Researchers Find Machine Learning Can Improve Patient Care

Using “deep learning,” the system makes real-time predictions by learning from earlier ICU cases to offer suggestions for critical care, along with providing the reasoning for the decisions.

ICU Intervene makes hourly predictions of five different interventions covering various critical care needs, like breathing assistance, improving cardiovascular function, reducing blood pressure, and fluid therapy. The system extracts values from the data representing the vital signs, along with clinical notes and other data points every hour. All this data is represented with values indicating how far away a patient is from the average (for evaluating further treatment).

What is particularly notable is that ICU Intervene can also make future predictions. For instance, the model can predict if a patient will require a ventilator six hours later instead of only 30 minutes or an hour later. The researchers found that the system outperformed the earlier work done in predicting interventions and was particularly good in predicting the requirement for vasopressors. The researchers will focus on improving ICU Intervene in the future to allow for more individualized care and provide more advanced reasoning for decisions, like why the dosage of steroids can be gradually reduced for a patient, or why a procedure such as an endoscopy may be required in the case of another patient.

“The system could potentially be an aid for doctors in the ICU, which is a high-stress, high-demand environment,” according to Harini Suresh, a PhD student and lead author of the paper on ICU Intervene presented in August 2017 at the Machine Learning for Healthcare Conference held in Boston. “The goal is to leverage data from medical records to improve health care and predict actionable interventions.”

Another team of CSAIL researchers has developed “EHR Model Transfer,” an approach to facilitate the application of predictive models on an electronic health record (EHR) system, in spite of being trained on data from a different EHR system. The team used this approach to demonstrate that predictive models for mortality and prolonged length of stay can be trained on one EHR system and used to make predictions in another. The approach can be adopted across different versions of EHR platforms, using natural language processing to identify clinical concepts that are encoded differently across systems and then mapping them to a common set of clinical concepts (such as “blood pressure” and “heart rate”). For instance, in the case of a patient in one EHR platform who could be changing hospitals and would need their data transferred to a different type of platform, the EHR Model Transfer can ensure that the model will still predict aspects of that patient’s ICU visit, such as their chances of a prolonged stay or even of dying in the unit. The researchers plan to evaluate the EHR Model Transfer system on data and EHR systems from other hospitals and care settings in the future.

Related Links:
CSAIL