New machine learning model to identify treatments that pose a higher risk

The system could help physicians select the least risky treatments in urgent situations.


Sepsis is a potentially life-threatening condition, occurs when your body has an unusually severe response to an infection. Some sepsis treatments lead to a patient’s deterioration. Hence, selecting the optimal therapy is a challenging task.

In recent years, Machine learning has successfully framed many sequential decision-making problems. Scientists at MIT and elsewhere do the same. To help clinicians avoid remedies that may potentially contribute to a patient’s death, scientists have developed a machine-learning model to identify treatments that pose a higher risk than other options.

The model also indicates when a septic patient is approaching a medical dead end. This helps them intervene before it is too late.

Scientists applied their model to a dataset of sepsis patients in a hospital intensive care unit. The model indicated that about 12 percent of treatments given to patients were harmful. The study likewise uncovers that around 3% of patients who didn’t endure entered a medical impasse as long as 48 hours before they died.

Taylor Killian, a graduate student in the Healthy ML group of the Computer Science and Artificial Intelligence Laboratory (CSAIL), said, “We see that our model is almost eight hours ahead of a doctor’s recognition of a patient’s deterioration. This is powerful because every minute counts in these susceptible situations, and being aware of how the patient is evolving. The risk of administering certain treatment at any given time is significant.”

To develop their model, scientists used reinforcement learning. They trained it using the limited data from a hospital ICU to identify treatments to avoid. They aimed to prevent patients from entering a medical dead end.

Scientists dubbed their approach as Dead-end Discovery (DVD). They developed their approach by creating two copies of a neural network. The first neural network focuses only on adverse outcomes — when a patient dies. The second network only focuses on positive results — when a patient survives.

They used both networks separately to detect a risky treatment in one and then confirm it using the other. Each network was fed with patient health statistics and proposed treatment.

The network predicted an estimated value of that treatment. It also evaluates the probability the patient will enter a medical dead end. Comparing these estimates helps scientists set thresholds to see if the situation raises any flags.

The yellow flag indicates that a patient is entering an area of concern. The red flag indicates a situation where it is very likely the patient will not recover.

When scientists tested their model, they found that 20 to 40 percent of patients who did not survive raised at least one yellow flag before their death. Most of the patients raised that flag almost 48 hours before they died.

Killian said, “The results also showed that, when comparing the trends of patients who survived versus patients who died, once a patient raises their first flag, there is a very sharp deviation in the value of administered treatments. The window of time around the first flag is a critical point when making treatment decisions.”

“This helped us confirm that treatment matters and the treatment deviates in terms of how patients survive and how patients do not. We found that upward of 11 percent of suboptimal treatments could have potentially been avoided because there were better alternatives available to doctors at those times. This is a pretty substantial number when you consider the worldwide volume of patients who have been septic in the hospital at any given time.”

Assistant Professor Marzyeh Ghassemi, head of the Healthy ML group and senior author, said“the model is intended to assist doctors, not replace them.”

“Human clinicians are who we want making care decisions, and advice about what treatment to avoid isn’t going to change that. We can recognize risks and add appropriate guardrails based on the outcomes of 19,000 patient treatments — that’s equivalent to a single caregiver seeing more than 50 septic patient outcomes every day for an entire year.”

Scientists are further planning to enhance the model to create uncertainty estimates around treatment values. This would help doctors make more informed decisions.

Journal Reference:

  1. Mehdi Fatemi, Taylor W. Killian, Jayakumar Subramanian, Marzyeh Ghassemi. Medical Dead-ends and Learning to Identify High-risk States and Treatments. (Paper)
Latest Updates