Graph neural networks for clinical risk prediction based on electronic health records: A survey.

OBJECTIVE: This study aims to comprehensively review the use of graph neural networks (GNNs) for clinical risk prediction based on electronic health records (EHRs). The primary goal is to provide an overview of the state-of-the-art of this subject, highlighting ongoing research efforts and identifying existing challenges in developing effective GNNs for improved prediction of clinical risks. METHODS: A search was conducted in the Scopus, PubMed, ACM Digital Library, and Embase databases to identify relevant English-language papers that used GNNs for clinical risk prediction based on EHR data. The study includes original research papers published between January 2009 and May 2023. RESULTS: Following the initial screening process, 50 articles were included in the data collection. A significant increase in publications from 2020 was observed, with most selected papers focusing on diagnosis prediction (n = 36). The study revealed that the graph attention network (GAT) (n = 19) was the most prevalent architecture, and MIMIC-III (n = 23) was the most common data resource. CONCLUSION: GNNs are relevant tools for predicting clinical risk by accounting for the relational aspects among medical events and entities and managing large volumes of EHR data. Future studies in this area may address challenges such as EHR data heterogeneity, multimodality, and model interpretability, aiming to develop more holistic GNN models that can produce more accurate predictions, be effectively implemented in clinical settings, and ultimately improve patient care.

Deep learning prediction models based on EHR trajectories: A systematic review.

BACKGROUND: Electronic health records (EHRs) are generated at an ever-increasing rate. EHR trajectories, the temporal aspect of health records, facilitate predicting patients' future health-related risks. It enables healthcare systems to increase the quality of care through early identification and primary prevention. Deep learning techniques have shown great capacity for analyzing complex data and have been successful for prediction tasks using complex EHR trajectories. This systematic review aims to analyze recent studies to identify challenges, knowledge gaps, and ongoing research directions. METHODS: For this systematic review, we searched Scopus, PubMed, IEEE Xplore, and ACM databases from Jan 2016 to April 2022 using search terms centered around EHR, deep learning, and trajectories. Then the selected papers were analyzed according to publication characteristics, objectives, and their solutions regarding existing challenges, such as the model's capacity to deal with intricate data dependencies, data insufficiency, and explainability. RESULTS: After removing duplicates and out-of-scope papers, 63 papers were selected, which showed rapid growth in the number of research in recent years. Predicting all diseases in the next visit and the onset of cardiovascular diseases were the most common targets. Different contextual and non-contextual representation learning methods are employed to retrieve important information from the sequence of EHR trajectories. Recurrent neural networks and the time-aware attention mechanism for modeling long-term dependencies, self-attentions, convolutional neural networks, graphs for representing inner visit relations, and attention scores for explainability were frequently used among the reviewed publications. CONCLUSIONS: This systematic review demonstrated how recent breakthroughs in deep learning methods have facilitated the modeling of EHR trajectories. Research on improving the ability of graph neural networks, attention mechanisms, and cross-modal learning to analyze intricate dependencies among EHRs has shown good progress. There is a need to increase the number of publicly available EHR trajectory datasets to allow for easier comparison among different models. Also, very few developed models can handle all aspects of EHR trajectory data.

Ventricular tachycardia caused by mesothelial cyst.

A 27 year-old- lady was evaluated due to recurrent ventricular tachycardia. After performing echocardiography and cardiac MRI, she was found to have large pericardial cyst. Pathologic examination confirmed it as mesothelial pericardial cyst. Up to our knowledge it is the first presentation of simple pericardial cyst as ventricular a tachycardia.

A Masked Language Model for Multi-Source EHR Trajectories Contextual Representation Learning.

Using electronic health records data and machine learning to guide future decisions needs to address challenges, including 1) long/short-term dependencies and 2) interactions between diseases and interventions. Bidirectional transformers have effectively addressed the first challenge. Here we tackled the latter challenge by masking one source (e.g., ICD10 codes) and training the transformer to predict it using other sources (e.g., ATC codes).

Developing a novel prediction model in opioid overdose using machine learning; a pilot analytical study.

Background and Aims: The opioid epidemic has extended to many countries. Data regarding the accuracy of conventional prediction models including the Simplified Acute Physiologic Score (SAPS) II and acute physiology and chronic health evaluation (APACHE) II are scarce in opioid overdose cases. We evaluate the efficacy of adding quantitative electroencephalogram (qEEG) data to clinical and paraclinical data in the prediction of opioid overdose mortality using machine learning. Methods: In a prospective study, we collected clinical/paraclinical, and qEEG data of 32 opioid-poisoned patients. After preprocessing and Fast Fourier Transform analysis, absolute power was computed. Also, SAPS II was calculated. Eventually, data analysis was performed using SAPS II as a benchmark at three levels to predict the patient's course in comparison with SAPS II. First, the qEEG data set was used alone, secondly, the combination of the clinical/paraclinical, SAPS II, qEEG datasets, and the SAPS II-based model was included in the pool of classifier models. Results: Seven out of 32 (22%) died. SAPS II (cut-off of 50.5) had a sensitivity/specificity/positive/negative predictive values of 85.7%, 84.0%, 60.0%, and 95.5% in predicting mortality, respectively. Adding majority voting on random forest with qEEG and clinical data, improved the model sensitivity, specificity, and positive and negative predictive values to 71.4%, 96%, 83.3%, and 92.3% (not significant). The model fusion level has 40% less prediction error. Conclusion: Considering the higher specificity and negative predictive value in our proposed model, it could predict survival much better than mortality. The model would constitute an indicator for better care of opioid poisoned patients in low resources settings, where intensive care unit beds are limited.

Mid-Term Patency in Radial Artery and Saphenous Vein After Coronary Artery Bypass Grafting in Asymptomatic Patients Using 128-Slice CT Coronary Angiography.

BACKGROUND: Patency of the revascularization conduit is an essential predictor of long-standing survival after coronary artery bypass grafting. OBJECTIVES: We have conducted this study to compare the mid-term patency rates of radial artery (RA), left internal thoracic artery (LITA) and also saphenous vein (SV) grafts in asymptomatic patients following coronary artery bypass graft surgery (CABG) undergoing total IV anesthesia. PATIENTS AND METHODS: In this study, 30 three-vessel disease patients with 104 RA, LITA, and SV grafts used concomitantly for primary isolated non-emergent CABG surgery were assessed. The primary end point was CT angiographic graft patency rate. After 53.5 (24-97) months' follow-up, graft patency was assessed using 128-slice CT coronary angiography. Logistic regression analysis was used to detect the independent predictors of graft failure. RESULTS: A total of 104 grafts, including 30 LITA, 44 SV, and 30 RA grafts, were studied. Cumulative graft patency rates were 93.3% in LITA, 83.3% in RA, and 70.5% in SV grafts. Statistically significant difference was found between the LITA and the SV graft patency rates (P = 0.019), whereas the difference between the RA conduit patency and the LITA or SV graft patency rates did not have any statistical significance (P = 0.424 and P = 0.273, respectively). Independent predictors of RA grafts occlusion were native coronary stenosis < 70% and female gender. CONCLUSIONS: In our patients, the RA grafts had an acceptable patency rate in 2 to 5 years' follow-up. Although the SV grafts had a relatively higher patency rate than RA grafts in our asymptomatic patients, the patency rates in RA and SV grafts were close to each other. The RA graft function was poor in the patients with a higher number of risk factors and in the females.