Paging Dr. Google: MedLM, the Latest Large Language Model to Hit the Market

Google Cloud launched MedLM, a family of healthcare-focused generative AI models.

The key word is “healthcare-focused.”

These models are specifically trained on healthcare data and primed for healthcare use cases, which I’ve discussed at length here.

In this article, I’ll break down Google Cloud’s MedLM, give you a quick lesson on LLMs and generative AI, and discuss the current healthcare AI arms race.

The Deets

Google’s MedLM is a family of healthcare-focused generative AI models tailored for the healthcare industry. One of the models is Med-PaLM 2, a model that has garnered significant attention for its impressive ability to accurately answer complex USMLE questions (it received way more hype than I did for passing my boards…). This achievement underscores the power of LLMs like Med-PaLM 2: by being trained on extensive medical datasets, they can not only understand intricate medical information but also generate precise, high-quality responses. Pretty neat.

MedLM has three main purposes:

1. Answering medical questions

2. Creating draft summaries from existing medical documentation (e.g. after visit summaries)

3. Generating insights from unstructured data

For now, the model is not intended for clinical purposes but rather for workflow, admin, and non-clinical research purposes.

Google has already partnered with several organizations to integrate MedLM into existing platforms and software.

• HCA Healthcare and Augmedix (backed by MedLM): Augmedix is a healthcare-focused ambient AI company that has been piloting Med-PaLM 2 and will integrate MedLM technology into its technology stack. HCA Healthcare and Augmedix have also partnered to roll out Augmedix’s technology in hospitals.

• BenchSci: BenchSci focuses on pre-clinical drug research and development by employing its AI-powered ASCEND platform and integrating MedLM, which collectively streamline the discovery process and enhance the accuracy and efficiency of scientific studies. The integration with MedLM further augments the platform's capabilities in identifying and analyzing novel biomarkers.

• Accenture: Accenture is working with Google to enhance healthcare organizations' use of generative AI, leveraging Accenture's industry expertise and Google's technology to improve patient access, experience, and outcomes, with solutions like Accenture's Solutions.AI for Processing and MedLM to automate processes and provide insights for better patient care.

• Deloitte: Deloitte and Google Cloud are collaborating with healthcare organizations to deploy generative AI and MedLM in interactive chatbots, aiding health plan members in navigating care options and assisting care teams in efficiently accessing provider and benefits information for optimal member support.

As we see, Google's MedLM is making significant strides in healthcare, from assisting in medical question answering to integrating with various platforms for enhanced healthcare services. With this innovative use of generative AI in the healthcare sector, it’s important to understand the broader concepts of Large Language Models and generative AI.

A Quick Lesson on LLMs and Generative AI

Large Language Models (LLM), such as Google’s MedLM, are advanced AI systems trained on vast datasets, enabling them to understand and generate human-like text. Now, when I say “vast” I mean extraordinarily large and diverse collections of data, consisting of billions of words.

These datasets include a vast array of medical texts—from clinical case reports and medical research papers to patient records and treatment guidelines. The size and diversity of these datasets are crucial for the AI to learn not just the nuances of medical language but also the complexities of medical knowledge and clinical decision-making processes. This extensive training enables the AI to provide more accurate, context-relevant responses and insights, a critical factor in healthcare applications.

Generative AI, a broader category under which LLMs fall, refers to AI that can create content, be it text, images, or even code, by learning from existing data. In healthcare, this means these AI models can draft medical summaries, provide insights from unstructured data like patient records or physician-patient conversations, and assist in answering complex medical queries.

Dashevsky’s Dissection

We’ve officially entered The Healthcare AI Arms Race.

Incumbent technology companies, including Amazon, Microsoft, and Google, are racing to develop (or acquire) generative AI technology that can improve healthcare. This includes AI technology that can not only enhance patient care and service delivery but also streamline both clinical and administrative workflows. Each of these companies is strategically focusing on developing AI solutions that facilitate clinical decision-making, automate the generation of medical notes, and offer copiloting capabilities—an integrated approach combining the first two functions:

• Google: launched MedLM and has partnered with Augmedix and HCA Healthcare. If you need any more details just read the above content.

• Amazon: launched HealthScribe, a generative AI service that can listen to physician-patient conversations and create text-based summaries or notes. They also partnered with 3M Health Information Systems to further develop Amazon’s HealthScribe.

• Microsoft: officially acquired ambient AI company Nuance in 2022. Nuance recently launched Nuance Dragon Ambient eXperience (DAX), which is a fully automated clinical documentation app, capturing clinician-patient conversations and translating them automatically into notes using ChatGPT-4. DAX is integrated into EHRs, streamlining workflows. Microsoft also partnered with Epic to develop and integrate generative AI into Epic’s EHRs.

Already, these technology companies are showing success in applying their expertise (technology) to healthcare, which has not always been the case for big tech and healthcare. Big tech’s healthcare initiatives have a remarkable history of not succeeding. Smaller companies end up beating big tech because smaller companies are solely focused on healthcare while big tech is doing a million other things. Hence, the reason Microsoft chose to acquire ambient clinical documentation company Nuance instead of building its own service from the bottom up!

However, my healthcare intuition tells me that this time around, big tech’s AI initiatives in healthcare won’t fail. Rather, these AI initiatives will catalyze the next epoch of healthcare and medicine. Bold statement, I know. But if you’ve used generative AI technology yet, for example ChatGPT, you can literally see and feel how the technology can be applied to healthcare. And the technology is already here at our fingertips!

Given the foreseen impact of AI in healthcare, there will be a race for big tech to develop the most robust AI solutions that can be integrated and deployed in EHRs and healthcare platforms throughout the U.S. This race isn't just about pioneering advanced technology, though. It's about reshaping the fabric of healthcare delivery. The potential for transformative change is immense—from enhancing patient experiences to enabling clinicians to focus more on care than paperwork. As AI becomes more sophisticated and ingrained in healthcare processes, we're likely to witness a significant shift in how medical services are provided, making healthcare more efficient, accessible, and personalized than ever before.

I will add a note of caution in regards to my optimism about AI in healthcare, which underscores the need for careful, ethical, and regulated integration of AI in healthcare:

1. Not fail proof: AI systems, no matter how advanced, are not infallible. They are trained on existing data, which can carry inherent biases, potentially leading to skewed or unfair outcomes in patient care.

2. Humans are complex: The complexity and variability in individual health cases mean that AI might not always capture the nuances of human health and disease. And, AI cannot replace the empathy, understanding, and the human touch that physicians provide.

3. Privacy and security: There's also the critical matter of privacy and security. Handling sensitive medical data requires stringent protocols, and the risk of breaches or unethical use of data is a significant concern.

In summary, Google Cloud's MedLM represents a significant advancement in healthcare-focused generative AI, offering capabilities like medical question answering, drafting summaries from medical documentation, and generating insights from unstructured data. This is part of a broader Healthcare AI Arms Race, with key players like Amazon, Microsoft, and Google developing AI technologies for clinical decision-making, note generation, and copiloting. Despite the potential of these AI systems, challenges like data bias, the complexity of human health, and data privacy and security concerns remain. Therefore, while AI holds great promise for revolutionizing healthcare, its integration must be approached thoughtfully, balancing technological advancement with ethical and practical considerations.