Septic shock prediction for ICU patients via coupled HMM walking on sequential contrast patterns

Background and objectiveCritical care patient events like sepsis or septic shock in intensive care units (ICUs) are dangerous complications which can cause multiple organ failures and eventual death. Preventive prediction of such events will allow clinicians to stage effective interventions for averting these critical complications.MethodsIt is widely understood that physiological conditions of patients on variables such as blood pressure and heart rate are suggestive to gradual changes over a certain period of time, prior to the occurrence of a septic shock. This work investigates the performance of a novel machine learning approach for the early prediction of septic shock. The approach combines highly informative sequential patterns extracted from multiple physiological variables and captures the interactions among these patterns via coupled hidden Markov models (CHMM). In particular, the patterns are extracted from three non-invasive waveform measurements: the mean arterial pressure levels, the heart rates and respiratory rates of septic shock patients from a large clinical ICU dataset called MIMIC-II.Evaluation and resultsFor baseline estimations, SVM and HMM models on the continuous time series data for the given patients, using MAP (mean arterial pressure), HR (heart rate), and RR (respiratory rate) are employed. Single channel patterns based HMM (SCP-HMM) and multi-channel patterns based coupled HMM (MCP-HMM) are compared against baseline models using 5-fold cross validation accuracies over multiple rounds. Particularly, the results of MCP-HMM are statistically significant having a p-value of 0.0014, in comparison to baseline models. Our experiments demonstrate a strong competitive accuracy in the prediction of septic shock, especially when the interactions between the multiple variables are coupled by the learning model.ConclusionsIt can be concluded that the novelty of the approach, stems from the integration of sequence-based physiological pattern markers with the sequential CHMM model to learn dynamic physiological behavior, as well as from the coupling of such patterns to build powerful risk stratification models for septic shock patients.     

Comparison of hospital charge prediction models for colorectal cancer patients: neural network vs. decision tree models

Analysis and prediction of the care charges related to colorectal cancer in Korea are important for the allocation of medical resources and the establishment of medical policies because the incidence and the hospital charges for colorectal cancer are rapidly increasing. But the previous studies based on statistical analysis to predict the hospital charges for patients did not show satisfactory results. Recently, data mining emerges as a new technique to extract knowledge from the huge and diverse medical data. Thus, we built models using data mining techniques to predict hospital charge for the patients. A total of 1,022 admission records with 154 variables of 492 patients were used to build prediction models who had been treated from 1999 to 2002 in the Kyung Hee University Hospital. We built an artificial neural network (ANN) model and a classification and regression tree (CART) model, and compared their prediction accuracy. Linear correlation coefficients were high in both models and the mean absolute errors were similar. But ANN models showed a better linear correlation than CART model (0.813 vs. 0.713 for the hospital charge paid by insurance and 0.746 vs. 0.720 for the hospital charge paid by patients). We suggest that ANN model has a better performance to predict charges of colorectal cancer patients.     

A causal model of blood enzyme data and visualization of tissue conditions

Quantitative diagnostics is an important field in which clinical data are converted into medical information. A variety of approaches to obtain medical diagnoses have been developed and multivariate statistical analysis supports the diagnostic process. Although many clinical data are affected by body conditions such as disease and functional failure, only a few models take this phenomenon into consideration. The correlation between laboratory test results can be understood as a causal relationship between body conditions and clinical test data variations. A multivariate statistical method, factor analysis, expresses a causal relationship between latent variables and observed variables. We developed a causal model for blood enzyme data using factor analysis. The latent variables were assumed to be organ specific regarding 9 enzyme data. This causal model expressed clinical knowledge within blood enzymes and allowed visualization of organ conditions. The visualization of laboratory data is useful to screen patient's pathological states.     

Integrating regression formulas and kernel functions into locally adaptive knowledge-based neural networks: a case study on renal function evaluation

OBJECTIVE: In many medical areas, there exist different regression formulas to predict/evaluate a medical outcome on the same problem, each of them being efficient only in a particular sub-space of the problem space. The paper aims at the development of a generic, incremental learning model that includes all available regression formulas for a particular prediction problem to define local areas of the problem space with their best performing formula along with useful explanation rules. Another objective of the paper is to develop a specific model for renal function evaluation using nine existing formulas. METHODS AND MATERIALS: We have used a connectionist neuro-fuzzy approach and have developed a knowledge-based neural network model (KBNN) which incorporates and adapts incrementally several existing regression formulas and kernel functions. The model incorporates different non-linear regression functions as neurons in its hidden layer and adapts these functions through incremental learning from data in particular local areas of the space. More specifically, each hidden neural node has a pair of functions associated with it--one regression formula, that represents existing knowledge and one Gaussian kernel function, that defines the sub-space of the whole problem space, in which the formula is locally adapted to new data. All these functions are aggregated and changed through incremental learning. The proposed KBNN model is illustrated using a medical dataset of observed patient glomerular filtration rate (GFR) measurements for renal function evaluation. In this case study, the regression function for each cluster is selected by the model from nine formulas commonly used by medical practitioners to predict GFR. 441 GFR data vectors from 141 patients taken from 12 sites in Australia and New Zealand have been used as a case study experimental data set. RESULTS: The proposed GFR prediction model, based on the proposed generic KBNN model, outperforms at least by 10% accuracy any of the individual regression formulas or a standard neural network model. Furthermore, we have derived locally adapted regression formulas to perform best on local clusters of data along with useful explanatory rules. CONCLUSION: The proposed KBNN model manifests better accuracy then existing regression formulas or neural network models for renal function evaluation and extracts modified formulas that perform well in local areas of the problem space.     

Regression

Regression models are widely used for addressing scientific questions of interest regarding the associations among a set of variables. In particular, linear regression models describe how part of the natural individual-to-individual variation in a continuous response variable can be explained by one or more explanatory variables. In this article we provide a general overview of regression concepts, emphasizing the two most common goals of regression analysis: explanation and prediction. We discuss various aspects of interpretation of regression coefficients. We also consider the notions of confounding and interaction within regression analyses. Finally, we consider important generalizations of linear regression to handle the case where the response variable is binary (logistic regression) and also settings with correlated responses (e.g., repeated measurements on individuals over time). We conclude by discussing how linear and logistic regression are special cases of a broad and useful collection of regression models known as generalized linear models.     

Prediction of drug’s Anatomical Therapeutic Chemical (ATC) code by integrating drug–domain network

Predicting Anatomical Therapeutic Chemical (ATC) code of drugs is of vital importance for drug classification and repositioning. Discovering new association information related to drugs and ATC codes is still difficult for this topic. We propose a novel method named drug–domain hybrid (dD-Hybrid) incorporating drug–domain interaction network information into prediction models to predict drug’s ATC codes. It is based on the assumption that drugs interacting with the same domain tend to share therapeutic effects. The results demonstrated dD-Hybrid has comparable performance to other methods on the gold standard dataset. Further, several new predicted drug-ATC pairs have been verified by experiments, which offer a novel way to utilize drugs for new purposes effectively.     

Parametric Comparisons of Intracranial Mechanical Responses from Three Validated Finite Element Models of the Human Head

A number of human head finite element (FE) models have been developed from different research groups over the years to study the mechanisms of traumatic brain injury. These models can vary substantially in model features and parameters, making it important to evaluate whether simulation results from one model are readily comparable with another, and whether response-based injury thresholds established from a specific model can be generalized when a different model is employed. The purpose of this study is to parametrically compare regional brain mechanical responses from three validated head FE models to test the hypothesis that regional brain responses are dependent on the specific head model employed as well as the region of interest (ROI). The Dartmouth Scaled and Normalized Model (DSNM), the Simulated Injury Monitor (SIMon), and the Wayne State University Head Injury Model (WSUHIM) were selected for comparisons. For model input, 144 unique kinematic conditions were created to represent the range of head impacts sustained by male collegiate hockey players during play. These impacts encompass the 50th, 95th, and 99th percentile peak linear and rotational accelerations at 16 impact locations around the head. Five mechanical variables (strain, strain rate, strain × strain rate, stress, and pressure) in seven ROIs reported from the FE models were compared using Generalized Estimating Equation statistical models. Highly significant differences existed among FE models for nearly all output variables and ROIs. The WSUHIM produced substantially higher peak values for almost all output variables regardless of the ROI compared to the DSNM and SIMon models (p < 0.05). DSNM also produced significantly different stress and pressure compared with SIMon for all ROIs (p < 0.05), but such differences were not consistent across ROIs for other variables. Regardless of FE model, most output variables were highly correlated with linear and rotational peak accelerations. The significant disparities in regional brain responses across head models regardless of the output variables strongly suggest that model-predicted brain responses from one study should not be extended to other studies in which a different model is utilized. Consequently, response-based injury tolerance thresholds from a specific model should not be generalized to other studies either in which a different model is used. However, the similar relationships between regional responses and the linear/rotational peak accelerations suggest that each FE model can be used independently to assess regional brain responses to impact simulations in order to perform statistical correlations with medical images and/or well-selected experiments with documented injury findings.     

An extension to the discriminant analysis of near-infrared spectra

Partial least squares discriminant analysis (PLS-DA) is widely used in multivariate calibration method. Very often, only one single quantitative model is constructed to predict the relationship between the response and the independent variables. This approach can easily misidentify, under or over estimate the important features contained in the independent variables. The results obtained by a single prediction model are thus unstable or correlated to spurious spectral variance, particularly when the training set for PLS-DA is relatively small. A new algorithm developed by applying the Monte Carlo method to PLS-DA, namely MC–PLS-DA, is proposed to classify spectral data obtained from near-infrared blood glucose measurement. Noise in the data is removed by randomly selecting different subsets from the whole training dataset to generate a large number of models. The mean sensitivity and specificity of these models are then calculated to determine the model with the best classification rate. The results show that the MC–PLS-DA method gives more accurate prediction results when compared with other classification methods used for classifying near infrared spectroscopic data of blood glucose. Also, the stability of the PLS-DA model is enhanced.     

糖尿病视网膜病变的风险揭示与关键因素分析

目的:通过构建组合模型对糖尿病并发视网膜病变（DR）的患病风险进行预测,为DR的预防和诊断提供参考。方法:基于3 000例糖尿病患者的生化检测数据,运用互信息作为评价标准筛选出与DR有关的特征因素,将其作为入模变量构建5种常见的模型,以准确率、精确率、召回率和AUC作为评价标准筛选出预测能力较优的3种模型,并运用Stacking方法构建组合模型。结果:通过互信息筛选出39个特征因素,发现随机森林模型、SVM模型以及Logistic回归模型这3种模型表现较优;构建的3种组合模型中,发现以SVM、Logistic为初级分类器,随机森林为次级分类器的组合模型预测能力最好,其AUC高达0.877。结论:组合模型相比单一模型具有更好的DR风险预测能力,更有助于DR的临床诊断。     

Technologies for measuring HIV-1 drug resistance

Drug resistance testing significantly improves response to antiretroviral treatment in HIV-1-infected patients, therefore it has recently been implemented into current guidelines for the management of antiretroviral therapy. Knowledge about technologies for measuring drug resistance is important for several reasons: (a) differences exist between different technologies and also between assays based on the same technology; (b) the results of resistance testing are strongly dependent on the reliability and precision of the technology used; and (c) technical aspects have to be considered for a clinically relevant interpretation of drug resistance. The spectrum of genotypic and phenotypic technologies as well as the technical quality is increasing, which shifts the emphasis to the interpretation of resistance profiles. The interpretation is based on the knowledge of drug resistance-associated mutations as well as correlations between genotype and phenotype and clinical response, which are incorporated into rules-based systems. Bioinformatic techniques are used to generate mathematical models for the prediction of drug resistance from genotype. Both approaches are converging toward the prediction of clinical response. Because therapy response is dependent on many additional variables, further efforts are required for the generation of a large clinical database. This will be the basis of a prediction system that will optimize the antiretroviral therapy for each individual patient.     

Characterizing observation volumes and the role of excitation saturation in one-photon fluorescence fluctuation spectroscopy

Fluorescence correlation spectroscopy (FCS) and related distribution analysis techniques have become extremely important and widely used research tools for analyzing the dynamics, kinetics, interactions, and mobility of biomolecules. However, it is not widely recognized that photophysical dynamics can dramatically influence the calibration of fluctuation spectroscopy instrumentation. While the basic theories for fluctuation spectroscopy methods are well established, there have not been quantitative models to characterize the photophysical-induced variations observed in measured fluctuation spectroscopy data under varied excitation conditions. We introduce quantitative models to characterize how the fluorescence observation volumes in one-photon confocal microscopy are modified by excitation saturation as well as corresponding models for the effect of the volume changes in FCS. We introduce a simple curve fitting procedure to model the role of saturation in FCS measurements and demonstrate its accuracy in fitting measured correlation curves over a wide range of excitation conditions.     

Parametric recursive system identification and self-adaptive modeling of the human energy metabolism for adaptive control of fat weight

A mathematical model has been developed to facilitate indirect measurements of difficult to measure variables of the human energy metabolism on a daily basis. The model performs recursive system identification of the parameters of the metabolic model of the human energy metabolism using the law of conservation of energy and principle of indirect calorimetry. Self-adaptive models of the utilized energy intake prediction, macronutrient oxidation rates, and daily body composition changes were created utilizing Kalman filter and the nominal trajectory methods. The accuracy of the models was tested in a simulation study utilizing data from the Minnesota starvation and overfeeding study. With biweekly macronutrient intake measurements, the average prediction error of the utilized carbohydrate intake was ?23.2 ± 53.8 kcal/day, fat intake was 11.0 ± 72.3 kcal/day, and protein was 3.7 ± 16.3 kcal/day. The fat and fat-free mass changes were estimated with an error of 0.44 ± 1.16 g/day for fat and ?2.6 ± 64.98 g/day for fat-free mass. The daily metabolized macronutrient energy intake and/or daily macronutrient oxidation rate and the daily body composition change from directly measured serial data are optimally predicted with a self-adaptive model with Kalman filter that uses recursive system identification.     

Bayesian networks: a new method for the modeling of bibliographic knowledge

Falls in geriatry are associated with important morbidity, mortality and high healthcare costs. Because of the large number of variables related to the risk of falling, determining patients at risk is a difficult challenge. The aim of this work was to validate a tool to detect patients with high risk of fall using only bibliographic knowledge. Thirty articles corresponding to 160 studies were used to modelize fall risk. A retrospective case–control cohort including 288 patients (88 ± 7 years) and a prospective cohort including 106 patients (89 ± 6 years) from two geriatric hospitals were used to validate the performances of our model. We identified 26 variables associated with an increased risk of fall. These variables were split into illnesses, medications, and environment. The combination of the three associated scores gives a global fall score. The sensitivity and the specificity were 31.4, 81.6, 38.5, and 90 %, respectively, for the retrospective and the prospective cohort. The performances of the model are similar to results observed with already existing prediction tools using model adjustment to data from numerous cohort studies. This work demonstrates that knowledge from the literature can be synthesized with Bayesian networks.     

Emergency total thyroidectomy for bleeding anaplastic thyroid carcinoma: a viable option for palliation

Anaplastic thyroid carcinoma (ATC) is a rare and highly aggressive thyroid neoplasm. Bleeding from tumor is an uncommon, but potentially life-threatening complication requiring sophisticated intervention facilities which are not usually available at odd hours in emergency. We report the case of a 45-year-old woman who presented with exsanguinating hemorrhage from ATC and was treated by emergency total thyroidectomy. The patient is well three months postoperatively. Emergency total thyroidectomy is a viable option for palliation in ATC presenting with bleeding.     

Technologies for measuring HIV-1 drug resistance

Abstract

Drug resistance testing significantly improves response to antiretroviral treatment in HIV-1-infected patients, therefore it has recently been implemented into current guidelines for the management of antiretroviral therapy. Knowledge about technologies for measuring drug resistance is important for several reasons: (a) differences exist between different technologies and also between assays based on the same technology; (b) the results of resistance testing are strongly dependent on the reliability and precision of the technology used; and (c) technical aspects have to be considered for a clinically relevant interpretation of drug resistance. The spectrum of genotypic and phenotypic technologies as well as the technical quality is increasing, which shifts the emphasis to the interpretation of resistance profiles. The interpretation is based on the knowledge of drug resistance-associated mutations as well as correlations between genotype and phenotype and clinical response, which are incorporated into rules-based systems. Bioinformatic techiques are used to generate mathematical models for the prediction of drug resistance from genotype. Both approaches are converging toward the prediction of clinical response. Because therapy response is dependent on many additional variables, further efforts are required for the generation of a large clinical database. This will be the basis of a prediction system that will optimize the antiretroviral therapy for each individual patient.     

Transport-Reaction Model of Mural Thrombogenesis: Comparisons of Mathematical Model Predictions and Results from Baboon Models

Thrombogenesis depends on biochemical reactions affected by blood flow dynamics. While mathematical models of mural thrombogenesis provide a means of understanding how blood flow affects thrombus growth, comparisons to experimental data are needed to validate the models and enable prediction of thrombus growth under diverse conditions. In this paper, we present mathematical models of mural thrombogenesis under flow and validation of the models with experimental data collected from a thrombogenic vascular graft segment. The grafts were placed in exteriorized high-flow arteriovenous (AV) shunts in baboons. Radiolabeled platelet deposition onto the thrombogenic segment, a marker of thrombus size, and plasma thrombin-antithrombin (TAT) concentration downstream of the graft, a marker of local thrombin generation, were monitored over time. The mathematical model of mural thrombogenesis consisted of transport-reaction equations in which platelets and thrombin were explicitly considered. We found that the transport-reaction model captured the order of magnitude of TAT sampled levels, while calculated rates of platelet deposition agreed well with radioimaging results. Analysis of experimental and modeling data indicates that, at least during part of thrombus growth progression, thrombin generation is in excess and platelet adhesion rates would be sustained even at lower local thrombin concentrations.     

Lipogranuloma with osseous metaplasia in the breast that developed after "Bu-Hwang" oriental medicine treatment

A lipogranuloma is an inflammatory reactive process associated with exogenous or endogenous lipids, and it's occurrence in the breast has rarely been reported. Osseous metaplasia, which is used to describe bone formation in abnormal locations, can develop from several conditions such as trauma or a tumor. However, few studies have reported benign breast lesions that have been seen as osseous metaplasia. We present a case of a benign calcified breast lesion that developed after a traumatic treatment process called "Bu-Hwang", and it was confirmed as a lipogranuloma with osseous metaplasia. To the best of our knowledge, this is the first reported case of a lipogranuloma with osseous metaplasia in the breast.     

Urinary tract infection of mice to model human disease: Practicalities,implications and limitations

Abstract

Urinary tract infections (UTIs) are among the most common bacterial infections in humans. Murine models of human UTI are vital experimental tools that have helped to elucidate UTI pathogenesis and advance knowledge of potential treatment and infection prevention strategies. Fundamentally, several variables are inherent in different murine models, and understanding the limitations of these variables provides an opportunity to understand how models may be best applied to research aimed at mimicking human disease. In this review, we discuss variables inherent in murine UTI model studies and how these affect model usage, data analysis and data interpretation. We examine recent studies that have elucidated UTI host–pathogen interactions from the perspective of gene expression, and review new studies of biofilm and UTI preventative approaches. We also consider potential standards for variables inherent in murine UTI models and discuss how these might expand the utility of models for mimicking human disease and uncovering new aspects of pathogenesis.