Predict Heart Failure with Deep Learning & EHR - Medical Journal Publication

I was one of the authors involved in the study "Electronic Health Record-Based Deep Learning Prediction of Death or Severe Decompensation in Heart Failure Patients" conducted at Washington University & Barnes Hospital.

In this research, we addressed the critical issue of timely identification of medical therapy nonresponders in heart failure patients, aiming to assist clinicians in accurate and prompt decision-making. We utilized a state-of-the-art deep learning approach that incorporated a Keras deep neural network, including Word2Vec and LSTM layers, to analyze commonly-available electronic health record (EHR) variables.

To support our research, I would like to express our gratitude to NVIDIA for providing the powerful RTX A6000 GPU that enabled the efficient training of our neural network.

Our study cohort comprised all patients admitted to a single tertiary care institution from January 2009 to December 2018, diagnosed with heart failure according to the International Classification of Diseases coding. We developed ensemble deep learning models using time-series and densely-connected networks, leveraging the vast amount of EHR data available. The positive class included observations resulting in severe progression, such as death from any cause or referral for heart failure surgical intervention within one year.

With an impressive dataset of over 79,850 distinct admissions from 52,265 heart failure patients, contributing to more than 350 million EHR datapoints for model training, validation, and testing, our deep learning models achieved remarkable accuracy. Approximately 20% of the model observations met the criteria for the positive class. The model's C-statistic, a measure of its predictive performance, reached an outstanding score of 0.91.

The results of our research have significant implications for clinical practice. The demonstrated accuracy of EHR-based deep learning model prediction for one-year all-cause death or referral for heart failure surgical therapy highlights its clinical relevance. This technology holds tremendous potential to assist heart failure clinicians in making informed decisions and improving the application of advanced surgical therapy in medical therapy nonresponders.

I am excited to delve into the details of our research, explaining the methodology, results, and implications. Join me in exploring the remarkable possibilities that deep learning and electronic health records bring to the field of heart failure management.

Note: This research paper was published in the Journal of the American College of Cardiology Heart Failure, 2022, and is copyrighted by the American College of Cardiology Foundation, published by Elsevier. All rights reserved.

McGilvray, M. M., Heaton, J., Guo, A., Masood, M. F., Cupps, B. P., Damiano, M., ... & Foraker, R. (2022). Electronic health record-based deep learning prediction of death or severe decompensation in heart failure patients. Heart Failure, 10(9), 637-647.

https://www.jacc.org/doi/10.1016/j.jchf.2022.05.010

~~~~~~~~~~~~~~~ CONNECT ~~~~~~~~~~~~~
🖥️ Website: https://www.heatonresearch.com/
🐦 Twitter - https://twitter.com/jeffheaton
😸🐙 GitHub - https://github.com/jeffheaton
📸 Instagram - https://www.instagram.com/jeffheatondotcom/
🦾 Discord: https://discord.gg/3bjthYv
▶️ Subscribe: https://www.youtube.com/c/heatonresearch?sub_confirmation=1
~~~~~~~~~~~~~~ SUPPORT ME 🙏~~~~~~~~~~~~~~
🅿 Patreon - https://www.patreon.com/jeffheaton
🙏 Other Ways to Support (some free) - https://www.heatonresearch.com/support.html
~~~~~~~~~~~~~~~~~~~~~~~~~~~~
#asus #deeplearning #proart #review #machinelearning

Видео Predict Heart Failure with Deep Learning & EHR - Medical Journal Publication канала Jeff Heaton