Background: Artificial intelligence (AI)-enabled analysis of 12-lead electrocardiograms (ECGs) may facilitate efficient estimation of incident atrial fibrillation (AF) risk. However, it remains unclear whether AI provides meaningful and generalizable improvement in predictive accuracy beyond clinical risk factors for AF. Methods: We trained a convolutional neural network ("ECG-AI") to infer 5-year incident AF risk using 12-lead ECGs in patients receiving longitudinal primary care at Massachusetts General Hospital (MGH). We then fit three Cox proportional hazards models, each composed of: a) ECG-AI 5-year AF probability, b) the Cohorts for Heart and Aging in Genomic Epidemiology AF (CHARGE-AF) clinical risk score, and c) terms for both ECG-AI and CHARGE-AF ("CH-AI"). We assessed model performance by calculating discrimination (area under the receiver operating characteristic curve, AUROC) and calibration in an internal test set and two external test sets (Brigham and Women's Hospital and UK Biobank). Models were recalibrated to estimate 2-year AF risk in the UK Biobank given limited available follow-up. We used saliency mapping to identify ECG features most influential on ECG-AI risk predictions and assessed correlation between ECG-AI and CHARGE-AF linear predictors. Results: The training set comprised 45,770 individuals (age 55±17 years, 53% women, 2,171 AF events), and the test sets comprised 83,162 individuals (age 59±13 years, 56% women, 2,424 AF events). AUROC was comparable using CHARGE-AF (MGH 0.802, 95% CI 0.767-0.836; BWH 0.752, 95% CI 0.741-0.763; UK Biobank 0.732, 95% CI 0.704-0.759) and ECG-AI (MGH 0.823, 95% CI 0.790-0.856; BWH 0.747, 95% CI 0.736-0.759; UK Biobank 0.705, 95% CI 0.673-0.737). AUROC was highest using CH-AI: MGH 0.838, 95% CI 0.807-0.869; BWH 0.777, 95% CI 0.766-0.788; UK Biobank 0.746, 95% CI 0.716-0.776). Calibration error was low using ECG-AI (MGH 0.0212; BWH 0.0129; UK Biobank 0.0035) and CH-AI (MGH 0.012; BWH 0.0108; UK Biobank 0.0001). In saliency analyses, the ECG P-wave had the greatest influence on AI model predictions. ECG-AI and CHARGE-AF linear predictors were correlated (Pearson r MGH 0.61, BWH 0.66, UK Biobank 0.41). Conclusions: AI-based analysis of 12-lead ECGs has similar predictive utility to a clinical risk factor model for incident AF and both approaches are complementary. ECG-AI may enable efficient quantification of future AF risk.

High-grade gliomas (HGGs) include the most common and the most aggressive primary brain tumor of adults and children. Despite multimodality treatment, most high-grade gliomas eventually recur and are ultimately incurable. Several studies suggest that the initiation, progression, and recurrence of gliomas are driven, at least partly, by cancer stem-like cells. A defining characteristic of these cancer stem-like cells is their capacity to self-renew. We have identified a hypoxia-induced pathway that utilizes the Hypoxia Inducible Factor 1α (HIF-1α) transcription factor and the JAK1/2-STAT3 (Janus Kinase 1/2 - Signal Transducer and Activator of Transcription 3) axis to enhance the self-renewal of glioma stem-like cells. Hypoxia is a commonly found pathologic feature of HGGs. Under hypoxic conditions, HIF-1α levels are greatly increased in glioma stem-like cells. Increased HIF-1α activates the JAK1/2-STAT3 axis and enhances tumor stem-like cell self-renewal. Our data further demonstrate the importance of Vascular Endothelial Growth Factor (VEGF) secretion for this pathway of hypoxia-mediated self-renewal. Brefeldin A and EHT-1864, agents that significantly inhibit VEGF secretion, decreased stem cell self-renewal, inhibited tumor growth, and increased the survival of mice allografted with S100β-v-erbB/p53−/− glioma stem-like cells. These agents also inhibit the expression of a hypoxia gene expression signature that is associated with decreased survival of HGG patients. These findings suggest that targeting the secretion of extracellular, autocrine/paracrine mediators of glioma stem-like cell self-renewal could potentially contribute to the treatment of HGGs.

Electronic health record (EHR) datasets are statistically powerful but are subject to ascertainment bias and missingness. Using the Mass General Brigham multi-institutional EHR, we approximated a community-based cohort by sampling patients receiving longitudinal primary care between 2001-2018 (Community Care Cohort Project [C3PO], n = 520,868). We utilized natural language processing (NLP) to recover vital signs from unstructured notes. We assessed the validity of C3PO by deploying established risk models for myocardial infarction/stroke and atrial fibrillation. We then compared C3PO to Convenience Samples including all individuals from the same EHR with complete data, but without a longitudinal primary care requirement. NLP reduced the missingness of vital signs by 31%. NLP-recovered vital signs were highly correlated with values derived from structured fields (Pearson r range 0.95–0.99). Atrial fibrillation and myocardial infarction/stroke incidence were lower and risk models were better calibrated in C3PO as opposed to the Convenience Samples (calibration error range for myocardial infarction/stroke: 0.012–0.030 in C3PO vs. 0.028–0.046 in Convenience Samples; calibration error for atrial fibrillation 0.028 in C3PO vs. 0.036 in Convenience Samples). Sampling patients receiving regular primary care and using NLP to recover missing data may reduce bias and maximize generalizability of EHR research.