During HLTH 2023, Google unveiled the addition of powerful generative AI capabilities to its Vertex AI platform for healthcare and life sciences companies. These enhancements enable organizations to search patient data, including clinical sources such as FHIR data and clinical notes, in conjunction with their medical large language model, Med-PaLM 2.
Vertex AI functions as a search engine, empowering organizations to develop their generative AI-powered search engines for customers. Med-PaLM, on the other hand, is a generative AI technology that leverages Google’s Large Language Models (LLMs) to provide medical answers.
This newly introduced Vertex AI Search feature can be seamlessly integrated with Med-PaLM 2. This synergy enables healthcare providers to locate answers to patient-specific medical queries within their medical records while also addressing general medical inquiries.
These capabilities expand upon VertexAI’s capacity to establish conversational search applications and question-answering functions.
Notably, this feature prioritizes HIPAA compliance and seamlessly integrates with Google Cloud’s Healthcare API, Healthcare Data Engine, and Google Health’s pilot product, Care Studio. It is currently available for trial use by healthcare and life sciences companies.
Google emphasizes that this offering aims to tackle the increasing workforce shortages in healthcare, reduce administrative burdens and physician burnout, and empower customers to access accurate clinical information, ultimately leading to more informed decisions.
Burak Gokturk, VP and General Manager of Cloud AI and Industry Solutions at Google Cloud, expressed, “Expanding Google-quality generative AI search capabilities across an organization’s ecosystem, including Electronic Health Records (EHRs), has the potential to significantly enhance efficiency, offer clinical decision support, and improve the quality of care provided to patients. Enhancing Vertex AI Search for healthcare and life science organizations remains a top priority for us because we understand that timely access to theright information and insights can be pivotal for healthcare.”
In the larger context, Med-PaLM 2 was subjected to rigorous testing, demonstrating “expert” test-taker level performance with over 85% accuracy on U.S. Medical Licensing Examination-style questions in March. It also received a passing score on the MedMCQA dataset, which is designed to address real-world medical entrance exam questions.
In July, Google researchers published a study in Nature showing that Med-PaLM provided long-form answers that aligned with the scientific consensus on 92.6% of questions submitted, closely matching clinician-generated responses at 92.9%.