AI's Role in Clinical Care: Opportunities and Challenges

At the beginning of the 2000s, the electronic medical record (EMR) movement began to reshape clinical care. As a division chief at a major children’s hospital, I witnessed the initial impact of EMRs. Now, as a seasoned veteran, I’m observing the even more profound effects of artificial intelligence (AI) in healthcare.

The EMR movement promised streamlined medical record transfers and improved patient care. However, in practice, many of these promises have yet to be fully realized. One major issue is the incompatibility between different institutions' customized EMR systems, even when the same vendor is used. Additionally, clinical notes often contain outdated and inaccurate information due to the frequent use of “cut-and-paste” practices, leading to serious errors.

When transitioning EMR data to paper records, the process often results in disorganized and unusable documents. For instance, each blood pressure reading, pulse, lab value, and medication administration is printed on a separate page, making it difficult to manage and interpret the information.

AI is now poised to have a far greater impact on clinical care than the EMR movement. It is being used to enhance diagnosis, predict clinical outcomes, inform decision-making, and manage patient care. AI is also playing a crucial role in research and teaching.

Studies have shown that AI can match or even exceed the diagnostic accuracy of experienced specialists, often in a fraction of the time. For example, a recent study examining over 3 million emergency room visits found that AI can accurately predict the likelihood of agitation and violence, confirming the clinical truism that past behavior is a strong predictor of future behavior.

AI can also detect subtle behaviors that are difficult for human clinicians to recognize. In neonatal intensive care units, AI models have been used to analyze cell phone videos of infants, quantifying head and hand movements to identify the best times for care and feeding.

However, the rapid adoption of AI in healthcare has also brought significant challenges. One major issue is the phenomenon of AI “hallucinations,” where AI programs generate false factual statements, including fabricated data. These errors can occur up to 75% of the time, raising concerns about the reliability of AI in clinical settings.

There is also a risk of overreliance on AI by clinicians, which could lead to misrepresentations of risks and benefits to patients. Some studies suggest that AI-based health care searches can generate substantial amounts of disinformation, including convincing but fabricated medical images.

The Veterans Administration (VA) has been a leader in using AI to develop clinical predictive models and integrate their results into routine patient care. One key lesson from the VA is that AI predictive models should be recalibrated annually to maintain accuracy. However, AI models are often poor at predicting low-frequency events, leading to a high rate of false alarms.

One of the most valuable functions of AI in healthcare is its ability to search patient EMRs for relevant symptoms and link them to the appropriate scientific literature. The VA is pioneering technologies such as “ambient listening,” which uses AI to record, transcribe, and analyze patient-clinician conversations, automating documentation and allowing clinicians to focus more on patient care.

Stephan Fihn, a pioneer in the VA’s work on EMRs and AI, believes that the future of AI in healthcare will look more like a Google search—pulling together all relevant information with a single click. If this vision is realized, AI may finally help EMRs achieve their full potential.

In conclusion, while AI holds immense promise for improving clinical care, it also presents significant challenges that must be addressed to ensure its safe and effective use.