MIT researchers develop new machine-learning models that help physicians make decisions
MIT researchers developed two new machine-learning models, called ICU Intervene and EHR Model Transfer that can help physicians make better medical decisions.
Both models were trained using data from a critical care database that was developed by the MIT Lab for Computation Physiology. It includes de-identified data from about 40,000 critical care patients.
ICU Intervene takes large amounts of ICU data from vitals, labs, notes and demographics to determine the types of treatments that are needed for different symptoms.
It leverages deep learning techniques to make real-time predictions for breathing assistance, improving cardiovascular function, lowering blood pressure and fluid therapy. It also learns from previous ICU cases to make suggestions for critical care and explains the reasoning behind the decisions.
Every hour, the system extracts values from the vital signs signs, as well as clinical notes and other data points. These are used to indicate how far off a patient is from the average.
For instance, it can predict whether a patient will need a ventilator six hours later instead of 30 minutes or an hour later.
The research team plans to improve ICU Intervene so that it can provide more individualized care and more advanced reasoning for decisions, such as why one patient may need a procedure like endoscopy.
The storage of that ICU data and what happens when the storage method is changed is also important to consider. For that, the MIT researchers created the EHR Model Transfer.
The model works across different versions of EHR platforms and uses natural language processing to identify clinical concepts that are encoded differently across systems. It then maps them to a common set of clinical concepts such as blood pressure and heart rate.
The team tested its ability to predict mortality and a patient's need for a prolonged stay. They trained it on one EHR platform and then tested its predictions on a different platform.
They found that EHR Model Transfer can outperform baseline approaches and can better transfer predictive models across different EHR platforms, compared to using EHR-specific events alone.
The researchers are planning to test the model on data and EHR systems from other hospitals and care settings.
MTCHT
ICT
TECHNOLOGICAL INNOVATIONS
POST
ABOUT US
NEWS
INTERESTING
INTERVIEW
ANALYSIS
ONLAIN LESSONS
