Tpeak‐Tend and QTC Interval in Healthy Male Subjects

S. Han^1,2,3, S. Choi^1,2,3, S. Han^1,2,3, K.‐W. Sung¹, D.‐S. Yim^1,2,3

¹The Catholic University of Korea, Seoul, Republic of Korea; ²St. Mary's Hospital, The Catholic University of Korea, Seoul, Republic of Korea; ³PIPET (Pharmacometrics Institute for Practical Education & Training), the Catholic University of Korea, Seoul, Republic of Korea

Background

Prolonged QTc (heart rate corrected QT) interval is a sensitive electrocardiogram (ECG) marker of torsade de pointes (TdP), but it is not specific in that drugs that have balanced ion channel blocks cause QT prolongation without TdP. The CiPA initiatives demonstrated that analysis of J‐Tpeak (early repolarization) and Tpeak‐Tend interval (late repolarization) can identify QTc prolonging drugs with true TdP risk and proposed them as a new cardiac safety marker for new drug candidates.

Methods

In this prospective, randomized, open‐label, parallel clinical trial approved by CMC IRB, 12 healthy Korean male subjects aged 20–44 received multiple oral doses of a pure hERG potassium channel blocker (moxifloxacin 400 mg qd, 3 times) or drugs that block hERG and calcium currents (verapamil 80 mg bid, 5 times). Peripheral blood samples were collected frequently before and after dose. Pre‐dose 12‐lead ECGs (5 times: 0, 2, 4, 8, 12 h) were taken to record diurnal changes and post‐dose 12‐lead ECGs (5 times: 0, 2, 4, 8, 12 h) were also recorded. An automated ECG measurement methodology using ecglib was applied for assessment of the J‐Tpeak and Tpeak‐Tend intervals.¹

Results

The PK/PD models were developed and the effect of moxifloxacin and verapamil on ECG markers was well explained by the nonlinear dose‐response (Emax) model. ECG analysis showed that both moxifloxacin and verapamil prolonged QTc interval. Moxifloxacin prolonged both J‐Tpeak and Tpeak‐Tend whereas verapamil which had additional inward current block preferentially shortened J‐Tpeak. These results are in line with previous results by Johannesen et al.²

Conclusion

Human cardiac repolarization using ECG markers, J‐Tpeak, Tpeak‐Tend and QTc interval in healthy male subjects was successfully performed. Our findings may improve the utility of the ECG for evaluating proarrhythmic risks.

References

1. Johannesen, L. et al. Automated algorithm for J‐Tpeak and Tpeak‐Tend assessment of drug‐induced proarrhythmia. PLoS One. 30, e0166925 (2016).

2. Johannesen, L. et al. Differentiating drug‐induced multichannel block on the electrocardiogram: Randomized study of dofetilide, quinidine, ranolazine, and verapamil. Clin. Pharmacol. Ther. 96, 549–558 (2014).

13

Pharmacokinetic Simulation of a Continuous Infusion of Lorazepam and Propylene Glycol in the Development of a Paediatric Clinical Trial Protocol

P. Healy¹, M. Marano², M. Montibeller², B. M. Goffredo², G. Pontrelli², O. Della Pasqua¹

¹University College London, London, UK; ²Ospedale Pediatrico Bambino Gesù (OPBG), Rome, Italy

Background

Sedation and analgesia are necessary for patients admitted to pediatric intensive care units (PICU). Benzodiazepines are used off‐label due to their excellent efficacy profiles. Of interest is the role of lorazepam (LZ), as its effects are not long‐lasting. However, LZ formulations contain propylene glycol (PG), an excipient which can accumulate in tissues, leading to toxicity. A model‐based approach was used to establish an optimised dosing regimen for analgosedation in ventilated patients, i.e., assess the efficacious and safe dose range of LZ for a prospective clinical study with pediatric patients in PICU.

Methods

A published PK model¹ with body weight as a covariate describing the disposition of LZ was used along with a one‐compartment model describing profiles of PG. Clinical trial simulations (CTS) including up to three cohorts (n = 54, weights: 10–70 kg) were implemented to assess the impact of intermittent boluses over continuous infusion. Model performance was evaluated against literature data. Dosing regimens to be used in the prospective study were based on a pre‐defined target exposure range (LZ) and safety threshold (PG).

Results

Three doses and dosing regimens were identified for prospective evaluation of efficacy of LZ in a clinical study. LZ doses should be administered as 30‐min bolus every 4 hours over the first two days, then a continuous infusion:

Conclusion

CTS can be used to optimize protocol design and support dose rationale for LZ. Moreover, our analysis shows that the use of continuous infusions, often used in clinical practice, may lead unacceptable PG exposure.

TABLE

Cohort 1
Day 1	0.2 mg/kg
Day 2	0.1 mg/kg
Day 3	0.03 mg/kg/h over 24 h
Cohort 2
Day 1	0.1 mg/kg
Day 2	0.2 mg/kg
Day 3	0.025 mg/kg/h over 24 h
Cohort 3
Day 1	0.3 mg/kg
Day 2	0.1 mg/kg
Day 3	0.025 mg/kg/h over 24 h

Reference

1. Gonzalez, D. et al. Population pharmacokinetics and exploratory pharmacodynamics of lorazepam in pediatric status epilepticus. Clin Pharmacokinet.56, 941–951 (2017).

14

Structuring Pharmacokinetic Literature with Natural Language Processing

F. Gonzalez Hernandez^1,2, S. J. Carter¹, W. Lilaonitku^1,3, J. Iso‐Sipilä⁴, P. Goldsmith⁵, F. Kloprogge¹, J. Standing¹

¹University College London, London, UK; ²The Alan Turing Institute, London, UK; ³Health Data Research, London, UK; ⁴BenevolentAI, London, UK; ⁵Eli Lilly and Company, London, UK

Background

Pharmacokinetic (PK) predictions are critical in early drug development. Manually curated PK databases are costly and time‐consuming to maintain. The availability of a standardized, large‐scale and up‐to‐date database would enable the use of data‐driven approaches for PK parameter prediction. An automated pipeline is required to retrieve and structure the growing PK literature. In this study, an open‐access Natural Language Processing pipeline has been developed to retrieve and characterize publications that report PK parameters.

Methods

A corpus of 3,952 journal articles from PubMed was labeled as “Relevant” (reporting PK parameters) and “Not relevant”. Supervised machine learning classifiers were trained to perform the binary classification task. Different document‐encoding methods, including Bag‐of‐Words (BoW) and word embeddings, and decoding architectures were compared. The BERN¹ algorithm, was used to perform named‐entity recognition and normalization of drugs, species and diseases mentioned in the abstract to facilitate document search.

Results

A BoW and averaged word embeddings for encoding documents, combined with an Extreme Gradient Boosting classifier yielded the highest F1‐score (94.3%) on our test set. A total of 82,555 PK studies were identified from PubMed and characterized by drugs, species and diseases mentioned. Preliminary analyses of 200 retrieved papers gave a precision of 93%.

Conclusions

A large corpus of publications reporting PK parameters was successfully retrieved and categorized from PubMed, and it will soon be released Open Access through a web interface at www.pkpdai.com. This first step makes large‐scale PK data easily accessible to use in big‐data PK prediction.

Reference

1. Kim, D., et al. A Neural named entity recognition and multi‐type normalization tool for biomedical text Mining. IEEE Access 7, 73729–73740 (2019).

15

Intravenous and Oral (PO) Fosfomycin Pharmacokinetics (Plasma and Cerebrospinal Fluid) in Neonates with Suspected Clinical Sepsis

Z. Kane¹, S. Gastine¹, P. Williams^2,3, S. Ellis⁴, J. A. Berkley^2,3, M. Sharland⁵, J. Standing¹

¹University College London, London, UK; ²KEMRI‐Wellcome Trust Research Programme, Kilifi, Kenya; ³University of Oxford, UK; ⁴Global Antibiotic Research and Development Partnership, Geneva, Switzerland; ⁵University of London, London, UK

Background

Antimicrobial resistance (AMR) is a global health priority and neonates are especially vulnerable. Fosfomycin could be part of an empiric treatment for neonatal sepsis in settings where there are high levels of resistance to empirical ampicillin/gentamicin (SOC) and/or ceftriaxone therapy.¹ Fosfomycin PK data in neonates is limited, entirely lacking PO data. Here we report model‐based estimation of fosfomycin CL, V, T1/2, F% and CSF penetration in hospitalized neonates.

Methods

The NeoFosfo study (ClinicalTrials.gov: NCT03453177) was conducted at Kilifi County Hospital, Kenya. 61 neonates received (SOC) plus 100 mg/kg q12 h fosfomycin, sample size determined by simulation‐estimation. A cross‐over design with 2 samples after 1st IV and 1st oral dose (timing randomized to cover expected dose interval) gave >80% power for CL, V and F% estimation. CSF samples were opportunistic. Pharmacokinetic (PK) modeling was performed using NONMEM 7.3. Covariates were selected on biological plausibility or correlation with model parameters.

Results

Of 238 plasma and 15 CSF concentrations, IV and PO plasma levels ranged (mean) from 7‐576 (202) and 7‐206 (70), respectively, with CSF from 16–66 (38) μg/mL. A three‐compartment model with first order absorption had covariates of weight, PMA, PNA and CSF protein. Population estimates (%RSE) for CL and Vc were 8.9 L/hr (7.3) and 18.8 L (4.1). Inter‐individual variability was 26% (28.6) and 14% (48.3) respectively. Residual error was 12% (21.9) and 11.8 μg/mL (30.6) on plasma and CSF predictions. F% was estimated at 50% (13.1), Ka =0.1 /hr (23.1) and CSF penetration 0.32 (9.5).

Conclusion

A Population PK model adequately describes IV and oral fosfomycin plasma and CSF levels in neonates, dose simulations in the context of two PD targets; AUC/MIC ratio and %T>MIC performed.

Reference

1. Williams, P. C. M. et al. The potential of fosfomycin for multi‐drug resistant sepsis: An analysis of in vitro activity against invasive paediatric gram‐negative Bacteria. J. Med. Microbiol.68, 711–719 (2019).

16

Earlier and Increased Loss of Anti‐Mycobacterial Effect After Repeated Exposure to INH‐RIF

F. Kloprogge¹, J. Ortiz Canseco¹, Z. Sadouki¹, N. Stoker¹, A. Witney², L. Phee¹, T. D McHugh¹

¹Institute for Global Health, University College London, UK; ²St George's University of London, UK

Background

Isoniazid‐rifampicin forms the backbone for drug sensitive Tuberculosis (TB) treatment, although the potent anti‐mycobacterial effect reduces over the course of treatment. In order to increase the anti‐mycobacterial effect rifampicin dosages have been escalated up to 40 mg/kg O.D. This study aimed to obtain a holistic understanding of the combined isoniazid‐rifampicin drug effects, emergence of persistence and resistance.

Methods

In vitro incubation experiments were conducted in duplicate using Mycobacterium tuberculosis H37Rv. Each experiment comprised a drug exposure phase (one week; 0‐32 X MIC for isoniazid, rifampicin, or in combination), a regrowth phase (two weeks) and a second drug exposure phase (one week). Data was parameterized using a single ODE with a turn‐over growth model and proportional EMAX drug effect that can reduce over time. Stochastic simulations at in vivo mimicking PK conditions were evaluated in vitro using the hollow‐fiber infection model.

Results

The model adequately described the MGIT Time To Positivity – time data. Bi‐phasic killing characteristics were more pronounced during the second drug exposure of the kill‐curve characteristics. Monte‐Carlo simulations mimicking standard INH‐RIF dosing and intensified T.I.D. dosing, to ensure elevated trough levels were confirmed with the hollow‐fiber infection model.

Conclusion

Earlier and increased emergence of persistence occurred after repeated exposure to INH‐RIF and the hollow‐fiber model confirmed model predicted bacillary killing at in vivo mimicking PK conditions.

17

Amalgamating Knowledge from Translational Bottom‐Up and Top‐Down Approaches to Elucidate Complex Pharmacokinetics: The Voriconazole Example

F. Kluwe^1,2, J. Schulz¹, W. Huisinga³, M. Zeitlinger⁴, G. MikUS; ⁵, R. Michelet^1,*, C. Kloft^1,*

¹Freie Universitaet Berlin, Germany; ²PharMetrX, Berlin, Germany; ³University of Potsdam, Potsdam, Germany; ⁴University of Vienna, Vienna, Austria; ⁵University Hospital Heidelberg, Heidelberg, Germany*shared senior authorship

Background

Voriconazole (VRC) is used for prophylaxis and treatment of invasive fungal infections, despite variable, nonlinear and not fully understood pharmacokinetics (PK). We developed an innovative research platform amalgamating different (i) knowledge sources and (ii) modeling methodologies, to elucidate VRC PK and to serve as a blueprint for similar well‐established compounds with complex PK.

Methods

A literature search identified knowledge gaps in VRC PK.¹ An in vivo PK database for VRC comprising individual‐level (9 trials) and summary‐level PK data (>60 literature datasets), after single/multiple intravenous/per oral administration, with/without comedication, determined in (total/unbound) plasma, urine and interstitial fluid in patients/volunteers with known/unknown CYP2C19 genotype, was developed and augmented with in vitro experiments. Top‐down nonlinear mixed‐effects and bottom‐up physiologically‐based PK modeling were applied and combined in a sequential middle‐out approach.

Results

Combining in vitro,in silico, and in vivo data and different methodologies, the proposed middle‐out approach enabled filling identified knowledge gaps in VRC PK through exploration and evaluation of various scenarios. Implementation of saturable enzyme kinetics, suspected VRC autoinhibition and CYP2C19 genotype was crucial to adequately describe VRC PK. Developed models were used to simulate different dosing scenarios to elucidate VRC PK and optimize its clinical use.

Conclusion

The developed VRC research platform was successfully established and can be adapted for other compounds. Data sharing, amalgamating knowledge from different sources and combining different methodologies are promising approaches to tackle complex clinical challenges.

Reference

1. Schulz, J., Kluwe, F., Mikus, G., Michelet, R., & Kloft, C. Novel insights into the complex pharmacokinetics of voriconazole: a review of its metabolism. Drug Metab. Rev. 51, 247–265 (2019).

18

Therapeutic Drug Monitoring: Adaptation of Literature Models to a Target Population Using the Prior Approach

A. Chan Kwong^1,2, A. O'Jeanson², P. Nolain³, F. Gattacceca¹, S. Khier²

¹Aix‐Marseille University, Marseille, France; ²Montpellier University, Montpellier, France; ³University of Limoges, Limoges, France

Background

Pharmacokinetic models from literature can be used for therapeutic drug monitoring (TDM). It may be of interest to adapt these models to the target population with the Prior approach¹ (tweaked models). We compared the predictive ability of both literature and tweaked models on TDM concentrations of meropenem.

Methods

Blood samples of meropenem were collected from 2017 to 2019 in patients of the intensive care unit of Montpellier hospital (France). The study protocol was approved by the Ethics Committee (2019_IRB‐MTP_03‐01). Data were split into an “estimation” (80%) and a “prediction” (20%) dataset. Population pharmacokinetic models for meropenem were selected from literature. These models were run on the “estimation” dataset with the $PRIOR subroutine in NONMEM to get tweaked models. Bayesian predictions of the “prediction dataset” were performed using the literature and the tweaked models to compare their predictive ability.

Results

The “estimation” and the “prediction” datasets consisted respectively in 85 and 30 concentrations from 46 and 12 patients. Quality criteria improved with the Prior approach only for one out of six models (Table).

Conclusion

For these sparse data from clinical practice, the Prior approach did not always improve the predictive ability of the literature models. Sharing model code could facilitate this approach.

Reference

1. Gisleskog P.O., Karlsson M.O., & Beal S.L. Use of prior information to stabilize a population data analysis. J. Pharmacokinet. Pharmacodyn.29, 473–505 (2002).

TABLE: Mean Prediction Error (MPE) and Root Mean Square Error (RMSE) of each model on the “prediction dataset”

	Literature model	Tweaked model
Dhaese 2019
MPE	3.13	2.33
RMSE	11.8	11.2
Jaruratanasirikul 2015
MPE	0.87	3.00
RMSE	9.64	10.0
Li 2006
MPE	0.370	1.76
RMSE	9.29	9.33
Mattioli 2016
MPE	5.80	8.64
RMSE	13.6	14.9
Roberts 2009
MPE	0.301	1.40
RMSE	9.11	10.0
Ulldemolins 2015
MPE	1.43	6.32
RMSE	12.0	13.9

19

Population Pharmacokinetics of Primaquine and Carboxy‐Primaquine in Korean Population

W. Yul Lee^1,2, D. Chae¹, C. Kim¹, K. Park¹

¹Yonsei University College of Medicine, Seoul, Korea; ²Brain Korea 21 Plus Project for Medical Science, Yonsei University, Seoul, Korea

Background

This study was conducted to investigate pharmacokinetic characteristics of primaquine in Korean population to be used as a basis of optimal dosage regimen design.

Methods

Data were acquired from a prospective, open label, parallel designed clinical trial conducted in 24 healthy subjects who received primaquine (PQ) 15 mg QD for 4 days coadministered with chloroquine. Blood samples were taken at 0 (pre‐dose), 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 12, 24 h after the last dose. Along with polymorphism analysis of CYP2D6, blood concentrations of PQ and carboxy‐primaquine(cPQ) were modeled with a minimal PBPK model, where liver blood flow was fixed to 90L/hr and flow and volume parameters were allometrically scaled to body weight of 70 kg. Covariates were tested using stepwise covariate modeling (SCM) at significance levels of P < 0.05 for forward addition and P < 0.01 for backward delition. All analyses were performed using R ver 3.5.2 and NONMEM ver 7.3.

Results

Clearance (CL) of PQ via CYP2D6 (CLCYP) and that via monoamine oxidase (CLMAO) were 26.6 L/hr and 16.9 L/hr, respectively. CL of cPQ was 1 L/hr. The volume of distribution (Vd) for PQ and cPQ were 268.8 L and 21.6 L, respectively. No significant covariate was found during SCM; although the CYP2D6 activity score significantly influenced CLCYP, it did so only in forward selection step. Inter‐individual variabilities (in CV%) were 36.9% for CLCYP and 37.8% for CLMAO, 28.4% for Vd of PQ and 135.9% for absorption rate constant, KA.

Conclusion

Minimal PBPK model succesfully described pharmacokinetics of PQ and cPQ, No significant covariate was found with the subjects studied.

20

Predicting Aqueous and Intrinsic Solubility of Pharmaceutical Molecules with Neural Networks

N. Melillo, C. Podrecca, R. Bartolucci, P. Magni

Università degli Studi di Pavia, Pavia, Italy.

Background

One of the most important applications of machine learning in drug discovery is to relate the chemical structure of compounds to physicochemical properties.¹ This task is known as Quantitative Structure Properties Relationship (QSPR). These predicted properties can then be used, for example, as inputs in physiologically‐based pharmacokinetic (PBPK) models to predict compounds PK during early drug discovery.² The objective of this work was to develop QSPR models able to predict the aqueous solubility (Sw) and the intrinsic solubility (Sint) of pharmaceutical compounds from their molecular structure. This work was done in the context of the drug solubility challenge 2.0.³

Methods

A dataset (DS) of 6340 compounds with their experimental Sw was built starting from the EPI Suite™.⁴ 2D molecular descriptors were calculated for each compound with the Python package Mordred.⁵ Another DS was constructed by converting Sw to Sint with the Henderson Hasselbalch equation. Given the lack of information on the experimental conditions, hypotheses were made for the solution temperature (25°C) and solvent pH (7). The pKa was either experimental or predicted with a neural network (NN). Finally, Sw and Sint DS were used to train two NN.

Results

The R² of the NN trained on the Sw DS was equal to 0.86 (0.84–0.88, 95% CI) and 0.84 (0.75–0.88, 95% CI) for the training and test set, respectively. Concerning the NN trained on the Sint DS, R² was equal to 0.73 (0.7–0.75, 95% CI) for the training and to 0.72 (0.61–0.77, 95% CI) for the test set.

Conclusion

The predictive performances of the Sw NN were better than those of Sint. This is due to the quality of data used to train the two models. In fact, from our experience, in the literature there is a lack of information regarding Sint values and regarding the experimental condition used for the derivation of Sw.

References

1. Lo, Y.‐C., Rensi, S. E., Torng, W. & Altman, R. B. Machine learning in chemoinformatics and drug discovery. Drug Discovery Today 23, 1538–1546 (2018).

2. Daga, P. R., Bolger, M. B., Haworth, I. S., Clark, R. D. & Martin, E. J. Physiologically based pharmacokinetic modeling in lead optimization. 1. Evaluation and adaptation of GastroPlus to predict bioavailability of Medchem Series. Mol. Pharmaceutics 15, 821–830 (2018).

3. Llinas, A. & Avdeef, A. solubility challenge revisited after ten years, with multilab shake‐flask data, using tight (SD ∼ 0.17 log) and loose (SD ∼ 0.62 log) test sets. J. Chem. Inf. Model.59, 3036–3040 (2019).

4. US EPA, O. EPI Suite™‐Estimation Program Interface. US EPA https://www.epa.gov/tsca‐screening‐tools/epi‐suitetm‐estimation‐program‐interface (2015).

5. Moriwaki, H., Tian, Y.‐S., Kawashita, N. & Takagi, T. Mordred: a molecular descriptor calculator. J. Cheminformatics 10, 4 (2018).

21

Predicting the Dose that Can Lead to 95% Sputum Culture Conversion by Two Weeks

L. Najjemba, J. Musaazi, B. Castlenuovo, B. Kuteesa, C. Sekaggya

Infectious Disease Institute, Makerere University, Kampala, Uganda

Background

Treatment of tuberculosis (TB) infection is an important component of TB control. Treatment with antimicrobial drugs requires that we achieve an adequate dose that can safely achieve maximal kill of the organism. Patients receiving TB treatment are presumed to be sputum smear negative after 2 weeks of therapy, however about 20% remain positive and may continue to transmit the mycobacteria especially among those with low rifampicin concentrations which is one of the cornerstone drugs. Several studies have demonstrated that higher doses of rifampicin are more effective, leading to faster sputum smear conversion in patients with tuberculosis, and are also safe. There are several on‐going trials aiming at evaluating higher doses including 35 mg/kg and 50 mg/kg doses. We sought to predict the dose that could lead to 95% conversion of sputum from positive to negative by week two on anti‐tuberculosis drugs.

Methods

Between May 2013 and November 2015 we enrolled HIV‐infected Ugandan adults with pulmonary TB and initiated them on fixed dose combinations of TB drug. Patients received anti‐TB therapy dosed according to WHO weight bands and also underwent pharmacokinetic sampling (1, 2, and 4 hours after drug intake) at 2, 8, and 24 weeks after treatment initiation. High‐performance liquid chromatography was used to quantify drug concentrations. We tested one‐ and two‐compartment disposition models with first‐order elimination to describe pharmacokinetic parameters for rifampicin. We developed a binary model to determine the probability of conversion by week two and performed simulations while adjusting doses to see changes in the probability of conversion by week 2, then plotted a dose probability curve using R version 3.6.1 Population PK modeling was implemented in NONMEM (version 7.4).

Results

Between May 2013 and November 2015, we enrolled 268 patients (148 males & 120 females) with median weight 53.5 (IQR: 47.5–59) kg and age 35 (IQR: 29–40) years. Before simulations, there was 80% conversion to full treatment by week two. Upon adjusting the doses, 900 mg (15 mg/kg) led to 82% conversion by week two, 1200 mg (20 mg/kg) led to 85% conversion, 1800 mg (30 mg/kg) led to 90% conversion and then a 2100 mg (35 mg/kg) led to 95% conversion to full treatment by week two (Figure 1).

Conclusion

A dose of rifampicin of 2100 mg (35 mg/kg) leads to 95% conversion of sputum culture from positive to negative by week two. This supports on going trials evaluating this dose for patients with tuberculosis.FIGURE 1: A dose probability graph showing 95% conversion by 2 weeks of treatment

22

Pharmacodynamics of the Macrophage Activation Marker Neopterin Following Miltefosine Treatment of Visceral Leishmaniasis in Eastern African Patients

S. Palić¹, A. D. R. Huitema^1,2, F. Alves³, J. H. Beijnen¹, T. P. C. Dorlo¹

¹The Netherlands Cancer Institute ‐ Antoni van Leeuwenhoek Hospital, Amsterdam, The Netherlands; ²Utrecht University, The Netherlands; ³Drugs for Neglected Diseases initiative, Geneva, Switzerland

Background

Visceral leishmaniasis (VL) is a deadly neglected tropical disease caused by the Leishmania parasite. The parasites live and replicate within macrophages, massively increasing neopterin production. Previously, we described neopterin dynamics in Eastern African children.¹ In the current study we pooled both adult and pediatric data, to further characterize neopterin dynamics in response to miltefosine treatment.

Methods

78 Eastern African VL patients (age 4–41) from two miltefosine trials were included in this study. Data were analyzed using the first‐order conditional estimation method with interaction in NONMEM (version 7.3.0). An indirect response model was applied to estimate treatment effect of miltefosine on neopterin using a second‐order rate constant in combination with an estimate of endogenous neopterin concentrations.

Results

Neopterin declined during treatment, with an estimated population concentration at start of treatment of 80.7 nmol/L, (inter‐individual variability 56.7%, residual standard error (RSE) 1.5%), and endogenous steady‐state concentration of 16.3 nmol/L (RSE 6.8%). Recrudescence of neopterin was captured by a first‐order rate constant λ, that approached zero in non‐recrudescing patients and median 0.0033 day⁻¹ in recrudescing patients.

Conclusion

We developed a semi‐mechanistic pharmacokinetic‐pharmacodynamic model for neopterin in VL and captured the recrudescence of neopterin during follow‐up in a subset of patients. Additional studies are required to elucidate whether increasing neopterin after the end of treatment could indicate asymptomatic VL patients.

Reference

1. Palić et al. PAGE 28 (2019) Abstr 9170 <www.page‐meeting.org/?abstract=9170>

23

Therapeutic Drug Monitoring of Adalimumab in Psoriasis: Integrated Statistical and Pharmacometric Analyses Using Real‐World Data

S. Pan¹, T. Tsakok^1,2, N. Wilkinson³, R. B. Warren⁴, J. F. Standing⁵, C. H. Smith^1,2, on behalf of the PSORT consortium

¹St John's Institute of Dermatology, Guy's and St Thomas’ NHS Foundation Trust, London, UK; ²King's College London, London, UK; ³Newcastle University, Newcastle‐upon‐Tyne, UK; ⁴University of Manchester, Manchester, UK; ⁵University College London, Great Ormond Street Institute of Child Health, London, UK

Background

Using large clinical data from a real‐world setting¹, we aimed to establish the relationship between adalimumab serum levels and treatment response in psoriasis as measured by Psoriasis Area Severity Index (PASI), and to evaluate a proactive therapeutic drug monitoring (TDM) strategy for adalimumab.

Methods

544 patients on adalimumab provided 950 serum samples and 1157 PASI measurements within the first year. Using multivariable logistic regression, a therapeutic range was derived by relating early and steady‐state serum levels with a 75% decrease in PASI (PASI 75) at 6 months. All data were pooled for PK/PD analysis; a one‐compartment linear PK model of adalimumab was linked with the turnover mechanism for PASI change over time. Patient stratification by response was considered. Real‐time stochastic simulation was performed for a proactive TDM strategy: trough levels at week 5 were assessed against the lower range for dose escalation, and at week 17 were re‐assessed by the upper range and PASI 90 response for dose reduction or escalation. At 6 months, response rates of PASI 75 and 90 and dose costs were compared to those in standard of care.

Results

A therapeutic range of 3.2–7.0 μg/mL was defined for adalimumab. Within the pharmacokinetic‐pharmacodyanmic model, in addition to a priori allometric scaling, waist size, females, hypertension, and time‐varying anti‐drug antibodies significantly increased clearance and volume. Large variabilities in PD parameters were not reduced by any covariates or response stratification. In comparison to standard of care, the proposed TDM strategy improved PASI 75 and 90 response rates by 8.7% and 25.8% respectively, with 14.3% increase of dose costs.

Conclusion

In future adalimumab levels could be monitored using a Bayesian algorithm for individual patients with psoriasis.

Reference

1. Griffiths, C.E.M., et al. Establishing an academic‐industrial stratified medicine consortium: Psoriasis stratification to optimize relevant therapy. J. Invest. Dermatol.135, 2903–2907 (2015).

24

Population Pharmacokinetic Model of Lamivudine in 40 Ugandan Postpartum Mothers

H. Pertinez¹, S. Nakalema², A. Amara¹, I. Kyohairwe², R. Nakijoba², M. Lamorde², S. Khoo¹, C. Waitt^1,2

¹University of Liverpool, Liverpool, UK; ²Makerere University College of Health Sciences, Kampala, Uganda

Background

With uptake of the World Health Organization ‘test and treat’ policy for antiretroviral therapy (ART), up to 1.5 million infants per year will be exposed to ART through pregnancy and breastfeeding. Gaps in understanding of pharmacokinetic transfer from mother to breastfed infant include understanding the changes which occur in both maternal and breastmilk (BM) PK in the early postpartum period.

Methods

40 Ugandan mothers on ART (NVP+AZT‐3TC or EFV‐TDF‐3TC) had intensive pharmacokinetic (PK) profiles in both plasma and BM undertaken at 1–2, 4–6 and 10–12 weeks postpartum (each mother‐infant pair randomized to 2 out of 3 occasions). Approvals were obtained from University of Liverpool and Kampala (JCRC) ethics committees and from the Uganda National Council of Science and Technology. Drug concentrations were measured with a validated LC‐MS/MS assay. Concentration data in plasma and breastmilk (516 observations total) were fitted simultaneously with a population PK model using NONMEM 7.4.1.

Results

3TC plasma PK data was adequately described with a one‐compartment PK disposition model with first order absorption. BM data was modeled as an excretion compartment with an input rate estimated as a fraction of the total plasma clearance of 3TC. Due to the lack of data re. volume of BM produced and the precise times of feeding/emptying of the BM compartment, the volume of distribution for the BM compartment was fixed to a physiologically plausible 0.15L, and the BM compartment allowed to empty via a first order elimination process.

Conclusion

3TC transfer into breastmilk can be described using a pop‐PK model. There were no significant differences in transfer into breastmilk between 1–2, 4–6 and 10–12 weeks postpartum, with creatinine clearance being a significant covariate for clearance.

25

Model‐Based Extrapolation of the Early Bacterial Activity (EBA) of Linezolid for the Treatment of Pulmonary Tuberculosis

F. Romano¹, M. Muliaditan², O. Della Pasqua^1,3

¹University College London, London, UK; ²GlaxoSmithKline, Stevenage, UK; ³GlaxoSmithKline, Uxbridge, UK

Background

A semi‐mechanistic two‐state model was developed¹ to include fast‐(F) and slow‐(S) growing subpopulations to describe the heterogeneity of M. tuberculosis infection. This approach allows for the evaluation of drug effects on bacterial growth dynamics, disentangling it from the underlying growth process. Predictive performance was assessed using standard of care drugs, but novel regimens have yet to be tested to confirm the generalizability of the model as a translational tool in early drug development.

Methods

We characterized the concentration‐effect relationship of linezolid (LZD) in a C57BL/6J murine infection model and scaled relevant model parameters to predict early bactericidal activity (EBA) in patients with pulmonary tuberculosis . The data analysis was implemented in NONMEM v.7.3. All animal studies were ethically reviewed and carried out in accordance with European Directive 2010/63/EEC and the GSK Policy on the Care, Welfare and Treatment of Animals.

Results

The pharmacokinetics (PK) of LZD was best described by a two‐compartment model with Michaelis‐Menten elimination (MM) and dose‐dependent oral bioavailability. The estimated IC50 values for the F and S subpopulations were 2.54 and 204 mg/L, respectively. Scaling of system‐specific parameters to humans was implemented using a one‐compartment PK model with parallel linear and MM elimination processes. VPCs showed good agreement between model‐predicted and observed CFU counts over time after two different LZD dosing regimens. Median predicted EBA was 0.17 vs. 0.22 CFU/ml/day.

Conclusion

Our findings add credence to this model‐based approach to predict EBA. If further validated for other monotherapies, it has the potential for ranking novel single and multi‐drug regimens prior to progression into the clinic.

Reference

1. Muliaditan, M. Development and dose rationale for drug combinations for the treatment of tuberculosis. PhD Thesis.

26

New Version of PFIM for Optimal Design in Nonlinear Mixed Effects Models Using R S4

J. Seurat, Y. Tang, T. T. Nguyen, F. Mentré, H. Le Nagard, on behalf of PFIM group

Université de Paris, IAME, INSERM, Paris, France

Background

Using the Fisher Information Matrix (FIM) to optimize the design of longitudinal studies is an efficient alternative to clinical trial simulation. PFIM 4.0 (1) is a R program devoted to the design evaluation and optimisation. Programmed in R S4 language, the new version of PFIM aims to increase the simplicity of use, the comprehensibility and the modularity of PFIM.

Methods

The conception of the new PFIM is based on multiple classes and inheritances. The different classes and objects are conceived to be easily used or modified for programmers and users of PFIM. The FIM is evaluated by first order linearization of the model, as in PFIM 4.0. Based on the D‐criterion, the design is optimized using a multiplicative algorithm which is a new feature. Several examples on the new PFIM are used and the results are compared to those obtained with PFIM 4.0.

Results

The use of the new PFIM is closer to most of R packages than PFIM 4.0. According to these different examples, the evaluated FIM is consistent with those of PFIM 4.0. The different elements of a project can be stored as objects. Moreover, the project can be easily saved. The design optimization by the multiplicative algorithm allows one to optimize the number of arms, measuring times and doses at the same time. After a design evaluation and/or optimization, the results can be presented in a summary, with the different elements that could be manipulated in R.

Conclusion

The use of optimal design approaches can anticipate ‘fatal’ studies. The new version of PFIM fulfill some needs by its usability. Nevertheless, this PFIM version is not final: some features of PFIM 4.0. have to be also implemented in the new PFIM. The perspectives are to implement new features such as alternative methods to evaluate the FIM for discrete response models.

Reference

1. Dumont, C., et al. PFIM 4.0, an extended R program for design evaluation and optimization in nonlinear mixed‐effect models. Comput. Methods Programs Biomed.156, 217–229 (2018).

27

Logical Modeling of Tumor‐Immune Cross‐Talk Netowrk to Predict the Response to Therapy for Muscle Invasive Bladder Cancer (MIBC)

S. D. Shah^1,2, B. A. Foster², D. E. Mager¹

¹State University at Buffalo, Buffalo, NY, USA; ²Roswell Park, Buffalo, NY, USA

Background

Chemotherapy like Cisplatin perturbs the tumor immune system, contributing to the over‐all response to treatment. The tumor‐immune system relationship is extremely dynamic and the tumor post treatment constantly re‐equilibrates the micro‐environment to support its growth. The ability to capture the dynamic changes of the immune microenvironment will allow us to design treatments that can modulate the immune cells for a sustained anti‐tumor effect.

Methods

The tumor‐ immune cross‐talk Boolean model was constructed using publicly available MIBC expression data, linked to the macrophage polarization and CD4 T‐cell differentiation network via “extra‐cellular” cytokines. Dynamic profiles of the tumor‐immune network model were simulated using Odefy under different treatment conditions. Lastly attractor analysis was performed using BoolNet R package to predict the immune cell phenotype for different treatment conditions. The initial condition was described by the cytokine profile of the “extra‐cellular” cytokine and the immune profile was described by the macrophage polarized state and the CD4 T‐cell differentiated cell.

Results

The Boolean model simulated towards Th1 and M1 polarized state in the presence of pro‐inflammatory “extracellular” cytokines and for anti‐inflammatory cytokines the model simulated a Th2 and M2 polarized state. Depending on the immune phenotype, attractor analysis with Cisplatin showed variable response represented by a Th1 v/s Th2 immune environment.

Conclusion

Network analysis suggests that response to treatment is dependent on the tumor‐immune sub‐type of the patients at the time of treatment enrollment. Our Boolean model can be used to study the dynamic behavior of the tumor‐immune crosstalk and identify novel combinations that will support sustained tumor inhibition.

28

Modeling the Impact of Adherence on Antiretroviral Drugs Concentration in the Lymphoid Tissues of HIV‐Positive Pregnant Women

B. Shenkoya¹, S. Atoyebi¹, D. Anweh², C. Ugboho², A. Olagunju^1,3

¹Obafemi Awolowo Uni, Ile‐Ife, Nigeria; ²Federal Medical Centre, Makurdi, Nigeria; ³University of Liverpool, Liverpool, UK

Background

Lymphoid tissues (LT) is a major Human Immunodeficiency Virus reservoir, data on its antiretroviral (ARV) penetration is limited. Poor ARV penetration into LT is hypothesized to cause virological failure among non‐adherent patients. We aim to estimate ARV concentration in LT of HIV‐positive pregnant women (HPPW) using physiologically‐based pharmacokinetic (PBPK) under different adherence scenarios.

Methods

We assessed adherence in 187 HPPW, blood samples were collected for viral load and the data was analyzed using SPSS. The study was approved by clinicaltrials.gov (NCT03284645). A PBPK model was built in Simbiology (MATLAB R2018b), drug‐specific and lymphatic parameters were incorporated and Fick's passive diffusion was used to describe ARV flow into LT. Model predictions within 0.5–1.5 absolute average fold error (AAFE) with clinical data were considered acceptable. Concentration of ARV in LT was simulated in 100 virtual patients taking efavirenz (EFV), dolutegravir (DTG) and rilpivirine (RPV), the results were compared with their 95% Inhibitory Concentration (IC95).

Results

Regression analysis showed that viral load ≤ 1,000 copies/mL was associated with low risk of HIV vertical transmission (aOR 0.846, 95% CI: 0.719–0.996, P = 0.045). The model predictions were within AAFE of 1.00–1.21, the LT Cmin‐to‐IC95 ratio for EFV, DTG and RPV was 74.8, 0.27 and 5.17 μg/mL respectively. Concentration in LT decreased with non‐adherence (−42.7% to −88.8%), patients missing doses consecutively have lowest LT penetration.

Conclusion

HPPW should ensure high adherence to increase HIV exposure to lethal ARVs in tissue reservoirs. While current clinical ARV assessment in LT of human is yet acceptable, PBPK modeling presents a useful tool to estimate LT concentration of drugs currently under development.

29

Pharmacometric Approach to Support Dosing Strategy of Nimotuzumab in Patients with Autosomal Dominant Polycystic Kidney Disease (ADPKD)

N. de Castro Suárez¹, M. N. Trame², V. Mangas Sanjuán^3,4, M. Ramos Suzarte⁵, J. M. Dávalos⁶, R. Bacallao⁶, A. Regla Maceo Sinabele⁷, L. Rodríguez Vera¹

¹University of Havana, Cuba; ²Novartis Institutes for BioMedical Research, Inc., Cambridge, MA, US; ³University of Valencia, Valencia, Spain; ⁴Polytechnic University of Valencia, Valencia, Spain; ⁵Center of Molecular Immunology, Havana, Cuba; ⁶National Institute of Nephrology, Havana, Cuba; ⁷National Center Coordinating of Clinical Trials

Background

Autosomal dominant polycystic kidney disease is a genetic disease characterized by an overexpression of epidermal growth factor receptor (EGFR). Nimotuzumab is a recombinant humanized monoclonal antibody against human EGFR. The aims of this study are to develop a population pharmacokinetic (PopPK) model for nimotuzumab as well as to identify demographic and clinical predictive factors of the PK variability.

Methods

Data from a single‐center phase 1 clinical trial were used for the PopPK analysis. This clinical trial was authorized in 2009 (number 442/05.014.08‐B, volume 06, folio 000402). Five patients with ADPKD were enrolled at each of the following fixed dose levels: 50, 100, 200 and 400 mg. Blood samples were drawn during 28 days for pharmacokinetic (PK) assessments. PopPK analysis of 409 concentration‐time data from 20 patients was performed using the nonlinear mixed‐effect model approach with NONMEM 7.3. Impact of patient demographics and clinical indices on PK parameters were explored using automated stepwise covariate model‐building technique in PsN.

Results

QSS approximation of the full TMDD model with constant target concentration best described the concentration‐time profiles. The final model estimates were 0.0102 L/h for linear clearance, 2.32 L for central volume (Vc), 0.0126 L/h for inter‐compartmental clearance (Q), 4.27 L for peripheral volume (Vp), 7.27 mg/L for steady‐state constant, 0.299 h⁻¹ for internalization rate, and 0.432 mg/L for total target concentration. Interindividual variability was associated with Vc, Q, Vp.

Conclusion

This is the first PopPK study of nimotuzumab in non‐oncological disease. The model was able to describe the effect of the mAb–target binding, and target and mAb–target complex turnovers on nimotuzumab pharmacokinetics.

30

Open‐Label Study to Evaluate Potential Pharmacokinetic Interactions of Mefloquine and Dihydroartemisinin‐Piperaquine in Healthy Adult Subjects

A. Sujintawong¹, R. M. Hoglund^1,2, J. l. Tarning^1,2

¹Mahidol University, Bangkok, Thailand; ²University of Oxford, Oxford, UK

Background

Malaria is a life‐threatening infectious disease, endemic in many areas worldwide. Artemisinin‐based combination therapy (ACT) is the recommended first‐line therapy for the treatment of malaria.¹ Recently, multi‐drug resistant parasites have emerged in the greater Mekong Sub‐region, reducing the efficacy of ACTs substantially.² One option to counteract the emerging resistance is to add one more drug, with a different mechanism of action, to existing drug combinations. A proposed triple combination is dihydroartemisinin (DHA) – piperaquine (PPQ) – mefloquine (MQ). However, safety and drug‐drug interactions of this novel combination have to be evaluated before implementation in field trials. The objective of this study was to investigate the pharmacokinetic properties of dihydroartemisinin and piperaquine when given alone and in combination with mefloquine, with a specific focus on potential drug‐drug interactions.

Methods

The population pharmacokinetic properties of dihydroartemisinin and piperaquine were investigated in 15 healthy Thai volunteers, whom were given a single oral dose of dihydroartemisinin‐ piperaquine (3×40 mg DHA/120 mg PPQ) alone and in combination with mefloquine (2×250 mg MQ). Plasma samples were collected at 0 (pre‐dose), 0.25, 0.5, 1, 1.5, 2, 3, 4, 6, 8, 10, 12, and 24 hours, and at day 3, 4, 7, 11, 15, 22, and 36. The collected pharmacokinetic data were analyzed using nonlinear mixed‐ effects modeling (NONMEM v.7.4.3).

Results

Dihydroartemisinin was described by a two‐compartment disposition model with four transit compartments in the absorption phase. Co‐administration with mefloquine increased the absorption time (31%) and decreased the relative bioavailability (19%), resulting in a decreased exposure to dihydroartemisinin (19%). Piperaquine was described by a three‐compartment disposition model with two‐transit compartment in the absorption phase. Mefloquine did not alter the pharmacokinetic properties of piperaquine when co‐administrated.

Conclusion

The developed population pharmacokinetic models described the pharmacokinetic properties of dihydroartemisinin and piperaquine adequately. Concomitant treatment with mefloquine was shown to lower the exposure to dihydroartemisinin by 19%.

References

1. World Health Organisation. Malaria, Key facts.2019, March 27; Available from: <https://who.int/news‐room/fact‐sheets/detail/malaria>

2. Imwong, M., Suwannasin, K., Kunasol, C., Sutawong, K., Mayxay, M., Rekol, H. & Dondorp, A. M. The spread of artemisinin‐resistant Plasmodium falciparum in the Greater Mekong subregion: a molecular epidemiology observational study. Lancet Infect. Dis.17, 491–497 (2017).

31

A Shiny App for Pediatric Dosing

T. Tikiso¹, H. Mcilleron¹, N. Sugandhi², T. R. Cressey³, M. Mirochnick⁴, E. V. Caparelli⁵, M. Penazzato⁶, P. Denti¹, for the WHO Paediatric Anti‐retroviral Working Group PAWG

¹University of Cape Town, South Africa; ²Columbia University, New York, NY, USA; ³Programme for HIV Prevention and Treatment, Chiang Mai, Thailand; ⁴Boston Med Centre, Boston, MA, USA; ⁵University of California San Diego, La Jolla, CA, USA; ⁶World Health Organization, Geneva, Switzerland

Background

Pediatric dosing of anti‐infectives aims to match adult exposure. Allometry and maturation generally predict paediatric doses down to around 2 years of age, their nonlinearity can be confusing for non‐modellers. The WHO Paediatric Anti‐retroviral Working Group has recently created a generic pediatric dosing tool based on MS Excel to ease the calculation of expected AUCs in children relative to adult targets adjusting for allometric scaling and maturation.¹ While helpful, the current tool has limitations, it only compares overall AUC, it can still be confusing with its multiple sheets and intimidating formulae, and there is a possibility that the user may edit the wrong cells. The objective of this work was to improve on the tool by re‐implementing it with a Shiny app and add additional functionalities.

Methods

The app contains two major sections. 1) the “generic form” which is designed similarly as the old tool and 2) a library of published drug‐specific models that allows for simulation of between‐subject and –occasion variability as well as other important drug‐specific covariates. Dose frequency and in silico population for the simulations (standard WHO growth charts, malnourished children, or custom data) can be changed.

Results

The “generic form” of the app has the same functionality as the old tool but better usability, while the drug specific models can simulate the entire PK profile and thus provide AUC, Cmax, and Cmin. Results are displayed in a graphical and tabular form and an R markdown report can be generated at the end.

Conclusions

The app is easier to use compared to the excel tool and maintains the same basic functionalities. It can be accessed online with a web browser, and the addition of the library of models and specific in silico populations makes it a much more powerful tool.

Reference

1. Denti, P., Sugandhi, N., Cressey, T.R., Mirochnick, M., Capparelli, E.V., Penazzato, M. An easy‐to‐use paediatric dosing tool: one mg/kg dose does not fit all. In: 24th Conference on Retroviruses and Opportunistic Infections. 2017. Abstract. 809

32

Cancer Cachexia: A Dynamic Energy Budget (DEB)‐Based Model for Tumor‐Related Anorexia and Sarcopenia

E. M. Tosca¹, A. Montanaro¹, M. Rocchetti², P. Magni¹

¹University of Pavia, Pavia, Italy; ²Consultant, Milan, Italy

Background

Cancer cachexia is a multifactorial syndrome defined as an ongoing loss of skeletal muscle (sarcopenia) with or without loss of fat mass. Body weight loss (BWL) is caused by a negative energy balance due to reduced food intake (anorexia) and altered metabolism. Based on its clinical relevance (20% of cancer deaths are directly caused by cachexia), specific animal models were developed. In this work, we tested our tumor‐in‐host DEB‐based model¹ in describing BWL, anorexia and body changes in tumor‐bearing (TB) animals.

Methods

Rodent studies involving different tumor models were taken from literature. Experiments included tumor‐free (TF) and TB animals with ad libitum feeding and pair fed (PF) healthy animals miming tumor‐induced anorexia. The DEB‐model was identified on host BW and tumor weights of TF and TB groups. Tumor‐anorexia model was tested on daily food intake and BW of PF group. Predictions of energy reserve (We) and structural biomass (Wv) were compared to experimental adipose tissue (epidydimal mass) and skeletal muscle (gastrocnemius weight) data.

Results

The model excellently predicted food intake reduction due to tumor progression in cell lines with different grade of anorexia. BWL predictions in PF group highlighted the model ability to discern body alterations caused by reduced feeding from the ones stemming from tumor energy consumption. A qualitative agreement between model predictions of host body composition dynamics and experimental data was observed. In particular, model showed that in TB animals both We (adipose tissue) and Wv (skeletal muscle) were decreased whereas, in TF animals, fasting affected We (adipose tissue) more than Wv (muscle).

Conclusion

Obtained results confirmed that the tumor‐in‐host DEB‐based model could be a useful tool to further investigate the complexities of cancer cachexia.

Reference

1. Tosca, E. M., et al. A population dynamic energy budget‐based tumor growth inhibition model for etoposide effects on Wistar rats. Pharmaceutical Res. 36, 38 (2019).

33

A Mechanistic Pharmacokinetic/Pharmacodynamic Model to Evaluate Safe Dosing of Primaquine in Mass Drug Administration

S. W. van Beek¹, E. M. Svensson^1,2, A. Tiono³, J. Okebe⁴, U. D'Alessendro⁴, B. Goncalves⁵, C. Drakeley⁵, T. Bousema⁶, R. ter Heine¹

¹Radboud University Medical Center, Nijmegen, The Netherlands; ²Uppsala University, Uppsala, Sweden; ³National Center for Research and Training on Malaria (CNRFP), Burkina Faso; ⁴Medical Research Council Unit The Gambia at the London School of Hygiene & Tropical Medicine, The Gambia; ⁵London School of Hygiene & Tropical Medicine, London, UK; ⁶Radboud University Medical Center, Nijmegen, The Netherlands

Background

Primaquine (PQ) is crucial in the treatment of malaria. Toxicity is aggravated in patients with G6PD deficiency (G6PDd), sometimes causing lethal hemolysis. Although a clear relationship between PQ dose and hemolysis has been established, the relationship between PQ pharmacokinetics (PK) and hemolysis has not been elucidated. Our aim was to identify this relationship to evaluate the possibility of safe PQ dosing in mass drug administration.

Methods

Pharmacokinetic (PK) analysis was performed by means of non‐linear mixed effects modeling. Single dose PQ PK and hemoglobin (Hb) data from an adult (n = 16, G6PDd) and a pediatric study (n = 40, G6PD normal) were used. The PK model was based on previous work.¹ A PD model was introduced in a sequential way and described the PQ‐dependent production and elimination of erythrocytes by means of a lifespan model. The developed model was used to explore the efficacy of dose splitting to increase safety.

Results

The PK model was adapted to our data by including allometric scaling using body weight and re‐estimation of the parameters. The lifespan model included a four‐compartment transit model and a concentration‐slope effect on the transit compartments to describe the PQ‐induced hemolysis. The estimated lifespan of the erythrocytes was 184 hours (RSE=27%). A single 0.4 mg/kg dose of primaquine in a typical G6PDd adult resulted in a drop of 7% in Hb from baseline. Splitting the dose over day 1, 3 and 5 resulted in less decrease in Hb from baseline (5%).

Conclusion

By describing the relationship between the PK of PQ and hemolysis, we show that dose splitting results in decreased toxicity. This encourages prospective clinical trials of efficacy and safety with this strategy.

Reference

1. Goncalves, B.P., et al. Age, weight, and CYP2D6 genotype are major determinants of primaquine pharmacokinetics in African children. Antimicrob Agents Chemother.61, e02590‐16 (2016).

34

Pharmacokinetics of Para‐Aminosalicylic Acid in Children Treated for Multidrug‐Resistant Tuberculosis

L. E. van der Laan^1,2, A. J. Garcia‐Prats², H. S. Schaaf², M. Chirehwa¹, J. L. Winckler², H. R. Draper², L. Wiesner¹, J. Norman¹, H. McIlleron¹, P. R. Donald², A. C. Hesseling^2,**, P. Denti^1,**

¹University of Cape Town, South Africa; ²Stellenbosch University, Cape Town, South Africa

*^*Senior authors

Background

The optimal para‐aminosalicylic acid (PAS) dose for children treated for multidrug‐resistant (MDR)‐TB is uncertain, specifically for delayed release granule preparations. We describe the pharmacokinetics (PK) of PAS in HIV‐infected and uninfected children routinely treated for MDR‐TB.

Methods

A population PK model was developed describing the PK of PAS in children (n = 27) receiving routine MDR‐TB treatment in combination with various regimens including levofloxacin/moxifloxacin/ofloxacin, linezolid, clofazimine, terizidone, pyrazinamide, ethambutol, ethionamide, high‐dose isoniazid, amikacin and capreomycin. As part of an observational PK study (MDR PK1; N11/03/059 Stellenbosch Ethics approval).

Results

The median (interquartile range) age and weight were 3.87 (1.78, 5.23) years, and 13.3 (10.9, 17.0) kg, respectively with 4 (14.8 %) HIV‐infected. A one‐compartment model with first‐order elimination and transit compartment absorption, described the PK of PAS. The typical clearance (CL) in a 13‐kg child was estimated at 9.09 L/h. Increased PAS CL was observed in both PK profiles from a single only patient on efavirenz (EFV). This was retained in the model as this interaction was previously reported in adults(1). No effect of renal function, sex, ethnicity, nutritional status, HIV status, other antiretrovirals (lamivudine, abacavir, lopinavir/ritonavir) or MDR‐TB drugs was detected.

Conclusion

A transit compartment adequately describes the absorption for the slow release PAS formulation. We confirmed a previously reported increase in CL due to EFV co‐administration, likely related to N‐acetyltransferase 1 (NAT 1) or NAT 2 induction by EFV.

Reference

1. de Kock, L., et al. Pharmacokinetics of para‐aminosalicylic acid in HIV‐uninfected and HIV‐coinfected tuberculosis patients receiving antiretroviral therapy, managed for multidrug‐resistant and extensively drug‐resistant tuberculosis. Antimicrob. Agents Chemother.58, 242–250 (2014).

35

Constructing a Representative In‐Silico Population for Pediatric Simulations

R. Wasmann¹, E. M. Svensson^2,3, S. Walker⁴, M. Clements⁴, P. Denti¹

¹University of Cape Town, Cape Town, South Africa; ²Radboud University Medical Center, Nijmegen, The Netherlands; ³Uppsala University, Uppsala, Sweden; ⁴Medical Research Council Clinical Trials Unit, London, UK

Background

Pediatric population pharmacokinetic (PK) models often include highly‐correlated covariates, e.g. age (for maturation of clearance), and allometry (scaling volume and clearance) with weight or fat‐free mass, which itself depends on weight, height, age, and sex. Because of these correlations, constructing a realistic pediatric population is crucial to the outcome of the simulations. International growth references can aid the process but are often not representative of specific populations, e.g. underweight/stunted children with Human Immunodeficiency Virus (HIV), tuberculosis, or malaria. We present a method to construct such an in‐silico population using data from the ARROW trial, a large cohort of HIV+ children from Sub‐Sahara Africa.¹

Methods

We collated demographic data from 1206 HIV+ children (63,321 measurements, range of 3.8–63 kg and 0.4–22 years) and used it to adjust the WHO and CDC international growth charts. The ARROW data was used to compute: 1) correlation between weight and height and 2) construct an age‐based correction factor for weight‐ and height‐for‐age. Z‐scores from a multivariate normal distribution with the calculated weight/height correlation simulated using the LMS‐formula and then adjusted with an age‐based correction factor. Agreement between simulated and observed subjects was assessed visually.

Results

Simulation from growth charts substantially overestimates both weight and height in our cohort. A weight/height correlation of 0.72, a correction factor of 0.92 for height (independent of age), and a piece‐wise linear function to correct weight‐for‐age downwards produced good agreement between simulated and observed data.

Conclusion

We suggested a method to adjust the WHO/CDC growth charts to a specific population and successfully applied it to an African pediatric HIV+ cohort.

36

Contribution of Machine Learning to Clinical Tumor Growth Inhibition Modeling

M. Wilbaux¹, D. Demanse¹, Y. Gu², A. Jullion¹, T. Kakizume³, A. Myers⁴, C. Meille¹

¹Novartis, Switzerland; ²Novartis, US; ³Novartis Pharma K.K., Japan; ⁴Novartis, China

Background

Machine learning opens new perspectives in identifying predictive factors of efficacy in oncology phase 1 studies and can contribute to improve predictions of tumor growth inhibition (TGI) pharmacokinetic/pharmacodynamic (PK/PD) models.

Methods

Once‐daily oral doses of FGF401 (FGFR4 kinase inhibitor) have been evaluated in a phase 1/2 study (NCT02325739 approved by Ethics Committee) including 127 hepatocellular carcinoma patients. Clinical efficacy, as longitudinal sum of the longest diameter data, was described by a TGI PK/PD model applying nonlinear mixed‐effects modeling. Penalized Cox regression using elastic net model, combined with cross‐validation was used to derive a risk score predictive of time to progression (TTP) from 82 patients’ baseline factors. This score was evaluated as a covariate on PK/PD model parameters. The final model allowed simulating TGI profiles for patients with low and high‐risk score.

Results

A two‐compartment model with a delayed 0‐order absorption and linear elimination was selected to describe PK data and derive individual PK parameters. Tumor growth was best described by a first‐order process, and PK linked to a tumor‐killing rate through an effect compartment. A resistance term was included to describe the observed relapse under treatment.¹ Elastic net model resulted in four baseline predictive factors of TTP: lymphocyte count, number of metastases, age and portal vein invasion. The derived score was found as a significant covariate (P < 0.02) on the TGI resistance parameter with 30% reduction of its variability.

Conclusion

The proposed methodology, combining machine learning and PK/PD modeling, allows including a large range of baseline factors to improve PK/PD model predictivity.

Reference

1. Claret, L. et al. Model‐based prediction of phase III overall survival in colorectal cancer on the basis of phase II tumor dynamics. J. Clin. Oncol.27, 4103–4108 (2009).

37

mPBPK and TMDD Modeling and Simulation to Predict Drug Exposure and Target Engagement at Site of Action and Guide the Design of a Phase 2 Dose‐Ranging Study

C. Zecchin, S. Flint, S. Zamuner

GlaxosmithKline

Background

GSK2330811 is a humanized IgG1 monoclonal antibody (mAb) that blocks oncostatin M (OSM). A potential indication is Crohn's disease (CD). Data from a single dose escalation study in healthy volunteers (HV) (NCT02386436) is available. This analysis aims to use the PK/PD model developed from single dose data¹ to simulate target engagement (TE) in plasma and colon, site of action for CD, for repeat dosing and inform the selection of doses for a possible phase 2 study.

Methods

The mPBPK model¹ with target mediated drug disposition (TMDD) in plasma and leaky tissues described mAb and OSM concentration in plasma and skin blister fluid and was used to simulate PK and TE for repeat dosing and different dose levels. Exposure and TE in colon were simulated isolating colon from the leaky tissues compartment, using physiological values for lymph flow to and from colon, volume and permeability. Baseline OSM plasma and colon concentration were estimated from plasma and colon biopsies of CD patients. The mAb formulation is 150 mg/mL in a pre‐filled syringe, thus dose‐ranging by frequency of administration was desirable. Doses covering 10x range of exposure and TE between 40% and >95% were investigated.

Results

Measured OSM baseline concentration was 2.5x higher in CD than HV. Simulated steady state mAb concentration in colon was 0.4x plasma concentration and TE in colon was 0.8x‐1x TE in plasma. Doses of 150 mg every 1, 2, 4, 8 weeks were selected to test the mechanism of action and dose‐response relationship.

Conclusion

This analysis shows how an mPBPK model with TMDD developed on single dose plasma and skin blister fluid data can be used to simulate repeat dose drug exposure and TE in colon (target tissue) and plasma, providing a rational basis for selection of doses for a phase 2 study.

Reference

1. Reid, J. et al. In vivo affinity and target engagement in skin and blood in a first‐time‐in‐human study of an anti‐oncostatin M monoclonal antibody. Br. J. Clin. Pharm. 84, 2280–2291 (2018).

38

Population Pharmacokinetics of Dapsone in Healthy Volunteers in Nigeria

O. Kotila^1,2, D. Ajayi¹, C. Masimirembwa², R. Thelingwani², A. Odetunde¹, A. Falusi¹, C. Babalola^1;

¹University of Ibadan, Ibadan, Nigeria; ²African Institute for Biomedical Sciences and Technology (AiBST), Harare, Zimbabwe

Background

Nigeria has a high burden of HIV and leprosy, therefore dapsone is frequently used in the population. However, data on population pharmacokinetic (popPK) analysis of dapsone in Nigerians are lacking. The aim of this study was to develop, on pilot scale, a popPK model for dapsone in Nigerians and to identify factors that account for variability in PK parameters.

Methods

The study group was composed of Nigerian healthy volunteers. Each participant received a 50 mg oral dose of dapsone tablet. Plasma concentrations of dapsone and its monoacetyl derivative were determined by high‐performance liquid chromatography. PopPk analysis involved nonlinear mixed effects modeling. One‐ and two‐compartment models with first‐order absorption (with or without lag time) and first‐order elimination were tested to determine the structural base model. Covariates investigated were sex, N‐acetyltransferase 2 isoform (NAT2) phenotype, and weight. The study was approved by the University of Ibadan/University College Hospital Ethics Committee with protocol number UI/EC/10/0021.

Results

A total of 11 participants were included in the analysis. The model was parameterized as Tlag (lag time), ka (absorption rate constant), Cl/F (apparent clearance), and V/F (apparent volume of distribution). The popPK parameter estimates for dapsone with inter‐individual variability were Tlag = 0.25 h (fixed parameter); ka = 1.20 h⁻¹ (38.6%); V/F = 59.90 L (28.3%); and Cl/F = 1.14 L/h (32.3%). Sex was significantly associated with Cl/F (P = 0.002).

Conclusion

The results show that dapsone PK in Nigerians follows one‐compartment model with lag time, first‐order absorption, and elimination. Sex significantly influences the clearance of dapsone in Nigerians, whereas NAT2 phenotype and weight do not.

39

Can Population Pharmacokinetic Models of Tacrolimus be Extrapolated Between Different Transplant Populations?

R. Kirubakaran^1,2, S. Hennig^3,4, B. Maslen¹, J. E. Carland^1,2, R. O. Day^1,2, S. L. Stocker^1,2

¹University of New South Wales, Sydney, Australia; ²St Vincent's Hospital, Sydney, Australia; ³The University of Queensland, Brisbane, Australia; ⁴Certara, Princeton, NJ, USA

Background

Over 60 population pharmacokinetic (popPK) models of tacrolimus in adult transplant recipients have been published. However, data on the accuracy of Bayesian forecasting with concomitant azole therapy or extrapolation to other transplant cohorts are scarce. The aim of our study is to externally evaluate the predictive performances of relevant popPK models of tacrolimus in adult heart transplant (HTX) recipients with and without azole therapy.

Methods

Published popPK models of tacrolimus (n = 66) were identified, and a subset was selected based on specific criteria. Models were transcribed and predictions were performed using NONMEM v7.4. Data from 39 HTX recipients (1712 concentrations) in 2017 treated with tacrolimus at St Vincent's Hospital, Sydney were obtained immediately post‐HTX up to 3 months post‐azole cessation. Bayesian forecasting was used to establish the predictive performances (bias [median prediction error] and precision [median absolute prediction error]) of the models to predict tacrolimus concentrations. Clinically acceptable bias was between ±20% and precision was ≤20%.

Results

Of the 15 models evaluated, 3 models^1–3 displayed the best predictive performances with concomitant azole therapy (1345 concentrations). However, all models were unsatisfactory in predicting TAC concentrations without azole therapy (367 concentrations).

Conclusion

The predictive performances of population PK models for tacrolimus in post‐HTX recipients varied substantially without azole therapy. The incorporation of azole therapy as a covariate may improve the accuracy of Bayesian forecasting. The applicability of extrapolating popPK models between different transplant populations warrants further investigation.

References

1. Lu, Z., Bonate, P., Keirns, J. Population pharmacokinetics of immediate‐ and prolonged‐release tacrolimus formulations in liver, kidney and heart transplant recipients. Br. J. Clin. Pharmacol. 85, 1692–1703 (2019).

2. Storset, E., Holford, N., Midtvedt, K., Bremer, S., Bergan, S., Asberg, A. Importance of hematocrit for a tacrolimus target concentration strategy. Eur. J. Clin. Pharmacol.70, 65–77 (2014).

3. Storset, E., et al. Improved prediction of tacrolimus concentrations early after kidney transplantation using theory‐based pharmacokinetic modelling. Br. J. Clin. Pharmacol. 78, 509–523 (2014).

40

Using RxODE for Inductive Linearization ODE Solving

M. Fidler¹, R. Schoemaker², J. Wilkins², R. Hooijmaijers³, T. Post³, Y. Xiong⁴, W. Wang⁵

¹Novartis, Fort Worth, TX, USA; ²Occams, Amstelveen, The Netherlands; ³LAP&P, Leiden, The Netherlands; ⁴Certara, Princeton, NJ, USA; ⁵Novartis, East Hanover, NJ, USA

Background

RxODE is an R package that implements fast and convenient ODE solving by converting R code to C. Inductive linearization (IL) is a new ODE solving methodology¹ which requires separating the ODE system into a linear and a non‐linear matrix. RxODE now allows inductive linearization ODE solving and selects these matrices automatically.

Methods

In RxODE, ODE systems are specified in Leibnitz notation and solved with event tables. This method internally separates inductive matrices by a combination of symbolic algebra and parsing. In this example, both the classic Michaelis–Menten kinetics and stiff van der Pol equations are solved. The results are compared to those from the lsoda solver.

Results

The ODE solving for the Michaelis‐Menten kinetics using IL are comparable to lsoda both in terms of speed (IL˜890 ms vs. lsoda 719 ms) and result (within 1e‐4). Similarly, with the classic stiff van der Pol system (with µ = 1000), the solvers are similar both for the speed (IL˜1457 µs vs lsoda ˜818 µs) and solutions (within 1e‐4). When extending the time‐mesh to large times with relatively sparse samples for a highly variable surface, the ODE solving using IL gives a much a smoother solution than the classic lsoda solution, and takes more time (IL˜13 s vs. lsoda˜13 ms).

Conclusion

The inductive linearization solver has different properties than the lsoda solver, which can be explored in further research. In the past, clever initialization of the inductive matrix led to faster solve times in optimization than using lsoda, which sped up optimization in nlmixr².

References

1. Hasegawa, C., Duffull, S.B. Exploring inductive linearization for pharmacokinetic‐pharmacodynamic systems of nonlinear ordinary differential equations. J Pharmacokinet Pharmacodyn.45, 35–47 (2018).

2. Fidler, M., et al. Exploring inductive linearization in population pharmacokinetic and pharmacodynamic models. ACoP9 2018.

41

A Semi‐Mechanistic Pharmacokinetic Model for Depot Medroxyprogesterone Acetate and Drug‐Drug Interactions with Commonly Used Antiretrovirals and Rifampicin

J. Francis¹, P. Denti¹, H. McIlleron¹, M. A. Kendall², X. Wu², K. E. Dooley³, C. Firnhaber⁴, C. Godfrey⁵, S. E. Cohn⁶, R. Mngqibisa⁷, the A5093, A5283, and A5338 Study Teams

¹University of Cape Town, Cape Town, South Africa; ²Harvard T.H. Chan School of Public Health, Boston, MA, USA; ³Johns Hopkins University School of Medicine, Baltimore, MD, USA; ⁴University of Colorado, Aurora, CO, USA; ⁵National Institutes of Allergy and Infectious Diseases, Bethesda MD, USA; ⁶Northwestern University, Feinberg School of Medicine, Chicago, IL, USA; ⁷Enhancing Care Foundation, Durban International CRS, Durban, South Africa

Background

Depot medroxyprogesterone acetate (DMPA) is a hormonal contraceptive administered as 150‐mg IM injection every 3 months. It is a CYP3A4 substrate, so drug‐drug interactions (DDIs) with HIV/TB treatment may lead to sub‐therapeutic MPA levels (<0.1 ng/mL) resulting in unwanted pregnancies.

Methods

Pharmacokinetic data from ACTG studies A5093 (DMPA alone, or with nelfinavir [NFV], efavirenz [EFV], or nevirapine), A5283 (with lopinavir/ritonavir [LPV/r]), and A5338 (with rifampicin [RIF]+EFV) were analyzed using NONMEM 7.4. MPA concentrations were measured at weeks 2, 4, 6, 8, 10, and 12 after injection. Allometry adjusted for body size and the effects of DDIs were studied.

Results

138 women with HIV (44 with HIV and TB), contributing 744 MPA concentration observations, were included. A one‐compartment model with first‐order elimination characterized DMPA disposition. Release of MPA from a micro‐crystalline suspension was described semi‐mechanistically using a two‐way absorption pathway that applied a logit transformed parameter. Ten percent of the dose was swiftly available in the systemic circulation, while the rest was released slowly with transit‐compartment absorption over a mean transit time of 278 hours. RIF+EFV and EFV alone increased clearance (CL) of MPA by 52% & 25%, respectively; LPV/r and NFV decreased CL by 29% & 16%, respectively. In the LPV/r cohort, the slow release of MPA into systemic circulation was accelerated. The model predicted at week 12, a typical 60‐kg woman on RIF+EFV or EFV would have double the risk of having sub‐therapeutic concentrations. Simulations regardless of weight found that re‐dosing every 8–10 weeks could overcome the risk.

Conclusion

A semi‐mechanistic PK model was developed, characterizing important DDIs, and suggesting alternative dosing scenarios that should be evaluated in future studies.

42

Target Concentration Intervention is Superior to Therapeutic Drug Monitoring for Warfarin Dosing

G. Ma, N. Holford, J. Hannam

The University of Auckland, Auckland, New Zealand

Background

All anticoagulant medications require dose individualization. Target concentration intervention (TCI) provides a more precise approach to adaptive dose individualization compared to therapeutic drug monitoring (TDM). It is unclear whether a TCI approach is necessary for anticoagulation therapy with warfarin. A TDM approach, where dose is not adjusted when measured INR is within 0.5 units of target INR, is advocated for in clinical guidelines. A TCI approach to warfarin uses each INR to inform dose adjustment to the target INR.

Methods

A simulation‐estimation cycle underpinned by a previously published warfarin PKPD model¹ was used to evaluate warfarin adaptive dosing in a simulated cohort (n = 1000) using INR measurements on days 3, 7, 10, 14, 21, 28, 35, 42, 49, and 56. TCI guided warfarin dosing was compared to TDM in three scenarios:

S1: TCI with empirical warfarin model (2) (target INR: 2.5)S2: TCI with warfarin PKPD model (target INR: 2.5)S3: TDM with warfarin PKPD model (acceptable INR: 2–3)

Predictive performance of each scenario at day 56 was summarized using mean prediction error (MPE) and root mean square error (RMSE).

Results

Using the PKPD model to determine dose, TCI was more accurate and precise than TDM (S2 vs. S3). TCI using an empirical model performed poorly compared to the PKPD model (S1 v S2 & 3). See Table.

Conclusion

These findings show that TCI is superior to TDM for warfarin dosing and highlight the need to implement theory‐based rather than empirical models to guide dose individualization.

TABLE

Scenario	MPE (95% CI)	RMSE (mg/day)
S1: TCI Empirical Model	−0.48 (−0.57, −0.38)	1.57
S2: TCI PKPD Model	0.05 (0.02, 0.09)	0.55
S3: TDM PKPD Model	−0.09 (−0.14, v0.04)	0.75

References

1. Xue L., et al. Theory‐based pharmacokinetics and pharmacodynamics of S‐ and R‐warfarin and effects on international normalized ratio: influence of body size, composition and genotype in cardiac surgery patients. Br. J. Clin. Pharmacol.83, 823‐835 (2017).

2. Ryan, P.J., Gilbert, M., Rose, P.E. Computer control of anticoagulant dose for therapeutic management. BMJ 299, 1207‐1209 (1989).

43

Modeling of the Pharmacokinetic Profiles of Piroxicam from Nanovesicles Delivered Transdermally Using Monolix Software

C. C. Mbah, J. I. Ogbonna, N. C. Obitte, A. A. Attama, M. U. Adikwu, S. I. Ofoefule, J. O. Onyechi

University of Nigeria, Nsukka, Nigeria

Background

To overcome adverse effects of oral piroxicam, a nanovesicular formulation (0.5% gel) was designed for transdermal delivery. Monolix software (version 5.1.1) was used to model the permeation profiles, alongside the branded (Feldene^®, 0.5%), to estimate bioavailability.

Methods

Male Wister rat (220–260 g body weight) skin (sample size, 10), and Franz diffusion cell were used for the assessment at 8 sampling time intervals within 0.5–24.0 hours, and the data were analyzed by ANOVA. The flux of piroxicam (j) and permeation coefficient (p) from the 2 products were modeled using 1 compartment, ±log(likelihood) (2LL) between subject variability and additive residual model, and effect of weight as covariate estimated.

Results

The visual predictive check (VPC) showed no significant difference (little error (red) region) between the model and actual observed values, implying good modelling. However, the population estimates obtained for the test formulation were 1.55 ± 0.83 µg/cm²h and 6.2 × 10⁻³ ± 0.05 for the fixed effects, j and p, respectively, with relative standard error (RSE) of 24%; and 0.61 ± 0.46 µg/cm²h, 1.2 × 10⁻³ ± 0.02 and 26%, respectively for the branded, indicating variability of the products. The goodness‐of‐fit (GOF) plot showed most of the points around the line of identity and the statistic (‐2LL) dropped on introduction of weight as a covariate. The structural model diagnostic plot showed the residuals scattered evenly around the zero line, indicating good precision.

Conclusion

Results of the model prediction showed that the test had better bioavailability potentials than the branded. Monolix could be used for modelling permeation pharmacokinetic profiles of transdermally delivered piroxicam.

44

Population Pharmacokinetic Modeling of Esomeprazole for Treatment of Early‐Onset Preeclampsia

M. S. GebreyesUS; ¹, C. A. Cluver², E. H. Decloedt², S. Tong³, S. Walker³, N. G. M. Hunfeld⁴, R. Wasmann¹, P. Denti¹

¹University of Cape Town, Cape Town, South Africa; ²Stellenbosch University, Stellenbosch, South Africa; ³University of Melbourne, Melbourne, Australia; ⁴Erasmus University, Rotterdam, The Netherlands

Background

Esomeprazole is a proton pump inhibitor with preclinical efficacy data showing it lowers soluble fms‐like tyrosine kinase (sFlt1) concentrations, a pathognomonic target identified in preeclampsia. A randomized controlled trial was conducted in South African women diagnosed with early‐onset preeclampsia to investigate efficacy, but it found no change in clinical outcome or sFlt concentrations. It was hypothesized that the 40 mg daily oral dose used is not enough to achieve therapeutic exposure. Esomeprazole is primarily metabolized by CYP2C19, which is polymorphic, and whose activity may be affected by pregnancy. This study aims to investigate the PK of esomeprazole in early‐onset preeclampsia.

Methods

PK data from ten patients in the treatment arm, (median age 30 [range 21–43] years, weight 98.8 [56–126] kg, and gestation age 29 [26–31] weeks), were included. Patients were treated with 40 mg esomeprazole daily and were sampled at 0.25, 0.5, 0.75, 1, 1.5, 2, 4, 8, 10, and 24 hours after the first dose. PK data was analyzed using nonlinear mixed effect modelling with allometric scaling on clearance (CL) and volume of distribution (Vd).

Results

A one‐compartment PK model with first‐order elimination and transit compartment absorption best described the data. Typical values of 19.2 L/h for CL and 44.2 L for Vd were estimated using this model.

Conclusion

We propose a PK model characterising the PK of esomeprazole in pregnant patients with early‐onset preeclampsia. Genotype information was not available in this study, thus making comparisons difficult; however, the estimated CL and Vd are higher than those previously reported for healthy non‐pregnant subjects.

45

Between‐Occasion Variability is Essential to Model Pre‐Dose Concentrations in Repeated‐Dose Pharmacokinetic Studies

A. N. Kawuma, R. E. Wasmann, M. T. Abdelwahab, P. Denti

University of Cape Town, Cape Town, South Africa

Background

Outpatient pharmacokinetic (PK) studies often rely on self‐reports to collect information on the timing of doses prior to the PK visit. The reported times and adherence to medication in general are often inaccurate, leading to unexpected variability in drug concentration prior to the observed dose in the clinic (i.e., the pre‐dose concentration). Additional factors, including diurnal variation in PK or differences in food intake with medications may also affect pre‐dose concentrations. The objective of this analysis is to explore how this variability is best handled.

Methods

A 1‐compartment model with a clearance (CL), volume (V), absorption rate constant (ka), and lag time (ALAG) of 4 L/h, 10 L, 1.2 h⁻¹, and 0.5 h, respectively, was used to simulate steady‐state concentrations for twice‐daily (morning and evening) dosing in 50 subjects. We included 20% between‐subject variability (BSV) on CL, V, and 30% between‐occasion (between‐dose) variability (BOV) on ka, and ALAG. The evening dose was administered 2 h later (to simulate slower absorption when the dose is taken with dinner). A sampling schedule of pre‐dose, 0.5, 1, 2, 4, 6, 8, and 12 h post‐dose was used around the morning dose. Different models were fitted to the simulated data, comparing parameter values, OFV, and plots of BOV on absorption parameters versus occasion.

Results

If only BSV is included, one concludes that a 2‐compartment model is the best fit to the data. However, if BOV in absorption for each dose is also included, one can identify trends showing that there is a systematic difference in the speed of absorption of the evening dose, and correctly account for that in the model.

Conclusion

Neglecting to include BOV for each dose may lead to the identification of an incorrect structural model.

46

Pharmacokinetics of SQ109 in Plasma and Human‐Like Tuberculosis Lesions in Rabbits

O. Egbelowo¹, V. Dartois², M. Gengenbacher², M. Zimmerman², M. Chirehwa¹, P. Denti¹

¹University of Cape Town, Cape Town, South Africa; ²Hackensack Meridian Health Center for Discovery, Nutley, NJ, USA

Background

SQ109 is a TB drug candidate with a novel mechanism of action. The aim of this study is characterizing the penetration of SQ109 in human‐like lesions in TB‐infected rabbits.

Methods

A group of uninfected rabbits received a single dose of SQ109 at 22 or 40 mg/kg orally, or 5 mg/kg IV. Another group of rabbits was infected with TB for 12 to 16 weeks until they developed human‐like lesions, after which SQ109 was administered at 25 or 40 mg/kg, either as single dose, or daily for 6 to 7 days. On the last day, the TB‐infected rabbits were euthanized at 2, 6, or 24 h after the last dose to extract the lung lesions. Blood was collected from the central ear artery pre‐dose and at several time points between drug administration and necropsy. The concentration of SQ109 was determined in plasma and different lesions (normal, cellular, caseous) using LC‐MS/MS. NONMEM was used to interpret the PK data, with the lesions modeled as effect compartments and estimating the time rate constant for plasma‐to‐lesions transfer and the penetration coefficient (ratio of lesions‐to‐plasma concentration at equilibrium).

Results

A two‐compartment model with first‐order elimination and lagged absorption described the plasma PK, which was not different between infected and uninfected rabbits. Oral absorption was found to be very slow and with low bioavailability (˜6%). The model estimated the penetration factor into lesions to be 4570, 4010, and 4020 for the normal, cellular and caseous lesions, respectively. These values show that SQ109 accumulate in lesions at much higher concentrations than plasma. The model also estimated that normal tissue lesions equilibrate faster (equilibrium half‐life 6 h) than caseous (11 h) or cellular lesions (19 h).

Conclusion

We developed a PK model describing the penetration of SQ109 in different lesions, which can be used to estimate the probability of lesion‐centric target attainment.

47

Slower Absorption of Rifampicin in Hospitalized TB‐HIV Patients

N. Abdelgawad¹, M. Chirehwa¹, C. Schutz², D. Barr³, A. Ward², S. Janssen⁴, R. Burton^2,5, R. J. Wilkinson^2,6,7, M. Shey², L. Wiesner¹, H. McIlleron^1,2, G. Maartens^1,2, G. Meintjes², P. Denti^1;

¹University of Cape Town, Cape Town, South Africa; ²University of Cape Town, Observatory, South Africa; ³University of Liverpool, Liverpool, UK; ⁴University of Amsterdam, Amsterdam, The Netherlands; ⁵Khayelitsha Hospital, Cape Town, South Africa; ⁶Imperial College, London, UK; ⁷The Francis Crick Institute, London, UK

Background

Hospitalized HIV‐TB patients typically have worse treatment outcomes than TB outpatients, in particular higher early mortality. To elucidate whether this could be due to worse drug exposure, we investigated the PK of rifampicin in hospitalized and non‐hospitalized TB patients.

Methods

TB‐HIV inpatients at Khayelitsha Hospital in South Africa and a control cohort of TB outpatients in the same area were recruited.¹ Standard TB treatment was given as per national guidelines, including daily weight‐adjusted doses of rifampicin. PK sampling was performed on the third day of treatment at 0, 1, 2.5, 4, 6, and 8 h post‐dose. Plasma rifampicin was quantified using HPLC‐MS/MS. Data were analyzed using NONMEM, and the effect of physiologically plausible covariates was assessed.

Results

60 in‐ and 48 outpatients, contributing 108 PK profiles, were available for analysis. Median (range) weight, fat‐free mass (FFM), and age were 56 (35–88) kg, 43.2 (26.4–64.2) kg, and 37 (19–77) years, respectively. Rifampicin PK was described by a one‐compartment model with first‐order elimination and transit compartments absorption. The typical values of CL and V, which were best allometrically scaled with FFM, were 8.8 L/h and 57 L, respectively. Inpatients were found to have slower absorption (ka 0.725 vs. 1.36 h⁻¹ t1/2 absorption 0.956 h vs. 0.510 h and MTT 0.49 vs. 0.27 h) than outpatients. The few patients (n = 2) taking rifampicin as individual tablets (as opposed to FDC) had significantly lower bioavailability (<50% of reference).

Conclusion

Hospitalized HIV‐TB patients were found to have slower absorption, but otherwise similar overall rifampicin exposures to outpatients.

Reference

1. Schutz, C. et al. Early antituberculosis drug exposure in hospitalized patients with human immunodeficiency virus associated tuberculosis. Br. J. Clin. Pharmacol.86, 966–978 (2020)

48

Model Based Stochastic Simulation and Estimation Method to Use Prior Information for Population Analysis of Sparse Data and Developing Integrated Population Model

S. Choi, S. Han

The Catholic University of Korea, Seoul, South Korea

Background

Developing the robust population PK model of tacrolimus is challenging, and combining available information in reference is important. In this study, a tacrolimus PK model was established by model‐based stochastic simulation and estimation (SSE) method to build the integrated model using published information.

Methods

2377 concentration data (n = 259) of three institutions was collected and combined with the virtual dataset simulated from the model in literature. The simulation and dataset integration were repeated 500 times, and integrated dataset were analyzed using nonlinear mixed effect model analysis. As population/individual parameters were estimated for each dataset, 500 sets of parameters were estimated, and the median was calculated. For patients with long observation (>2 years), half of their samples were used for estimation (training set) and the other half (test set) were used for parameter evaluation.

Results

The structural model was a one‐compartment model with first‐order absorption and elimination. The median of clearance (CL/F), volume of distribution (V/F), and absorption rate (Ka) were 6.23 L/h, 307.97 L, 3.6/h respectively. POD, weight, drug, and institution were selected as covariates. CL/F increases with POD for 2 weeks and decreases afterwards, while V/F decreases with POD for 1 month and increases afterward, possibly due to hypercatabolic state after the surgery. Visual predictive check showed the estimate is appropriate. Also, test set data were close to prediction value simulated using individual parameters.

Conclusion

The integrated population PK model combining both prior information and new data was developed by model‐based SSE method showing the possibility of developing integrated population model without the prior raw data.

49

Population Pharmacokinetics of Amikacin in Egyptian Pediatric Cancer Patients: Dose Individualization Approach

M. A. Afifi¹, M. Nagy¹, A. El‐zeiny¹, M. Ibrahim², L. Shalaby¹

¹Children Cancer Hospital Egypt – CCHE, Cairo, Egypt; ²Helwan University, Helwan, Egypt

Background

Amikacin is an antibiotic that exhibits a concentration‐dependent bactericidal activity. In vitro and in vivo studies showed that the development of bacterial resistance is linked to sub‐optimal dosing of amikacin. Current literature suggests that 10 to 15 mg/kg dose is not sufficient, and a higher daily dose of ≥40 mg/kg was recommended in patients with pediatric cancer. The increase in recommended dose is likely due to the documented increase in amikacin MIC90 from 4 to 8 mg/L against Pseudomonas aeruginosa over the past decades. Amikacin MIC90 was reported at 16 mg/L. This study aims to investigate and analyze the demographics and clinical factors that may influence the pharmacokinetic behavior in Egyptian pediatric cancer patients from routine TDM data.

Methods

This retrospective study included pediatric cancer patients admitted to the in‐patient ward of CCHE, Cairo, Egypt. All patients were treated empirically with amikacin 7.5 mg/kg q12H (total 15 mg/kg) infused intravenously (IV) over 0.5 h for suspected or documented gram‐negative infection. One compartment IV infusion model with nonlinear mixed effects was implemented using Monolix v5.1.1 (Lixoft, France). Age, gender, serum creatinine, comorbidities, weight, and all liver function tests were collected from the local hospital registry. The choice among the competing covariate and error models was based on Akaike information criteria (AIC), where a reduction of 3.84 of AIC was considered significant (P < 0.05). A log‐normal distribution was hypothesized to characterize the interindividual variability of the estimated pharmacokinetic parameters.

Results

A total of 2151 patients were included, with 2292 amikacin plasma concentration sample with one sample or more per each subject. The average age was 9.4 ± 5.1 years, and body weight was 25.9 ± 16.8 kg. The average amikacin total daily dose was 213.1 ± 141.3 mg. The peak and trough concentration that resulted from this dose were 21.05 ± 9.9 and 3.07 ± 5.2 mg/L, respectively. The final covariate model included only age and weight for volume of distribution, while the clearance model included age and serum creatinine. Proportional error model demonstrated the best fit.

Conclusion

This population pharmacokinetic model could be used to individualize the dose of amikacin in pediatric cancer patients. The integration of MIC is further warranted to find the most suitable dose for different MICs.

TABLE

Model	Parameter values ± SE
log(V) = log( Θv) + βage * Age + βweight * weight	Θv = 3.12 ± 0.0858βage = 0.0371 ± 0.00482βweight = 0.016 ± 0.00141
log(Cl) = log ( Θcl) + βage * Age + βserum creatinine * serum creatinine	Θcl = 1.17 ± 0.0615βage = 0.108 ± 0.00464βserum creatinine = −1.16 ± 0.0731

50

An Online Clinical Pharmacology and Pharmacometrics Training Course in Africa: Results of a Proof of Concept Implementation

A. N. Kawuma^1,2, L. C. Nalule¹, P. Denti², L. Aarons³, C. Pillai², the Pharmacometrics Africa Team

¹Makerere University, Kampala, Uganda; ²University of Cape Town, Cape Town, South Africa; ³University of Manchester, Manchester, UK

Background

Pharmacometrics Africa is a platform for open access quantitative clinical pharmacology educational programs run in partnership with local research organizations and academic groups. A 10‐week online clinical pharmacology and pharmacometrics course was developed as part of a set of training activities to support the development of pharmacometrics competency and expertise in low and middle‐income countries (LMICs). This paper describes our experiences and invites collaborations from the pharmacometrics community.

Methods

Course materials that we previously taught at two distance learning institutions in Europe were transferred to the Virtual Learning Environment (VLE) of the Infectious Diseases Institute, Makerere University, Uganda. The curriculum started with concepts in pharmacokinetics (PK), pharmacodynamics (PD), and biostatistics and then proceeded to exploring PK and PD models in modeling software. Each lesson was comprised of videos, guided reading, and exercises and were released weekly to the class for self‐study. At the end of the week, participants joined a live webinar tutorial during which a faculty member discussed the week's materials and exercises. Tutorials were recorded for off‐line viewing for those who might have had difficulties with internet connectivity. The VLE allowed for monitoring of student engagement with the course materials. Students were assigned to small groups under the mentorship of an international and local team of experts who monitored their progress and provided guidance when needed.

Results

The first course included 61 participants mainly from African countries including Ghana, Rwanda, South Africa, Malawi, Uganda, and Nigeria. Participants included: pharmacists, medical officers, statisticians, mathematicians, and scientists with backgrounds in biology, drug regulation, pharmaceutics, and pharmacogenetics. All 61 were retained in the program, with 43 (70.5%) meeting the criteria for award of a certificate of completion. Participants reported spending 5 to 8 h per week on the course. All those who received certificates of completion were invited to attend one of two hands‐on non‐linear mixed effects modelling workshops hosted in Uganda and Nigeria. Following the successful completion of the initial course, the content was re‐written, and thre further programs are either underway or planned.

Conclusion

We successfully implemented an online clinical pharmacology and pharmacometrics course in an African setting.

Pharmacometrics Africa team: Gary Maartens, Maxwell Tawanda Chirehwa, Anthony Afum‐Adjei Awuah, Adeniyi Olagunju, Kayode Ogungbenro, Kamunkwala Gausi, Eva Maria Hodel, Edmund Ekuadzi, Pius Fasinu, Ahmed Abulfathi, Linda Chaba, Eliford Kitabi Ngaimisa, Emmanuel Chigutsa, Simbarashe Zvada.

51

An Online Clinical Pharmacology and Pharmacometrics Training Course in Africa: Incorporation into University Post‐Graduate Degree Programs

E. Ekuadzi¹, A. Olagunju², C. Masimirembwa³, L. Chaba⁴, B. Ogutu⁴, P. Denti⁵, L. Aarons⁶, C. Pillai⁵, the Pharmacometrics Africa Team

¹Kwame Nkrumah University of Science and Technology, Kumasi, Ghana; ²Obafemi Awolowo University, Ife Ife, Nigeria; ³African Institute of Biomedical Science and Technology, AiBST, Harare, Zimbabwe; ⁴Strathmore University, Nairobi, Kenya; ⁵Department of Medicine, University of Cape Town, Cape Town, South Africa; ⁶University of Manchester, Manchester, UK

Background

A new online clinical pharmacology and pharmacometrics course in Africa; ¹ supports pharmacometrics capacity development in low‐ and middle‐income countries. Under the assumption that long‐term sustainability might be facilitated if incorporated into university degree programs, this paper describes our efforts at universities in Zimbabwe, Nigeria, Kenya, and Ghana.

Methods

Local subject matter experts were involved in curriculum development, course delivery, and delegate selection as the program evolved. Each country representative also sought to host related training programs as webinars and face‐to‐face workshops. This provided opportunities for incorporation of local case examples, relevant country specifics, and awareness generation.

Results

The procedures for approval across the countries are varied, time consuming, and asynchronous. The modular online course structure¹ allows for easier incorporation into existing degree programs. The aim was for high quality, engaging and rigorous learning experiences that complement the traditional classroom format. In Zimbabwe, the African Institute of Biomedical Science and Technology, and the Chinhoyi University of Technology, have initiated an MSc Genomics and Precision Medicine and MSc Pharmaceutical Medicine, which will include this content. The first intake of students will be in August 2020. In Ghana, the Kwame Nkrumah University of Science and Technology, Faculty of Pharmacy hosted a 5‐day Clinical Pharmacokinetics workshop with local and international faculty that also generated interest with the University authorities to incorporate pharmacometrics training into the Doctor of Pharmacy program. In Nigeria, the Translational Pharmacokinetics Research Group at Obafemi Awolowo University, Ife Ife is progressing a proposal through the relevant University structures for an MSc in Pharmacometrics. In Kenya, the Master of Science in Biomathematics at Strathmore University, which focuses on methods of applied mathematics and modeling used to analyze practical questions in biomedical research, is currently introducing this content as a module in quantitative clinical pharmacology. This will constitute 12% of the course.

Conclusion

While the processing timelines are slow and bureaucratic, the individual country champions are optimistic of success.

Reference

1. Kawuma, A., et al. An on‐line clinical pharmacology and pharmacometrics training course in Africa: results of a proof of concept implementation. Poster at WCoP2020

Pharmacometrics Africa team: Gary Maartens, Maxwell Tawanda Chirehwa, Anthony Afum‐Adjei Awuah, Adeniyi Olagunju, Kayode Ogungbenro, Kamunkwala Gausi, Eva Maria Hodel, Edmund Ekuadzi, Pius Fasinu, Ahmed Abulfathi, Linda Chaba, Eliford Kitabi Ngaimisa, Emmanuel Chigutsa, Simbarashe Zvada.