AI can help doctors make better decisions and save lives

Implementing and evaluating machine learning interventions to improve clinical care and patient outcomes is a critical step in moving clinical deterioration models from the lab to the bedside, according to a June 13 editorial. Intensive Care Medicine This is a comment on a Mount Sinai study published in the same issue.

A key study found that when care teams received AI-generated alerts of a hospitalized patient's deteriorating health, the patient was 43% more likely to receive enhanced care and was significantly less likely to die.

“We wanted to see whether rapid alerts created by AI and machine learning trained on different types of patient data could help reduce both how often patients need intensive care and their likelihood of dying in the hospital,” said lead study author Matthew A. Levin, MD, professor of Anesthesiology, Perioperative and Pain Medicine, and Genetics and Genomic Sciences at Icahn Mount Sinai and director of Clinical Data Science at Mount Sinai Hospital.

“Traditionally, we have relied on older manual methods, such as the Modified Early Warning Score (MEWS), to predict clinical deterioration. However, our study shows that an automated machine learning algorithm score that triggers an assessment by a provider is better than these previous methods at accurately predicting this deterioration. Importantly, it could allow for earlier intervention, saving more lives.”

The nonrandomized, prospective study included 2,740 adult patients admitted to four medical-surgical wards at Mount Sinai Hospital in New York. Patients were divided into two groups: those who received real-time alerts based on the predictability of deterioration and sent directly to a “rapid response team” of nurses, physicians, or intensivists, and those in which alerts were generated but not sent. In wards where alerts were suppressed, patients who met standard criteria for deterioration received urgent intervention from the rapid response team.

Further findings from the intervention group showed that patients were more likely to receive medication to support their heart and circulation, doctors were showing signs of taking action earlier, and they were less likely to die within 30 days.

“Our study shows that real-time alerts using machine learning can significantly improve patient outcomes,” said David L. Reich, MD, chancellor of Mount Sinai Hospital and Mount Sinai Queens, the Horace W. Goldsmith Professor of Anesthesiology at Icahn Mount Sinai, and professor of Artificial Intelligence and Human Health, and senior author of the study.

“These models will support accurate and timely clinical decision-making, helping us dispatch the right team to the right patient at the right time. We think of them as 'augmented intelligence' tools that will speed up in-person clinical assessments by doctors and nurses and drive care that keeps patients safe. These are important steps toward our goal of becoming a learning health system.”

The study was closed early due to the COVID-19 pandemic. The algorithm is being deployed to all step-down units within Mount Sinai Hospital, using a simplified workflow. Step-down units are specialized areas within the hospital where patients who are stable but require close monitoring and care are placed. They are located halfway between the intensive care unit (ICU) and the general hospital area, ensuring that patients receive the appropriate level of care as they recover.

Each day, a team of intensivists visits the 15 patients with the highest prediction scores and recommends treatment to the doctors and nurses responsible for them. As the algorithm is continually retrained on more patients over time, the intensivists' assessment serves as the gold standard for accuracy, and the algorithm becomes more accurate through reinforcement learning.

In addition to the clinical deterioration algorithm, the researchers developed and deployed 15 additional AI-based clinical decision support tools across the Mount Sinai Health System.

The Mount Sinai paper is titled “Real-Time Machine Learning Alerts to Prevent Escalation of Care: A Non-Randomized Clustered Pragmatic Clinical Trial.” The remaining authors of the paper are all Icahn Mount Sinai unless otherwise noted: Arash Kia MD, MS, Prem Timsina PhD, Fu-yuan Cheng MS, Kim-Anh-Nhi Nguyen MS, Roopa Kohli-Seth MD, Hung-Mo Lin ScP (Yale University), Yuxia Ouyang PhD, and Robert Freeman RN, MSN, NE-BC.