AI and healthcare: What clinicians need to know

Already pressured by limited time and resources, clinicians often face the daunting task of untangling a web of information to resolve urgent clinical questions. Treating a patient can involve sifting through a multitude of medical articles, studies and guidelines — each demanding attention as clinicians navigate multiple tabs in search of pertinent insights.

Dr Louise Chang se abre en una nueva pestaña/ventana, Global VP of Clinical Solutions Strategy and Partnerships at Elsevier, drew on her own clinical experience to paint the picture:

Of all the ways AI has promised to transform healthcare, it’s here that Louise sees it making an immediate, vivid difference. Instead of laboriously searching for information across various platforms, physicians can now pose their complex clinical queries to a tool like ClinicalKey AI in a singular, coherent manner. The platform then compiles and synthesizes cited information relevant to the circumstances of the patient, quickly providing a comprehensive, succinct overview.

By amalgamating vast amounts of data and presenting it in a user-friendly format, generative AI has the potential to transform clinical practice experience, ultimately enhancing patient outcomes. As this technology continues to evolve, it heralds a new era where medical professionals can focus more on patient interaction and less on information retrieval, thereby bridging the gap between data and clinical expertise. As Louise explained:

We’re leveraging the power of generative AI to not just search this vast repository of information but also to synthesize the results and serve them up in a way that is easy to understand, to iterate on, and where you can follow references back to the source to verify information.

Overcoming reservations about GenAI in medicine

The promise of generative AI is a bold one, and it cannot come to pass without the buy-in of the clinician community. So far, that community has been circumspect about the possibilities of AI. According to Elsevier’s Clinician of the Future: AI Edition report, 96% of respondents believe AI will help accelerate knowledge discovery, yet only 26% have used it for work purposes. The survey also indicated that 82% of clinicians believed AI will cause critical errors.

Overcoming that reservation with robust guardrails is critically important, argued Dr Rhett Alden se abre en una nueva pestaña/ventana, Chief Technology Officer for Health Markets at Elsevier:

In our case, we leverage large language models to summarize information directly from high quality source materials, which is a distinctly different approach.

Rhett compared the difference to an open- and a closed-book exam; the former approach, analogous to the approach a tool like ClinicalKey AI uses, allows for the use of reference materials, while the latter, used by ChatGPT, relies solely on the training set of the LLM that can result in hallucinations.

That commitment to accuracy and credibility is paramount, especially in the medical field. As Rhett pointed out: “We offer very precise responses, very accurate responses — and just as importantly, referenced responses.”

Why genAI works faster than search tools

Rhett also noted that the conversational nature of genAI tools can help make them much more effective than search tools in clinical settings. He gave the example of treating a pregnant woman with diabetes and hypertension

“The information you need might be available in a search tool, but retrieving it is difficult because you have to piece together certain documents — a piece on pregnancy, on diabetes, on cardiovascular issues,” he explained. However, if a tool uses a conversational format, vector search and a RAG architecture, it can retrieve and summarize that information in seconds from multiple disparate clinical references:

“You go from a search methodology that might require 15 to 30 minutes, to one that happens in literally the time it takes to compose a question within ClinicalKey AI.”

Assessing AI systems for safety

For institutions considering adopting AI tools, Louise emphasized the importance of scrutinizing the accuracy of tools. “It's really important to be critical in how they assess these tools,” she stated, highlighting the dual imperatives of security and compliance inherent to the healthcare landscape. She urged stakeholders to inquire about the validation processes behind these technologies. “How are you validating the quality of what you’re delivering? How do you know the integrity of the data?” she said, underscoring the necessity for clinical evaluation and safety measures.

Moreover, Louise noted that the dynamic nature of healthcare — where no two patients are identical —demands a feedback loop between users and vendors.

“As you have more and more users using any kind of AI system, there are going to be questions or feedback that people will identify as an issue,” she explained. Therefore, the ability to foster a responsive relationship with vendors is paramount. “What kind of feedback mechanism can you develop ... that can allow you to provide feedback and have the vendor take action? That’s the kind of question people should be asking.”

Clinicians need to join the conversation

As AI tools gather speed and take shape, Louise recommended that clinicians ensure they have a voice in the conversation by sharing their thoughts with buyers and vendors:

Whenever I give talks, I encourage clinicians to share their thoughts on AI in a healthcare setting. Clinicians should be critical judges for these tools because that’s how we make progress. Everyone should take part.

That can mean looking within one’s own organization for opportunities to discuss and trial AI tools, or seeking out professional organizations that can act as a platform for your views. “In the US, we have consortia and coalitions that are starting to form around this topic,” Louise said. “Overall, I would recommend that you lean in and get involved.”

Both Rhett and Louise noted that this technology is still very much in its infancy. Louise again highlighted the importance of clinicians guiding its development:

What does AI do best?

Rhett sees AI as significantly enhancing diagnostic processes:

“I think that we’re going to see much more AI helping with pre-diagnosis, early diagnosis,” he said, emphasizing the way these technologies can transform how clinicians interact with clinical information.

Rhett used the example of the challenges posed by lengthy referral documents, noting that “those referral documents can be 100 pages long.” Given the extensive free text, clinical notes and patient information contained within, he highlighted the frustration faced by healthcare professionals in synthesizing such information:

“AI can be really indispensable on summarizing that information and providing ... key information that’s relevant to the intake and treatment of an individual,” he said.

But he emphasized that AI’s capabilities extend beyond mere summarization to include the detection and early diagnosis of diseases. “These documents can be readily reviewed and periodically reviewed by AI systems to really look at gaps in care and gaps in treatment,” he noted, highlighting the potential for AI to identify critical triage points.

He underscored the potential of AI to streamline patient care and mitigate risks associated with incomplete information. “If information is either excluded or ignored, that can be problematic,” Rhett cautioned, suggesting that AI could be pivotal in extracting crucial details that might otherwise be overlooked amid the substantial documentation in healthcare.

As Rhett concluded, the implications of these advancements are substantial, offering not only improved patient outcomes but also enhanced operational efficiency.

You’re going to see a lot of this happening ... a lot of optimization occurring as information moves around within a healthcare system.