Clinical decision support for chemotherapy-induced neutropenia using a hybrid pharmacodynamic/machine learning model.

Consensus guidelines recommend use of granulocyte colony stimulating factor in patients deemed at risk of chemotherapy-induced neutropenia, however, these risk models are limited in the factors they consider and miss some cases of neutropenia. Clinical decision making could be supported using models that better tailor their predictions to the individual patient using the wealth of data available in electronic health records (EHRs). Here, we present a hybrid pharmacokinetic/pharmacodynamic (PKPD)/machine learning (ML) approach that uses predictions and individual Bayesian parameter estimates from a PKPD model to enrich an ML model built on her data. We demonstrate this approach using models developed on a large real-world data set of 9121 patients treated for lymphoma, breast, or thoracic cancer. We also investigate the benefits of augmenting the training data using synthetic data simulated with the PKPD model. We find that PKPD-enrichment of ML models improves prediction of grade 3-4 neutropenia, as measured by higher precision (61%) and recall (39%) compared to PKPD model predictions (47%, 33%) or base ML model predictions (51%, 31%). PKPD augmentation of ML models showed minor improvements in recall (44%) but not precision (56%), and data augmentation required careful tuning to control overfitting its predictions to the PKPD model. PKPD enrichment of ML shows promise for leveraging both the physiology-informed predictions of PKPD and the ability of ML to learn predictor-outcome relationships from large data sets to predict patient response to drugs in a clinical precision dosing context.

Evaluation of Neonatal and Paediatric Vancomycin Pharmacokinetic Models and the Impact of Maturation and Serum Creatinine Covariates in a Large Multicentre Data Set.

BACKGROUND AND OBJECTIVE: Infants and neonates present a clinical challenge for dosing drugs with high interindividual variability due to these patients' rapid growth and the interplay between maturation and organ function. Model-informed precision dosing (MIPD), which can account for interindividual variability via patient characteristics and Bayesian forecasting, promises to improve individualized dosing strategies in this complex population. Here, we assess the predictive performance of published population pharmacokinetic models describing vancomycin in neonates and infants, and analyze the robustness of these models in the face of clinical uncertainty surrounding covariate values. METHODS: The predictive precision and bias of nine pharmacokinetic models were compared in a large multi-site data set (N = 2061 patients, 5794 drug levels, 28 institutions) of patients aged 0-365 days. The robustness of model predictions to errors in serum creatinine measurements and gestational age was assessed by using recorded values or by replacing covariate values with 0.3, 0.5 or 0.8 mg/dL or with 40 weeks, respectively. RESULTS: Of the nine models, two models (Dao and Jacqz-Aigrain) resulted in predicted concentrations within 2.5 mg/L or 15% of the measured values for at least 60% of population predictions. Within individual models, predictive performance often 2 differed in neonates (0-4 weeks) versus older infants (15-52 weeks). For preterm neonates, imputing gestational age as 40 weeks reduced the accuracy of model predictions. Measured values of serum creatinine improved model predictions compared to using imputed values even in neonates ≤1 week of age. CONCLUSIONS: Several available pharmacokinetic models are suitable for MIPD in infants and neonates. Availability and accuracy of model covariates for patients will be important for guiding dose decision-making.

Evaluating and Improving Neonatal Gentamicin Pharmacokinetic Models Using Aggregated Routine Clinical Care Data.

Model-informed precision dosing (MIPD) can aid dose decision-making for drugs such as gentamicin that have high inter-individual variability, a narrow therapeutic window, and a high risk of exposure-related adverse events. However, MIPD in neonates is challenging due to their dynamic development and maturation and by the need to minimize blood sampling due to low blood volume. Here, we investigate the ability of six published neonatal gentamicin population pharmacokinetic models to predict gentamicin concentrations in routine therapeutic drug monitoring from nine sites in the United State (n = 475 patients). We find that four out of six models predicted with acceptable levels of error and bias for clinical use. These models included known important covariates for gentamicin PK, showed little bias in prediction residuals over covariate ranges, and were developed on patient populations with similar covariate distributions as the one assessed here. These four models were refit using the published parameters as informative Bayesian priors or without priors in a continuous learning process. We find that refit models generally reduce error and bias on a held-out validation data set, but that informative prior use is not uniformly advantageous. Our work informs clinicians implementing MIPD of gentamicin in neonates, as well as pharmacometricians developing or improving PK models for use in MIPD.

Continuous Learning in Model-Informed Precision Dosing: A Case Study in Pediatric Dosing of Vancomycin.

Model-informed precision dosing (MIPD) leverages pharmacokinetic (PK) models to tailor dosing to an individual patient's needs, improving attainment of therapeutic drug exposure targets and thus potentially improving drug efficacy or reducing adverse events. However, selection of an appropriate model for supporting clinical decision making is not trivial. Error or bias in dose selection may arise if the selected model was developed in a population not fully representative of the intended MIPD population. One previously proposed approach is continuous learning, in which an initial model is used in MIPD and then updated as additional data becomes available. In this case study of pediatric vancomycin MIPD, the potential benefits of the continuous learning approach are investigated. Five previously published models were evaluated and found to perform adequately in a data set of 273 pediatric patients in the intensive care unit. Additionally, two predefined simple PK models were fitted on separate populations of 50-350 patients in an approach mimicking clinical implementation of automated continuous learning. With these continuous learning models, prediction error using population PK parameters could be reduced by 2-13% compared with previously published models. Sample sizes of at least 200 patients were found suitable for capturing the interindividual variability in vancomycin at this institution, with limited benefits of larger data sets. Although comprised mostly of trough samples, these sparsely sampled routine clinical data allowed for reasonable estimation of simulated area under the curve (AUC). Together, these findings lay the foundations for a continuous learning MIPD approach.

Population genetic simulation study of power in association testing across genetic architectures and study designs.

While it is well established that genetics can be a major contributor to population variation of complex traits, the relative contributions of rare and common variants to phenotypic variation remains a matter of considerable debate. Here, we simulate genetic and phenotypic data across different case/control panel sampling strategies, sequencing methods, and genetic architecture models based on evolutionary forces to determine the statistical performance of rare variant association tests (RVATs) widely in use. We find that the highest statistical power of RVATs is achieved by sampling case/control individuals from the extremes of an underlying quantitative trait distribution. We also demonstrate that the use of genotyping arrays, in conjunction with imputation from a whole-genome sequenced (WGS) reference panel, recovers the vast majority (90%) of the power that could be achieved by sequencing the case/control panel using current tools. Finally, we show that for dichotomous traits, the statistical performance of RVATs decreases as rare variants become more important in the trait architecture. Our results extend previous work to show that RVATs are insufficiently powered to make generalizable conclusions about the role of rare variants in dichotomous complex traits.