A Hybrid Model Associating Population Pharmacokinetics with Machine Learning: A Case Study with Iohexol Clearance Estimation.

BACKGROUND: Maximum a posteriori Bayesian estimation (MAP-BE) based on a limited sampling strategy and a population pharmacokinetic model is frequently used to estimate pharmacokinetic parameters in individuals, however with some uncertainty (bias). Recent works have shown that the performance in individual estimation or pharmacokinetic parameters can be improved by combining population pharmacokinetic and machine learning algorithms. OBJECTIVE: The objective of this work was to investigate the use of a hybrid machine learning/population pharmacokinetic approach to improve individual iohexol clearance estimation. METHODS: The reference iohexol clearance values were derived from 500 simulated profiles (samples collected between 0.1 and 24.7 h) using a population pharmacokinetic model we recently developed in Monolix and obtained using all the concentration timepoints available. Xgboost and glmnet algorithms able to predict the error of MAP-BE clearance estimates based on a limited sampling strategy (0.1 h, 1 h, and 9 h) versus reference values were developed in a training subset (75%) and were evaluated in a testing subset (25%) and in 36 real patients. RESULTS: The MAP-BE limited sampling strategy estimated clearance was corrected by the machine learning predicted error leading to a decrease in root mean squared error by 29% and 24%, and in the percentage of profiles with the mean prediction error out of the ± 20% bias by 60% and 40% in the external validation dataset for the glmnet and Xgboost machine learning algorithms, respectively. These results were attributable to a decrease in the eta-shrinkage (shrinkage for a MAP-BE limited sampling strategy = 32.4%, glmnet = 18.2%, and Xgboost = 19.4% in the external dataset). CONCLUSIONS: In conclusion, this hybrid algorithm represents a significant improvement in comparison to MAP-BE estimation alone.

Comparison of Tacrolimus Starting Doses Based on CYP3A5 Phenotype or Genotype in Kidney Transplant Recipients.

BACKGROUND: Selection of expected phenotypes (ie, expressers/non-expressers) is currently used in CYP3A5*3 genotype-based tacrolimus dosing. The authors assessed whether a dosing regimen based on the 3 CYP3A5 genotypes may reduce the occurrence of inadequate exposure. METHODS: Tacrolimus whole blood trough levels (C0) were retrieved from a retrospective cohort of 100 kidney transplant recipients treated with a starting dose of 0.15 (non-expressers) or 0.30 (expressers) mg/kg/d. The authors evaluated the occurrence of overexposures (12 < C0 < 20 ng/mL) or toxic concentrations (C0 ≥ 20 ng/mL). These results were used to set up a new strategy based on the 3 distinct CYP3A5 genotypes, which relevance was evaluated in a prospective cohort of 107 patients. RESULTS: In the retrospective cohort, non-expressers exhibited frequent overexposure (63.6%) or toxic C0 (20.8%). Among expressers, none of the homozygous *1 carriers exhibited overexposure contrary to 25% of the heterozygotes. Based on these results, new tacrolimus starting doses were set at 0.10, 0.20, and 0.30 mg/kg/d for CYP3A5*3/*3, CYP3A5*1/*3, and CYP3A5*1/*1 genotypes, respectively. Tacrolimus overexposure was reduced in the CYP3A5*3/*3 group (63.6% vs 40%, P = .0038). None of the heterozygous patients exhibited toxic tacrolimus C0. Clinical outcomes were not different between the 2 periods, whatever the genotype. Our results indicate that the best tacrolimus exposure was obtained for doses of 0.10, 0.20, and 0.20 mg/kg/d for CYP3A5*3/3, CYP3A5*1/*3, and CYP3A5*1/*1, respectively. CONCLUSIONS: Our results confirm that selecting tacrolimus dosing regimen according to the expected phenotype is appropriate, but that lower than currently recommended doses may be preferable.

Pharmacokinetic Studies in Neonates: The Utility of an Opportunistic Sampling Design.

BACKGROUND AND OBJECTIVE: The use of an opportunistic (also called scavenged) sampling strategy in a prospective pharmacokinetic study combined with population pharmacokinetic modelling has been proposed as an alternative strategy to conventional methods for accomplishing pharmacokinetic studies in neonates. However, the reliability of this approach in this particular paediatric population has not been evaluated. The objective of the present study was to evaluate the performance of an opportunistic sampling strategy for a population pharmacokinetic estimation, as well as dose prediction, and compare this strategy with a predetermined pharmacokinetic sampling approach. METHODS: Three population pharmacokinetic models were derived for ciprofloxacin from opportunistic blood samples (SC model), predetermined (i.e. scheduled) samples (TR model) and all samples (full model used to previously characterize ciprofloxacin pharmacokinetics), using NONMEM software. The predictive performance of developed models was evaluated in an independent group of patients. RESULTS: Pharmacokinetic data from 60 newborns were obtained with a total of 430 samples available for analysis; 265 collected at predetermined times and 165 that were scavenged from those obtained as part of clinical care. All datasets were fit using a two-compartment model with first-order elimination. The SC model could identify the most significant covariates and provided reasonable estimates of population pharmacokinetic parameters (clearance and steady-state volume of distribution) compared with the TR and full models. Their predictive performances were further confirmed in an external validation by Bayesian estimation, and showed similar results. Monte Carlo simulation based on area under the concentration-time curve from zero to 24 h (AUC24)/minimum inhibitory concentration (MIC) using either the SC or the TR model gave similar dose prediction for ciprofloxacin. CONCLUSION: Blood samples scavenged in the course of caring for neonates can be used to estimate ciprofloxacin pharmacokinetic parameters and therapeutic dose requirements.