Alexander Zlotnik

Building a Decision Support System for Inpatient Admission Prediction With the Manchester Triage System and Administrative Check-in Variables.

The usage of decision support tools in emergency departments, based on predictive models, capable of estimating the probability of admission for patients in the emergency department may give nursing staff the possibility of allocating resources in advance. We present a methodology for developing and building one such system for a large specialized care hospital using a logistic regression and an artificial neural network model using nine routinely collected variables available right at the end of the triage process.A database of 255.668 triaged nonobstetric emergency department presentations from the Ramon y Cajal University Hospital of Madrid, from January 2011 to December 2012, was used to develop and test the models, with 66% of the data used for derivation and 34% for validation, with an ordered nonrandom partition. On the validation dataset areas under the receiver operating characteristic curve were 0.8568 (95% confidence interval, 0.8508–0.8583) for the logistic regression model and 0.8575 (95% confidence interval, 0.8540–0. 8610) for the artificial neural network model. &khgr;2 Values for Hosmer-Lemeshow fixed “deciles of risk” were 65.32 for the logistic regression model and 17.28 for the artificial neural network model. A nomogram was generated upon the logistic regression model and an automated software decision support system with a Web interface was built based on the artificial neural network model.

Strengthening Web Based Learning through Software Quality Analysis

The Web is changing the way people access & exchange information. Specifically in the teaching & learning environment, we are witnessing that the traditional model of presence based magisterial classes is shifting towards Web Based Learning. This new model draws on remote access systems, knowledge sharing, and student mobility. In this context, pedagogical strategies are also changing, and for instance, Project- Based Learning (PBL) is seen as a potential driver for growth and development in this arena. This study is focused on a PBL oriented course with a Distributed Remote ACcess (DRAC) system. The objective is to analyze how quantitative methods can be leveraged to design and evaluate automatic diagnosis and feedback tools to assist students on quality-related pedagogical issues in DRAC enabled PBL courses. Main conclusions derived from this study are correlation-based and reveal that the development of automatic quality assessment and feedback requires further research.

Prediction of the Degree of Parkinson’s Condition Using Recordings of Patients’ Voices

This paper addresses the estimation of the degree of Parkinson’s Condition (PC) using exclusively the patient’s voice. Firstly, a new database with speech recordings of 25 Spanish patients with different degrees of PC is presented. Secondly, we propose to face this problem as a regression task using machine learning techniques. In particular, utilizing this database, we have developed several systems for predicting the PC degree from a set of acoustic characteristics extracted from the patients’ voice, being the most successful ones, those based on the Support Vector Regression (SVR) algorithm. To determine the optimal way of exploiting the data for our purposes, three kind of experiments have been considered: cross-speaker, leave-one-out-speaker and multi-speaker. From the results, it can be concluded that prediction systems based on acoustic features and machine learning algorithms can be applied for tracking the PC progression if enough validation/training speech data of the patient is available.

Effects of Maraviroc versus Efavirenz in Combination with Zidovudine-Lamivudine on the CD4/CD8 Ratio in Treatment-Naive HIV-Infected Individuals

ABSTRACT A low CD4/CD8 ratio during treated HIV infection reflects heightened immune activation and predicts death. The effects of different antiretroviral therapy regimens on CD4/CD8 ratio recovery remains unclear. We performed a post hoc analysis of the MERIT study, a randomized, double-blind trial of maraviroc versus efavirenz in combination with zidovudine-lamivudine in treatment-naive HIV-infected individuals. We found higher rates of CD4/CD8 ratio normalization with efavirenz, which was driven by a greater CD8+ T-cell decline.

Calculating classifier calibration performance with a custom modification of Weka

Calibration is often overlooked in machine-learning problem-solving approaches, even in situations where an accurate estimation of predicted probabilities, and not only a discrimination between classes, is critical for decision-making. One of the reasons is the lack of readily available open-source software packages which can easily calculate calibration metrics. In order to provide one such tool, we have developed a custom modification of the Weka data mining software, which implements the calculation of Hosmer-Lemeshow groups of risk and the Pearson chi-square statistic comparison between estimated and observed frequencies for binary problems. We provide calibration performance estimations with Logistic regression (LR), BayesNet, Naive Bayes, artificial neural network (ANN), support vector machine (SVM), k-nearest neighbors (KNN), decision trees and Repeated Incremental Pruning to Produce Error Reduction (RIPPER) models with six different datasets. Our experiments show that SVMs with RBF kernels exhibit the best results in terms of calibration, while decision trees, RIPPER and KNN are highly unlikely to produce well-calibrated models.