MIT researchers use machine learning to predict ICU interventions

Researchers at the Massachusetts Institute of Technology’s Computer Science and Artificial Intelligence Laboratory have developed a machine learning algorithm that leverages large amounts of intensive care unit (ICU) data to predict actionable interventions for patients and improve health outcomes.

By tapping into an MIT database of de-identified data for 40,000 critical care patients—including demographics, laboratory tests, medications and vital signs—the research team is able to use deep learning to determine what kinds of treatments are needed for different symptoms.

The approach—called ICU Intervene—was presented in a paper this past weekend at the Machine Learning for Healthcare Conference in Boston. According to the authors, their model is the first to use deep neural networks to predict both onset and weaning of interventions using all available modalities of ICU data.

Also See: MIT Medical to implement integrated system from Cerner

“The decisions that are made in the ICU are made in a particularly high-stress and high-demand environment,” says Harini Suresh, a PhD student and lead author on the paper, who adds that clinicians in these situations are bombarded with different types of data for many patients and as a result it can be difficult to make real-time treatment decisions.

Using a time stamp for the data, Suresh notes that at each hour ICU Intervene extracts values from the data that represent vital signs, clinical notes and other information. “We know for a given patient what happens at every point in their stay,” she observes. “What we’re trying to do is find insights and connections for patients which may not be immediately clear.”

Based on the data, the algorithm learns from past ICU cases in the database and provides an hourly prediction of five interventions: invasive ventilation, non-invasive ventilation, vasopressors, colloid boluses and crystalloid boluses.

“It predicts whether a patient will need any number of interventions six hours in the future,” including predicting whether a patient will need a ventilator six hours later, contends Suresh.

According to Suresh, the model was very effective at predicting the need for vasopressors, a medication that tightens blood vessels and raises blood pressure. “The accuracy was pretty good for all five of the interventions, but vasopressors is one in which there has been previous work, and we were able to outperform those baselines for intervention prediction,” she adds.

“Deep neural-network-based predictive models in medicine are often criticized for their black-box nature,” says Nigam Shah, an associate professor of medicine at Stanford University who was not involved in the paper. “However, these authors predict the start and end of medical interventions with high accuracy and are able to demonstrate interpretability for the predictions they make."

Going forward, Suresh says researchers will look to “make the algorithm as robust as possible” and improve ICU Intervene’s ability to provide advanced reasoning for the model’s predictions in the case of specific patients in order to give physicians more insight into their treatment decisions.

“One of the things we tried to focus on in this paper was the interpretability aspect—in addition to making a prediction, trying to explain why the model made that prediction,” she says. “Expanding on that will be important to make this more compelling for physicians, because incorporating these types of machine learning technologies into hospitals and having them work seamlessly is a big change.”

Ultimately, Suresh says the goal is to have the algorithm “tested with real doctors in real ICUs.”