Dear Doctors: If it's true that, according to a 2024 study, there are some 200,000 deaths (amazingly!) each year from medical errors in the United States, might AI be used to reduce them? I've read it can help with ventilator mishandling, patient instruction omissions and detecting diagnostic oversights.
Dear Reader: Your question about artificial intelligence broaches a topic that is generating both cautious enthusiasm and widespread concern. It may seem like an emerging debate, but the idea of AI as a medical tool dates back to the mid-20th century. What started as a largely theoretical idea in 1950 had, by the turn of the century, evolved into a range of algorithms and pattern-recognition models. These were designed to assist with imaging analysis, diagnosis and even some aspects of clinical care. It wasn't until a significant leap in data and computing resources in the 2010s that modern deep-learning approaches would be able to turn theory into practice.
Before we look at how AI may be useful in medical care, we should clarify the statistics you cited. Despite a widespread desire for an authoritative number, estimates of deaths due to medical error vary widely, depending on how those errors are defined. Some studies stick solely to point-of-care mistakes. Others include system-level failures, such as a lack of coordinated care, a lack of access to appropriate care, medical insurance practices, faulty or incomplete safety nets and inadequate follow-up care. The analysis associated with that 200,000 figure takes this broader view and does not reflect only point-of-care mistakes.
Proponents of AI believe these tools can be harnessed to improve medical care. Pattern-recognition models, in particular, have already proven effective at analyzing medical imaging scans, often with gains in both speed and accuracy. A study done here at UCLA found that AI models were able to identify pancreatic tumors in CT scans months, and sometimes years, before the human eye could see them. Another real-world use is screening for diabetic retinopathy, a complication of diabetes in which high blood sugar levels damage blood vessels in the retina. AI systems have also shown promise as early warning tools for sepsis, in detecting heart arrhythmias in ECGs and for reading mammograms in breast cancer screenings.
While acknowledging the possible benefits of these powerful tools, many clinicians and researchers advise a measured approach to adopting them. Medicine is, at its core, a deeply personal science, and context matters. For that reason, most experts emphasize that artificial intelligence should be used as an aid, not treated as an authority.
When incorporated thoughtfully and with proper oversight, AI systems have the potential to improve medical care. They can scan large amounts of data quickly and identify patterns associated with early-stage diseases or conditions that might otherwise go unnoticed. They can improve accuracy in imaging review, detect early signs that a patient’s vital signs are signaling future problems, identify gaps in follow-up care and reveal breakdowns in coordination before they lead to harm. As AI tools evolve, keeping humans firmly in the equation will be key not only to their success but to their eventual acceptance.
(Send your questions to askthedoctors@mednet.ucla.edu, or write: Ask the Doctors, c/o UCLA Health Sciences Media Relations, 10960 Wilshire Blvd., Suite 1955, Los Angeles, CA, 90024. Owing to the volume of mail, personal replies cannot be provided.)