Suchi Saria, PhD
Computer scientist Suchi Saria, PhD, was hard at work with a team at Johns Hopkins University on AI research to help better predict the onset of sepsis in U.S. hospitals, when tragedy struck. Her nephew in India, in his 20s and newly married, fell ill and developed a serious infection.
That led to the very condition Saria and her team were working to prevent: sepsis — a life-threatening immune response to an infection that can cause rapid organ failure and death within hours.
By the time doctors caught it, it was too late and — as is all too common in sepsis cases — Saria's nephew passed away.
That was 7 years ago. His death lit a fire under her, Saria says.
"I started to realize, 'Oh, my God, we're writing these results and they're getting a lot of research recognition.' The question is: How is this translating into something practical?"she says.
"I wanted to have an impact, and I realized, to have a meaningful impact, I needed to understand how to translate these tools into real-world technology that works and scales in hospitals big and small, rural and community hospitals where you need them most."
It's a technology that, were it available at the time, may well have helped to save her nephew's life, Saria says.
A Surprising Journey
Saria grew up in the 1980s and '90s in the foothills of Darjeeling, India, a small northeastern city perched on a narrow mountain ridge of the Sikkim Himalayas.
As a child, she became fascinated with computer programming, an uncommon pursuit for girls in Darjeeling. It's an area where girls are not typically encouraged to study science, technology, engineering, and math, Saria says, let alone earn graduate degrees or go into research.
Many women marry and become homemakers. Other girls in the area never attend school and go directly to work in the fields harvesting tea leaves at a young age. Women make up more than 50% of the Darjeeling tea industry labor force, according to research about the region.
But Saria had other plans. In fifth grade, she was reading a computer science textbook, when a passage piqued her interest. Programming computers to be "intelligent," the line read, would be the world's "next frontier." She never lost interest.
She attended college in the U.S. starting in 2000, eventually earning a PhD in artificial intelligence and computer science from Stanford University in 2011.
Her early AI work developing smart assistants for computers was interesting, Saria says, but she felt the impact wasn't big enough. Surely, artificial intelligence could do more.
Around 2008, Saria started exploring the research potential of electronic health records with Stanford colleagues.
That led her in 2010 to develop an algorithm that predicts major complications in premature babies with more than 90% accuracy. The algorithm, PhysiScore, incorporates data from monitors, birth weight, and time spent in the womb to estimate the likelihood that preterm infants develop serious illness.
Following that success, Saria set her target on another health problem where time is vital: sepsis. Sepsis is a leading cause of hospital deaths in the U.S., contributing to at least 1.7 million hospitalizations and 350,000 fatalities a year.
Though intravenous antibiotics can treat sepsis, they often come too late, says Saria, who now directs the Machine Learning and Healthcare Lab at Johns Hopkins University in Baltimore, and is founding research director of the university's Malone Center for Engineering in Healthcare.
She felt AI could help shorten the time to diagnosis.
Using physiological and laboratory data from intensive care unit patients, Saria developed an AI-driven algorithm that predicted patients at risk for septic shock, the last stage of sepsis. In 2015, she and her team showed the AI tool identified at-risk patients a median of 28 hours before the start of septic shock and worked better than traditional detection tools.
Today, a new refined tool helps predict sepsis much earlier in the process, sometimes before it even starts.
An AI Tool That Saves Lives
But the tool only works if doctors use it, says Saria. That's why, after her nephew's death, she and her team began studying why some doctors didn't adopt machine learning-based systems like the one they developed. Doctors seemed to view AI systems as inferior to human expertise and question its clinical value. They found that lack of explanation behind the machine's logic was a top reason for mistrust.
That's why Saria and her team designed the sepsis AI tool so health care professionals can review the data that led to the system's alert and understand its decision. If they agree with the alert, they can take action, but if they disagree, they can document their reasoning in the health record. The AI model even learns from doctors' input.
Starting in 2018, Saria and her team deployed the sepsis early warning system to hospitals around Maryland and Washington, D.C.
The tool monitored 590,736 patients across the five hospitals. It alerted medical staff to possible sepsis in 6,877 patients before doctors started antibiotic therapy.
In the most severe cases, the AI tool detected sepsis an average of 5.7 hours earlier than traditional methods. That led to an 18% fewer in-hospital deaths, according a series of studies published in Nature Medicine in 2022.
A big part of the tool's success is doctor buy-in. Of the 31,591 alerts generated by the AI system, 89% of them were evaluated by a doctor or advanced practice provider, the study found.
In 2023, the AI tool received "breakthrough" designation from the FDA. Breakthrough status expedites the development and review of devices that provide for more effective treatment or diagnosis of life-threatening diseases or conditions.
Every day, I think about it as: How many more lives have I saved?
-- Suchi Saria, PhD
Loading...
And Saria and her team hope to now expand the AI system to help predict other complications in hospitals and beyond.
"We had to think outside the box to make this happen. To me, it's very rewarding to see that we've moved from an idea to demonstration to now scaling.
"Every day, I think about it as: How many more lives have I saved? It makes waking up every day worthwhile."
More in This Series
David Fajgenbaum, MD
After finding the cure for his own deadly rare disease in an existing drug, David Fajgenbaum, MD, MFA, devoted himself to using AI to do the same for others.
Regina Barzilay, PhD
A breast cancer diagnosis seemed to come out of the blue for Regina Barzilay, PhD. That inspired her to develop an AI system to help better predict breast cancers before they start.