Nonparametric Adaptive Grid Algorithm Research Articles

Modeling Pharmacokinetics in Individual Patients Using Therapeutic Drug Monitoring and Artificial Population Quasi-Models: A Study with Piperacillin.

Population pharmacokinetic (pop-PK) models constructed for model-informed precision dosing often have limited utility due to the low number of patients recruited. To augment such models, an approach is presented for generating fully artificial quasi-models which can be employed to make individual estimates of pharmacokinetic parameters. Based on 72 concentrations obtained in 12 patients, one- and two-compartment pop-PK models with or without creatinine clearance as a covariate were generated for piperacillin using the nonparametric adaptive grid algorithm. Thirty quasi-models were subsequently generated for each model type, and nonparametric maximum a posteriori probability Bayesian estimates were established for each patient. A significant difference in performance was found between one- and two-compartment models. Acceptable agreement was found between predicted and observed piperacillin concentrations, and between the estimates of the random-effect pharmacokinetic variables obtained using the so-called support points of the pop-PK models or the quasi-models as priors. The mean squared errors of the predictions made using the quasi-models were similar to, or even considerably lower than those obtained when employing the pop-PK models. Conclusion: fully artificial nonparametric quasi-models can efficiently augment pop-PK models containing few support points, to make individual pharmacokinetic estimates in the clinical setting.

Model-Informed Rationale for Early Therapeutic Drug Monitoring of Colistin in Critically Ill Patients.

Adequate colistin exposure is important for microbiological clearance. This study was performed in critically ill patients >18 years old to develop a simplified nonparametric pharmacokinetic (PK) model of colistin for routine clinical use and to determine the role of dose optimization. The Non-Parametric Adaptive Grid algorithm within the Pmetrics software package for R was used to develop a PK model from 47 patients, and external validation of the final model was performed in 13 patients. A 1-compartment multiplicative gamma error model with 0-order input and first-order elimination of colistin was developed with creatinine clearance and serum albumin as covariates on elimination rate constant. An R2 for observed vs individual predicted colistin concentrations of 0.92 was obtained in the validation cohort. High interindividual variability in colistin steady-state area under the plasma concentration-time curve (AUC) from from 120 hours to 144 hours (coefficient of variation=80.1%) and a high interoccasion variability (median coefficient of variation of AUC from time 0 to hours predicted every 8hours for initial 96hours after starting colistin=23.8) was predicted in patients who received this antibiotic for a period of over 152hours (n=22). With the model-suggested dose regimen, only 20% of simulated profiles achieved AUC from time 0 to 24 hours in the range of 50 to 60mg• h/L due to high variability in population PK. In this group of patients, steady-state colistin concentrations were predicted to be achieved >96hours after initiation of colistimethate sodium. This study advocates the need for early and repeated therapeutic drug monitoring and dose optimization in critically ill patients to achieve adequate therapeutic concentration of colistin.

Imipenem/cilastatin/relebactam pharmacokinetics in critically ill patients with augmented renal clearance.

Imipenem and relebactam are predominantly excreted via glomerular filtration. Augmented renal clearance (ARC) is a common syndrome in critically-ill patients with sepsis, and sub-therapeutic antibiotic concentrations are of concern. Herein, we describe the pharmacokinetics of imipenem/relebactam in critically-ill patients with ARC. Infected patients in the ICU with ARC (CLCR ≥ 130 mL/min) received a single dose of imipenem/cilastatin/relebactam 1.25 g as a 30 min infusion. Blood samples were collected over 6 h for concentration determination. Protein binding was assessed by ultrafiltration. An 8 h urine creatinine collection confirmed ARC. Population pharmacokinetic models with and without covariates were fit using the non-parametric adaptive grid algorithm in Pmetrics. A 5000 patient Monte Carlo simulation assessed joint PTA using relebactam fAUC/MIC ≥8 and imipenem ≥40% fT>MIC. Eight patients with ARC completed the study. A base population pharmacokinetic model with two-compartments fitted the data best. The mean ± SD parameters were: CL, 17.31 ± 5.76 L/h; Vc, 16.15 ± 7.75 L; k12, 1.62 ± 0.99 h-1; and k21, 3.53 ± 3.31 h-1 for imipenem, and 11.51 ± 4.79 L/h, 16.54 ± 7.43 L, 1.59 ± 1.12 h-1, and 2.83 ± 2.91 h-1 for relebactam. Imipenem/cilastatin/relebactam 1.25 g as a 30 min infusion every 6 h achieved 100% and 93% PTA at MICs of 1 and 2 mg/L, respectively. Despite enhanced clearance of both imipenem and relebactam, the currently approved dosing regimen for normal renal function was predicted to achieve optimal exposure in critically-ill patients with ARC sufficient to treat most susceptible pathogens.

1573. Population Pharmacokinetic Analyses for Cefepime in Adult and Pediatric Patients

BackgroundCefepime (CEF) is commonly used for adult and pediatric infections. Several studies have examined CEF’s pharmacokinetics (PK) in various populations; however, a unifying PK model for adult and pediatric subjects does not yet exist. We developed a combined population model for adult and pediatric patients and validated the model.MethodsThe initial model includes adult and pediatric patients with a rich cefepime sampling design. All adults received 2 g CEF while pediatric subjects received a mean of 49 (SD 5) mg/kg. One- and two-compartment models were considered as base models and were fit using a non-parametric adaptive grid algorithm within the Pmetrics package 1.5.2 (Los Angeles, CA) for R 3.5.1. Compartmental model selection was based on Akaike information criteria (AIC). Covariate relationships with PK parameters were visually inspected and mathematically assessed. Predictive performance was evaluated using bias and imprecision of the population and individual prediction models. External validation was conducted using a separate adult cohort.ResultsA total of 45 subjects (n = 9 adults; n = 36 pediatrics) were included in the initial PK model build and 12 subjects in the external validation cohort. Overall, the data were best described using a two-compartment model with volume of distribution (V) normalized to total body weight (TBW/70 kg) and an allometric scaled elimination rate constant (Ke) for pediatric subjects (AIC = 4,138.36). Final model observed vs. predicted plots demonstrated good fit (population R2 = 0.87, individual R2 = 0.97, Figure 1a and b). For the final model, the population median parameter values (95% credibility interval) were V0 (total volume of distribution), 11.7 L (10.2–14.6); Ke for adult, 0.66 hour−1 (0.38–0.78), Ke for pediatrics, 0.82 hour−1 (0.64–0.85), KCP (rate constant from central to peripheral compartment), 1.4 hour−1 (1.3–1.8), KPC (rate constant from peripheral to central compartment), 1.6 hour−1 (1.2–1.8). The validation cohort has 12 subjects, and the final model fit the data well (individual R2 = 0.75).ConclusionIn this diverse group of adult and pediatrics, a two-compartment model described CEF PK well and was externally validated with a unique cohort. This model can serve as a population prior for real-time PK software algorithms. Disclosures All authors: No reported disclosures.

Population Pharmacokinetics of Cefazolin in Serum and Adipose Tissue From Overweight and Obese Women Undergoing Cesarean Delivery.

The optimal antibiotic prophylaxis dosing regimen of cefazolin for cesarean delivery (CD) in overweight and obese women is unknown. This study was done to compare the duration that cefazolin concentrations remain above the minimum inhibitory concentration (MIC) in adipose tissue (AT). Serum and AT concentrations from 3 previous studies in CD patients were comodeled using the nonparametric adaptive grid algorithm in Pmetrics. AT concentrations for 5000 overweight and obese patients receiving 1-, 2-, and 3-g cefazolin regimens were simulated to calculate the probability that free drug concentrations remained above an MIC of 2μg/mL at 1, 1.5, and 2 hours after administration. Sixty-seven patients (mean body mass index 38.7 kg/m2 ; range 25.5-55.8 kg/m2 ) provided data. A 2-compartment model with 1 of the compartments representing AT fit the data best. Final model parameters were clearance 7.38 ± 5.34L/h, volume of central compartment 11.8±9.36 L, and AT volume of distribution 80.12 ± 55.47 L. The mean±SD (median) penetration ratio of cefazolin into AT was 0.81 ± 2.06 (0.62). At 1.5 and 2 hours, 1-, 2-, and 3-g regimens achieved AT concentrations above the MIC in 71.2%, 92.4%, and 94.7%, and 55.7%, 86.8%, and 91.7%, respectively, of simulated patients. Cefazolin achieved good penetration into AT. Because CD duration is commonly less than 1.5 hours, a 2-g dose has a high probability of providing AT concentrations above the target pathogens' MIC for overweight and obese females. A second dose may be considered for longer surgeries.

An exploratory analysis of the ability of a cefepime trough concentration greater than 22 mg/L to predict neurotoxicity.

Cefepime trough concentrations >22 mg/L (T(>22)) have been associated with neurotoxicity in a single study. Neurotoxicity outcomes for 28 cefepime-treated adult patients with febrile neutropenia were abstracted from the literature. The precision of T(>22) to predict neurotoxicity was quantified using 95% confidence intervals. Thirty-two cefepime-treated patients contributed serum concentrations for a pharmacokinetic model, fit using the Nonparametric Adaptive Grid algorithm within the Pmetrics package for R. Concentration-time curves were simulated for common dosing schemes and 3 renal dispositions. Probabilities of neurotoxicity and numbers needed to harm were calculated from simulations according to the proposed pharmacokinetic/toxicodynamic threshold of T(>22). Bayesian modeling was utilized to explore other pharmacokinetic parameters relationships with neurotoxicity. The mean probability of neurotoxicity at T(>22) was 51.4% (95% CI: 16.4-85.0%). Among the schemes and renal dispositions simulated, the combination of cefepime 2 g every 8 h and a creatinine clearance of 60 mL/min produced the greatest probability of neurotoxicity (48.3%). Estimated numbers needed to harm according to T(>22) ranged from 2.1 to 18.5 persons. Explorations of maximal serum concentration and area under the curve demonstrated high levels of collinearity, making it impossible to identify trough concentrations as the driver of neurotoxicity. T(>22) had low precision as a predictive neurotoxic threshold. When a neurotoxic threshold of T(>22) was assumed, projected neurotoxicity rates and numbers needed to harm greatly exceeded observed neurotoxicity rates in the general population and in high risk subpopulations. Other drug exposure metrics should be explored.

Study of relationship between volume of distribution and body weight application to amikacin

Amikacin use is difficult because of its narrow therapeutic and its pharmacokinetic variability. This variability of amikacin is not well known. To adapt amikacin the physician assumes that there is a linear and continuous relation between the volume of distribution and the body weight. The objective of our study was to evaluate the relationship between the volume of distribution (Vd) and the body weight (BW) using a non parametric statistical analysis of dependence so called Z method. Retrospective pharmacokinetic population study and statistic analysis. 872 patients receiving intravenous amikacin. The volume of distribution was modelled using the Non Parametric Adaptive Grid algorithm (NPAG) for a two-compartment model with intravenous infusion. Z coefficient was performed to evaluate the relationships between Vd and BW. For the 872 patients (mean age of 73 ± 17 years) dispatched as follow 53 % female and 47 % male, the analysis of the statistical relationships by the non parametric Z analysis showed a scattered linkage between Vd and BW. For the whole population, the relationship between Vd and BW was not linear (regression analysis). Z analysis demonstrated that only for 80 % of patients there is a relationship between Vd and BW. For these patients, regression analysis give a significant adjustment of a linear model (r = 0.47, p < 0.001). In the whole studied population there is not a continuous and linear relationship between Vd estimated by NPAG and the BW. These results underline the difficulties to adapt doses of amikacin with only BW information.

Population Modeling and Monte Carlo Simulation Study of the Pharmacokinetics and Antituberculosis Pharmacodynamics of Rifampin in Lungs

Little information exists on the pulmonary pharmacology of antituberculosis drugs. We used population pharmacokinetic modeling and Monte Carlo simulation to describe and explore the pulmonary pharmacokinetics and pharmacodynamics of rifampin (RIF; rifampicin). A population pharmacokinetic model that adequately described the plasma, epithelial lining fluid (ELF), and alveolar cell (AC) concentrations of RIF in a population of 34 human volunteers was made by use of the nonparametric adaptive grid (NPAG) algorithm. The estimated concentrations correlated well with the measured concentrations, and there was little bias and good precision. The results obtained with the NPAG algorithm were then imported into Matlab software to perform a 10,000-subject Monte Carlo simulation. The ability of RIF to suppress the development of drug resistance and to induce a sufficient bactericidal effect against Mycobacterium tuberculosis was evaluated by calculating the proportion of subjects achieving specific target values for the maximum concentration of drug (C(max))/MIC ratio and the area under the concentration-time curve from time zero to 24 h (AUC(0-24))/MIC ratio, respectively. At the lowest MIC (0.01 mg/liter), after the administration of one 600-mg oral dose, the rates of target attainment for C(max)/MIC (> or =175) were 95% in ACs, 48.8% in plasma, and 35.9% in ELF. Under the same conditions, the target attainment results for the killing effect were 100% in plasma (AUC(0-24)/MIC > or = 271) but only 54.5% in ELF (AUC(0-24)/MIC > or = 665). The use of a 1,200-mg RIF dose was associated with better results for target attainment. The overall results suggest that the pulmonary concentrations obtained with the standard RIF dose are too low in most subjects. This work supports the need to evaluate higher doses of RIF for the treatment of patients with tuberculosis.

Population pharmacokinetic modelling of repaglinide in healthy volunteers by using Non-Parametric Adaptive Grid Algorithm

To estimate population pharmacokinetic parameters of repaglinide in 121 healthy Malaysian volunteers. Each subject received 4 mg of oral repaglinide. Six blood samples were taken per individual (0, 30, 60, 120, 180 and 240 min) for repaglinide's serum concentration determination by using high-performance liquid chromatography. The parametric Iterative Two-Stage Bayesian Population Model (it2b) program followed by the Non-Parametric Adaptive Grid (npag) program was used to determine a population pharmacokinetic modelling of repaglinide. Using the npag program, the mean elimination rate constant (k(el)) of repaglinide was 0.58 +/- 0.27 h and the volume of distribution (V(d)) was 23.09 +/- 9.19 L/h. In this first report, specifically on the population pharmacokinetic modelling of repaglinide, the data generated should help us to better understand appropriate dosage-regimens for the drug.

A Population Pharmacokinetic Model of Gabapentin Developed in Nonparametric Adaptive Grid and Nonlinear Mixed Effects Modeling

Gabapentin is used in analgesic treatment of neuropathic pain, and large interindividual variation has been observed in the pharmacokinetics (PK) of the drug. The aim of this study was to develop a population PK model for gabapentin appropriate for monitoring patients with neuropathic pain and for individualizing their dose regimens. Steady-state serum concentrations of gabapentin, distributed over a dosage interval, were obtained from 16 adult patients. Data were analyzed with an iterative 2-stage Bayesian and a nonparametric adaptive grid algorithm (NPAG) (USC*PACK) and with nonlinear mixed effects modeling (NONMEM). Compartmental population models for gabapentin PK were developed in NPAG and NONMEM using creatinine clearance and body weight as covariates. Bioavailability was included in the models as a function of dose by using a hyperbolic function derived from data previously reported in the literature. The mean population parameter estimates from the final NPAG model predicted individual serum concentrations reasonably well. The models developed in NONMEM provided additional information about the relevance of the various possible covariates and also allowed for further evaluation by simulation from the model. The population PK model may be utilized in the MM-USCPACK monitoring software (MM: multiple model dosage design) for predicting and achieving individually optimized steady-state serum concentrations of gabapentin.

Pharmacokinetics of Aztreonam in Healthy Subjects and Patients with Cystic Fibrosis and Evaluation of Dose-Exposure Relationships Using Monte Carlo Simulation

Aztreonam (AZM) is a monobactam antibiotic with a high level of activity against gram-negative micro-organisms, including Pseudomonas aeruginosa. We evaluated AZM pharmacokinetics and pharmacokinetic-pharmacodynamic relationships in patients with cystic fibrosis (CF) and healthy subjects. Pharmacokinetic data in eight CF patients and healthy subjects that were matched for age, gender, weight, and height were obtained and analyzed by using the nonparametric adaptive grid algorithm. Probabilities of target attainment using percentages of time of unbound concentration above the MIC (fT>MIC) were obtained by using a Monte Carlo simulation. AZM total body clearance was significantly higher in CF patients (100.1 +/- 17.1 versus 76.2 +/- 7.4 ml/min in healthy subjects; P < 0.01). The pharmacokinetic parameter estimates for terminal half-life (1.54 +/- 0.17 h [mean +/- the standard deviation]) and volume of distribution (0.20 +/- 0.02 liters/kg in patients with CF patients were not different from those in healthy subjects. Monte Carlo simulations with a target of a fT>MIC of 50 to 60% at a dose of 1,000 mg every 8 h indicated a clinical breakpoint of 4 mg/liter and 1 to 2 mg/liter for healthy subjects and CF patients, respectively. This study using matched controls showed that AZM total body clearance and not the volume of distribution is higher in CF patients as a result of increased renal clearance. Pharmacokinetic parameter estimates in healthy subjects resulted in a clinical susceptibility breakpoint of < or =4 mg/liter for a dose of 1,000 mg every 8 h. Patients suspected of having high clearance rates, such as CF patients, should be monitored closely, with dosing regimens adjusted accordingly.

Development of a population pharmacokinetic model for carbamazepine based on sparse therapeutic monitoring data from pediatric patients with epilepsy

Population models can be important extensions of therapeutic drug monitoring (TDM), as they allow estimation of individual pharmacokinetic parameters based on a small number of measured drug concentrations. This study used a Bayesian approach to explore the utility of routinely collected and sparse TDM data (1 sample per patient) for carbamazepine (CBZ) monotherapy in developing a population pharmacokinetic (PPK) model for CBZ in pediatric patients that would allow prediction of CBZ concentrations for both immediate- and controlled-release formulations. Patient and TDM data were obtained from a pediatric neurology outpatient database. Data were analyzed using an iterative 2-stage Bayesian algorithm and a nonparametric adaptive grid algorithm. Models were compared by final log likelihood, mean error (ME) as a measure of bias, and root mean squared error (RMSE) as a measure of precision. Fifty-seven entries with data on CBZ monotherapy were identified from the database and used in the analysis (36 from males, 21 from females; mean [SD] age, 9.1 [4.4] years [range, 2-21 years]). Preliminary models estimating clearance (Cl) or the elimination rate constant (K(el)) gave good prediction of serum concentrations compared with measured serum concentrations, but estimates of Cl and K(el) were highly correlated with estimates of volume of distribution (V(d)). Different covariate models were then tested. The selected model had zero-order input and had age and body weight as covariates. Cl (L/h) was calculated as K(el) . V(d), where K(el) = [K(i) - (K(s) . age)] and V(d) = [V(i) + (V(s) . body weight)]. Median parameter estimates were V(i) (intercept) = 11.5 L (fixed); V(s) (slope) = 0.3957 L/kg (range, 0.01200-1.5730); K(i) (intercept) = 0.173 h(-1) (fixed); and K(s) (slope) = 0.004487 h(-1) . y(-1) (range, 0.0001800-0.02969). The fit was good for estimates of steady-state serum concentrations based on prior values (population median estimates) (R = 0.468; R(2) = 0.219) but was even better for predictions based on individual Bayesian posterior values (R(2) = 0.991), with little bias (ME = -0.079) and good precision (RMSE = 0.055). Based on the findings of this study, sparse TDM data can be used for PPK modeling of CBZ clearance in children with epilepsy, and these models can be used to predict Cl at steady state in pediatric patients. However, to estimate additional pharmacokinetic model parameters (eg, the absorption rate constant and V(d)), it would be necessary to combine sparse TDM data with additional well-timed samples. This would allow development of more informative PPK models that could be used as part of Bayesian dose-individualization strategies.