A paper published by Google on arXiv, titled "Scalable and accurate deep learning for electronic health records" by Alvin Rajkomar and colleagues, presents an innovative approach to using EHR data. The study leverages the FHIR (Fast Healthcare Interoperability Resources) format to represent raw patient records, enabling the use of deep learning techniques to accurately predict various medical events across multiple healthcare centers.
The research highlights how traditional predictive modeling often involves labor-intensive feature extraction from normalized EHR data, which can lead to loss of valuable information. In contrast, the proposed method uses a sequence-based representation of EHR data, capturing a vast amount of clinical information—over 46 billion data points—from more than 200,000 hospitalized patients.
The deep learning models developed in this study demonstrated exceptional performance in predicting outcomes such as in-hospital mortality (AUROC 0.93–0.94), unplanned readmissions (AUROC 0.75–0.76), extended hospital stays (AUROC 0.85–0.86), and final diagnoses (frequency-weighted AUROC 0.90). These results significantly outperformed conventional statistical models.
Additionally, the paper includes a case study showing how a neural network attribution system can provide transparency in model predictions, helping clinicians understand the reasoning behind AI-driven decisions. This approach not only enhances accuracy but also supports explainability, making it more practical for real-world clinical applications.
During the research, the team recognized the need for efficient data handling when scaling machine learning solutions. To address this, they proposed integrating a protocol buffer tool into the FHIR standard, allowing large-scale data to be serialized and stored efficiently for analysis.
Recently, Google announced that the protocol buffer tool has been open-sourced. According to their blog post, over the past decade, healthcare data has transitioned from paper-based to digital formats. However, challenges remain, including inconsistent data representations across vendors, varying coding practices for the same drugs, and fragmented data spread across multiple tables.
FHIR was introduced to address these issues by offering a standardized, scalable, and web-based framework for exchanging health data. While FHIR has become widely adopted, the researchers emphasized the need for additional tools—like the protocol buffer—to support large-scale machine learning.
Today, Google is excited to release the FHIR protocol buffer tool, initially supporting Java with future support for C++, Go, and Python. Additional features, such as configuration files and tools for converting legacy data to FHIR, are also in development. This move aims to further enhance interoperability and enable more efficient analysis of healthcare data at scale.
Pcb Pluggable Terminal Block Connector ,Pluggable Terminal Block,Contact Pluggable Terminal Blocks ,Pluggable Screw Terminal
Cixi Xinke Electronic Technology Co., Ltd. , https://www.cxxinke.com