Paul Walsh

A validated clinical model to predict the need for admission and length of stay in children with acute bronchiolitis.

Objective: To develop and validate a logistic regression model to predict need for admission and length of hospital stay in children presenting to the Emergency Department with bronchiolitis. Setting: Two childrens hospitals in Dublin, Ireland. Methods: We reviewed 118 episodes of bronchiolitis in 99 children admitted from the Emergency Department. Those discharged within 24 h by a consultant/attending paediatrician were retrospectively categorized as suitable for discharge. We then validated the model using a cohort of 182 affected infants from another paediatric Emergency Department in a bronchiolitis season 2 years later. In the validation phase actual admission, failed discharge, and age less than 2 months defined the need for admission. Results: The model predicted admission with 91% sensitivity and 83% specificity in the validation cohort. Age [odds ratio (OR) 0.86, 95% confidence interval (CI) 0.76–0.97], dehydration (OR 2.54, 95% CI 1.34–4.82), increased work of breathing (OR 3.39, 95% CI 1.29–8.92) and initial heart rate above the 97th centile (OR 3.78, 95% CI 1.05–13.57) predicted the need for admission and a longer hospital stay. Conclusion: We derived and validated a severity of illness model for bronchiolitis. This can be used for outcome prediction in decision support tools or severity of illness stratification in research/audit.

An artificial neural network ensemble to predict disposition and length of stay in children presenting with bronchiolitis.

Background: Artificial neural networks apply complex non-linear functions to pattern recognition problems. An ensemble is a ‘committee’ of neural networks that usually outperforms single neural networks. Bronchiolitis is a common manifestation of viral lower respiratory tract infection in infants and toddlers. Objective: To train artificial neural network ensembles to predict the disposition and length of stay in children presenting to the Emergency Department with bronchiolitis. Methods: A specifically constructed database of 119 episodes of bronchiolitis was used to train, validate, and test a neural network ensemble. We used EasyNN 7.0 on a 200 Mhz pentium PC with a maths co-processor. The ensemble of neural networks constructed was subjected to fivefold validation. Comparison with actual and predicted dispositions was measured using the kappa statistic for disposition and the Kaplan–Meier estimations and log rank test for predictions of length of stay. Results: The neural network ensembles correctly predicted disposition in 81% (range 75–90%) of test cases. When compared with actual disposition the neural network performed similarly to a logistic regression model and significantly better than various ‘dumb machine’ strategies with which we compared it. The prediction of length of stay was poorer, 65% (range 60–80%), but the difference between observed and predicted lengths of stay were not significantly different. Conclusion: Artificial neural network ensembles can predict disposition for infants and toddlers with bronchiolitis; however, the prediction of length of hospital stay is not as good.

PATIENTS WOULD PREFER WARD TO EMERGENCY DEPARTMENT BOARDING WHILE AWAITING AN INPATIENT BED

Abstract Boarding of admitted patients in the Emergency Department (ED), rather than in inpatient care areas, is widespread. We surveyed boarded patients, patients without a disposition, and visitors at a county hospital ED serving a mixed urban and rural population. Subjects were asked “If you needed to be admitted to the hospital but no inpatient bed is available, would you prefer to be kept in an ER hallway or a hallway on an inpatient ward?” Boarded patients said they would prefer ward to ED boarding, 117/213 (54.9%; 95% confidence interval [CI] 48.0%–61.7%). Patients without a disposition 314/477 (65.8%; 95% CI 61.4%–70.0%) and visitors 370/532 (69.5%; 95% CI 65.4%–73.4%) stated a preference for ward boarding in 314/477 (65.8%; 95% CI 61.4%–70.0%) and in 370/532 (69.5%; 95% CI 65.4%–73.4%), respectively. Common reasons for preferring inpatient ward boarding were privacy concerns and reduced noise levels. Those preferring ED boarding valued easy access to a doctor.

Explaining the output of ensembles in medical decision support on a case by case basis

The use of ensembles in machine learning (ML) has had a considerable impact in increasing the accuracy and stability of predictors. This increase in accuracy has come at the cost of comprehensibility as, by definition, an ensemble model is considerably more complex than its component models. This is of significance for decision support systems in medicine because of the reluctance to use models that are essentially black boxes. Work on making ensembles comprehensible has so far focused on global models that mirror the behaviour of the ensemble as closely as possible. With such global models there is a clear tradeoff between comprehensibility and fidelity. In this paper, we pursue another tack, looking at local comprehensibility where the output of the ensemble is explained on a case-by-case basis. We argue that this meets the requirements of medical decision support systems. The approach presented here identifies the ensemble members that best fit the case in question and presents the behaviour of these in explanation.

RETENTION OF URETHROVESICAL FOREIGN BODIES: CASE REPORT AND LITERATURE REVIEW

Insertion of foreign bodies into the urethram to obtain sexual pleasure may be complicated by their passage into the urinary bladder along with an inability for the patient to recover the foreign body. We present such a case, review the relevant literature, and discuss Emergency Department (ED) diagnosis and management.

AN EVALUATION OF THE USEFULNESS OF EXPLANATION IN A CASE‐BASED REASONING SYSTEM FOR DECISION SUPPORT IN BRONCHIOLITIS TREATMENT

The research presented here explores the hypothesis that the deployment and acceptance of decision support systems in medicine will be enhanced if the basis for the recommendation produced by the system is apparent. We describe a decision support system for advising on patients suffering from bronchiolitis. This system supports its recommendations with precedent cases selected to support the recommendation along with justification text that highlights aspects of these cases relevant to the query case. It also presents an estimate of its confidence in the recommendation. The main contribution of this paper is an evaluation of this system in a clinical context. The evaluation shows that this type of explanation does enhance the usefulness of the system for practitioners.

Use of V4R in patients who sustain blunt chest trauma

OBJECTIVE In blunt chest trauma, the right ventricle is more vulnerable than the left. The purpose of this study was to determine whether recording V4R in patients with blunt chest trauma would provide additional useful information to that already obtained from the standard 12-lead electrocardiogram (ECG). METHODS Forty-five patients with blunt chest trauma and 40 unmatched control subjects without blunt chest trauma had standard 12-lead ECG and right precordial leads recorded. The ECGs were read blindly by three physicians. RESULTS Patients with chest trauma were distinguishable from controls on the basis of the left-sided ECGs (odds ratio, 2.9; 95% confidence interval, 1.71-4.90). This was not the case using V4R (odds ratio, 1.23; 95% confidence interval, 0.59-2.0). CONCLUSION Patients with a significant mechanism and physical findings of blunt chest trauma were more likely than controls to have an abnormal ECG. They were not more likely to have abnormalities in V4R. We recommend that a 12-lead ECG, but not V4R, be routinely obtained on these patients.

Is the interpretation of rapid antigen testing for respiratory syncytial virus as simple as positive or negative

Objective To measure the performance characteristics of an immunochromatographic rapid antigen test for respiratory syncytial virus (RSV) and determine how its interpretation should be contextualised in patients presenting to the emergency department (ED) with bronchiolitis. Design Diagnostic accuracy study of a rapid RSV test. Setting County hospital ED. Intervention We took paired nasal samples from consecutively enrolled infants with bronchiolitis and tested them with a rapid immunochromatographic antigen test and reverse transcriptase PCR gold standard. Outcome measures Sensitivity, specificity, the effect of point prevalence, clinical findings and overall context on predictive values. We used these to construct a graphical contextual model to show how the results of RSV antigen tests from infants presenting within 24 h should influence interpretation of subsequent antigen tests. Results We analysed 607 patients. The sensitivity and specificity for immunochromatographic testing was 79.4% (95% CI 73.9% to 84.2%) and 67.1% (95% CI 61.9% to 72%) respectively. We found little evidence of spectrum bias. In our contextual model the best predictor of a positive RT-PCR test was a positive antigen test OR 5.47 (95% CI 3.65 to 8.18) and the number of other infants having positive tests within 24 h OR 1.48 (95% CI 1.26 to 1.72) per infant. Increasing numbers presenting to the ED with bronchiolitis in a given day increases the probability of RSV infection. Conclusions The RSV antigen test we examined had modest performance characteristics. The results of the antigen test should be interpreted in the context of the results of previous tests.

American Academy of Pediatrics 2014 Bronchiolitis Guidelines: Bonfire of the Evidence

N/A Supporting material: Figure Copyright Information: Copyright 2015 by the article author(s). This work is made available under the terms of the Creative Commons Attribution4.0 none 4.0 4.0 license, http://creativecommons.org/licenses/by/4.0 none 4.0 4.0/

Explaining Predictions from a Neural Network Ensemble One at a Time

This paper introduces a new method for explaining the predictions of ensembles of neural networks on a case by case basis. The approach of explaining individual examples differs from much of the current research which focuses on producing a global model of the phenomenon under investigation. Explaining individual results is accomplished by modelling each of the networks as a rule-set and computing the resulting coverage statistics for each rule given the data used to train the network. This coverage information is then used to choose the rule or rules that best describe the example under investigation. This approach is based on the premise that ensembles perform an implicit problem space decomposition with ensemble members specialising in different regions of the problem space. Thus explaining an ensemble involves explaining the ensemble members that best fit the example.