The following is a guest article by Julien Dubuis, Chief Commercial Officer at Nym
The excitement surrounding large language models (LLMs) and generative AI in healthcare is palpable. Predictions abound, heralding a new era where these technologies reshape patient care, administrative processes, and medical research. Yet, much like the early days of AI in other sectors, healthcare’s embrace of LLMs has sparked an initial wave of overestimation. While their transformative potential is immense, the journey to realizing it will be long, requiring both technological maturation and sector-specific adaptation.
The $1 Trillion Opportunity
McKinsey estimates that generative AI, including LLMs, could unlock up to $1 trillion in value for the healthcare sector. This value could manifest in many ways, from improving clinical outcomes and reducing operational inefficiencies to streamlining administrative tasks. Healthcare, however, is unlike finance or retail, where AI adoption has rapidly advanced. It is a field where a single error can have life-or-death consequences. Additionally, patient safety, privacy, and ethical considerations present significant challenges for seamless AI integration.
Healthcare’s diversity adds further challenges. Each discipline has unique workflows, data structures, and regulations. Many LLMs have been trained and fine-tuned on clinical data, each with its own performance scorecard. Some are better suited for specific healthcare tasks, such as aiding in drug discovery, while others excel at automating patient communication or interpreting medical images. Developing AI systems that effectively address this broad range of needs requires a deep understanding of each sub-sector’s intricacies. In this landscape, a “one-size-fits-all” solution simply doesn’t exist.
Early Wins in Administrative Efficiency
Despite these hurdles, LLMs are beginning to carve out meaningful roles, particularly in healthcare administration. Insurance claims processing, patient correspondence, and billing tasks often consume valuable resources that could be better spent on patient care. Reducing such administrative burdens is where LLMs can shine.
A particularly striking example of LLMs’ impact is in ambient scribing. LLMs can now produce detailed, structured clinical notes by listening to doctor-patient interactions, significantly reducing the time clinicians spend on note-taking. While the ROI and long-term impact are yet to be fully proven, this innovation in clinical documentation shows promise in improving provider satisfaction and potentially reducing burnout, allowing clinicians to focus more on patient care and spend less time on burdensome administrative tasks.
In medical coding, LLMs have shown potential, but, as highlighted in NEJM AI, off-the-shelf LLMs are not a “silver bullet.” Nym takes a more tailored approach by selectively using LLMs for specific tasks, such as parsing documentation and performing Named Entity Recognition (NER), a process that identifies medical terms within text. Nym fine-tunes clinical LLMs trained on biomedical literature to excel at tasks like negation, temporality, and subjectivity, ensuring higher accuracy. Nym’s coding engine then applies a rules-based system rooted in established coding guidelines from the American College of Emergency Physicians (ACEP), the American Academy of Professional Coders (AAPC), and Scripps Health Standard Operating Procedures (SOPS) to assign billing codes. This process occurs within seconds and with zero human intervention, delivering over 95% accuracy, streamlining workflows, reducing errors, and recovering revenue efficiently.
Payers are also adopting LLM-driven processes to improve claims adjudication, further streamlining the overall billing ecosystem. This kind of automation can save time and money, allowing healthcare providers to focus more on their core mission—delivering care.
Furthermore, LLMs are showing promise in automating repetitive tasks such as drafting appeal letters for insurance claims. Currently, hospitals and clinics invest significant resources in appealing denied claims, a process that often involves manually gathering medical records, summarizing clinical justifications, and drafting comprehensive responses. By employing LLMs to generate initial drafts of these letters, healthcare organizations can save time, reduce errors, and potentially recover more in reimbursements, enhancing their operational efficiency.
Another area showing progress is appointment scheduling. LLMs can assist in managing appointment requests and cancellations, streamlining communication between patients and providers. By automating these interactions, healthcare facilities reduce administrative strain and improve the patient experience with quicker response times.
Critical Care: Progress in Diagnostics and Safety
While administrative applications of LLMs are gaining traction, their deployment in direct clinical care is advancing more cautiously. That said, progress is being made in areas where the technology serves as an adjunct to human expertise, offering tools to enhance accuracy and speed, without replacing critical decision-makers.
Pathology is one field where LLMs are proving useful. Traditionally, pathologists manually review slides of tissue samples to diagnose conditions like cancer—a process that is time-consuming and susceptible to human error. In one study published in The Lancet, LLMs were trained to analyze medical images, flagging abnormalities that may be missed by even the most experienced pathologists. The LLM doesn’t replace the specialist, but provides a second set of “eyes.” This allows pathologists to focus their attention on the most critical cases, speeding up the diagnostic process and improving the overall accuracy of disease detection, especially in high-volume areas where human error is a risk.
Another emerging application is in pharmacy error detection. Medication mistakes, such as incorrect dosages or dangerous drug interactions, are a leading cause of preventable harm in healthcare. According to a 2024 study in Nature Medicine, LLMs are being used to review prescription orders for potential errors, such as contraindications or excessive doses, before the medication is dispensed. By flagging these mistakes in real time, LLMs can help reduce the incidence of adverse drug events and improve patient safety.
The Challenge of Product-Market Fit
Despite early successes, LLMs have yet to deliver the sweeping transformations some have anticipated. Achieving product-market fit in this space remains a significant challenge. One of the main hurdles is addressing the long tail of healthcare use cases—those low-frequency, high-complexity scenarios that require specialized knowledge and nuanced handling. While it’s possible to build solutions that effectively address 30-50% of more common cases, the long tail presents a unique problem. As Benedict Evans points out, many AI technologies initially perform well in narrow applications but struggle when scaling to more complex, real-world use cases. In healthcare, where the stakes are high, adapting LLMs to handle this wide range of scenarios demands extensive testing, fine-tuning, and collaboration across disciplines.
Regulatory compliance further complicates adoption. Technologies that interact with patient data must adhere to strict privacy and security laws, such as HIPAA in the U.S. There is also a need to mitigate algorithmic bias, which could exacerbate health disparities. Achieving true product-market fit will require not only technical innovation but also alignment with healthcare providers, technology developers, and regulatory bodies.
Building a Sustainable Future for LLMs in Healthcare
Though the road to widespread adoption may be slow, the long-term potential of LLMs in healthcare is immense. As the technology matures, it could become integral to areas such as personalized medicine, predictive analytics, and patient engagement, fundamentally transforming how care is delivered.
To realize this potential, stakeholders must focus on patient safety, ethical transparency, and fostering collaboration across the healthcare ecosystem. The promise of LLMs lies not in replacing human expertise but in augmenting it, creating a more efficient and accurate healthcare system, while safeguarding trust and well-being.
About Julien Dubuis
Julien Dubuis is Chief Commercial Officer (CCO) at Nym, where he leads the sales and marketing teams to drive commercial growth. Prior to Nym, Julien served as Vice President of Sales at Clarify Health and as a project leader at The Boston Consulting Group (BCG), where he focused on the intersection of technology and life sciences. He holds a PhD in physics from Princeton University and a BSc in physics from the Ecole Normale Superieure in France.
No comments:
Post a Comment