Q&A: AI in healthcare as a powerful assistant
Q: How is artificial intelligence (AI) being used in healthcare today, and what role can AI play in improving documentation and coding workflows while still requiring human oversight?
A: AI, at its heart, is fast pattern recognition. Think of it like a seasoned emergency room nurse who walks into a room and instantly knows the patient is sick before she even looks at the vital signs monitor because she has seen that pattern often enough. AI does this by using billions of data points in seconds. Unlike many cases of AI, professional healthcare AI is often designed as a closed system to ensure that the tool is only using high quality vetted medical data that it has been given. In fact, many healthcare systems take the extra step of turning off internet searching in their AI tools, allowing the AI to be entirely focused on an organization's policies, clinical guidelines, and the patient's record. It ultimately becomes a powerful tool that stays inside of a healthcare system. That can be a massive win in terms of security, as eliminating the risk of data leaking out to the public is very important.
The purpose of using AI is truly about taking robotic work away from humans so that trained professionals can get back to the essence of their jobs. There are two big areas that AI can be total game changers for: coding and CDI. Imagine scrolling through the work queue and seeing a chart for a patient who has been in the hospital for 40 days. The chart has hundreds of documents. It would take a person hours just to piece together the timeline of the care. Alternatively, AI can read the entire history and provide some type of summary in seconds, allowing coding and CDI teams to complete their work more quickly.
Even though AI is a clinical authority in terms of speed, it still requires human verification. AI may be incredible at scanning the entire history of the patient, but sometimes it forgets to look at the details and can make inappropriate suggestions, also known as AI hallucinations. Consider a patient who had a history of an amputation 10 years ago. The AI might see that detail in a deep archival search of the old record and pull it forward into a current active problem list, or maybe even suggest a complication condition because of the amputation. Another example is an AI seeing that a patient is tachycardic with a low-grade fever and hallucinating a sepsis diagnosis even though the provider has clearly documented that the patient is in a postoperative period and reacting normally to the surgery they recently had. Only a human can be the authority on how AI connects documentation information to the patient’s clinical picture for proper coding. Even the most advanced systems can hallucinate or be too overconfident, thus jumping to conclusions that do not fit the whole picture. This is why human verification is a professional guardrail, not a suggestion. AI can do heavy lifting by sorting through data, but healthcare has to keep humans at the helm to make the final call.
Editor’s note: This Q&A was originally published in JustCoding. The information was provided by Karen Lane, MSN.Ed, CCDS, CCDS-O, CDIP, RN, a clinical documentation integrity specialist III at an academic medical center in the Midwest, who answered this question on the ACDIS Podcast.