How generative AI and large language models can close the gap between data and outcomes in healthcare

• For decades, healthcare workers have been asked to document patient care in pursuit of better outcomes and more innovation, but this has resulted in a greater administrative burden and less patient-centric care.
• Artificial Intelligence (AI) can transform healthcare by creating end-to-end solutions that bridge the divide between documentation and outcomes.
• We must ensure these tools are safe, equitable and widely available to achieve better, higher-quality care for all.

Over 100 years ago, the staff of Mayo Clinic developed a simple yet transformative idea for healthcare: a unified health record where each patient’s health information would be stored, accessed and updated by the entire care team. The idea quickly proved essential to our understanding of exceptional healthcare – capturing high-quality data and documentation means better care for that patient and it lays the foundation for groundbreaking research and approaches that will benefit future patients.

In the last 30 years, electronic health records (EHR) have enhanced our ability to capture data and documentation. It is estimated that a single hospital produces 50 petabytes of data a year, the equivalent of 170 years of HD video footage. Healthcare professionals responsible for capturing, documenting and navigating this data have struggled under the increased administrative burden. One study found that physicians spend up to half their workday, and an additional 28 off-hours per month, completing EHR tasks. According to two other reports, in 2021, 60% of physicians reported at least one symptom of burnout and 70% of EHR users reported work stress from using health information technology.

We expect the volume of data in healthcare to grow exponentially with the addition of genomic and other omics data. We are already seeing a system at breaking point – straining healthcare workers to document and collect more data at a pace, format and structure that, instead of improving care and spurring breakthroughs, may be inhibiting care and innovation.

Optimizing generative AI to enhance healthcare

In recent years, researchers, physicians and providers have begun developing and using predictive AI-powered tools trained on large datasets to quietly transform healthcare. This is leading to better care for patients, better experiences for healthcare workers and more innovation and access across the sector. Predictive AI tools are helping Mayo Clinic and healthcare workers across the globe diagnose diseases earlier, personalize care and improve outcomes.

Last summer, the European Union announced a €220 million investment in AI solutions, part of it apportioned for healthcare AI. In Brazil, physicians are using AI to rapidly detect subtle changes in CT images, leading to earlier cancer diagnoses and better outcomes. Health tech startups in Rwanda are using AI to improve supply-chain management and ensure patients and providers have the essential resources they need. South Korea’s Ministry of Health has introduced new policies to speed the development of innovative medical devices, including those powered by AI. For nations with a critically low doctor-to-patient ratio, AI-powered tools have the potential to provide access to life-saving care and expertise to millions in need.

Since the debut of ChatGPT in November of 2022, generative AI tools, including large language models (LLM), have further inspired many data-rich fields, including healthcare, to reconceive their approaches to collaboration, innovation and problem-solving. LLMs, which excel at summarizing and analyzing large volumes of data, as well as generating cogent text, have seen rapid deployment across the healthcare sector.

In just one year, we and many healthcare providers have used LLMs to streamline and automate administrative documentation, draft messages to patients, summarize visit notes and detect potentially missed or inaccurate diagnoses. In one study published by Mayo Clinic Proceedings, surgeons used an LLM to write high-quality notes that meet clinical guidelines in five seconds, something which would otherwise take seven minutes to write. This is 84 times faster and at a quality of satisfactory outputs for patients and their physicians.

The speed of generative AI and LLM innovation and application in healthcare shows the profound potential of generative AI in healthcare, including in unlocking greater insights from imaging and genomics. It also highlights, however, the urgent need for an overarching strategy to ensure the safe, validated, effective and integrated use of solutions that can bridge the divide between data, providers and patients.

Using generative AI and LLMs to create a transformative future for healthcare

As healthcare continues to innovate and implement novel generative AI tools across the sector, research institutions, healthcare organizations and government agencies must work to make progress in four general areas:

1. Rules for Trustworthy AI and LLMs

Healthcare is a sector built on a foundation of trust between patients and their providers. Without that trust, even the most innovative tools will fall short of their potential impact. As well documented as the strengths of generative AI and LLMs are, just as well documented are their capacity to create misinformation, false information or make up answers. To take full advantage of LLMs and other AI tools, healthcare organizations, tech innovators, research institutions and government agencies must collaborate to create shared standards for trustworthy, ethical and equitable AI tools that are fit for every patient they encounter. Organizations, such as the Coalition for Health AI, have established initial guidelines for developing and using these transformative tools, including a Blueprint for Trustworthy AI.

2. Automation

By combining LLMs’ ability to create cogent text with other technologies, such as ambient listening and natural language processing, we can offload rote or tedious tasks and strengthen the relationship between patients and their care teams by removing intrusive technology — such as a keyboard and a screen — from patient visits. Automating routine tasks allows healthcare workers to focus on putting patients first.

3. A 'second opinion' for healthcare providers

To provide high-quality care, physicians and other healthcare professionals must navigate a complex web of information within each patient’s EHR. Generative AI specifically trained on clinical data, documentation and analysis could serve as a streamlined 'second opinion' for physicians, instantly analyzing the health record to suggest any alternative diagnoses or missed information, while highlighting other key information for the care team. This virtual assistant to physicians and other healthcare professionals could improve healthcare quality, reduce errors and save valuable time for patients and their care teams by arriving at the right diagnosis and treatment sooner.

4. Generative AI as an expert healthcare resource for the world

AI tools, such as LLMs, have the potential to remove the barriers to accessing high-quality healthcare created by time and distance. Specifically, a generative AI tool trained on the expertise of leading healthcare providers, and powered by decades of longitudinal clinical data, could make world-class, highly complex care and expertise readily available to people across the world.

We are only at the beginning of how AI tools will reshape and transform healthcare, especially with the rapidly advancing field of generative AI. We must recognize, however, that no individual AI tool or technology is enough on its own to solve the structural inefficiencies and inequities facing healthcare globally. To take advantage of these transformative tools, we must develop them within a dynamic platform model of care, centered on secure, de-identified data and innovative physical infrastructure designed to maximize it, this will spur the development of innovative tools and expand access to such tools and high-quality care to millions of people in need.