Sutter Health taps Ferrum Health for AI-based 'safety net' for lung cancer scans

The pilot showcases an approach to AI integration that keeps impact on the radiologist's workflow to a minimum.
By Jonah Comstock
07:00 AM

Northern California healthcare system Sutter Health announced today that has begun a pilot with startup Ferrum Health to use AI to monitor the work of radiologists and catch preventable medical errors in patients with lung cancer.

WHY IT MATTERS
Ferrum was founded by Pelu Tran, who previously founded and led Google Glass company Augmedix. It’s an AI company, but rather than building its business model on selling particular algorithms, it’s focused on deploying existing algorithms in a new way: as a background-monitoring “safety net” that only requires active attention from radiologists when it detects a discrepancy with their diagnoses.

"Coming out of my previous company and the strength that we had working with healthcare systems like Sutter, we realized that AI could be used in a few different ways, but that physicians, given all the pressures that they were under, were going to really struggle to carve out time in their workflow, in their day-to-day, to deploy solutions that are really focused on improving quality," said Tran.

Learn on-demand, earn credit, find products and solutions. Get Started >>

"So we thought that there was a better approach to trying to help them care for their patients," he explained. "And that was really by providing solutions that improve quality without disrupting workflow. And we did it here by deploying a monitoring system, a tool that runs in the background and and looks for these errors at a population enterprise level without really disrupting or requiring physician behavior change."

The initial deployment is focused specifically on finding nodules that can show up in the early stages of lung cancer. In the first 90 days, the technology has already reviewed 10,000 CT scans containing lung tissue and found 83 discrepancies that then prompted additional review and intervention by radiologists. The technology works fast too — radiologist are prompted to review flagged scans the same day they originally looked at them.

"When you look briefly at lung cancer, survival is all about early detection," said Dr. Jason Wiesner, medical director of the Sutter Health's Diagnostic Imaging Service Line, noting that the survival rate for lung cancer detected in stage one is 50%, compared with 4% or 5% of cases not detected until stage three. "Early detection at that stage one level is critical to survival for the population. So what we're doing here actually is measurably impacting the population health."

Sutter already has lower-than-average rates of preventable errors in its lung cancer detection. Wiesner says that’s due to a cultural focus on high reliability, which he hopes this technology will further improve.

"At Sutter Health, we've become laser-focused on this quality improvement and this no-harm, high reliability organizational commitment," he said. "It's become a CEO-level commitment to the fact that as just a brief sidebar there, in the last year and a half or so, almost 50,000 employees have gone through a safety training exercise to just look for potential sources of error and stop the line before they actually became errors.

"And because of that, our culture has really transformed," he added. "We actually share these sorts of good catches and prevention stories before all of our meetings. … But we can always do better. And again, I think that's where Ferrum for us in diagnostic imaging really takes us to that next level."

If the pilot continues to go well, Tran says he hopes to deploy many more similar screening programs "for breast cancer, for colon cancer, for fractures, for aneurysms or for things like missed followups just throughout the patient journey.

"It's challenging to envision a path forward where the rank and file radiologists are the ones who are interfacing with hundreds of different data algorithms in their day-to-day practice," he said. "And so for us, the most important part about the workflow is that it's essentially infinitely scalable. We actually can deploy hundreds of different point solution monitoring or use cases enterprise-wide across millions of patients. And we can do that too without really requiring additional burden on rank and file radiology."

Monitoring systems are the norm in other industries, Tran pointed out, including ones with arguably lower stakes than healthcare.

"In cybersecurity, you have antivirus that's always running," he said. "You have intrusion detection, you have anomaly detection. In financial services you have fraud detection, and in manufacturing you have defect detection. Healthcare is the only industry out there – and it actually happens to be the most critical one – and it just doesn't have anything. If the doctor makes a mistake, there's no safeguard."

THE LARGER TREND
Preventable medical errors are widely recognized to be one of the leading causes of death in hospitals. Meanwhile, radiology has become more complicated in recent years as technology has improved.

"CT scans, for example, used to be 60 to 100 images, and now they're three to six thousand images," said Wiesner. "The explosion of medical imaging data, all that comes down on the workflow and the desk of the radiologists. We're empowered by that. We can make diagnoses that we couldn't before with much better, better specificity. But the demands on our time have increased."

Meanwhile, AI algorithms for radiology are being developed all the time. Just this month saw two separate publications, from Google and Korean company Lunit, on very promising breast cancer detection algorithms.

But how to use algorithms like this in clinical practice is something health systems are still figuring out. Generally speaking, turning over diagnosis entirely to computers is still too high-risk to be a serious consideration. 

NYU Langone Health is currently considering using an AI technology as a virtual second opinion that radiologists can refer to if they feel they need it. Ferrum’s model is similar to that approach, but it automatically runs the check.

ON THE RECORD
"Our radiologists are so excited about this workflow and the results that we've shown," said Wiesner. "They definitely want to continue having this working right next to them and in the background, making sure that they're providing the highest quality care they can. And so we are we're just so satisfied at this point. It's one of the best projects that I have going right now in my leadership role in my clinical practice."

Want to get more stories like this one? Get daily news updates from Healthcare IT News.
Your subscription has been saved.
Something went wrong. Please try again.