Google announced the release of MedLM, two foundational models built off Med-PaLM 2, designed to answer medical questions, generate insights from unstructured data and summarize medical information.  

The tech giant said that through piloting its LLMs with healthcare organizations, it has learned the most effective AI models are designed to address specific use cases. Therefore, the large model of MedLM is made to address complex tasks, while the other is a medium model that can be fine-tuned and scaled across various tasks. 

Google partnered with clinical documentation startup Augmedix to utilize MedLM to convert data into medical notes and announced drug research and development company BenchSci is using the technology for drug discovery to speed up the drug development process. 

The company also collaborated with Accenture to increase enterprise adoption and help healthcare organizations utilize MedLM and Deloitte and healthcare providers to help care teams obtain information from provider directories and benefits documents to help contact center agents identify best-fit providers for members.  

MedLM is available to allowlisted Google Cloud customers via Vertex AI. 

“As we reflect on 2023 — and close out the year by expanding MedLM access to more healthcare organizations — we’re excited for the progress and potential ahead and our continuing work on pushing forward breakthrough research in health and life sciences,” Yossi Matias, VP of engineering and research at Google, and Aashima Gupta, global director of healthcare strategy and solutions at Google Cloud, said in a statement. 

THE LARGER TREND

Med-PaLM 2 was tested on U.S. Medical Licensing Examination-style questions in March and performed at an “expert” test-taker level with 85%+ accuracy. The LLM also received a passing score on the MedMCQA dataset, a multiple-choice dataset designed to address real-world medical entrance exam questions. 

“It’s hard to believe it was less than a year ago that we published the first Med-PaLM paper, our medically tuned LLM. And maybe I’ll just say that the thing that’s been a big takeaway from 2023, is the pace here,” Dr. Michael Howell, Google’s chief clinical officer, told MobiHealthNews in an interview earlier this month. 

When Med-PaLM was first released, it showed 67.6% accuracy in correctly answering MedQA (U.S. Medical Licensing Exam-style questions). A couple of months later, it increased to 85%+ accuracy, and at last report, it was at 92.6% accurate. 

“Med-PaLM has been quite remarkable, and the speed of improvement has been remarkable,” Howell said during a HIMSS TV interview. “What’s really exciting about this as we think about 2024 is we have all these new capabilities in hand.”