Prediction of emergency department revisits among child and youth mental health outpatients using deep learning techniques.

BACKGROUND: The proportion of Canadian youth seeking mental health support from an emergency department (ED) has risen in recent years. As EDs typically address urgent mental health crises, revisiting an ED may represent unmet mental health needs. Accurate ED revisit prediction could aid early intervention and ensure efficient healthcare resource allocation. We examine the potential increased accuracy and performance of graph neural network (GNN) machine learning models compared to recurrent neural network (RNN), and baseline conventional machine learning and regression models for predicting ED revisit in electronic health record (EHR) data. METHODS: This study used EHR data for children and youth aged 4-17 seeking services at McMaster Children's Hospital's Child and Youth Mental Health Program outpatient service to develop and evaluate GNN and RNN models to predict whether a child/youth with an ED visit had an ED revisit within 30 days. GNN and RNN models were developed and compared against conventional baseline models. Model performance for GNN, RNN, XGBoost, decision tree and logistic regression models was evaluated using F1 scores. RESULTS: The GNN model outperformed the RNN model by an F1-score increase of 0.0511 and the best performing conventional machine learning model by an F1-score increase of 0.0470. Precision, recall, receiver operating characteristic (ROC) curves, and positive and negative predictive values showed that the GNN model performed the best, and the RNN model performed similarly to the XGBoost model. Performance increases were most noticeable for recall and negative predictive value than for precision and positive predictive value. CONCLUSIONS: This study demonstrates the improved accuracy and potential utility of GNN models in predicting ED revisits among children and youth, although model performance may not be sufficient for clinical implementation. Given the improvements in recall and negative predictive value, GNN models should be further explored to develop algorithms that can inform clinical decision-making in ways that facilitate targeted interventions, optimize resource allocation, and improve outcomes for children and youth.

Data-driven approach to quantify trust in medical devices using Bayesian networks.

Bayesian networks are increasingly used to quantify the uncertainty of subjective and stochastic concepts such as trust. In this article, we propose a data-driven approach to estimate Bayesian parameters in the domain of wearable medical devices. Our approach extracts the probability of a trust factor being in a specific state directly from the devices (e.g. sensor quality). The strength of the relationship between related factors is defined by expert knowledge and incorporated into the model. We use propagation rules from requirements engineering to estimate how much each trust factor contributes to the related intermediate nodes in the network and ultimately compute the trust score. The trust score is a relative measure of trustworthiness when different devices are evaluated in the same test conditions and using the same Bayesian structure. To evaluate our approach, we developed Bayesian networks for the trust quantification of similar wearable devices from two manufacturers under identical test conditions and noise levels. The results demonstrated the learnability and generalizability of our approach.

Evaluation methodology for deep learning imputation models.

There is growing interest in imputing missing data in tabular datasets using deep learning. Existing deep learning-based imputation models have been commonly evaluated using root mean square error (RMSE) as the predictive accuracy metric. In this article, we investigate the limitations of assessing deep learning-based imputation models by conducting a comparative analysis between RMSE and alternative metrics in the statistical literature including qualitative, predictive accuracy, statistical distance, and descriptive statistics. We design a new aggregated metric, called reconstruction loss (RL), to evaluate deep learning-based imputation models. We also develop and evaluate a novel imputation evaluation methodology based on RL. To minimize model and dataset biases, we use a regression imputation model and two different deep learning imputation models: denoising autoencoders and generative adversarial nets. We also use two tabular datasets from different industry sectors: health care and financial. Our results show that the proposed methodology is effective in evaluating multiple properties of the deep learning-based imputation model's reconstruction performance.

From Group-Level Statistics to Single-Subject Prediction: Machine Learning Detection of Concussion in Retired Athletes.

There has been increased effort to understand the neurophysiological effects of concussion aimed to move diagnosis and identification beyond current subjective behavioral assessments that suffer from poor sensitivity. Recent evidence suggests that event-related potentials (ERPs) measured with electroencephalography (EEG) are persistent neurophysiological markers of past concussions. However, as such evidence is limited to group-level analyzes, the extent to which they enable concussion detection at the individual-level is unclear. One promising avenue of research is the use of machine learning to create quantitative predictive models that can detect prior concussions in individuals. In this paper, we translate the recent group-level findings from ERP studies of concussed individuals into a machine learning framework for performing single-subject prediction of past concussion. We found that a combination of statistics of single-subject ERPs and wavelet features yielded a classification accuracy of 81% with a sensitivity of 82% and a specificity of 80%, improving on current practice. Notably, the model was able to detect concussion effects in individuals who sustained their last injury as much as 30 years earlier. However, failure to detect past concussions in a subset of individuals suggests that the clear effects found in group-level analyses may not provide us with a full picture of the neurophysiological effects of concussion.

An Environmental Scan of the National and Provincial Diagnostic Reference Levels in Canada for Common Adult Computed Tomography Scans.

Several regulatory bodies have agreed that low-dose radiation used in medical imaging is a weak carcinogen that follows a linear, non-threshold model of cancer risk. While avoiding radiation is the best course of action to mitigate risk, computed tomography (CT) scans are often critical for diagnosis. In addition to the as low as reasonably achievable principle, a more concrete method of dose reduction for common CT imaging exams is the use of a diagnostic reference level (DRL). This paper examines Canada's national DRL values from the recent CT survey and compares it to published provincial DRLs as well as the DRLs in the United Kingdom and the United States of America for the 3 most common CT exams: head, chest, and abdomen/pelvis. Canada compares well on the international scale, but it should consider using more electronic dose monitoring solutions to create a culture of dose optimization.

Machine Learning and Mobile Health Monitoring Platforms: A Case Study on Research and Implementation Challenges.

Machine learning-based patient monitoring systems are generally deployed on remote servers for analyzing heterogeneous data. While recent advances in mobile technology provide new opportunities to deploy such systems directly on mobile devices, the development and deployment challenges are not being extensively studied by the research community. In this paper, we systematically investigate challenges associated with each stage of the development and deployment of a machine learning-based patient monitoring system on a mobile device. For each class of challenges, we provide a number of recommendations that can be used by the researchers, system designers, and developers working on mobile-based predictive and monitoring systems. The results of our investigation show that when developers are dealing with mobile platforms, they must evaluate the predictive systems based on its classification and computational performance. Accordingly, we propose a new machine learning training and deployment methodology specifically tailored for mobile platforms that incorporates metrics beyond traditional classifier performance.