It took four years of college, four years of medical school, five years of residency, and 13 years of your life to become a doctor, but AI is learning to diagnose and treat diseases in a fraction of that time.
AI is becoming increasingly important in healthcare, influencing decision-making, shaping workforce dynamics, and helping reduce burnout. This is because, as reported in one report, AI can perform 30-40% of medical tasks, freeing up employees to focus on complex patient-centered care. article About innovations in medicine by the National Library of Medicine.
For immunologists Julia CroninShe works at Sutter Health Palo Alto Medical Foundation, and her personal journey toward AI began about a year ago when she was first introduced to what is known as AI technology. abridge. Abridge is an app that listens to a patient’s consultation over the phone and uses AI at the end to generate feedback on possible diagnoses and summarize the conversation into notes.
“That way what I said will sound better and more organized than the way I actually said it,” Cronin said. “At the end of the visit, sometimes you don’t have to do much to assess or plan; it’s already done.”
Cronin initially didn’t see the need for Abridge. Because she was already working with human scribes. A scribe is a person who remotely listens to the details of a patient’s medical examination over the phone and summarizes the information. But about six months ago, she started using Abridge with her scribe.
“Instead, it just augments the work of my scribes, and on days and times when my scribes are not available,” Cronin said.
Abridge helped Cronin save time documenting, but it’s not perfect. According to article Regarding AI scribing technology by the National Library of Medicine, AI scribing applications are still in the early adoption stage. They may struggle with issues such as distinguishing sounds and interpreting context.
Cronin encountered these limitations firsthand, as Abridge sometimes struggled to differentiate and contextualize information between multiple families. For example, when explaining test results or conducting inspections, Cronin may have to edit the entire transcription generated by Abridge due to transcription errors.
Still, thanks to Abridge and her scribe, Cronin’s overall writing time was significantly reduced.
“One of my colleagues says that if he leaves his cell phone at his desk and forgets to bring it into his room, he wants to cry because he knows the visit will take five times as long,” Cronin said.
Cronin believes Sutter Health may be considering discontinuing the writing service because Abridge’s prices are significantly lower. According to National Library of MedicineAn AI scribe system costs about $100 per user per month, while the average monthly salary for a human medical scribe is about $2,800.
Furthermore, patients have expressed a preference for Abridge over human scribes as they feel uncomfortable sharing personal information with others.
“Scriveners get sick, they scream, they’re in foreign countries, and we’ve never gone a week or so without a scribe,” Cronin said.
According to article The shift to AI in health care raises concerns about the role of management in health care, as hospitals may face financial pressure to adopt more cost-effective methods, according to the National Library of Medicine’s report on the balance of innovation in health care. Without strong workforce protections, AI management technology could displace workers whose responsibilities are largely automated.
Meanwhile, AI scribes improve doctor-patient interactions by eliminating the need for clinicians to focus on computer screens during appointments. According to the National Library of Medicine, AI processing documents allows doctors to maintain eye contact, observe body language, and give patients their full attention.
Sutter Health is also piloting the use of AI to respond to patient messages, Cronin said, and the AI will provide doctors with draft responses to review and send.
“Sometimes we actually use them, and they’re really good, and sometimes they’re terrible,” Cronin said.
While Cronin’s specialty is just beginning to embrace AI, radiology has been leading the field for nearly two decades. The field reportedly accounted for 18% of all AI-related medical publications from 2003 to 2023, generating 14,228 papers between 2021 and 2023. article About emerging AI trends in healthcare from the National Library of Medicine.
Radiologist Justin Shafa first started noticing the emergence of AI platforms in 2019. Currently he uses tools such as: Rad AI and Grimer AI They can help with a variety of tasks, from detecting bone fractures with X-rays to identifying cerebral hemorrhages and pulmonary embolism to interpreting CT scans.
These platforms have helped Shafa capture details that he might have missed, saving him time overall.
“They’re supposed to make us faster and more accurate. They’re not 100% in any capacity. They’re not perfect,” Shafa said. “Sometimes they make embarrassing mistakes. Sometimes they miss a fracture, and sometimes they say it’s a fracture when it’s not.”
The flaws Cronin sees in Abridged mirror the obstacles Schaffa encounters in radiology. According to article A National Library of Medicine study on the balance of innovation in health care found that the challenges facing AI stem from how it learns.
AI analyzes large amounts of patient data to make decisions. Its accuracy highly depends on the diversity and representation of the data.
When certain groups are underrepresented, AI performance can deteriorate, leading to biased or unreliable results. for example, 2019 AI model developed by DeepMind Women performed worse in predicting kidney damage because they were underrepresented in the training data.
“There are a lot of false positives that the AI incorrectly labels, so human observation is still needed to properly diagnose the patient,” said Interventional Radiology Technologist Carlos Buitrago Pizon.
Cronin believes that AI will play an important role in the future of healthcare, and many physicians are grateful for the support it provides. However, we hope to see further improvements, such as domain-specific writing and AI-assisted advance charts.
“Specialized scribes would be great, and it would be even more helpful if AI could learn our templates and put what we say directly into the templates we use,” Cronin said. “Another thing is that AI might be able to create charts in advance. So when you’re looking at who’s coming in the day before, you might want to just have information about the records in the chart. AI might be able to help you with that.”
