AI in Healthcare: A Lot of Talk, Not Much Action…Yet

It’s all anyone in healthcare wants to talk about — AI. At least, you’d think so from an AI-evaluation of all the presentation titles at the recent HIMSS 2024 conference.

In one of the many AI presentations, a few executives at several prominent health systems, including Honor Health, Banner Health/Aetna, and Holston Medical Group, gave their take on AI use cases for healthcare — both big and small. I’ve summarized a couple of interesting ideas and considerations for you. I also added in a few AI hacks you could start doing today, although admittedly those are more administrative in nature and not specific to healthcare.

Using AI to Relieve Administrative Burden

There was a lot of talk about using AI to help relieve clinician's documentation burden. Here’s some small ideas that could have a big impact:

1. Streamlining documentation: AI-powered tools can assist in streamlining documentation processes by automatically extracting relevant information from recorded visits, medical records, lab reports, and other sources. Imagine a scribe that takes notes during a patient’s visit, then provides a written summary to the patient at the end of the visit with major takeaways for the patient. Perhaps the AI tool could surface a few recommendations that a clinician might give to the patient. For example, a diet recommendation for someone with high cholesterol, pre-diabetes, or diabetes. How about a weight loss regimen and/or recommended exercises for knee pain or back pain or sciatica?

Some practices are already offering something similar to this, but the key is that the AI software is deriving notes from what was said during the visit. It summarizes what the clinician said and can be given to the patient as they are leaving. Virtually no administrative burden.

Now, how about all the documentation burden facing clinicians? What if, after a visit, the AI software prompts the clinician with a few questions at the end that are derived from the conversation? Would you like a lab ordered for the patient? Patient mentioned they had a flu shot in December; would you like this added to the record?

2. Automating repetitive tasks: AI can automate routine administrative tasks such as appointment scheduling, patient registration, and billing processes.
3. Improving communication and coordination: AI-powered chatbots and virtual assistants can handle routine patient inquiries, provide basic medical information, and offer appointment reminders.
4. Combing through endless documentation: AI could go through all documentation on a patient and provide a chronological summary of a specific patient's condition over time. Then providing citations from the original text with page locations.

From a Quality angle. What if we could ask a bot where a code is in our EHR? Where do we document X in our EHR? What codes are available for this specific Value Set. Or even ask it to review a specification and tell you what the patient population is. What measures are we required to report in the IQR program this year?

What Should We Worry About?

“If we just use AI to say, well, you used to be able to see 20 patients a day, now you can see 35, then we've missed the point completely,” Dr. Scott Fowler, CEO of Holston Medical Group, said.

AI can provide significant value in healthcare, but the focus should be on enhancing patient care rather than increasing throughput. Dr. Robert Groves, CMO at Banner | Aetna, emphasized the importance of human relationships in healthcare. “How can we use AI as a tool to serve human relationships?” Dr. Jim Whitfill, Digital Healthcare Strategy and Operations at HonorHealth, echoed this sentiment, stating that while AI can help streamline certain processes, it should not replace the human element in patient care.

Using AI to Fulfill the Big Healthcare Dreams

The panel agreed that Clinical Decision Support (CDS) was at the forefront of healthcare’s most ambitious AI dreams.

1. Enhancing decision support: AI algorithms can analyze vast amounts of patient data, including medical history, symptoms, and test results, to provide clinicians with real-time decision support. This can help in diagnosing conditions, suggesting treatment plans, and identifying potential drug interactions, ultimately reducing the administrative burden of researching and analyzing information.

The panel talked about some really amazing results coming out of imaging right now. They referenced recent research that showed the AI software was better at correctly interpreting diagnostic images, and that unfortunately (scarily) the AI performed better without the assistance of clinician review.

2. Optimizing resource allocation: AI algorithms can analyze data on patient flow, resource utilization, and staff scheduling to optimize resource allocation within healthcare facilities. This helps in reducing administrative tasks related to managing patient queues, optimizing staff schedules, and ensuring efficient use of resources.
3. Enhancing retro-analytics: Quality leaders are always trying to figure out why patients are failing out of a measure. Is it a missing code that isn’t mapped? Is it non-compliance? Is it negation documentation? (<<It’s always negation.) Where did the fall-out happen? What unit? Which clinicians are causing the most patients to fall out of the measure — for whatever reason? If there were a bot that you could ask these questions to and it gave you reliable and accurate answers, you’d have such a powerful way of moving faster in improving patient care.

What should we worry about?

Boy, this all sounds amazing! But we don’t quite trust it yet. In fact, the panel chair, Paul Battle, Lenovo’s Executive Director, referenced a recent survey which showed trust in AI is way down. Half the U.S. population trusts AI in healthcare and half do not.

Nathan Bay, head of mergers and acquisitions at Citi Financial, pointed out that trust in AI also hinges on how organizations handle and use the data. “Governance is important because what you put in is going to impact what comes out. And how you think about the kind of the governance regime that sits over the inputs is important.”

Here’s my favorite quote from the session. Nathan said, “I think that in many new technologies there's an overestimation of the impact in the short term and an underestimation of the impact in the long term.”

We are currently in an AI hype cycle. AI will not solve all your problems today. It’s not really ready for you…yet. But the long-reaching consequences are not fully understood. In the same way, none of us could understand how the internet would revolutionize everything when it came into the mainstream. Now, looking at the incredible (and frightening) things the internet has provided for us in a couple of decades, we should use that to carefully govern how we adopt AI in our healthcare system.

So, what can you do with AI today?

Well, if you’re one of the lucky few whose organization has invested in AI resources — either through internal means or with an external AI vendor — you probably have a few ideas. If you don’t have designated AI resources available to you, you can still play with a few things.

• Record your next meeting (with willing participants) and upload the recording to an AI software program. ChatGPT is free, so you could start there. Ask it to summarize the meeting for you and provide you with action items, parking lot items, and an agenda for the next meeting.
• Take any document — it could be a report, a word doc, or even an email, and ask it to check it for spelling and grammar. Or ask it to re-write it to sound more or less professional. You could ask it to shorten it or make it longer.
• You could take some stream-of-consciousness bullets and ask it to re-write them in narrative form.
• Did someone send you an article or research paper to read? You can insert that URL and ask it to give you a bulleted summary.
• Test out what it knows. If you find yourself searching for something, pop it into the AI software to see if it knows the answer. “What are my 2024 IQR requirements.” Make sure you ask it to show you its sources.
• For something more fun, you can also ask it to do things like the following examples: “Plan me a 7-day road trip down the east coast.” “What do most people do for one day in New York City.”

What should we worry about?

Do not insert anything that is PHI. This should go without saying, but be careful. If your meeting recording has patients mentioned during the call, that would be PHI. ChatGPT is an open-source AI software program. There are a ton of issues with using this open-AI source model. There are copyright issues surfacing as well. So, make sure you aren’t using it to generate content that wasn’t derived from you to begin with. And always ask it to share its sources if it’s giving you a definitive answer so you can validate what it said.

In conclusion, while there is a lot of talk about AI in healthcare, implementation and impact are still in the early stages. The potential for AI to relieve administrative burden, enhance decision support, optimize resource allocation, and improve retro analytics is promising. However, trust in AI and the proper governance of data remain significant concerns. It is important to approach AI adoption in healthcare with caution, learning from the lessons of the internet revolution. While we may not have all the answers today, the long-term impact of AI in healthcare cannot be underestimated. As we continue to explore and develop AI capabilities, it is crucial to prioritize patient care and maintain the human element in healthcare interactions.

(^^I asked an AI bot to write me a concluding paragraph for this article I wrote. What do you think?)