Is AI The New Medical Expert?
The emergence of advanced LLMs is making a “technological consult” a clinical imperative.
GPT Summary: The development of advanced AI and language models is surpassing the cognitive skills of many physicians, leading to the use of AI as a new medical expert in areas such as diagnosis and predicting patient outcomes. Truveta Language Model (TLM) is a domain-specific LLM trained on de-identified medical records that can normalize healthcare data for medical research with high accuracy. The introduction of “generalist medical AI” (GMAI) in medicine can further revolutionize the field by carrying out diverse tasks using diverse medical modalities. LLMs can also improve accuracy and utility in other industries such as finance. As AI and LLMs continue to be developed and refined, they will undoubtedly have a significant impact on the medical field and beyond.
From hospital rounds to a telemedicine consultation, medicine commonly relies on the voice of the expert. The free exchange of information and the support of the “though leader” has helped dissect and define many aspects of care.
However, with the rapid advancement of technology in recent years, the role of the physician as the ultimate authority in medical decision-making has been called into question. With the development of increasingly sophisticated AI and language models, it is becoming increasingly clear that the cognitive skill set of many (if not most) physicians is being surpassed by technology. This raises the question: what does it mean to be an expert in medicine today, and how can physicians adapt to this new reality to optimize care?
The use of artificial intelligence and language models such as GPT in medicine has been steadily increasing in recent years. This technology has the potential to transform the way we diagnose and treat illnesses, with the increasing power of large language models (LLMs) bringing us to a point where the capabilities of technology exceed those of even the most experienced clinical experts.
A relevant example of a domain specific LLM is the newly announced Truveta Language Model (TLM). The model is trained on the largest collection of complete & de-identified medical records that represent the diversity of the US population. TLM transforms data from 80 million patient journeys, 5.5 billion diagnoses, 3.1 billion encounters, 2.4 billion medication orders, and 2.5 billion notes that are updated daily from 28 health systems.
Assisted by TLM, healthcare and life science researchers can study concepts that were previously inaccessible in clinician notes—importantly not found in claims data—that is now structured for analytics. TLM normalizes for clinical accuracy to currently support medical research and improved clinical outcomes.
TLM’s training is done with a tool custom-designed to train the AI for clinical accuracy. The results of the model are continuously checked as it runs through a workflow that involves dozens of clinicians continuously refining the accuracy of concepts being actively researched. When greater accuracy than clinical experts in a particular healthcare domain is achieved, we deploy the model into Truveta Embassies to start normalizing data. TLM is currently achieving more than 92% accuracy on diagnoses, medications, lab results, lab values, clinical observations, and more.
Referencing the chart above, Truveta’s clinical expert annotation team labels thousands of raw clinical terms, including misspellings and abbreviations, to train TLM to normalize healthcare data for clinical research. Note that this labeling of this chart is 100% focused on clinical accuracy with no commercial bias. With different types of data, TLM learns how to normalize raw medical text to the most appropriate medical information ontology. And of particular interest is the how the model accuracy is now above established clinical experts across key clinical parameters.
Further, a current paper is Nature leverages the value of LLMs with the introduction of the “generalist medical AI” (GMAI) across six medical use cases that the authors suggest will establish a better model than existing platforms.
The exceptionally rapid development of highly flexible, reusable artificial intelligence (AI) models is likely to usher in newfound capabilities in medicine. We propose a new paradigm for medical AI, which we refer to as generalist medical AI (GMAI). GMAI models will be capable of carrying out a diverse set of tasks using very little or no task-specific labelled data. Built through self-supervision on large, diverse datasets, GMAI will flexibly interpret different combinations of medical modalities, including data from imaging, electronic health records, laboratory results, genomics, graphs or medical text.
With this increase in processing power, we are now crossing a threshold where technology has begun to exceed the cognitive capabilities of even the most experienced clinical experts. This means that AI and LLMs can provide a level of diagnostic acumen that was previously thought impossible. This is particularly important in cases where rare diseases are concerned, where human clinicians may not have the experience or knowledge necessary to make an accurate diagnosis. However, the day to day utility of LLMs in clinical medicine may reflect its biggest opportunity and advantage.
But medicine is just a subset of domain-specific LLMs. From law to business, LLMs can develop a deep understanding of the terminology and concepts used within that field, leading to improved accuracy and utility. An example of this is BloombergGPT, an LLM developed by Bloomberg that is specifically designed to analyze and understand financial news. By training the model on a vast dataset of financial news articles, BloombergGPT can quickly and accurately extract insights and trends from large volumes of data. As we continue to develop more domain-specific LLMs, we can expect to see even greater improvements in accuracy and utility, leading to increased efficiency and innovation in various industries.
The increasing ability and value of AI and LLMs are undeniable. With the ability to analyze vast amounts of data and develop an understanding of complex concepts from medicine to business, AI has established a “trajectory of change” that will impact almost every aspect of our lives.
