Search

Hemodialysis patients frequently require ambulance transport to the hospital for dialysis. Some patients require urgent dialysis (UD) within 24 hours of transport to hospital to avoid morbidity and mortality. UD is not available in all hospitals; therefore, predicting patients who need UD prior to hospital transport can help paramedics with destination planning. In this paper, we developed machine learning models for paramedics to predict whether a patient needs UD based on patient characteristics available at the time of ambulance transport. This paper presented a study based on ambulance data collected in Halifax, Canada. Given that relatively few patients need UD, a class imbalance problem is addressed by up-sampling methods and prediction models are developed using multiple machine learning methods. The achieved prediction scores are F1-score=0.76, sensitivity=0.76, and specificity=0.97, confirming that models can predict UD with limited patient characteristics. [ABSTRACT FROM AUTHOR]

Voriconazole is a second-generation triazole antifungal agent with strong antifungal activity against a variety of clinically significant pathogens. Controlling blood concentrations within guideline limits through blood concentration monitoring can reduce the probability of hepatotoxicity in patients with voriconazole. However, statistical analysis based on real-world data found that there were still several patients who had blood concentration monitoring developed voriconazole induced hepatotoxicity. Therefore, it has important clinical significance to predict whether hepatotoxicity will occur in patients who meet the guidelines for voriconazole plasma concentration requirements. In this study, based on real-world data, the mixed-effects random forest was used to analyze the electronic medical record data of patients who met the guidelines for voriconazole blood concentration requirements during hospitalization, and a predictive model was constructed to predict whether patients would develop hepatotoxicity within 30 days after using voriconazole. [ABSTRACT FROM AUTHOR]

Continuous intraoperative monitoring with electroencephalo 2 graphy (EEG) is commonly used to detect cerebral ischemia in high-risk surgical procedures such as carotid endarterectomy. Machine learning (ML) models that detect ischemia in real time can form the basis of automated intraoperative EEG monitoring. In this study, we describe and compare two time-series aware precision and recall metrics to the classical precision and recall metrics for evaluating the performance of ML models that detect ischemia. We trained six ML models to detect ischemia in intraoperative EEG and evaluated them with the area under the precision-recall curve (AUPRC) using time-series aware and classical approaches to compute precision and recall. The Support Vector Classification (SVC) model performed the best on the time-series aware metrics, while the Light Gradient Boosting Machine (LGBM) model performed the best on the classical metrics. Visual inspection of the probability outputs of the models alongside the actual ischemic periods revealed that the time-series aware AUPRC selected a model more likely to predict ischemia onset in a timely fashion than the model selected by classical AUPRC. [ABSTRACT FROM AUTHOR]

Supervised predictive models require labeled data for training purposes. Complete and accurate labeled data is not al- ways available, and imperfectly labeled data may need to serve as an alternative. An important question is if the accuracy of the labeled data creates a performance ceiling for the trained model. In this study, we trained several models to recognize the presence of delirium in clinical documents using data with annotations that are not completely accurate. In the external evaluation, the support vector machine model with a linear kernel performed best, achieving an area under the curve of 89.3% and accuracy of 88%, surpassing the 80% accuracy of the training sample. We then generated a set of simulated data and carried out a series of experiments which demonstrated that models trained on imperfect data can (but do not always) outperform the accuracy of the training data. [ABSTRACT FROM AUTHOR]

Estimation of semantic similarity and relatedness between biomedical concepts has utility for many informatics applications. Automated methods fall into two categories: methods based on distributional statistics drawn from text corpora, and methods using the structure of existing knowledge resources. Methods in the former category disregard taxonomic structure, while those in the latter fail to consider semantically relevant empirical information. In this paper, we present a method that retrofits distributional context vector representations of biomedical concepts using structural information from the UMLS Metathesaurus, such that the similarity between vector representations of linked concepts is augmented. We evaluated it on the UMNSRS benchmark. Our results demonstrate that retrofitting of concept vector representations leads to better correlation with human raters for both similarity and relatedness, surpassing the best results reported to date. They also demonstrate a clear improvement in performance on this reference standard for retrofitted vector representations, as compared to those without retrofitting. [ABSTRACT FROM AUTHOR]

Extracting and understanding information, themes and relationships from large collections of documents is an important task for biomedical researchers. Latent Dirichlet Allocation is an unsupervised topic modeling technique using the bag-of-words assumption that has been applied extensively to unveil hidden thematic information within large sets of documents. In this paper, we added MeSH descriptors to the bag-of-words assumption to generate 'hybrid topics', which are mixed vectors of words and descriptors. We evaluated this approach on the quality and interpretability of topics in both a general corpus and a specialized corpus. Our results demonstrated that the coherence of 'hybrid topics' is higher than that of regular bag-of-words topics in the specialized corpus. We also found that the proportion of topics that are not associated with MeSH descriptors is higher in the specialized corpus than in the general corpus. [ABSTRACT FROM AUTHOR]

Automated identification provides an efficient way to categorize patient safety incidents. Previous studies have focused on identifying single incident types relating to a specific patient safety problem, e.g., clinical handover. In reality, there are multiple types of incidents reflecting the breadth of patient safety problems and a single report may describe multiple problems, i.e., it can be assigned multiple type labels. This study evaluated the abilty of multi-label classification methods to identify multiple incident types in single reports. Three multi-label methods were evaluated: binary relevance, classifier chains and ensemble of classifier chains. We found that an ensemble of classifier chains was the most effective method using binary Support Vector Machines with radial basis function kernel and bag-of-words feature extraction, performing equally well on balanced and stratified datasets, (F-score: 73.7% vs. 74.7%). Classifiers were able to identify six common incident types: falls, medications, pressure injury, aggression, documentation problems and others. [ABSTRACT FROM AUTHOR]

Automated wound detection has become a common issue in health care. A broad variety of image processing algorithms already exist, but they are very power consuming on mobile devices. Meanwhile the use of machine learning algorithms is on the rise and new frameworks have been developed to use these techniques with improved on-device-performance such as Apple Core Machine Learning Interface. In this paper, we evaluate the performance of libSVM for wound detection in practice. [ABSTRACT FROM AUTHOR]

We developed an automated toolchain to generate reports of i2b2 data. It is based on free open source software and runs on a Java Application Server. It is sucessfully used in an ED registry project. The solution is highly configurable and portable to other projects based on i2b2 or compatible factual data sources. [ABSTRACT FROM AUTHOR]

Chemical-induced disease relations (CID) are crucial in various biomedical tasks. In the CID task of Biocreative V, no classifiers with multiple kernels have been developed. In this study, a multiple kernel learning-boosting (MKLB) method is proposed. Different kernel functions according to different types of features were constucted and boosted, each of which were learned with multiple kernels. Our multiple kernel learning-boosting (MKLB) method achieved a F1 score of 0.5068 without incorporating knowledge bases. [ABSTRACT FROM AUTHOR]

We have used deep neural networks (DNNs) to generate clinical opinions from general blood test results. DNNs have overfitting problem in general. We believe the complex structure of DNN and insufficient data to be the major reasons of overfitting in our case. In this paper, we apply dropout and batch normalization to avoid overfitting. Experimental results show the improvement in the performance of the DNNs. [ABSTRACT FROM AUTHOR]

This study objectives to investigate a range of Partin table and several machine learning methods for pathological stage prediction and assess them with respect to their predictive model performance based on Koreans data. The data was used SPCDB and gathered records from 944 patients treated with tertiary hospital. Partin table has low accuracy (65.68%) when applied on SPCDB dataset for comparison on patients with OCD and NOCD conditions. SVM (75%) represents a promising alternative to Partin table from which pathology staging can benefit. [ABSTRACT FROM AUTHOR]

This paper presents a data-driven method to study the relationship of survival and clinical information of patients. The machine learning models were established to study the survival situation at the time of interest based on survival analysis. The way to determine the time of interest is an innovation of this paper. The distribution of survival time is considered, namely the three quartiles, as well as the traditional analysis experience is taken into consideration. [ABSTRACT FROM AUTHOR]

Pain is a significant public health problem, affecting an estimated 100 million Americans. Evidence has highlighted that patients with chronic pain often suffer from deficits in pain care quality (PCQ). Efforts to improve PCQ hinge on the identification of reliable PCQ indicators such as pain assessment. In this study, we developed a classifier that leverages narratives in clinical notes to derive indicators of pain assessment for patients with chronic pain. [ABSTRACT FROM AUTHOR]