Background COVID-19 has infected millions of people worldwide and is responsible for several hundred thousand fatalities. The COVID-19 pandemic has necessitated thoughtful resource allocation and early identification of high-risk patients. However, effective methods to meet these needs are lacking. Objective The aims of this study were to analyze the electronic health records (EHRs) of patients who tested positive for COVID-19 and were admitted to hospitals in the Mount Sinai Health System in New York City; to develop machine learning models for making predictions about the hospital course of the patients over clinically meaningful time horizons based on patient characteristics at admission; and to assess the performance of these models at multiple hospitals and time points. Methods We used Extreme Gradient Boosting (XGBoost) and baseline comparator models to predict in-hospital mortality and critical events at time windows of 3, 5, 7, and 10 days from admission. Our study population included harmonized EHR data from five hospitals in New York City for 4098 COVID-19–positive patients admitted from March 15 to May 22, 2020. The models were first trained on patients from a single hospital (n=1514) before or on May 1, externally validated on patients from four other hospitals (n=2201) before or on May 1, and prospectively validated on all patients after May 1 (n=383). Finally, we established model interpretability to identify and rank variables that drive model predictions. Results Upon cross-validation, the XGBoost classifier outperformed baseline models, with an area under the receiver operating characteristic curve (AUC-ROC) for mortality of 0.89 at 3 days, 0.85 at 5 and 7 days, and 0.84 at 10 days. XGBoost also performed well for critical event prediction, with an AUC-ROC of 0.80 at 3 days, 0.79 at 5 days, 0.80 at 7 days, and 0.81 at 10 days. In external validation, XGBoost achieved an AUC-ROC of 0.88 at 3 days, 0.86 at 5 days, 0.86 at 7 days, and 0.84 at 10 days for mortality prediction. Similarly, the unimputed XGBoost model achieved an AUC-ROC of 0.78 at 3 days, 0.79 at 5 days, 0.80 at 7 days, and 0.81 at 10 days. Trends in performance on prospective validation sets were similar. At 7 days, acute kidney injury on admission, elevated LDH, tachypnea, and hyperglycemia were the strongest drivers of critical event prediction, while higher age, anion gap, and C-reactive protein were the strongest drivers of mortality prediction. Conclusions We externally and prospectively trained and validated machine learning models for mortality and critical events for patients with COVID-19 at different time horizons. These models identified at-risk patients and uncovered underlying relationships that predicted outcomes.

Drug–drug interactions (DDIs) are an emerging threat to public health. Recent estimates indicate that DDIs cause nearly 74 000 emergency room visits and 195 000 hospitalizations each year in the USA. Current approaches to DDI discovery, which include Phase IV clinical trials and post-marketing surveillance, are insufficient for detecting many DDIs and do not alert the public to potentially dangerous DDIs before a drug enters the market. Recent work has applied state-of-the-art computational and statistical methods to the problem of DDIs. Here we review recent developments that encompass a range of informatics approaches in this domain, from the construction of databases for efficient searching of known DDIs to the prediction of novel DDIs based on data from electronic medical records, adverse event reports, scientific abstracts, and other sources. We also explore why DDIs are so difficult to detect and what the future holds for informatics-based approaches to DDI discovery.

Hip fractures are a leading cause of death and disability among older adults. Hip fractures are also the most commonly missed diagnosis on pelvic radiographs, and delayed diagnosis leads to higher cost and worse outcomes. Computer-aided diagnosis (CAD) algorithms have shown promise for helping radiologists detect fractures, but the image features underpinning their predictions are notoriously difficult to understand. In this study, we trained deep-learning models on 17,587 radiographs to classify fracture, 5 patient traits, and 14 hospital process variables. All 20 variables could be individually predicted from a radiograph, with the best performances on scanner model (AUC = 1.00), scanner brand (AUC = 0.98), and whether the order was marked “priority” (AUC = 0.79). Fracture was predicted moderately well from the image (AUC = 0.78) and better when combining image features with patient data (AUC = 0.86, DeLong paired AUC comparison, p = 2e-9) or patient data plus hospital process features (AUC = 0.91, p = 1e-21). Fracture prediction on a test set that balanced fracture risk across patient variables was significantly lower than a random test set (AUC = 0.67, DeLong unpaired AUC comparison, p = 0.003); and on a test set with fracture risk balanced across patient and hospital process variables, the model performed randomly (AUC = 0.52, 95% CI 0.46–0.58), indicating that these variables were the main source of the model’s fracture predictions. A single model that directly combines image features, patient, and hospital process data outperforms a Naive Bayes ensemble of an image-only model prediction, patient, and hospital process data. If CAD algorithms are inexplicably leveraging patient and process variables in their predictions, it is unclear how radiologists should interpret their predictions in the context of other known patient data. Further research is needed to illuminate deep-learning decision processes so that computers and clinicians can effectively cooperate.

MotivationThe biomedical community’s collective understanding of how chemicals, genes and phenotypes interact is distributed across the text of over 24 million research articles. These interactions offer insights into the mechanisms behind higher order biochemical phenomena, such as drug-drug interactions and variations in drug response across individuals. To assist their curation at scale, we must understand what relationship types are possible and map unstructured natural language descriptions onto these structured classes. We used NCBI’s PubTator annotations to identify instances of chemical, gene and disease names in Medline abstracts and applied the Stanford dependency parser to find connecting dependency paths between pairs of entities in single sentences. We combined a published ensemble biclustering algorithm (EBC) with hierarchical clustering to group the dependency paths into semantically-related categories, which we annotated with labels, or ‘themes’ (‘inhibition’ and ‘activation’, for example). We evaluated our theme assignments against six human-curated databases: DrugBank, Reactome, SIDER, the Therapeutic Target Database, OMIM and PharmGKB.ResultsClustering revealed 10 broad themes for chemical-gene relationships, 7 for chemical-disease, 10 for gene-disease and 9 for gene–gene. In most cases, enriched themes corresponded directly to known database relationships. Our final dataset, represented as a network, contained 37 491 thematically-labeled chemical-gene edges, 2 021 192 chemical-disease edges, 136 206 gene-disease edges and 41 418 gene–gene edges, each representing a single-sentence description of an interaction from somewhere in the literature.Availability and implementationThe complete network is available on Zenodo (https://zenodo.org/record/1035500). We have also provided the full set of dependency paths connecting biomedical entities in Medline abstracts, with associated sentences, for future use by the biomedical research community.Supplementary information Supplementary data are available at Bioinformatics online.