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.

Recent advancements in computing have led to the development of artificial intelligence (AI) enabled healthcare technologies. AI-assisted clinical decision support (CDS) integrated into electronic health records (EHR) was demonstrated to have a significant potential to improve clinical care. With the rapid proliferation of AI-assisted CDS, came the realization that a lack of careful consideration of socio-technical issues surrounding the implementation and maintenance of these tools can result in unanticipated consequences, missed opportunities, and suboptimal uptake of these potentially useful technologies. The 48-h Discharge Prediction Tool (48DPT) is a new AI-assisted EHR CDS to facilitate discharge planning. This study aimed to methodologically assess the implementation of 48DPT and identify the barriers and facilitators of adoption and maintenance using the validated implementation science frameworks. The major dimensions of RE-AIM (Reach, Effectiveness, Adoption, Implementation, Maintenance) and the constructs of the Consolidated Framework for Implementation Research (CFIR) frameworks have been used to analyze interviews of 24 key stakeholders using 48DPT. The systematic assessment of the 48DPT implementation allowed us to describe facilitators and barriers to implementation such as lack of awareness, lack of accuracy and trust, limited accessibility, and transparency. Based on our evaluation, the factors that are crucial for the successful implementation of AI-assisted EHR CDS were identified. Future implementation efforts of AI-assisted EHR CDS should engage the key clinical stakeholders in the AI tool development from the very inception of the project, support transparency and explainability of the AI models, provide ongoing education and onboarding of the clinical users, and obtain continuous input from clinical staff on the CDS performance.

Introduction: Racial and ethnic categories are frequently used in pharmacogenetics literature to stratify patients; however, these categories can be inconsistent across different studies. To address the ongoing debate on the applicability of traditional concepts of race and ethnicity in the context of precision medicine, we aimed to review the application of current racial and ethnic categories in pharmacogenetics and its potential impact on clinical care. Methods: One hundred and three total pharmacogenetics papers involving the CYP2C9, CYP2C19, and CYP2D6 genes were analyzed for their country of origin, racial, and ethnic categories used, and allele frequency data. Correspondence between the major continental racial categories promulgated by National Institutes of Health (NIH) and those reported by the pharmacogenetics papers was evaluated. Results: The racial and ethnic categories used in the papers we analyzed were highly heterogeneous. In total, we found 66 different racial and ethnic categories used which fall under the NIH race category “White”, 47 different racial and ethnic categories for “Asian”, and 62 different categories for “Black”. The number of categories used varied widely based on country of origin: Japan used the highest number of different categories for “White” with 17, Malaysia used the highest number for “Asian” with 24, and the US used the highest number for “Black” with 28. Significant variation in allele frequency between different ethnic subgroups was identified within 3 major continental racial categories. Conclusion: Our analysis showed that racial and ethnic classification is highly inconsistent across different papers as well as between different countries. Evidence-based consensus is necessary for optimal use of self-identified race as well as geographical ancestry in pharmacogenetics. Common taxonomy of geographical ancestry which reflects specifics of particular countries and is accepted by the entire scientific community can facilitate reproducible pharmacogenetic research and clinical implementation of its results.

Summary: There has been a call for evidence‐based oral healthcare guidelines, to improve precision dentistry and oral healthcare delivery. The main challenges to this goal are the current lack of up‐to‐date evidence, the limited integrative analytical data sets, and the slow translations to routine care delivery. Overcoming these issues requires knowledge discovery pipelines based on big data and health analytics, intelligent integrative informatics approaches, and learning health systems. This article examines how this can be accomplished by utilizing big data. These data can be gathered from four major streams: patients, clinical data, biological data, and normative data sets. All these must then be uniformly combined for analysis and modelling and the meaningful findings can be implemented clinically. By executing data capture cycles and integrating the subsequent findings, practitioners are able to improve public oral health and care delivery.