This article presents a scalable and accurate deep learning approach for predictive modeling using electronic health records (EHRs). The authors propose a representation of patients' entire raw EHR records based on the Fast Healthcare Interoperability Resources (FHIR) format, which allows deep learning models to accurately predict multiple medical events without site-specific data harmonization. The approach was validated using de-identified EHR data from two US academic medical centers with 216,221 adult patients hospitalized for at least 24 hours. The data were unrolled into a total of 46,864,534,945 data points, including clinical notes. Deep learning models achieved high accuracy for tasks such as predicting in-hospital mortality (AUROC 0.93–0.94), 30-day unplanned readmission (AUROC 0.75–0.76), prolonged length of stay (AUROC 0.85–0.86), and all of a patient's final discharge diagnoses (frequency-weighted AUROC 0.90). These models outperformed traditional, clinically-used predictive models in all cases. The authors also demonstrated that neural networks can be used to identify relevant information from the patient's chart. The study highlights the potential of deep learning to improve healthcare quality and personalize medicine by leveraging the vast amount of information in EHRs. The approach is scalable and can be applied to a variety of clinical scenarios. The study also discusses the challenges of using EHR data for predictive modeling, including the need for data standardization and the complexity of the data. The authors conclude that their approach can be used to create accurate and scalable predictions for a variety of clinical scenarios.This article presents a scalable and accurate deep learning approach for predictive modeling using electronic health records (EHRs). The authors propose a representation of patients' entire raw EHR records based on the Fast Healthcare Interoperability Resources (FHIR) format, which allows deep learning models to accurately predict multiple medical events without site-specific data harmonization. The approach was validated using de-identified EHR data from two US academic medical centers with 216,221 adult patients hospitalized for at least 24 hours. The data were unrolled into a total of 46,864,534,945 data points, including clinical notes. Deep learning models achieved high accuracy for tasks such as predicting in-hospital mortality (AUROC 0.93–0.94), 30-day unplanned readmission (AUROC 0.75–0.76), prolonged length of stay (AUROC 0.85–0.86), and all of a patient's final discharge diagnoses (frequency-weighted AUROC 0.90). These models outperformed traditional, clinically-used predictive models in all cases. The authors also demonstrated that neural networks can be used to identify relevant information from the patient's chart. The study highlights the potential of deep learning to improve healthcare quality and personalize medicine by leveraging the vast amount of information in EHRs. The approach is scalable and can be applied to a variety of clinical scenarios. The study also discusses the challenges of using EHR data for predictive modeling, including the need for data standardization and the complexity of the data. The authors conclude that their approach can be used to create accurate and scalable predictions for a variety of clinical scenarios.