Estimation of drug exposure by machine learning based on simulations from published pharmacokinetic models: The example of tacrolimus

Evaluation and application of population pharmacokinetic models for optimising linezolid treatment in non-adherence multidrug-resistant tuberculosis patients

BACKGROUND

Population pharmacokinetic (popPK) models can optimise linezolid dosage regimens in patients with multidrug-resistant tuberculosis (MDR-TB); however, unknown cross-centre precision and poor adherence remain problematic. This study aimed to assess the predictive ability of published models and use the most suitable model to optimise dosage regimens and manage compliance.

METHODS

One hundred fifty-eight linezolid plasma concentrations from 27 patients with MDR-TB were used to assess the predictive performance of published models. Prediction-based metrics and simulation-based visual predictive checks were conducted to evaluate predictive ability. Individualised remedial dosing regimens for various delayed scenarios were optimised using the most suitable model and Monte Carlo simulations. The influence of covariates, scheduled dosing intervals, and patient compliance were assessed.

RESULTS

Seven popPK models were identified. Body weight and creatinine clearance were the most frequently identified covariates influencing linezolid clearance. The model with the best performance had a median prediction error (PE%) of -1.62 %, median absolute PE of 29.50 %, and percentages of PE within 20 % (F20, 36.97 %) and 30 % (F30, 51.26 %). Monte Carlo simulations indicated that a twice-daily 300 mg linezolid dose may be more efficient than 600 mg once daily. For the 'typical' patient treated with 300 mg twice daily, half the dosage should be taken after a delay of ≥ 3 h.

CONCLUSIONS

Monte Carlo simulations based on popPK models can propose remedial regimens for delayed doses of linezolid in patients with MDR-TB. Model-based compliance management patterns are useful for balancing efficacy, adverse reactions, and resistance suppression.

Application of machine-learning models to predict the ganciclovir and valganciclovir exposure in children using a limited sampling strategy

Intravenous ganciclovir and oral valganciclovir display significant variability in ganciclovir pharmacokinetics, particularly in children. Therapeutic drug monitoring currently relies on the area under the concentration-time (AUC). Machine-learning (ML) algorithms represent an interesting alternative to Maximum-a-Posteriori Bayesian-estimators for AUC estimation. The goal of our study was to develop and validate an ML-based limited sampling strategy (LSS) approach to determine ganciclovir AUC0-24 after administration of either intravenous ganciclovir or oral valganciclovir in children. Pharmacokinetic parameters from four published population pharmacokinetic models, in addition to the World Health Organization growth curve for children, were used in the mrgsolve R package to simulate 10,800 pharmacokinetic profiles of children. Different ML algorithms were trained to predict AUC0-24 based on different combinations of two or three samples. Performances were evaluated in a simulated test set and in an external data set of real patients. The best estimation performances in the test set were obtained with the Xgboost algorithm using a 2 and 6 hours post dose LSS for oral valganciclovir (relative mean prediction error [rMPE] = 0.4% and relative root mean square error [rRMSE] = 5.7%) and 0 and 2 hours post dose LSS for intravenous ganciclovir (rMPE = 0.9% and rRMSE = 12.4%). In the external data set, the performance based on these two sample LSS was acceptable: rMPE = 0.2% and rRMSE = 16.5% for valganciclovir and rMPE = -9.7% and rRMSE = 17.2% for intravenous ganciclovir. The Xgboost algorithm developed resulted in a clinically relevant individual estimation using only two blood samples. This will improve the implementation of AUC-targeted ganciclovir therapeutic drug monitoring in children.

Getting Tacrolimus Dosing Right

ABSTRACT

Tacrolimus (TAC) dosing is typically guided by the trough concentration (C0). Yet, significant relationships between TAC C0 and clinical outcomes have seldom been reported or only with adverse events. Large retrospective studies found a moderate correlation between TAC C0 and the area under the curve (AUC), where, for any given C0 value, the AUC varied 3- to 4-fold between patients (and vice versa). However, no randomized controlled trial evaluating the dose adjustment based on TAC AUC has been conducted yet. A few observational studies have shown that the AUC is associated with efficacy and, to a lesser extent, adverse effects. Other studies showed the feasibility of reaching predefined target ranges and reducing underexposure and overexposure. TAC AUC0-12 h is now most often assessed using Bayesian estimation, but machine learning is a promising approach. Microsampling devices are well accepted by patients and represent a valuable alternative to venous blood sample collection during hospital visits, especially when a limited sampling strategy is required. As AUC monitoring cannot be proposed very frequently, C0 monitoring has to be used in the interim, which has led to fluctuating doses in patients with an AUC/C0 ratio far from the population mean, because of different dose recommendations between the 2 biomarkers. We proposed estimating the individual AUC/C0 ratio and derived individual C0 targets to be used in between or as a replacement for AUC monitoring. Existing technology and evidence are now sufficient to propose AUC monitoring interspersed with individualized-C0 monitoring for all patients with kidney transplants while collecting real-world data to strengthen the evidence.

Prediction of teicoplanin plasma concentration in critically ill patients: a combination of machine learning and population pharmacokinetics

BACKGROUND

Teicoplanin has been widely used in patients with infections caused by Staphylococcus aureus, especially for critically ill patients. The pharmacokinetics (PK) of teicoplanin vary between individuals and within the same individual. We aim to establish a prediction model via a combination of machine learning and population PK (PPK) to support personalized medication decisions for critically ill patients.

METHODS

A retrospective study was performed incorporating 33 variables, including PPK parameters (clearance and volume of distribution). Multiple algorithms and Shapley additive explanations were employed for feature selection of variables to determine the strongest driving factors.

RESULTS

The performance of each algorithm with PPK parameters was superior to that without PPK parameters. The composition of support vector regression, categorical boosting and a backpropagation neural network (7:2:1) with the highest R2 (0.809) was determined as the final ensemble model. The model included 15 variables after feature selection, of which the predictive performance was superior to that of models considering all variables or using only PPK. The R2, mean absolute error, mean squared error, absolute accuracy (±5 mg/L) and relative accuracy (±30%) of external validation were 0.649, 3.913, 28.347, 76.12% and 76.12%, respectively.

CONCLUSIONS

Our study offers a non-invasive, fast and cost-effective prediction model of teicoplanin plasma concentration in critically ill patients. The model serves as a fundamental tool for clinicians to determine the effective plasma concentration range of teicoplanin and formulate individualized dosing regimens accordingly.

A Machine Learning Algorithm to Predict the Starting Dose of Daptomycin

BACKGROUND AND OBJECTIVE

The dosage of daptomycin is usually based on body weight. However, it has been shown that this approach yields too high an exposure in obese patients. Pharmacokinetic and pharmacodynamic indexes (PK/PD) have been proposed for daptomycin's antibacterial effect (AUC/CMI >666) and toxicity (C0 > 24.3 mg/L). We previously developed machine learning (ML) algorithms to predict starting doses based on Monte Carlo simulations. We propose a new way to perform probability of target attainment based on an ML algorithm to predict the daptomycin starting dose.

METHODS

The Dvorchik model of daptomycin was implemented in the mrgsolve R package and 4950 pharmacokinetic profiles were simulated with doses ranging from 4 to 12 mg/kg. We trained and benchmarked four machine learning algorithms and selected the best to iteratively search for the optimal dose of daptomycin maximizing the event (AUC/CMI > 666 and C0 < 24.3 mg/L). The ML algorithm was evaluated in simulations and an external database of real patients in comparison with population pharmacokinetics.

RESULTS

The performance of the Xgboost algorithms developed to predict the event (ROC AUC) in the training and test set were 0.762 and 0.761, respectively. The most important prediction variables were dose, creatinine clearance, body weight and sex. In the external database of real patients, the starting dose administered based on the ML algorithm significantly improved the target attainment by 7.9% (p-value = 0.02929) in comparison with the dose administered based on body weight.

CONCLUSION

The developed algorithm improved the target attainment for daptomycin in comparison with weight-based dosing. We built a Shiny app to calculate the optimal starting dose.

Estimation of linezolid exposure in patients with hepatic impairment using machine learning based on a population pharmacokinetic model

PURPOSE

To investigate the pharmacokinetic changes of linezolid in patients with hepatic impairment and to explore a method to predict linezolid exposure.

METHODS

Patients with hepatic impairment who received linezolid were recruited. A population pharmacokinetic model (PPK) was then built using NONMEM software. And based on the final model, virtual patients with rich concentration values was constructed through Monte Carlo simulations (MCS), which were used to build machine learning (ML) models to predict linezolid exposure levels. Finally, we investigated the risk factors for thrombocytopenia in patients included.

RESULTS

A PPK model with population typical values of 3.83 L/h and 34.1 L for clearance and volume of distribution was established, and the severe hepatic impairment was identified as a significant covariate of clearance. Then, we built a series of ML models to predict the area under 0 -24 h concentration-time curve (AUC0-24) of linezolid based on virtual patients from MCS. The results showed that the Xgboost models showed the best predictive performance and were superior to the methods for estimating linezolid AUC0-24 based on though concentration or daily dose. Finally, we found that baseline platelet count, linezolid AUC0-24, and combination with fluoroquinolones were independent risk factors for thrombocytopenia, and based on this, we proposed a method for calculating the toxicity threshold of linezolid.

CONCLUSION

In this study, we successfully constructed a PPK model for patients with hepatic impairment and used ML algorithm to estimate linezolid AUC0-24 based on limited data. Finally, we provided a method to determine the toxicity threshold of linezolid.

Machine Learning Approach in Dosage Individualization of Isoniazid for Tuberculosis

INTRODUCTION

Isoniazid is a first-line antituberculosis agent with high variability, which would profit from individualized dosing. Concentrations of isoniazid at 2 h (C2h), as an indicator of safety and efficacy, are important for optimizing therapy.

OBJECTIVE

The objective of this study was to establish machine learning (ML) models to predict the C2h, that can be used for establishing an individualized dosing regimen in clinical practice.

METHODS

Published population pharmacokinetic (PopPK) models for adults were searched based on PubMed and ultimately four reliable models were selected for simulating individual C2h datasets under different conditions (demographics, genotype, ethnicity, etc.). Machine learning models were trained on simulated C2h obtained from the four PopPK models. Five different algorithms were used for ML model building to predict C2h. Real-world data were used for predictive performance evaluations. Virtual trials were used to compare ML-optimized doses with PopPK model-optimized doses.

RESULTS

Categorical boosting (CatBoost) exhibited the highest prediction ability. Target C2h can be predicted using the ML model combined with the dosing regimen and three covariates (N-acetyltransferase 2 [NAT2] genotypes, weight and race [Asians and Africans]). Real-world data validation results showed that the ML model can achieve an overall prediction accuracy of 93.4%. Using the final ML model, the mean absolute prediction error value decreased by 45.7% relative to the average of PopPK models. Using the ML-optimized dosing regimen, the probability of target attainment increased by 43.7% relative to the PopPK model-optimized dosing regimens.

CONCLUSION

Machine learning models were developed with great predictive performance, which can be used to determine the individualized initial dose of isoniazid in adult patients.

Improving prediction of tacrolimus concentration using a combination of population pharmacokinetic modeling and machine learning in chinese renal transplant recipients

Aims

The population pharmacokinetic (PPK) model-based machine learning (ML) approach offers a novel perspective on individual concentration prediction. This study aimed to establish a PPK-based ML model for predicting tacrolimus (TAC) concentrations in Chinese renal transplant recipients.

Methods

Conventional TAC monitoring data from 127 Chinese renal transplant patients were divided into training (80%) and testing (20%) datasets. A PPK model was developed using the training group data. ML models were then established based on individual pharmacokinetic data derived from the PPK basic model. The prediction performances of the PPK-based ML model and Bayesian forecasting approach were compared using data from the test group.

Results

The final PPK model, incorporating hematocrit and CYP3A5 genotypes as covariates, was successfully established. Individual predictions of TAC using the PPK basic model, postoperative date, CYP3A5 genotype, and hematocrit showed improved rankings in ML model construction. XGBoost, based on the TAC PPK, exhibited the best prediction performance.

Conclusion

The PPK-based machine learning approach emerges as a superior option for predicting TAC concentrations in Chinese renal transplant recipients.

A machine learning approach to predict daptomycin exposure from two concentrations based on Monte Carlo simulations

Daptomycin is a concentration-dependent lipopeptide antibiotic for which exposure/effect relationships have been shown. Machine learning (ML) algorithms, developed to predict the individual exposure to drugs, have shown very good performances in comparison to maximum a posteriori Bayesian estimation (MAP-BE). The aim of this work was to predict the area under the blood concentration curve (AUC) of daptomycin from two samples and a few covariates using XGBoost ML algorithm trained on Monte Carlo simulations. Five thousand one hundred fifty patients were simulated from two literature population pharmacokinetics models. Data from the first model were split into a training set (75%) and a testing set (25%). Four ML algorithms were built to learn AUC based on daptomycin blood concentration samples at pre-dose and 1 h post-dose. The XGBoost model (best ML algorithm) with the lowest root mean square error (RMSE) in a 10-fold cross-validation experiment was evaluated in both the test set and the simulations from the second population pharmacokinetic model (validation). The ML model based on the two concentrations, the differences between these concentrations, and five other covariates (sex, weight, daptomycin dose, creatinine clearance, and body temperature) yielded very good AUC estimation in the test (relative bias/RMSE = 0.43/7.69%) and validation sets (relative bias/RMSE = 4.61/6.63%). The XGBoost ML model developed allowed accurate estimation of daptomycin AUC using C0, C1h, and a few covariates and could be used for exposure estimation and dose adjustment. This ML approach can facilitate the conduct of future therapeutic drug monitoring (TDM) studies.

Machine Learning: A New Approach for Dose Individualization

The application of machine learning (ML) has shown promising results in precision medicine due to its exceptional performance in dealing with complex multidimensional data. However, using ML for individualized dosing of medicines is still in its early stage, meriting further exploration. A systematic review of study designs and modeling details of using ML for individualized dosing of different drugs was performed. We have summarized the status of the study populations, predictive targets, and data sources for ML modeling, the selection of ML algorithms and features, and the evaluation and validation of their predictive performance. We also used the Prediction model Risk of Bias Assessment Tool (PROBAST) to assess the risk of bias of included studies. Currently, ML can be used for both a priori and a posteriori dose selection and optimization, and it can also assist the implementation of therapeutic drug monitoring. However, studies are mainly focused on drugs with narrow therapeutic windows, predominantly immunosuppressants (N = 23, 35.9%) and anti-infectives (N = 21, 32.8%), and there is currently only very limited attention for special populations, such as children (N = 22, 34.4%). Most studies showed poor methodological quality and a high risk of bias. The lack of external validation and clinical utility evaluation currently limits the further clinical implementation of ML for dose individualization. We therefore have proposed several ways to improve the clinical relevance of the studies and facilitate the translation of ML models into clinical practice.

Optimization of Ganciclovir and Valganciclovir Starting Dose in Children by Machine Learning

BACKGROUND AND OBJECTIVES

Ganciclovir (GCV) and valganciclovir (VGCV) show large interindividual pharmacokinetic variability, particularly in children. The objectives of this study were (1) to develop machine learning (ML) algorithms trained on simulated pharmacokinetics profiles obtained by Monte Carlo simulations to estimate the best ganciclovir or valganciclovir starting dose in children and (2) to compare its performances on real-world profiles to previously published equation derived from literature population pharmacokinetic (POPPK) models achieving about 20% of profiles within the target.

MATERIALS AND METHODS

The pharmacokinetic parameters of four literature POPPK models in addition to the World Health Organization (WHO) growth curve for children were used in the mrgsolve R package to simulate 10,800 pharmacokinetic profiles. ML algorithms were developed and benchmarked to predict the probability to reach the steady-state, area-under-the-curve target (AUC0-24 within 40-60 mg × h/L) based on demographic characteristics only. The best ML algorithm was then used to calculate the starting dose maximizing the target attainment. Performances were evaluated for ML and literature formula in a test set and in an external set of 32 and 31 actual patients (GCV and VGCV, respectively).

RESULTS

A combination of Xgboost, neural network, and random forest algorithms yielded the best performances and highest target attainment in the test set (36.8% for GCV and 35.3% for the VGCV). In actual patients, the best GCV ML starting dose yielded the highest target attainment rate (25.8%) and performed equally for VGCV with the Franck model formula (35.3% for both).

CONCLUSION

The ML algorithms exhibit good performances in comparison with previously validated models and should be evaluated prospectively.

Bridging the Worlds of Pharmacometrics and Machine Learning

Precision medicine requires individualized modeling of disease and drug dynamics, with machine learning-based computational techniques gaining increasing popularity. The complexity of either field, however, makes current pharmacological problems opaque to machine learning practitioners, and state-of-the-art machine learning methods inaccessible to pharmacometricians. To help bridge the two worlds, we provide an introduction to current problems and techniques in pharmacometrics that ranges from pharmacokinetic and pharmacodynamic modeling to pharmacometric simulations, model-informed precision dosing, and systems pharmacology, and review some of the machine learning approaches to address them. We hope this would facilitate collaboration between experts, with complementary strengths of principled pharmacometric modeling and flexibility of machine learning leading to synergistic effects in pharmacological applications.

Artificial Intelligence in Pharmaceutical Technology and Drug Delivery Design

Artificial intelligence (AI) has emerged as a powerful tool that harnesses anthropomorphic knowledge and provides expedited solutions to complex challenges. Remarkable advancements in AI technology and machine learning present a transformative opportunity in the drug discovery, formulation, and testing of pharmaceutical dosage forms. By utilizing AI algorithms that analyze extensive biological data, including genomics and proteomics, researchers can identify disease-associated targets and predict their interactions with potential drug candidates. This enables a more efficient and targeted approach to drug discovery, thereby increasing the likelihood of successful drug approvals. Furthermore, AI can contribute to reducing development costs by optimizing research and development processes. Machine learning algorithms assist in experimental design and can predict the pharmacokinetics and toxicity of drug candidates. This capability enables the prioritization and optimization of lead compounds, reducing the need for extensive and costly animal testing. Personalized medicine approaches can be facilitated through AI algorithms that analyze real-world patient data, leading to more effective treatment outcomes and improved patient adherence. This comprehensive review explores the wide-ranging applications of AI in drug discovery, drug delivery dosage form designs, process optimization, testing, and pharmacokinetics/pharmacodynamics (PK/PD) studies. This review provides an overview of various AI-based approaches utilized in pharmaceutical technology, highlighting their benefits and drawbacks. Nevertheless, the continued investment in and exploration of AI in the pharmaceutical industry offer exciting prospects for enhancing drug development processes and patient care.

A novel approach for pharmacological substantiation of safety signals using plasma concentrations of medication and administrative/healthcare databases: a case study using Danish registries for an FDA warning on lamotrigine

PHARMACOM-EPI is a novel framework to predict plasma concentrations of drugs at the time of occurrence of clinical outcomes. In early 2021, the U.S. Food and Drug Administration (FDA) issued a warning on the antiseizure drug lamotrigine claiming that it has the potential to increase the risk of arrhythmias and related sudden cardiac death due to a pharmacological sodium channel-blocking effect. We hypothesized that the risk of arrhythmias and related death is due to toxicity. We used the PHARMACOM-EPI framework to investigate the relationship between lamotrigine's plasma concentrations and the risk of death in older patients using real-world data. Danish nationwide administrative and healthcare registers were used as data sources and individuals aged 65 years or older during the period 1996 - 2018 were included in the study. According to the PHARMACOM-EPI framework, plasma concentrations of lamotrigine were predicted at the time of death and patients were categorized into non-toxic and toxic groups based on the therapeutic range of lamotrigine (3-15mg/L). Over 1 year of treatment, the incidence rate ratio (IRR) of all-cause mortality was calculated between the propensities score matched toxic and non-toxic groups. In total, 7286 individuals were diagnosed with epilepsy and were exposed to lamotrigine, 432 of which had at least one plasma concentration measurement The pharmacometric model by Chavez et al. was used to predict lamotrigine's plasma concentrations considering the lowest absolute percentage error among identified models (14.25%, 95% CI: 11.68-16.23). The majority of lamotrigine associated deaths were cardiovascular-related and occurred among individuals with plasma concentrations in the toxic range. The IRR of mortality between the toxic group and non-toxic group was 3.37 [95% CI: 1.44-8.32] and the cumulative incidence of all-cause mortality exponentially increased in the toxic range. Application of our novel framework PHARMACOM-EPI provided strong evidence to support our hypothesis that the increased risk of all-cause and cardiovascular death was associated with a toxic plasma concentration level of lamotrigine among older lamotrigine users.

Individual dose recommendations for drug interaction between tacrolimus and voriconazole in adult liver transplant recipients: A semiphysiologically based population pharmacokinetic modeling approach

The magnitude of drug-drug interaction between tacrolimus and voriconazole is highly variable, and individually tailoring the tacrolimus dose when concomitantly administered with voriconazole remains difficult. This study aimed to develop a semiphysiologically based population pharmacokinetic (semi-PBPK) model and a web-based dashboard to identify the dynamic inhibition of tacrolimus metabolism caused by voriconazole and provide individual tacrolimus regimens for Chinese adult liver transplant recipients. A total of 264 tacrolimus concentrations and 146 voriconazole concentrations were prospectively collected from 32 transplant recipients. A semi-PBPK model with physiological compartments including the gut wall, portal vein, and liver was developed using the nonlinear mixed-effects modeling software NONMEM (version 7.4). A web-based dashboard was established in R software (version 3.6.1) to recommend the individual tacrolimus regimens when concomitantly administered with voriconazole. The reversible inhibition of tacrolimus metabolism caused by voriconazole was investigated in both the liver and the gut wall. Moreover, voriconazole could highly inhibit the CYP3A activity in the gut wall more than in the liver. BMI and postoperative days were identified as significant covariates on intrinsic intestinal and hepatic clearance of tacrolimus, respectively. Age and postoperative days were identified as significant covariates on the volume of distribution of voriconazole. The individual tacrolimus regimens when concomitantly administered with voriconazole could be recommended in the dashboard (https://tac-vor-ddi.shinyapps.io/shinyapp3/). In conclusion, the semi-PBPK model successfully described the dynamic inhibition process between tacrolimus and voriconazole, and the web-based dashboard could provide individual tacrolimus regimens when concomitantly administered with voriconazole.

Handling Delayed or Missed Dose of Antiseizure Medications: A Model-Informed Individual Remedial Dosing

BACKGROUND AND OBJECTIVES

Delayed or missed antiseizure medications (ASMs) doses are common during long-term or lifelong antiepilepsy treatment. This study aims to explore optimal individualized remedial dosing regimens for delayed or missed doses of 11 commonly used ASMs.

METHODS

To explore remedial dosing regimens, Monte Carlo simulation was used based on previously identified and published population pharmacokinetic models. Six remedial strategies for delayed or missed doses were investigated. The deviation time outside the individual therapeutic range was used to evaluate each remedial regimen. The influences of patients' demographics, concomitant medication, and scheduled dosing intervals on remedial regimens were assessed. RxODE and Shiny in R were used to perform Monte Carlo simulation and recommend individual remedial regimens.

RESULTS

The recommended remedial regimens were highly correlated with delayed time, scheduled dosing interval, and half-life of the ASM. Moreover, the optimal remedial regimens for pediatric and adult patients were different. The renal function, along with concomitant medication that affects the clearance of the ASM, may also influence the remedial regimens. A web-based dashboard was developed to provide individualized remedial regimens for the delayed or missed dose, and a user-defined module with all parameters that could be defined flexibly by the user was also built.

DISCUSSION

Monte Carlo simulation based on population pharmacokinetic models may provide a rational approach to propose remedial regimens for delayed or missed doses of ASMs in pediatric and adult patients with epilepsy.

Antimicrobial Pharmacokinetics and Pharmacodynamics in Critical Care: Adjusting the Dose in Extracorporeal Circulation and to Prevent the Genesis of Multiresistant Bacteria

Critically ill patients suffering from severe infections are prone to pathophysiological pharmacokinetic changes that are frequently associated with inadequate antibiotic serum concentrations. Minimum inhibitory concentrations (MICs) of the causative pathogens tend to be higher in intensive care units. Both pharmacokinetic changes and high antibiotic resistance likely jeopardize the efficacy of treatment. The use of extracorporeal circulation devices to support hemodynamic, respiratory, or renal failure enables pharmacokinetic changes and makes it even more difficult to achieve an adequate antibiotic dose. Besides a clinical response, antibiotic pharmacokinetic optimization is important to reduce the selection of strains resistant to common antibiotics. In this review, we summarize the present knowledge regarding pharmacokinetic changes in critically ill patients and we discuss the effects of extra-corporeal devices on antibiotic treatment together with potential solutions.

Application of machine learning to predict tacrolimus exposure in liver and kidney transplant patients given the MeltDose formulation

PURPOSE

Machine Learning (ML) algorithms represent an interesting alternative to maximum a posteriori Bayesian estimators (MAP-BE) for tacrolimus AUC estimation, but it is not known if training an ML model using a lower number of full pharmacokinetic (PK) profiles (= "true" reference AUC) provides better performances than using a larger dataset of less accurate AUC estimates. The objectives of this study were: to develop and benchmark ML algorithms trained using full PK profiles to estimate MeltDose^®-tacrolimus individual AUCs using 2 or 3 blood concentrations; and to compare their performance to MAP-BE.

METHODS

Data from liver (n = 113) and kidney (n = 97) transplant recipients involved in MeltDose-tacrolimus PK studies were used for the training and evaluation of ML algorithms. "True" AUC0-24 h was calculated for each patient using the trapezoidal rule on the full PK profile. ML algorithms were trained to estimate tacrolimus true AUC using 2 or 3 blood concentrations. Performances were evaluated in 2 external sets of 16 (renal) and 48 (liver) transplant patients.

RESULTS

Best estimation performances were obtained with the MARS algorithm and the following limited sampling strategies (LSS): predose (0), 8, and 12 h post-dose (rMPE = - 1.28%, rRMSE = 7.57%), or 0 and 12 h (rMPE = - 1.9%, rRMSE = 10.06%). In the external dataset, the performances of the final ML algorithms based on two samples in kidney (rMPE = - 3.1%, rRMSE = 11.1%) or liver transplant recipients (rMPE = - 3.4%, rRMSE = 9.86%) were as good as or better than those of MAP-BEs based on three time points.

CONCLUSION

The MARS ML models developed using "true" MeltDose^®-tacrolimus AUCs yielded accurate individual estimations using only two blood concentrations.

Tacrolimus pharmacokinetics in pediatric nephrotic syndrome: A combination of population pharmacokinetic modelling and machine learning approaches to improve individual prediction

Background and Aim: Tacrolimus (TAC) is a first-line immunosuppressant for the treatment of refractory nephrotic syndrome (RNS), but the pharmacokinetics of TAC varies widely among individuals, and there is still no accurate model to predict the pharmacokinetics of TAC in RNS. Therefore, this study aimed to combine population pharmacokinetic (PPK) model and machine learning algorithms to develop a simple and accurate prediction model for TAC.

Methods: 139 children with RNS from August 2013 to December 2018 were included, and blood samples of TAC trough and partial peak concentrations were collected. The blood concentration of TAC was determined by enzyme immunoassay; CYP3A5 was genotyped by polymerase chain reaction-restriction fragment length polymorphism method; MYH9, LAMB2, ACTN4 and other genotypes were determined by MALDI-TOF MS method; PPK model was established by nonlinear mixed-effects method. Based on this, six machine learning algorithms, including eXtreme Gradient Boosting (XGBoost), Random Forest (RF), Extra-Trees, Gradient Boosting Decision Tree (GBDT), Adaptive boosting (AdaBoost) and Lasso, were used to establish the machine learning model of TAC clearance.

Results: A one-compartment model of first-order absorption and elimination adequately described the pharmacokinetics of TAC. Age, co-administration of Wuzhi capsules, CYP3A5 *3/*3 genotype and CTLA4 rs4553808 genotype were significantly affecting the clearance of TAC. Among the six machine learning models, the Lasso algorithm model performed the best (R2 = 0.42).

Conclusion: For the first time, a clearance prediction model of TAC in pediatric patients with RNS was established using PPK combined with machine learning, by which the individual clearance of TAC can be predicted more accurately, and the initial dose of administration can be optimized to achieve the goal of individualized treatment.

Effect of Meal Timings and Meal Content on the AUC0-12h of Mycophenolic Acid: A Simulation Study

Meal timings and content related to gallbladder emptying in the enterohepatic circulation are important for explaining the high variability in mycophenolic acid exposure. The limited sampling strategy (LSS) was established to estimate the area under the plasma concentration-time curve from time 0 to 12 hours (AUC_0-12h ) of mycophenolic acid in therapeutic drug monitoring. The aim of this study is to investigate the effect of meal timings and content on the AUC_0-12h of mycophenolic acid and to assess the influence of meals on LSS. A mycophenolic acid pharmacokinetic model with a mechanism-based enterohepatic circulation process was employed to perform simulations under various assumed meal scenarios. The simulations were compared to evaluate the effect of meal timings and meal content on mycophenolic acid AUC_0-12h . Monte Carlo simulations were performed using the meal parameter with the greatest impact on mycophenolic acid AUC_0-12h as a variable. The corresponding LSS equations were established, and the predictive performance was assessed. Both the meal timings and meal content affected the mycophenolic acid AUC_0-12h , and the postdose fasting period had the greatest impact. The predictive performance of the LSS is sensitive to the postdose fasting period. Therefore, meal timings may improve the estimation of mycophenolic acid AUC_0-12h and the efficacy of therapeutic drug monitoring.

Artificial intelligence in pharmacology research and practice

In recent years, the use of artificial intelligence (AI) in health care has risen steadily, including a wide range of applications in the field of pharmacology. AI is now used throughout the entire continuum of pharmacology research and clinical practice and from early drug discovery to real-world datamining. The types of AI models used range from unsupervised clustering of drugs or patients aimed at identifying potential drug compounds or suitable patient populations, to supervised machine learning approaches to improve therapeutic drug monitoring. Additionally, natural language processing is increasingly used to mine electronic health records to obtain real-world data. In this mini-review, we discuss the basics of AI followed by an outline of its application in pharmacology research and clinical practice.

Machine learning advances the integration of covariates in population pharmacokinetic models: Valproic acid as an example

Background and Aim: Many studies associated with the combination of machine learning (ML) and pharmacometrics have appeared in recent years. ML can be used as an initial step for fast screening of covariates in population pharmacokinetic (popPK) models. The present study aimed to integrate covariates derived from different popPK models using ML.

Methods: Two published popPK models of valproic acid (VPA) in Chinese epileptic patients were used, where the population parameters were influenced by some covariates. Based on the covariates and a one-compartment model that describes the pharmacokinetics of VPA, a dataset was constructed using Monte Carlo simulation, to develop an XGBoost model to estimate the steady-state concentrations (Css) of VPA. We utilized SHapley Additive exPlanation (SHAP) values to interpret the prediction model, and calculated estimates of VPA exposure in four assumed scenarios involving different combinations of CYP2C19 genotypes and co-administered antiepileptic drugs. To develop an easy-to-use model in the clinic, we built a simplified model by using CYP2C19 genotypes and some noninvasive clinical parameters, and omitting several features that were infrequently measured or whose clinically available values were inaccurate, and verified it on our independent external dataset.

Results: After data preprocessing, the finally generated combined dataset was divided into a derivation cohort and a validation cohort (8:2). The XGBoost model was developed in the derivation cohort and yielded excellent performance in the validation cohort with a mean absolute error of 2.4 mg/L, root-mean-squared error of 3.3 mg/L, mean relative error of 0%, and percentages within ±20% of actual values of 98.85%. The SHAP analysis revealed that daily dose, time, CYP2C19*2 and/or *3 variants, albumin, body weight, single dose, and CYP2C19*1*1 genotype were the top seven confounding factors influencing the Css of VPA. Under the simulated dosage regimen of 500 mg/bid, the VPA exposure in patients who had CYP2C19*2 and/or *3 variants and no carbamazepine, phenytoin, or phenobarbital treatment, was approximately 1.74-fold compared to those with CYP2C19*1/*1 genotype and co-administered carbamazepine + phenytoin + phenobarbital. The feasibility of the simplified model was fully illustrated by its performance in our external dataset.

Conclusion: This study highlighted the bridging role of ML in big data and pharmacometrics, by integrating covariates derived from different popPK models.

An interpretable stacking ensemble learning framework based on multi-dimensional data for real-time prediction of drug concentration: The example of olanzapine

Background and Aim: Therapeutic drug monitoring (TDM) has evolved over the years as an important tool for personalized medicine. Nevertheless, some limitations are associated with traditional TDM. Emerging data-driven model forecasting [e.g., through machine learning (ML)-based approaches] has been used for individualized therapy. This study proposes an interpretable stacking-based ML framework to predict concentrations in real time after olanzapine (OLZ) treatment.

Methods: The TDM-OLZ dataset, consisting of 2,142 OLZ measurements and 472 features, was formed by collecting electronic health records during the TDM of 927 patients who had received OLZ treatment. We compared the performance of ML algorithms by using 10-fold cross-validation and the mean absolute error (MAE). The optimal subset of features was analyzed by a random forest-based sequential forward feature selection method in the context of the top five heterogeneous regressors as base models to develop a stacked ensemble regressor, which was then optimized via the grid search method. Its predictions were explained by using local interpretable model-agnostic explanations (LIME) and partial dependence plots (PDPs).

Results: A state-of-the-art stacking ensemble learning framework that integrates optimized extra trees, XGBoost, random forest, bagging, and gradient-boosting regressors was developed for nine selected features [i.e., daily dose (OLZ), gender_male, age, valproic acid_yes, ALT, K, BW, MONO#, and time of blood sampling after first administration]. It outperformed other base regressors that were considered, with an MAE of 0.064, R-square value of 0.5355, mean squared error of 0.0089, mean relative error of 13%, and ideal rate (the percentages of predicted TDM within ± 30% of actual TDM) of 63.40%. Predictions at the individual level were illustrated by LIME plots, whereas the global interpretation of associations between features and outcomes was illustrated by PDPs.

Conclusion: This study highlights the feasibility of the real-time estimation of drug concentrations by using stacking-based ML strategies without losing interpretability, thus facilitating model-informed precision dosing.

Optimization of Vancomycin Initial Dose in Term and Preterm Neonates by Machine Learning

INTRODUCTION

Vancomycin is one of the antibiotics most used in neonates. Continuous infusion has many advantages over intermittent infusions, but no consensus has been achieved regarding the optimal initial dose. The objectives of this study were: to develop a Machine learning (ML) algorithm based on pharmacokinetic profiles obtained by Monte Carlo simulations using a population pharmacokinetic model (POPPK) from the literature, in order to derive the best vancomycin initial dose in preterm and term neonates, and to compare ML performances with those of an literature equation (LE) derived from a POPPK previously published.

MATERIALS AND METHODS

The parameters of a previously published POPPK model of vancomycin in children and neonates were used in the mrgsolve R package to simulate 1900 PK profiles. ML algorithms were developed from these simulations using Xgboost, GLMNET and MARS in parallel, benchmarked and used to calculate the ML first dose. Performances were evaluated in a second simulation set and in an external set of 82 real patients and compared to those of a LE.

RESULTS

The Xgboost algorithm yielded numerically best performances and target attainment rates: 46.9% in the second simulation set of 400-600 AUC/MIC ratio vs. 41.4% for the LE model (p = 0.0018); and 35.3% vs. 28% in real patients (p = 0.401), respectively). The Xgboost model resulted in less AUC/MIC > 600, thus decreasing the risk of nephrotoxicity.

CONCLUSION

The Xgboost algorithm developed to estimate the initial dose of vancomycin in term or preterm infants has better performances than a previous validated LE and should be evaluated prospectively.

Machine learning algorithms to estimate everolimus exposure trained on simulated and patient pharmacokinetic profiles

Everolimus is an immunosuppressant with a small therapeutic index and large between‐patient variability. The area under the concentration versus time curve (AUC) is the best marker of exposure but measuring it requires collecting many blood samples. The objective of this study was to train machine learning (ML) algorithms using pharmacokinetic (PK) profiles from kidney transplant recipients, simulated profiles, or both types, and compare their performance for everolimus AUC_0‐12h estimation using a limited number of predictors, as compared to an independent set of full PK profiles from patients, as well as to the corresponding maximum a posteriori Bayesian estimates (MAP‐BE). XGBoost was first trained on 508 patient interdose AUCs estimated using MAP‐BE, and then on 500–10,000 rich interdose PK profiles simulated using previously published population PK parameters. The predictors used were: predose, ~1 h, and ~2 h whole blood concentrations, differences between these concentrations, relative deviations from theoretical sampling times, morning dose, patient age, and time elapsed since transplantation. The best results were obtained with XGBoost trained on 5016 simulated profiles. AUC estimation achieved in an external dataset of 114 full‐PK profiles was excellent (root mean squared error [RMSE] = 10.8 μg*h/L) and slightly better than MAP‐BE (RMSE = 11.9 μg*h/L). Using more profiles (n = 10,035) did not improve the ML algorithm performance. The contribution of mixing patient and simulated profiles was significant only when they were in balanced numbers, with ~500 for each (RMSE = 12.5 μg*h/L), compared with patient data alone (RMSE = 18.0 μg*h/L).

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

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

Therapeutic drug monitoring of immunosuppressive drugs in hepatology and gastroenterology

Immunosuppressive drugs have been key to the success of liver transplantation and are essential components of the treatment of inflammatory bowel disease (IBD) and autoimmune hepatitis (AIH). For many but not all immunosuppressants, therapeutic drug monitoring (TDM) is recommended to guide therapy. In this article, the rationale and evidence for TDM of tacrolimus, mycophenolic acid, the mammalian target of rapamycin inhibitors, and azathioprine in liver transplantation, IBD, and AIH is reviewed. New developments, including algorithm-based/computer-assisted immunosuppressant dosing, measurement of immunosuppressants in alternative matrices for whole blood, and pharmacodynamic monitoring of these agents is discussed. It is expected that these novel techniques will be incorporate into the standard TDM in the next few years.