Machine learning can help doctors treat patients better MIT researchers have discovered efficient ways to deploy computers as tools of the physician in charge.
Much information
Doctors are often inundated with information coming from charts, test results and other metrics. It can be difficult to integrate and monitor all this data for multiple patients while having to make real-time treatment choices. It becomes even more difficult when data is documented inconsistently by different hospitals.
In a new study, researchers at the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) have discovered ways that computers can help doctors make better medical decisions.
ICU Intervene
One team created a machine learning approach called 'ICU Intervene', which uses large amounts of data from intensive care units to determine what forms of treatment are needed for different symptoms. The system uses 'deep learning' to make real-time predictions. In doing so, it uses old situations from intesive care. "The system has the potential to be an auxiliary tool for intensive care doctors, who have to deal with a stressful and demanding environment," said lead author Harini Suresh. "The goal is to leverage data from medical records to improve healthcare and predict actionable interventions."
EHR Model Transfer
Another team developed an approach called 'EHR Model Transfer'. This model can facilitate the application of predictive models on an electronic health record (EHR) system despite being trained with data from another EHR system.
Importance of both systems
Integrated data from Intensive Care is important for making the process of predicting patients' health outcomes autonomous.
ICU Intervene focuses on predicting five different interventions covering a wide range of critical care needs, such as ventilation, improving cardiovascular function and lowering blood pressure.
Every hour, the system extracts values from data representing vital signs, as well as clinical notes and other data. All data are represented with values that show how far the patient is from the averages (after which further treatment is evaluated). Importantly, ICU Intervene can make predictions far into the future. For example, the model can predict whether a patient needs a ventilator as early as six hours in advance. In addition, the researchers made sure that the system also provides an explanation for the predictions it makes, giving doctors more insight into the situation.
Existing machine-learning models need consistent data storage. The fact that hospitals regularly change their EHR system can cause major problems when it comes to data analysis and prediction. This is where the EHR Model Transfer comes in.
The approach works with different versions of electronic platforms; by using natural language processing, clinical concepts are identified that are integrated into different systems. They are then mapped to common clinical concepts (such as blood pressure and heart rate). This would therefore mean that changing an electronic system in a hospital or transferring a patient to another hospital would not affect data integration.
Both systems still need future improvements to make individual patient care in particular as optimal as possible.
By: Kelly Bakker
Source + photo: MIT/Pixabay
