Nicolás Serrano

Using Bayesian networks in the construction of a bi-level multi-classifier. A case study using intensive care unit patients data

Combining the predictions of a set of classifiers has shown to be an effective way to create composite classifiers that are more accurate than any of the component classifiers. There are many methods for combining the predictions given by component classifiers. We introduce a new method that combine a number of component classifiers using a Bayesian network as a classifier system given the component classifiers predictions. Component classifiers are standard machine learning classification algorithms, and the Bayesian network structure is learned using a genetic algorithm that searches for the structure that maximises the classification accuracy given the predictions of the component classifiers. Experimental results have been obtained on a datafile of cases containing information about ICU patients at Canary Islands University Hospital. The accuracy obtained using the presented new approach statistically improve those obtained using standard machine learning methods.

Serum soluble CD40 Ligand levels are associated with severity and mortality of brain trauma injury patients

BACKGROUND Serum soluble CD40 Ligand (sCD40L) levels, which exhibit prothrombotic and proinflammatory properties, have not been studied in patients with traumatic brain injury (TBI). Thus, the objective of this study was to determine whether serum sCD40L levels are associated with severity and mortality in patients with severe TBI. METHODS This was a prospective, observational and multicenter study carried out in six Spanish Intensive Care Units. Patients with severe TBI defined as Glasgow Coma Scale (GCS) lower than 9 were included, while those with Injury Severity Score (ISS) in non-cranial aspects higher than 9 were excluded. Serum levels of sCD40L were measured on the day of TBI. Endpoint was established in 30-day mortality. RESULTS We found higher serum sCD40L levels (P<0.001) in non-surviving TBI patients (N=27) than in survivor ones (N=73). Logistic regression analysis showed that serum sCD40L levels were associated with 30-day mortality (OR=1.58; 95% CI=1.12-2.21; P=0.008) controlling for APACHE-II score and computer tomography findings. The area under the curve (AUC) for serum sCD40L levels as predictor of 30-day mortality was 0.79 (95% CI=0.70-0.86; P<0.001). Survival analysis showed that patients with serum sCD40L levels higher than 2.11 ng/mL presented increased 30-day mortality than patients with lower levels (Hazard ratio=9.0; 95% CI=4.25-19.27; P<0.001). We found an association between serum sCD40L levels and APACHE-II (rho=0.33; P=0.001), and GCS score (rho=-0.21; P=0.04). CONCLUSIONS To our knowledge, this is the first study reporting data on serum sCD40L levels in patients with severe TBI. The most relevant and newer findings of our study are that serum sCD40L levels in non-surviving patients with severe TBI are higher than in surviving ones, and that there are an association between serum sCD40L levels and TBI severity and mortality.

Acute adrenal insufficiency after cardiac surgery.

Adrenal insufficiency after cardiac surgery can easily be confused during the course of an immediate unstable postoperative period. If unrecognized, this condition may cause serious morbidity and can be fatal. We report on a 43-yr-old female patient with chronic known adrenal insufficiency, who, despite her adequate preoperative replacement therapy, presented with one episode of acute hypoadrenal crisis after elective open heart surgery, which could serve as a model to illustrate the salient clinical features and possible problems in this setting for diagnosing this problem to patients in whom chronic adrenal insufficiency remains unknown.

Machine Learning Inspired Approaches to Combine Standard Medical Measures at an Intensive Care Unit

There are many standard methods used at Intensive Care Units (ICU) in order to overview patients situation. We present in this paper a new method that outperforms the prediction accuracy of each medical standard method by combining them using Machine Learning (ML) inspired classification approaches. We have used different Machine Learning algorithms to compare the accuracy of our new method with other existing approaches used by ML community. The new method is an hybrid made between the Nearest Neighbour and the Naive Bayes classification methods. Experimental results show that this new approach is better than any standard method used in the prediction of survival of ICU patients, and better than the combination of these medical approaches done by using standard ML algorithms.

Ether suicide poisoning by intravenous injection.

cortex or the brain stem [4]. There was no hypoglycemia contrary to reports of propranolol intoxication in children [2]. This is the first report, to our knowledge, of such a pure form without dream-like features [3]. Propranolol-related low cardiac output could be responsible for transient hypoxic consequences in the peduncular areas [1±3]. In our case a direct pharmacological effect of this lipid-soluble beta-blocker that crosses the brain-blood barrier would explain the hallucinosis because of the history, the effectiveness of the antidote with the disappearance of hallucinations before the haemodynamic improvement and the membrane stabilizing effect of propranolol [1±3]. A serum level of 1,324 nmol/l two hours after exposure accounted for major propranolol intoxication [1]. The direct neurologic effectiveness of glucagon might be explained by the presence of glucagon receptors in some region of the brain and glucagon crossing the brain-blood barrier [5].