Applying Neural ODEs to Derive a Mechanism-Based Model for Characterizing Maturation-Related Serum Creatinine Dynamics in Preterm Newborns.

Serum creatinine in neonates follows complex dynamics due to maturation processes, most pronounced in the first few weeks of life. The development of a mechanism-based model describing complex dynamics requires high expertise in pharmacometric (PMX) modeling and substantial model development time. A recently published machine learning (ML) approach of low-dimensional neural ordinary differential equations (NODEs) is capable of modeling such data from newborns automatically. However, this efficient data-driven approach in itself does not result in a clinically interpretable model. In this work, an approach to deriving an interpretable model with reasonable PMX-type functions is presented. This "translation" was applied to derive a PMX model for serum creatinine in neonates considering maturation processes and covariates. The developed model was compared to a previously published mechanism-based PMX model whereas both models had similar mechanistic structures. The developed model was then utilized to simulate serum creatinine concentrations in the first few weeks of life considering different covariate values for gestational age and birth weight. The reference serum creatinine values derived from these simulations are consistent with observed serum creatinine values and previously published reference values. Thus, the presented NODE-based ML approach to model complex serum creatinine dynamics in newborns and derive interpretable, mathematical-statistical components similar to those in a conventional PMX model demonstrates a novel, viable approach to facilitate the modeling of complex dynamics in clinical settings and pediatric drug development.

Pharmacometric in silico studies used to facilitate a national dose standardisation process in neonatology - application to amikacin.

BACKGROUND AND AIMS: Pharmacometric in silico approaches are frequently applied to guide decisions concerning dosage regimes during the development of new medicines. We aimed to demonstrate how such pharmacometric modelling and simulation can provide a scientific rationale for optimising drug doses in the context of the Swiss national dose standardisation project in paediatrics using amikacin as a case study. METHODS: Amikacin neonatal dosage is stratified by post-menstrual age (PMA) and post-natal age (PNA) in Switzerland and many other countries. Clinical concerns have been raised for the subpopulation of neonates with a post-menstrual age of 30-35 weeks and a post-natal age of 0-14 days ("subpopulation of clinical concern"), as potentially oto-/nephrotoxic trough concentrations (Ctrough >5 mg/l) were observed with a once-daily dose of 15 mg/kg. We applied a two-compartmental population pharmacokinetic model (amikacin clearance depending on birth weight and post-natal age) to real-world demographic data from 1563 neonates receiving anti-infectives (median birth weight 2.3 kg, median post-natal age six days) and performed pharmacometric dose-exposure simulations to identify extended dosing intervals that would ensure non-toxic Ctrough (Ctrough <5 mg/l) dosages in most neonates. RESULTS: In the subpopulation of clinical concern, Ctrough <5 mg/l was predicted in 59% versus 79-99% of cases in all other subpopulations following the current recommendations. Elevated Ctrough values were associated with a post-natal age of less than seven days. Simulations showed that extending the dosing interval to ≥36 h in the subpopulation of clinical concern increased the frequency of a desirable Ctrough below 5 mg/l to >80%. CONCLUSION: Pharmacometric in silico studies using high-quality real-world demographic data can provide a scientific rationale for national paediatric dose optimisation. This may increase clinical acceptance of fine-tuned standardised dosing recommendations and support their implementation, including in vulnerable subpopulations.

Low-dimensional neural ODEs and their application in pharmacokinetics.

Machine Learning (ML) is a fast-evolving field, integrated in many of today's scientific disciplines. With the recent development of neural ordinary differential equations (NODEs), ML provides a new tool to model dynamical systems in the field of pharmacology and pharmacometrics, such as pharmacokinetics (PK) or pharmacodynamics. The novel and conceptionally different approach of NODEs compared to classical PK modeling creates challenges but also provides opportunities for its application. In this manuscript, we introduce the functionality of NODEs and develop specific low-dimensional NODE structures based on PK principles. We discuss two challenges of NODEs, overfitting and extrapolation to unseen data, and provide practical solutions to these problems. We illustrate concept and application of our proposed low-dimensional NODE approach with several PK modeling examples, including multi-compartmental, target-mediated drug disposition, and delayed absorption behavior. In all investigated scenarios, the NODEs were able to describe the data well and simulate data for new subjects within the observed dosing range. Finally, we briefly demonstrate how NODEs can be combined with mechanistic models. This research work enhances understanding of how NODEs can be applied in PK analyses and illustrates the potential for NODEs in the field of pharmacology and pharmacometrics.

Evaluation of machine learning methods for covariate data imputation in pharmacometrics.

Missing data create challenges in clinical research because they lead to loss of statistical power and potentially to biased results. Missing covariate data must be handled with suitable approaches to prepare datasets for pharmacometric analyses, such as population pharmacokinetic and pharmacodynamic analyses. To this end, various statistical methods have been widely adopted. Here, we introduce two machine-learning (ML) methods capable of imputing missing covariate data in a pharmacometric setting. Based on a previously published pharmacometric analysis, we simulated multiple missing data scenarios. We compared the performance of four established statistical methods, listwise deletion, mean imputation, standard multiple imputation (hereafter "Norm"), and predictive mean matching (PMM) and two ML based methods, random forest (RF) and artificial neural networks (ANNs), to handle missing covariate data in a statistically plausible manner. The investigated ML-based methods can be used to impute missing covariate data in a pharmacometric setting. Both traditional imputation approaches and ML-based methods perform well in the scenarios studied, with some restrictions for individual methods. The three methods exhibiting the best performance in terms of least bias for the investigated scenarios are the statistical method PMM and the two ML-based methods RF and ANN. ML-based approaches had comparable good results to the best performing established method PMM. Furthermore, ML methods provide added flexibility when encountering more complex nonlinear relationships, especially when associated parameters are suitably tuned to enhance predictive performance.

Model enhanced reinforcement learning to enable precision dosing: A theoretical case study with dosing of propofol.

Extending the potential of precision dosing requires evaluating methodologies offering more flexibility and higher degree of personalization. Reinforcement learning (RL) holds promise in its ability to integrate multidimensional data in an adaptive process built toward efficient decision making centered on sustainable value creation. For general anesthesia in intensive care units, RL is applied and automatically adjusts dosing through monitoring of patient's consciousness. We further explore the problem of optimal control of anesthesia with propofol by combining RL with state-of-the-art tools used to inform dosing in drug development. In particular, we used pharmacokinetic-pharmacodynamic (PK-PD) modeling as a simulation engine to generate experience from dosing scenarios, which cannot be tested experimentally. Through simulations, we show that, when learning from retrospective trial data, more than 100 patients are needed to reach an accuracy within the range of what is achieved with a standard dosing solution. However, embedding a model of drug effect within the RL algorithm improves accuracy by reducing errors to target by 90% through learning to take dosing actions maximizing long-term benefit. Data residual variability impacts accuracy while the algorithm efficiently coped with up to 50% interindividual variability in the PK and 25% in the PD model's parameters. We illustrate how extending the state definition of the RL agent with meaningful variables is key to achieve high accuracy of optimal dosing policy. These results suggest that RL constitutes an attractive approach for precision dosing when rich data are available or when complemented with synthetic data from model-based tools used in model-informed drug development.

Introduction of an artificial neural network-based method for concentration-time predictions.

Pharmacometrics and the application of population pharmacokinetic (PK) modeling play a crucial role in clinical pharmacology. These methods, which describe data with well-defined equations and estimate physiologically interpretable parameters, have not changed substantially during the past decades. Although the methods have proven their usefulness, they are often resource intensive and require a high level of expertise. We investigated whether a method based on artificial neural networks (ANNs) may provide an alternative approach for the prediction of concentration-time curve to supplement the gold standard methods. In this work, we used simulated data to overcome the requirement for a large clinical training data set, implemented a pharmacologically reasonable network architecture to improve extrapolation to different dosing schemes, and used transfer learning to quickly adapt the predictions to new patient groups. We demonstrate that ANNs are able to learn the shape of concentration-time curves and make individual predictions based on a short sequence of PK measurements. Furthermore, an ANN trained on simulated data was applied to real clinical data and was demonstrated to extrapolate to different dosing schemes. We also adapted the ANN trained on simulated healthy subjects to simulated hepatic impaired patients through transfer learning. In summary, we demonstrate how ANNs could be leveraged in a PK workflow to efficiently make individual concentration-time predictions, and we discuss the current limitations and advantages of such an ANN-based method.

National Poison Center Calls Before vs After Availability of High-Dose Acetaminophen (Paracetamol) Tablets in Switzerland.

Importance: Acetaminophen (paracetamol) is among the most widely used pain medications worldwide; while safe within the therapeutic range, intake exceeding 4000 mg/d can lead to hepatotoxicity. Prior evidence suggests that limiting the availability of large quantities of acetaminophen is associated with decreased acetaminophen-related poisonings and mortality; in Switzerland, 500-mg tablets are available over-the-counter (OTC) and, as of October 2003, 1000-mg tablets are available with prescription. Objective: To evaluate the association of adding 1000-mg acetaminophen tablets to the Swiss market with utilization and poisonings. Design, Setting, and Participants: This cross-sectional study used a quasi-experimental interrupted time series analysis to evaluate 15 790 acetaminophen poison records from January 1, 2000, to December 31, 2018. All calls for acetaminophen-related poisonings identified from the National Swiss Poisons Centre and all sales for oral acetaminophen tablets (prescription and OTC) dispensed between January 2000 and December 2018 were included. Exposure: October 3, 2003 (Q4 2003), was defined as the intervention date, corresponding to the date of market entry for 1000-mg acetaminophen tablets in Switzerland. Main Outcomes and Measures: The primary outcome was the number of quarterly acetaminophen-related poison calls to the National Poison Centre. Additional outcomes included quarterly sales for acetaminophen and change in poisoning circumstances, stratified by preintervention and postintervention periods and by formulation (ie, 500-mg and 1000-mg tablets). Results: Between 2000 and 2018, 15 790 acetaminophen-related poisoning calls were identified, of which 10 628 (67.3%) were regarding women, and the mean (SD) age of patients was 25.2 (18.2) years. The interrupted time series analysis identified a significant increase in the slope for the number of reported poisonings following the intervention point, particularly for accidental circumstances (z score, -3.62; P < .001). In the preintervention period, 120 of 961 poisonings (15.3%) involved a dose greater than 10 000 mg, while for the postintervention period, 1140 of 5696 (30.6%) had a dose larger than 10 000 mg (P < .001). There was a rapid uptake in 1000-mg acetaminophen sales, while sales of the 500-mg tablet decreased slightly. Since 2012, a mean (SD) of 20.7 million (1.4 million) 1000-mg tablets were dispensed quarterly compared with 2.7 million (0.5 million) 500-mg tablets. Conclusions and Relevance: This study found a significant increase in acetaminophen dispensing and acetaminophen-related poisonings in Switzerland following the approval of 1000-mg tablets in 2003. The availability of 1000-mg acetaminophen should be re-evaluated to minimize the potential for accidental poisonings.