Using seconds-resolved pharmacokinetic datasets to assess pharmacokinetic models encompassing time-varying physiology

AIM

Pharmacokinetics have historically been assessed using drug concentration data obtained via blood draws and bench-top analysis. The cumbersome nature of these typically constrains studies to at most a dozen concentration measurements per dosing event. This, in turn, limits our statistical power in the detection of hours-scale, time-varying physiological processes. Given the recent advent of in vivo electrochemical aptamer-based (EAB) sensors, however, we can now obtain hundreds of concentration measurements per administration. Our aim in this paper was to assess the ability of these time-dense datasets to describe time-varying pharmacokinetic models with good statistical significance.

METHODS

We used seconds-resolved measurements of plasma tobramycin concentrations in rats to statistically compare traditional one- and two-compartmental pharmacokinetic models to new models in which the proportional relationship between a drug's plasma concentration and its elimination rate varies in response to changing kidney function.

RESULTS

We found that a modified one-compartment model in which the proportionality between the plasma concentration of tobramycin and its elimination rate falls reciprocally with time either meets or is preferred over the standard two-compartment pharmacokinetic model for half of the datasets characterized. When we reduced the impact of the drug's rapid distribution phase on the model, this one-compartment, time-varying model was statistically preferred over the standard one-compartment model for 80% of our datasets.

CONCLUSIONS

Our results highlight both the impact that simple physiological changes (such as varying kidney function) can have on drug pharmacokinetics and the ability of high-time resolution EAB sensor measurements to identify such impacts.

A Novel Method for Identifying a Parsimonious and Accurate Predictive Model for Multiple Clinical Outcomes

BACKGROUND AND OBJECTIVE

Most methods for developing clinical prognostic models focus on identifying parsimonious and accurate models to predict a single outcome; however, patients and providers often want to predict multiple outcomes simultaneously. As an example, for older adults one is often interested in predicting nursing home admission as well as mortality. We propose and evaluate a novel predictor-selection computing method for multiple outcomes and provide the code for its implementation.

METHODS

Our proposed algorithm selected the best subset of common predictors based on the minimum average normalized Bayesian Information Criterion (BIC) across outcomes: the Best Average BIC (baBIC) method. We compared the predictive accuracy (Harrell's C-statistic) and parsimony (number of predictors) of the model obtained using the baBIC method with: 1) a subset of common predictors obtained from the union of optimal models for each outcome (Union method), 2) a subset obtained from the intersection of optimal models for each outcome (Intersection method), and 3) a model with no variable selection (Full method). We used a case-study data from the Health and Retirement Study (HRS) to demonstrate our method and conducted a simulation study to investigate performance.

RESULTS

In the case-study data and simulations, the average Harrell's C-statistics across outcomes of the models obtained with the baBIC and Union methods were comparable. Despite the similar discrimination, the baBIC method produced more parsimonious models than the Union method. In contrast, the models selected with the Intersection method were the most parsimonious, but with worst predictive accuracy, and the opposite was true in the Full method. In the simulations, the baBIC method performed well by identifying many of the predictors selected in the baBIC model of the case-study data most of the time and excluding those not selected in the majority of the simulations.

CONCLUSIONS

Our method identified a common subset of variables to predict multiple clinical outcomes with superior balance between parsimony and predictive accuracy to current methods.

Prediction of regulatory targets of alternative isoforms of the epidermal growth factor receptor in a glioblastoma cell line

Background

The epidermal growth factor receptor (EGFR) is a major regulator of proliferation in tumor cells. Elevated expression levels of EGFR are associated with prognosis and clinical outcomes of patients in a variety of tumor types. There are at least four splice variants of the mRNA encoding four protein isoforms of EGFR in humans, named I through IV. EGFR isoform I is the full-length protein, whereas isoforms II-IV are shorter protein isoforms. Nevertheless, all EGFR isoforms bind the epidermal growth factor (EGF). Although EGFR is an essential target of long-established and successful tumor therapeutics, the exact function and biomarker potential of alternative EGFR isoforms II-IV are unclear, motivating more in-depth analyses. Hence, we analyzed transcriptome data from glioblastoma cell line SF767 to predict target genes regulated by EGFR isoforms II-IV, but not by EGFR isoform I nor other receptors such as HER2, HER3, or HER4.

Results

We analyzed the differential expression of potential target genes in a glioblastoma cell line in two nested RNAi experimental conditions and one negative control, contrasting expression with EGF stimulation against expression without EGF stimulation. In one RNAi experiment, we selectively knocked down EGFR splice variant I, while in the other we knocked down all four EGFR splice variants, so the associated effects of EGFR II-IV knock-down can only be inferred indirectly. For this type of nested experimental design, we developed a two-step bioinformatics approach based on the Bayesian Information Criterion for predicting putative target genes of EGFR isoforms II-IV. Finally, we experimentally validated a set of six putative target genes, and we found that qPCR validations confirmed the predictions in all cases.

Conclusions

By performing RNAi experiments for three poorly investigated EGFR isoforms, we were able to successfully predict 1140 putative target genes specifically regulated by EGFR isoforms II-IV using the developed Bayesian Gene Selection Criterion (BGSC) approach. This approach is easily utilizable for the analysis of data of other nested experimental designs, and we provide an implementation in R that is easily adaptable to similar data or experimental designs together with all raw datasets used in this study in the BGSC repository, https://github.com/GrosseLab/BGSC.

Electronic supplementary material

The online version of this article (10.1186/s12859-019-2944-9) contains supplementary material, which is available to authorized users.

Connections between physics, mathematics, and deep learning

Starting from Fermat's principle of least action, which governs classical and quantum mechanics and from the theory of exterior differential forms, which governs the geometry of curved manifolds, we show how to derive the equations governing neural networks in an intrinsic, coordinate-invariant way, where the loss function plays the role of the Hamiltonian. To be covariant, these equations imply a layer metric which is instrumental in pretraining and explains the role of conjugation when using complex numbers. The differential formalism clarifies the relation of the gradient descent optimizer with Aristotelian and Newtonian mechanics. The Bayesian paradigm is then analyzed as a renormalizable theory yielding a new derivation of the Bayesian information criterion. We hope that this formal presentation of the differential geometry of neural networks will encourage some physicists to dive into deep learning and, reciprocally, that the specialists of deep learning will better appreciate the close interconnection of their subject with the foundations of classical and quantum field theory.

Predicting Productivity Losses from Health-Related Quality of Life Using Patient Data

OBJECTIVE

This paper estimates productivity loss using the health of the patient in order to allow indirect estimation of these costs for inclusion in economic evaluation.

METHODS

Data from two surveys of inpatients [Health outcomes data repository (HODaR) sample (n = 42,442) and health improvement and patient outcomes (HIPO) sample (n = 6046)] were used. The number of days off paid employment or normal activities (excluding paid employment) was modelled using the health of the patients measured by the EQ-5D, international classification of diseases (ICD) chapters, and other health and sociodemographic data. Two-part models (TPMs) and zero-inflated negative binomial (ZINB) models were identified as the most appropriate specifications, given large spikes at the minimum and maximum days for the dependent variable. Analysis was undertaken separately for the two datasets to account for differences in recall period and identification of those who were employed.

RESULTS

Models were able to reflect the large spike at the minimum (zero days) but not the maximum, with TPMs doing slightly better than the ZINB model. The EQ-5D was negatively associated with days off employment and normal activities in both datasets, but ICD chapters only had statistically significant coefficients for some chapters in the HODaR.

CONCLUSIONS

TPMs can be used to predict productivity loss associated with the health of the patient to inform economic evaluation. Limitations include recall and response bias and identification of who is employed in the HODaR, while the HIPO suffers from a small sample size. Both samples exclude some patient groups.

Bayesian model selection in logistic regression for the detection of adverse drug reactions

Spontaneous adverse event reports have a high potential for detecting adverse drug reactions. However, due to their dimension, the analysis of such databases requires statistical methods. In this context, disproportionality measures can be used. Their main idea is to project the data onto contingency tables in order to measure the strength of associations between drugs and adverse events. However, due to the data projection, these methods are sensitive to the problem of coprescriptions and masking effects. Recently, logistic regressions have been used with a Lasso type penalty to perform the detection of associations between drugs and adverse events. On different examples, this approach limits the drawbacks of the disproportionality methods, but the choice of the penalty value is open to criticism while it strongly influences the results. In this paper, we propose to use a logistic regression whose sparsity is viewed as a model selection challenge. Since the model space is huge, a Metropolis-Hastings algorithm carries out the model selection by maximizing the BIC criterion. Thus, we avoid the calibration of penalty or threshold. During our application on the French pharmacovigilance database, the proposed method is compared to well-established approaches on a reference dataset, and obtains better rates of positive and negative controls. However, many signals (i.e., specific drug-event associations) are not detected by the proposed method. So, we conclude that this method should be used in parallel to existing measures in pharmacovigilance. Code implementing the proposed method is available at the following url: https://github.com/masedki/MHTrajectoryR.

A Consistent Information Criterion for Support Vector Machines in Diverging Model Spaces

Information criteria have been popularly used in model selection and proved to possess nice theoretical properties. For classification, Claeskens et al. (2008) proposed support vector machine information criterion for feature selection and provided encouraging numerical evidence. Yet no theoretical justification was given there. This work aims to fill the gap and to provide some theoretical justifications for support vector machine information criterion in both fixed and diverging model spaces. We first derive a uniform convergence rate for the support vector machine solution and then show that a modification of the support vector machine information criterion achieves model selection consistency even when the number of features diverges at an exponential rate of the sample size. This consistency result can be further applied to selecting the optimal tuning parameter for various penalized support vector machine methods. Finite-sample performance of the proposed information criterion is investigated using Monte Carlo studies and one real-world gene selection problem.

Hybrid fusion of linear, non-linear and spectral models for the dynamic modeling of sEMG and skeletal muscle force: an application to upper extremity amputation

Estimating skeletal muscle (finger) forces using surface Electromyography (sEMG) signals poses many challenges. In general, the sEMG measurements are based on single sensor data. In this paper, two novel hybrid fusion techniques for estimating the skeletal muscle force from the sEMG array sensors are proposed. The sEMG signals are pre-processed using five different filters: Butterworth, Chebychev Type II, Exponential, Half-Gaussian and Wavelet transforms. Dynamic models are extracted from the acquired data using Nonlinear Wiener Hammerstein (NLWH) models and Spectral Analysis Frequency Dependent Resolution (SPAFDR) models based system identification techniques. A detailed comparison is provided for the proposed filters and models using 18 healthy subjects. Wavelet transforms give higher mean correlation of 72.6 ± 1.7 (mean ± SD) and 70.4 ± 1.5 (mean ± SD) for NLWH and SPAFDR models, respectively, when compared to the other filters used in this work. Experimental verification of the fusion based hybrid models with wavelet transform shows a 96% mean correlation and 3.9% mean relative error with a standard deviation of ± 1.3 and ± 0.9 respectively between the overall hybrid fusion algorithm estimated and the actual force for 18 test subjects' k-fold cross validation data.

Comparison of phenotypes of childhood wheeze and cough in 2 independent cohorts

BACKGROUND

Among children with wheeze and recurrent cough there is great variation in clinical presentation and time course of the disease. We previously distinguished 5 phenotypes of wheeze and cough in early childhood by applying latent class analysis to longitudinal data from a population-based cohort (original cohort).

OBJECTIVE

To validate previously identified phenotypes of childhood cough and wheeze in an independent cohort.

METHODS

We included 903 children reporting wheeze or recurrent cough from an independent population-based cohort (validation cohort). As in the original cohort, we used latent class analysis to identify phenotypes on the basis of symptoms of wheeze and cough at 2 time points (preschool and school age) and objective measurements of atopy, lung function, and airway responsiveness (school age). Prognostic outcomes (wheeze, bronchodilator use, cough apart from colds) 5 years later were compared across phenotypes.

RESULTS

When using a 5-phenotype model, the analysis distinguished 3 phenotypes of wheeze and 2 of cough as in the original cohort. Two phenotypes were closely similar in both cohorts: Atopic persistent wheeze (persistent multiple trigger wheeze and chronic cough, atopy and reduced lung function, poor prognosis) and transient viral wheeze (early-onset transient wheeze with viral triggers, favorable prognosis). The other phenotypes differed more between cohorts. These differences might be explained by differences in age at measurements.

CONCLUSIONS

Applying the same method to 2 different cohorts, we consistently identified 2 phenotypes of wheeze (atopic persistent wheeze, transient viral wheeze), suggesting that these represent distinct disease processes. Differences found in other phenotypes suggest that the age when features are assessed is critical and should be considered carefully when defining phenotypes.