Bayesian Posteriors Research Articles

Bayesian system ID: optimal management of parameter, model, and measurement uncertainty

We evaluate the robustness of a probabilistic formulation of system identification (ID) to sparse, noisy, and indirect data. Specifically, we compare estimators of future system behavior derived from the Bayesian posterior of a learning problem to several commonly used least squares-based optimization objectives used in system ID. Our comparisons indicate that the log posterior has improved geometric properties compared with the objective function surfaces of traditional methods that include differentially constrained least squares and least squares reconstructions of discrete time steppers like dynamic mode decomposition (DMD). These properties allow it to be both more sensitive to new data and less affected by multiple minima --- overall yielding a more robust approach. Our theoretical results indicate that least squares and regularized least squares methods like dynamic mode decomposition and sparse identification of nonlinear dynamics (SINDy) can be derived from the probabilistic formulation by assuming noiseless measurements. We also analyze the computational complexity of a Gaussian filter-based approximate marginal Markov Chain Monte Carlo scheme that we use to obtain the Bayesian posterior for both linear and nonlinear problems. We then empirically demonstrate that obtaining the marginal posterior of the parameter dynamics and making predictions by extracting optimal estimators (e.g., mean, median, mode) yields orders of magnitude improvement over the aforementioned approaches. We attribute this performance to the fact that the Bayesian approach captures parameter, model, and measurement uncertainties, whereas the other methods typically neglect at least one type of uncertainty.

Sampling of Bayesian posteriors with a non-Gaussian probabilistic learning on manifolds from a small dataset

This paper tackles the challenge presented by small-data to the task of Bayesian inference. A novel methodology, based on manifold learning and manifold sampling, is proposed for solving this computational statistics problem under the following assumptions: (1) neither the prior model nor the likelihood function are Gaussian and neither can be approximated by a Gaussian measure; (2) the number of functional input (system parameters) and functional output (quantity of interest) can be large; (3) the number of available realizations of the prior model is small, leading to the small-data challenge typically associated with expensive numerical simulations; the number of experimental realizations is also small; (4) the number of the posterior realizations required for decision is much larger than the available initial dataset. The method and its mathematical aspects are detailed. Three applications are presented for validation: The first two involve mathematical constructions aimed to develop intuition around the method and to explore its performance. The third example aims to demonstrate the operational value of the method using a more complex application related to the statistical inverse identification of the non-Gaussian matrix-valued random elasticity field of a damaged biological tissue (osteoporosis in a cortical bone) using ultrasonic waves.

A probabilistic approach to phase calibration – I. Effects of source structure on fringe-fitting

ABSTRACT We propose a probabilistic framework for performing simultaneous estimation of source structure and fringe-fitting parameters in very long baseline interferometry (VLBI) observations. As a first step, we demonstrate this technique through the analysis of synthetic short-duration Event Horizon Telescope observations of various geometric source models at 230 GHz, in the presence of baseline-dependent thermal noise. We perform Bayesian parameter estimation and model selection between the different source models to obtain reliable uncertainty estimates and correlations between various source and fringe-fitting related model parameters. We also compare the Bayesian posteriors with those obtained using widely used VLBI data reduction packages such as casa and aips, by fringe-fitting 200 Monte Carlo simulations of each source model with different noise realizations, to obtain distributions of the maximum a posteriori estimates. We find that, in the presence of resolved asymmetric source structure and a given array geometry, the traditional practice of fringe-fitting with a point source model yields appreciable offsets in the estimated phase residuals, potentially biasing or limiting the dynamic range of the starting model used for self-calibration. Simultaneously estimating the source structure earlier in the calibration process with formal uncertainties improves the precision and accuracy of fringe-fitting and establishes the potential of the available data, especially when there is little prior information. We also note the potential applications of this method to astrometry and geodesy for specific science cases and the planned improvements to the computational performance and analyses of more complex source distributions.

Dynesty: a dynamic nested sampling package for estimating Bayesian posteriors and evidences

ABSTRACT We present dynesty, a public, open-source, python package to estimate Bayesian posteriors and evidences (marginal likelihoods) using the dynamic nested sampling methods developed by Higson et al. By adaptively allocating samples based on posterior structure, dynamic nested sampling has the benefits of Markov chain Monte Carlo (MCMC) algorithms that focus exclusively on posterior estimation while retaining nested sampling’s ability to estimate evidences and sample from complex, multimodal distributions. We provide an overview of nested sampling, its extension to dynamic nested sampling, the algorithmic challenges involved, and the various approaches taken to solve them in this and previous work. We then examine dynesty’s performance on a variety of toy problems along with several astronomical applications. We find in particular problems dynesty can provide substantial improvements in sampling efficiency compared to popular MCMC approaches in the astronomical literature. More detailed statistical results related to nested sampling are also included in the appendix.

Predicting the global far-infrared SED of galaxies via machine learning techniques

Context. Dust plays an important role in shaping a galaxy’s spectral energy distribution (SED). It absorbs ultraviolet (UV) to near-infrared radiation and re-emits this energy in the far-infrared (FIR). The FIR is essential to understand dust in galaxies. However, deep FIR observations require a space mission, none of which are still active today. Aims. We aim to infer the FIR emission across six Herschel bands, along with dust luminosity, mass, and effective temperature, based on the available UV to mid-infrared (MIR) observations. We also want to estimate the uncertainties of these predictions, compare our method to energy balance SED fitting, and determine possible limitations of the model. Methods. We propose a machine learning framework to predict the FIR fluxes from 14 UV–MIR broadband fluxes. We used a low redshift sample by combining DustPedia and H-ATLAS, and extracted Bayesian flux posteriors through SED fitting. We trained shallow neural networks to predict the far-infrared fluxes, uncertainties, and dust properties. We evaluated them on a test set using a root mean square error (RMSE) in log-space. Results. Our results (RMSE = 0.19 dex) significantly outperform UV–MIR energy balance SED fitting (RMSE = 0.38 dex), and are inherently unbiased. We can identify when the predictions are off, for example when the input has large uncertainties on WISE 22 μm, or when the input does not resemble the training set. Conclusions. The galaxies for which we have UV–FIR observations can be used as a blueprint for galaxies that lack FIR data. This results in a “virtual FIR telescope”, which can be applied to large optical-MIR galaxy samples. This helps bridge the gap until the next FIR mission.

Learning Bayesian Posteriors with Neural Networks for Gravitational-Wave Inference.

We seek to achieve the holy grail of Bayesian inference for gravitational-wave astronomy: using deep-learning techniques to instantly produce the posterior p(θ|D) for the source parameters θ, given the detector data D. To do so, we train a deep neural network to take as input a signal + noise dataset (drawn from the astrophysical source-parameter prior and the sampling distribution of detector noise), and to output a parametrized approximation of the corresponding posterior. We rely on a compact representation of the data based on reduced-order modeling, which we generate efficiently using a separate neural-network waveform interpolant [A. J. K. Chua, C. R. Galley, and M. Vallisneri, Phys. Rev. Lett. 122, 211101 (2019)PRLTAO0031-900710.1103/PhysRevLett.122.211101]. Our scheme has broad relevance to gravitational-wave applications such as low-latency parameter estimation and characterizing the science returns of future experiments.

Bayesian analysis for social data: A step-by-step protocol and interpretation

The paper proposes Bayesian analysis as an alternative approach for the conventional frequentist approach in analyzing social data. A step-by-step protocol of how to implement Bayesian multilevel model analysis with social data and how to interpret the result is presented. The article used a dataset regarding religious teachings and behaviors of lying and violence as an example. An analysis is performed using R statistical software and a bayesvl R package, which offers a network-structured model construction and visualization power to diagnose and estimate results.•The paper provides guidance for conducting a Bayesian multilevel analysis in social sciences through constructing directed acyclic graphs (DAGs, or "relationship trees") for different models, basic and more complex ones.•The method also illustrates how to visualize Bayesian diagnoses and simulated posterior.•The interpretations of visualized diagnoses and simulated posteriors of Bayesian inference are also discussed.

Nonparametric Bayesian posterior contraction rates for scalar diffusions with high-frequency data

We consider inference in the scalar diffusion model $\,\mathrm{d}X_{t}=b(X_{t})\,\mathrm{d}t+\sigma(X_{t})\,\mathrm{d}W_{t}$ with discrete data $(X_{j\Delta_{n}})_{0\leq j\leq n}$, $n\to\infty$, $\Delta_{n}\to0$ and periodic coefficients. For $\sigma$ given, we prove a general theorem detailing conditions under which Bayesian posteriors will contract in $L^{2}$-distance around the true drift function $b_{0}$ at the frequentist minimax rate (up to logarithmic factors) over Besov smoothness classes. We exhibit natural nonparametric priors which satisfy our conditions. Our results show that the Bayesian method adapts both to an unknown sampling regime and to unknown smoothness.

Augmented renal clearance of aminoglycosides using population-based pharmacokinetic modelling with Bayesian estimation in the paediatric ICU

To evaluate augmented renal clearance (ARC) using aminoglycoside clearance (CLAMINO24h) derived from pharmacokinetic (PK) modelling. A retrospective study at two paediatric hospitals of patients who received tobramycin or gentamicin from 1999 to 2016 was conducted. Compartmental PK models were constructed using the Pmetrics package, and Bayesian posteriors were used to estimate CLAMINO24h. ARC was defined as a CLAMINO24h of ≥130 mL/min/1.73 m2. Risk factors for ARC were identified using multivariate logistic regression. The final population model was fitted to 275 aminoglycoside serum concentrations. Overall clearance (L/h) was=CL0×(TBW/70)0.75×AGEH/(TMH + AGEH) + CL1 (0.5/SCr), where TBW is total body weight, H is the Hill coefficient, TM is a maturation term and SCr is serum creatinine. Median CLAMINO24h in those with versus without ARC was 157.36 and 93.42 mL/min/1.73 m2, respectively (P<0.001). ARC was identified in 19.5% of 118 patients. For patients with ARC, median baseline SCr was lower than for those without ARC (0.38 versus 0.41 mg/dL, P=0.073). Risk factors for ARC included sepsis [adjusted OR (aOR) 3.77, 95% CI 1.01-14.07, P=0.048], increasing age (aOR 1.11, 95% CI 1-1.23, P=0.04) and low log-transformed SCr (aOR 0.16, 95% CI 0.05-0.52, P=0.002). Median 24 h AUC (AUC24h) was significantly lower in patients with ARC at 45.27 versus 56.95 mg·h/L, P<0.01. ARC was observed in one of every five patients. Sepsis, increasing age and low SCr were associated with ARC. Increased clearance was associated with an attenuation of AUC24h in this population. Future studies are needed to define optimal dosing in paediatric patients with ARC.

1568. The Pharmacodynamic–Toxicodynamic Relationship of AUC and CMAX in Vancomycin-Induced Kidney Injury

BackgroundVancomycin induces exposure-related acute kidney injury [AKI]; yet the toxicodynamic (TD) driver for toxicity (area under the curve [AUC], vs. maximum concentration [CMAX] vs. trough concentrations [CMIN]) remains unclear. Recent vancomycin guidelines now recommend monitoring AUC over troughs. Here we employed our translational rat model with intensive dose fractionation and sensitive FDA qualified urinary biomarkers to better understand TD relationship.MethodsMale Sprague-Dawley rats received intravenous (IV) vancomycin via an internal jugular vein catheter. Total daily doses of 300 and 400 mg/kg/day were administered as a single, twice, thrice and four times a day injection over 24 hours. Controls received IV saline. Plasma sampling was conducted via a second dedicated jugular catheter, with up to 8 samples in 24 hours. Twenty-four-hour urine was collected during this time and assayed for kidney injury molecule 1 (KIM-1) and other urinary biomarkers using the MILLIPLEX MAP Rat Kidney Panel. Vancomycin in plasma was quantified via LC-MS/MS. PK analyses were conducted using Pmetrics for R. PK exposures during the first 24 hours (i.e., AUC0-24hours, CMAX0-24hours) were calculated from Bayesian posteriors. Pharmacokinetic-toxicodynamic (PK-TD) relationships were assessed with the best fit mathematic model (e.g., exposure-response curve fitting in GraphPad v.7).ResultsSixty-four vancomycin treated and control rats contributed to PK-TD data. An exposure-response (via 4-parameter Hill) model was identified for AUC0-24hoursvs KIM-1 (R2 = 0.62, Figure 1a). Convergence was not obtained for exposure-response models for CMAX0-24hours and CMIN, which was also verified visually (Figure 1b and c). Despite the intensive fractionation, AUC0-24hoursand CMAX0-24hours were highly correlated (P < 0.001, rho = 0.89) and this correlation was consistent across KIM-1 values (Figure 2).ConclusionVancomycin-induced kidney injury was driven by AUC0-24 hours and not CMIN. Continuous infusion studies are needed to understand if changing the infusion profile can improve safety; however, these studies suggest that isometric AUCs may result in similar toxicity. When using intermittent infusion schemes, clinicians should focus on AUC to prevent AKI. Disclosures All authors: No reported disclosures.

Bayesian and frequentist approaches to resonance searches

We investigate Bayesian and frequentist approaches to resonance searches using a toy model based on an ATLAS search for the Higgs boson in the diphoton channel. We draw pseudo-data from the background only model and background plus signal model at multiple luminosities, from 10−3/ fb to 107/ fb. We chart the change in the Bayesian posterior of the background only model and the global p-value. We find that, as anticipated, the posterior converges to certainty about the model as luminosity increases. The p-value, on the other hand, randomly walks between 0 and 1 if the background only model is true, and otherwise converges to 0. After briefly commenting on the frequentist properties of the posterior, we make a direct comparison of the significances obtained in Bayesian and frequentist frameworks. We find that the well-known look-elsewhere effect reduces local significances by about 1σ. We furthermore find that significances from our Bayesian framework are typically about 1−2σ smaller than the global significances, though the reduction depends on the prior, global significance and integrated luminosity. This suggests that even global significances could significantly overstate the evidence against the background only model. We checked that this effect—the Bayes effect—was robust with respect to fourteen choices of prior and investigated the Jeffreys-Lindley paradox for three of them.

Semantic Information G Theory and Logical Bayesian Inference for Machine Learning

An important problem in machine learning is that, when using more than two labels, it is very difficult to construct and optimize a group of learning functions that are still useful when the prior distribution of instances is changed. To resolve this problem, semantic information G theory, Logical Bayesian Inference (LBI), and a group of Channel Matching (CM) algorithms are combined to form a systematic solution. A semantic channel in G theory consists of a group of truth functions or membership functions. In comparison with the likelihood functions, Bayesian posteriors, and Logistic functions that are typically used in popular methods, membership functions are more convenient to use, providing learning functions that do not suffer the above problem. In Logical Bayesian Inference (LBI), every label is independently learned. For multilabel learning, we can directly obtain a group of optimized membership functions from a large enough sample with labels, without preparing different samples for different labels. Furthermore, a group of Channel Matching (CM) algorithms are developed for machine learning. For the Maximum Mutual Information (MMI) classification of three classes with Gaussian distributions in a two-dimensional feature space,only 2–3 iterations are required for the mutual information between three classes and three labels to surpass 99% of the MMI for most initial partitions For mixture models, the Expectation-Maximization (EM) algorithm is improved to form the CM-EM algorithm, which can outperform the EM algorithm when the mixture ratios are imbalanced, or when local convergence exists. The CM iteration algorithm needs to combine with neural networks for MMI classification in high-dimensional feature spaces. LBI needs further investigation for the unification of statistics and logic.

Bayesian cross validation for gravitational-wave searches in pulsar-timing array data

ABSTRACT Gravitational-wave data analysis demands sophisticated statistical noise models in a bid to extract highly obscured signals from data. In Bayesian model comparison, we choose among a landscape of models by comparing their marginal likelihoods. However, this computation is numerically fraught and can be sensitive to arbitrary choices in the specification of parameter priors. In Bayesian cross validation, we characterize the fit and predictive power of a model by computing the Bayesian posterior of its parameters in a training data set, and then use that posterior to compute the averaged likelihood of a different testing data set. The resulting cross-validation scores are straightforward to compute; they are insensitive to prior tuning; and they penalize unnecessarily complex models that overfit the training data at the expense of predictive performance. In this article, we discuss cross validation in the context of pulsar-timing-array data analysis, and we exemplify its application to simulated pulsar data (where it successfully selects the correct spectral index of a stochastic gravitational-wave background), and to a pulsar data set from the NANOGrav 11-yr release (where it convincingly favours a model that represents a transient feature in the interstellar medium). We argue that cross validation offers a promising alternative to Bayesian model comparison, and we discuss its use for gravitational-wave detection, by selecting or refuting models that include a gravitational-wave component.

Sequential Monte Carlo for Hierarchical Bayes with Large Datasets

Practical use of Hierarchical Bayes models require the availability of efficient methods for posterior inference. Sequential Monte Carlo methods have appeared as an extremely robust way to simulate complicated Bayesian posteriors. The simplest version successively weights and resample draws called particles from a sequence of target distributions. The method has a main weakness keeping it from being used in complex hierarchical models and large data sets: sample impoverishment. This issue is usually alleviated through the use of a refreshing MCMC step. However, to preserve effectiveness, the computational costs are quadratically increasing with the total number of observations. We propose a new SMC-within-SMC method. In a first step, each individual-level parameter is estimated separately using standard SMC and a non-hierarchical auxiliary prior. In a second step, we use weighting methods to replace the auxiliary prior with the hierarchical one without the need to recompute any likelihood. In addition to allowing for the separate processing of individual data, this approach drastically reduces the computational costs. A MATLAB package is provided.

Identification and sampling of Bayesian posteriors of high-dimensional symmetric positive-definite matrices for data-driven updating of computational models

We propose a probabilistic methodology for data-driven updating of non-Gaussian high-dimensional symmetric positive-definite matrices involved in computational models. We cast the data-driven updating as a Bayesian identification of the symmetric positive-definite matrices. The posterior thus obtained exhibits several hyperparameters that control the dispersion of the prior and the weight of the weighted distance that represents the model-data misfit in the likelihood function. Using an identification criterion that quantifies the agreement between the predictions and the data, we identify these hyperparameters so as to obtain not only improved predictions but also a probabilistic representation of model uncertainties. The numerical implementation of the Bayesian inversion by using a Markov chain Monte Carlo (MCMC) method is computationally challenging because the support of the posterior is restricted to a set of symmetric positive-definite matrices and the dimensionality of the problem grows with the square of the matrix dimension and hence can be high. We thus use a transformation of measure to set up the Markov chain in terms of real-valued state variables whose distribution is Gaussian under the prior but non-Gaussian under the posterior. This transformation of measure allows us to sample the posterior using an Itô-SDE-based MCMC method that inherits computational efficiency in high dimension from leveraging the gradient of the posterior. We apply this methodology to a problem of a data-driven updating of a reduced-order model in computational structural dynamics.

Electron density and velocity functions in a low beta Polywell

Langmuir probe data from a low beta Polywell device at the University of Sydney are analysed to compare models describing its electron velocity distribution function (EVDF). Three models are compared using the Bayesian evidence across datasets measured at different spatial points in the device at varying coil voltage biases and coil currents. The best performing model of the EVDF across all datasets is found to be a mixture of Maxwellian and Gaussian distributions. This outperforms a model with a pure Maxwellian distribution and another model with an advanced non-thermalised distribution, indicating that this device partially thermalises its confined electron population. From the same analysis, the Bayesian posterior is used to give statistical distributions of plasma parameters, such as the plasma potential and density. Compared across the different datasets, these parameters indicate the successful formation of an electric potential well, crucial to the operation of the Polywell, and the conditions in which it appears. They also indicate that the electron density forms a dense shell near the centre of the device, which contracts with increasing bias voltage and coil current.

Postauditing and Cost Estimation Applications: An Illustration of MCMC Simulation for Bayesian Regression Analysis

Often in Bayesian anlysis closed-form posteriors cannot be derived for complex models. However, it is important to be able to do Bayesian analysis relatively easily. This article presents an alternative, the more general Markov chain Monte Carlo (MCMC) simulation approach, which permits the efficient development of posterior distributions. MCMC simulation methods are now becoming the state of the art in numerous empirical and analytical applications in applied mathematics, biostatistics, marketing, economics, and other areas, but those methods are noticeably absent in the engineering economic analysis literature. The purpose of this article is to introduce MCMC simulation methods to the engineering economics research and practitioner community. Using postaudits and cost estimation as application areas, the article focuses on what MCMC simulation entails, its advantages, and its disadvantages and highlights the usefulness and versatility of the approach.

Fully Bayesian estimation under informative sampling

Survey data are often collected under informative sampling designs where subject inclusion probabilities are designed to be correlated with the response variable of interest. The data modeler seeks to estimate the parameters of a population model they specify from these data. Sampling weights constructed from marginal inclusion probabilities are typically used to form an exponentiated pseudo likelihood as a plug-in estimator in a partially Bayesian pseudo posterior. We introduce the first fully Bayesian alternative, based on a Bayes rule construction, that simultaneously performs weight smoothing and estimates the population model parameters in a construction that treats the response variable(s) and inclusion probabilities as jointly randomly generated from a population distribution. We formulate conditions on known marginal and pairwise inclusion probabilities that define a class of sampling designs where $L_{1}$ consistency of the joint posterior is guaranteed. We compare performances between the two approaches on synthetic data. We demonstrate that the credibility intervals under our fully Bayesian method achieve nominal coverage. We apply our method to data from the National Health and Nutrition Examination Survey to explore the relationship between caffeine consumption and systolic blood pressure.

1394. A Translational Pharmacokinetic Rat Model of Cerebral Spinal Fluid (CSF) and Plasma Concentrations of Cefepime

BackgroundFor treatment of central nervous system infections caused by GNB, adequate cefepime concentrations are required in the cerebral spinal fluid (CSF) and brain. However, high plasma cefepime exposures have resulted in neurotoxicity. There is a need to understand the real-time pharmacokinetic (PK) relationship between plasma and CSF concentrations as serial CSF sampling is not regularly performed.MethodsMale Sprague-Dawley rats received cefepime via an internal jugular vein catheter. A total daily dose of 150 mg/kg/day was administered as a single injection every 24 hours for 4 days. Plasma samples (mean n = 5 per rat) was obtained via a second dedicated catheter, with up to five samples obtained on a single concentration–time curve. CSF sampling occurred via an intracisternal catheter, with up to two samples taken every 24 hours. Cefepime in plasma and CSF were quantified via LC–MS/MS. PK analyses were conducted using Pmetrics for R. Multiple physiologic compartmental models were fit, with the ultimate model selected by Akaike score and rule of parsimony. Each rat’s concentrations were predicted from the final model, and predictive performance was evaluated with bias and imprecision of the population and Bayesian posterior prediction models. PK parameters were estimated, and PK exposures during the first 24 hours (i.e., AUC0–24, CMAX0–24, CMIN0–24) were calculated for each rat from Bayesian posteriors.ResultsEleven rats contributed PK data. A three-compartment model fit the data well [Figure 1: Plasma, Population (a) and Bayesian (b); Figure 2: CSF, Population (a) and Bayesian (b)]. Population median parameter values (CV%) for rate constant (Ke), volume central compartment (V1), volume CSF compartment (V3), rate to/from central/peripheral compartment (KCP, KPC), rate to/from central/CSF compartment (K13, K31) were: 5.04 hour−1 (43.4%), 0.069 L (39.24%), 0.28 hour−1 (52.11%), 17.42 hour−1(34.83%), 0.32 hour−1 (165.3), 0.31 hour−1 (79.89), respectively. Noncompartmental analysis of the Bayesian posteriors revealed a plasma median half-life, AUC0–24, CMAX0–24, and CMIN0–24 of 2.6 hours, 158.1 mg hour/L, 189.3 mg/L, and 0.0003 mg/L from the first dose.ConclusionCefepime transit to the CSF is rapid and highly predictable in the rat model. This model will be highly useful for understanding neurotoxicity outcomes for cefepime. Disclosures J. Liu, Merck: Grant fund from Merck, Research grant.

1406. Augmented Renal Clearance Using Aminoglycoside Population-Based Pharmacokinetic Modeling with Bayesian Estimation in Children in the Pediatric Intensive Care Unit

BackgroundAugmented renal clearance (ARC) in critically ill pediatric patients has been evaluated in limited studies. We evaluated ARC using clearance of aminoglycosides (CLAMINO) derived from population-based pharmacokinetic modeling.MethodsA retrospective, cohort study was conducted at two pediatric hospitals in patients who received aminoglycosides from 1999 to 2016. ARC was defined as a CLAMINO of ≥130 mL/minute/1.73 m2 within the first 24 hours of therapy. Pharmacokinetic (PK) models with nonparametric parameter estimation were constructed using Pmetrics in R, with the ultimate model selected by Akaike score and rule of parsimony. Covariate modifiers considered included: age, total body weight (TBW), serum creatinine (SCr) and sex. Noncompartmental analysis was performed on the Bayesian posteriors from the first dose to generate CLAMINO within the first 24 hours and other PK exposure metrics (i.e., area under the curve for first 24 hours [AUC24], maximum concentration [CMAX]). Summary of patient demographics and statistical analysis were performed using GraphPad Prism version 7.ResultsARC was identified in 34 of 117 (29%) subjects using 275 aminoglycoside serum concentrations. A two-compartment model fit the data well (See Figure 1: Population [a], Bayesian [b]). Allometric scaling of CLAMINO utilized a fixed exponent of 0.75 and volume of distribution (VD) scaling utilized a fixed exponent of 1 in the final model. The final population model for CLAMINO (L/hour) was 3.45 × (TBW/40)0.75 + 0.05 × 10(SCr/AGE) and VD was 10.64 × (TBW/40)1. Median age and baseline SCr were similar in those with and without ARC (13 [IQR 10–16] vs. 11.0 [5.0–15.0] years, P = 0.11, and 0.37 [0.27–0.49] vs. 0.38 [0.28–0.50] mg/dL, P = 0.67, respectively). Median TBW was found to be significantly higher in those with vs. without ARC (44.9 [26.9–61.7] vs. 34 [17.6–54.9] kg P = 0.04). Median 24 hours CLAMINO was also found to be significantly higher in those with vs. without ARC (147.3 [138.7–163.9] vs. 94.5 [79.4–112.9], mL/minute/1.73 m2, P < 0.001). Patients with vs. without ARC had significantly lower AUC24 and CMAX (40.7 [33.3–54.4] vs. 55.7[46.7–66.4] mg hour/L, P ≤ 0.001 and 5.06 [4.11–6.76] vs. 6.32 [5–7.44], µg/mL, P = 0.01).ConclusionThe incidence of ARC observed was similar to adult studies. Patients that exhibited ARC had lower AUC24 and CMAX; thus, higher doses may be warranted. Disclosures All authors: No reported disclosures.