Arkaitz Artetxe

Bridging challenges of clinical decision support systems with a semantic approach. A case study on breast cancer

The integration of Clinical Decision Support Systems (CDSS) in nowadays clinical environments has not been fully achieved yet. Although numerous approaches and technologies have been proposed since 1960, there are still open gaps that need to be bridged. In this work we present advances from the established state of the art, overcoming some of the most notorious reported difficulties in: (i) automating CDSS, (ii) clinical workflow integration, (iii) maintainability and extensibility of the system, (iv) timely advice, (v) evaluation of the costs and effects of clinical decision support, and (vi) the need of architectures that allow the sharing and reusing of CDSS modules and services. In order to do so, we introduce a new clinical task model oriented to clinical workflow integration, which follows a federated approach. Our work makes use of the reported benefits of semantics in order to fully take advantage of the knowledge present in every stage of clinical tasks and the experience acquired by physicians. In order to introduce a feasible extension of classical CDSS, we present a generic architecture that permits a semantic enhancement, namely Semantic CDSS (S-CDSS). A case study of the proposed architecture in the domain of breast cancer is also presented, pointing some highlights of our methodology.

IMPACT OF REFLEXIVE ONTOLOGIES IN SEMANTIC CLINICAL DECISION SUPPORT SYSTEMS

Ontology processing is arguably a time-consuming process with high associated computational costs. Query actions constitute a crucial part of the reasoning process and are a primary source of time consumption. Reflexive ontologies (ROs) is a novel approach intended to reduce time consumption problems while providing a fast reaction from ontology-based applications. In this article we present the implementation of a knowledge-based clinical decision support system (CDSS) for the diagnosis of Alzheimers disease, which was the benchmark used to evaluate the impact of RO in the overall performance of the system. The implementation details and the definition of the implementation methodology are exposed in this article, along with the results of the evaluation. Some novel techniques that aim to optimize the performance of ROs are also presented with highlights of the test application introduced in our previous work.

Predicting 30-Day Emergency Readmission Risk

Objective: Predicting Emergency Department (ED) readmissions is of great importance since it helps identifying patients requiring further post-discharge attention as well as reducing healthcare costs. It is becoming standard procedure to evaluate the risk of ED readmission within 30 days after discharge. Methods. Our dataset is stratified into four groups according to the Kaiser Permanente Risk Stratification Model. We deal with imbalanced data using different approaches for resampling. Feature selection is also addressed by a wrapper method which evaluates feature set importance by the performance of various classifiers trained on them. Results. We trained a model for each scenario and subpopulation, namely case management (CM), heart failure (HF), chronic obstructive pulmonary disease (COPD) and diabetes mellitus (DM). Using the full dataset we found that the best sensitivity is achieved by SVM using over-sampling methods (40.62 % sensitivity, 78.71 % specificity and 71.94 accuracy). Conclusions. Imbalance correction techniques allow to achieve better sensitivity performance, however the dataset has not enough positive cases, hindering the achievement of better prediction ability. The arbitrary definition of a threshold-based discretization for measurements which are inherently is an important drawback for the exploitation of the data, therefore a regression approach is considered as future work.

Emergency Department Readmission Risk Prediction: A Case Study in Chile

Short time readmission prediction in Emergency Departments (ED) is a valuable tool to improve both the ED management and the healthcare quality. It helps identifying patients requiring further post-discharge attention as well as reducing healthcare costs. As in many other medical domains, patient readmission data is heavily imbalanced, i.e. the minority class is very infrequent, which is a challenge for the construction of accurate predictors using machine learning tools. We have carried computational experiments on a dataset composed of ED admission records spanning more than 100000 patients in 3 years, with a highly imbalanced distribution. We employed various approaches for dealing with this highly imbalanced dataset in combination with different classification algorithms and compared their predictive power for the estimation of the ED readmission probability within 72 h after discharge. Results show that random undersampling and Bagging (RUSBagging) in combination with Random Forest achieves the best results in terms of Area Under ROC Curve (AUC).

Balanced training of a hybrid ensemble method for imbalanced datasets: a case of emergency department readmission prediction

Dealing with imbalanced datasets is a recurrent issue in health-care data processing. Most literature deals with small academic datasets, so that results often do not extrapolate to the large real-life datasets, or have little real-life validity. When minority class sample generation by interpolation is meaningless, the recourse to undersampling the majority class is mandatory in order to reach some acceptable results. Ensembles of classifiers provide the advantage of the diversity of their members, which may allow adaptation to the imbalanced distribution. In this paper, we present a pipeline method combining random undersampling with bootstrap aggregation (bagging) for a hybrid ensemble of extreme learning machines and decision trees, whose diversity improves adaptation to the imbalanced class dataset. The approach is demonstrated on a realistic greatly imbalanced dataset of emergency department patients from a Chilean hospital targeted to predict patient readmission. Computational experiments show that our approach outperforms other well-known classification algorithms.

Design and Development of a Mobile Cardiac Rehabilitation System

In this article we present the design and implementation of a mobile cardiac monitoring system oriented to patients in Phase II and III of cardiac rehabilitation. The complete monitoring system involves both hardware and software design perspectives. At the hardware level, we present a T-shirt with a 12-lead ECG system and an embedded inertial sensor for the monitoring of activity and energy expenditure. At the software level, a modular cloud platform performs data processing to detect relevant cardiac events and to provide advanced visualization capabilities. As a case study, we have implemented our system at the Cardiac Rehabilitation program at Donostia University Hospital (Spain). Finally, the validation of the 12-lead ECG recording system is also presented and discussed.

Predicting 30-day Readmission in Heart Failure using Machine Learning Techniques.

Heart Failure (HF) is a syndrome that reduces patients’ quality of life, and has severe impacts on healthcare systems worldwide, such as the high rate of readmissions. In order to reduce the readmissions and improve patients’ quality of life, several studies are trying to assess the risk of a patient to be readmitted, so that taking right actions clinicians can prevent patient deterioration and readmission. Predictive models have the ability to identify patients at high risk. Henceforth, this paper studies predictive models to determine the risk of a HF patient to be readmitted in the next 30 days after discharge. We present two different approaches. In the first one, we combine unsupervised and supervised classification and achieved AUC score of 0.64. In the second one, we combine decision tree and Naïve Bayes classifiers and achieved AUC score of 0.61. Additionally, we discover that the results improve when training the predictive models with different readmission’s threshold outcome, reaching the AUC score of 0.73 when applying the first approach.

Experience-based SEEG planning: from retrospective data to automated electrode trajectories suggestions

StereoElectroEncephaloGraphy (SEEG) is a minimally invasive technique that consists of the insertion of multiple intracranial electrodes to precisely identify the epileptogenic focus. The planning of electrode trajectories is a cumbersome and time-consuming task. Current approaches to support the planning focus on electrode trajectory optimisation based on geometrical constraints but are not helpful to produce an initial electrode set to begin with the planning procedure. In this work, the authors propose a methodology that analyses retrospective planning data and builds a set of average trajectories, representing the practice of a clinical centre, which can be mapped to a new patient to initialise planning procedure. They collected and analysed the data from 75 anonymised patients, obtaining 30 exploratory patterns and 61 mean trajectories in an average brain space. A preliminary validation on a test set showed that they were able to correctly map 90% of those trajectories and, after optimisation, they have comparable or better values than manual trajectories in terms of distance from vessels and insertion angle. Finally, by detecting and analysing similar plans, they were able to identify eight planning strategies, which represent the main tailored sets of trajectories that neurosurgeons used to deal with the different patient cases.

Heart Failure Readmission or Early Death Risk Factor Analysis: A Case Study in a Telemonitoring Program

Heart Failure (HF) is a clinical syndrome caused by a structural and/or functional cardiac abnormality that imposes tremendous burden on patients and on the healthcare systems worldwide. In this context, predictive models may facilitate the identification of patients at high risk of death or unplanned hospital readmissions and potentially enable direct specific interventions. Currently a plethora of studies in this field is discussing whether hospital readmission and mortality can be effectively predicted in patients with HF. In this work, we present a preliminary study for identifying risk factors for unplanned readmission or death, using a clinical dataset with 119 patients and 60 features. Different classification algorithms and feature selection approaches were employed in order to increase the prediction ability of the models and reduce their complexity in terms of number of features. Results show that sequential feature selection methods along with SVM achieve the best scores in terms of accuracy for predicting 30-day readmission or death risk.

Gaining Insight from Physical Activity Data using a Similarity-based Interactive Visualization

This paper presents a new interactive visualization approach which aims to help and support the user in gaining insight over his physical activity data. The main novelty of the proposed visualization approach is the representation of similarities in the physical activity patterns in time using data clustering techniques, in addition to the continuous physical activity representation over a circular chart. This grouping of similar activity patterns helps identifying meaningful events or behaviors, combined with the periodicity highlighting circular charts. The user is able to interact with the visualization during the knowledge discovery process by changing the represented time-scale, time-frame and the number of clusters used for the user’s physical activity pattern categorization. Additionally, the proposed visualization approach allows to easily report and store the insights gained during the visual data analysis process, by adding a textual description linked to the particular user tailored visualization configuration which led to that insight.