Science is never static, and neither are we. Providence Genomics uses learning partnerships to improve care and move the field of personalized medicine forward. Working together, we are building an evidence engine to uncover new therapies, integrate innovations into standard care, and guide care transformations with evidence that is grounded in real-world care experiences.
AI is an important part of this strategy, but it’s important we get it right. We support the full R&D life cycle to help move ideas from conception to large-scale reach and impact with this approach:
Develop: Use research and data science as an engine to develop AI applications that can improve health care.
Test: Develop & test AI tools in real care settings to help improve them and assess overall impact & value.
Scale for Impact: Develop integrated strategies to get the best tools to as many patients as possible.
Prov-AI Models
AI Foundation Models
As an early leader in developing healthcare-focused foundation models, Providence has built a solid base for advancing AI tool development. Our foundation models are designed to learn general skills—such as language understanding, image recognition, and pattern processing—before being tailored to support specific AI-driven tasks. Researchers use these as the building blocks for specialized task models. Examples include:
Prov-GigaPath: This open-source foundational model is based on a massive dataset (Prov-Path) that rapidly analyzes large pathology images to aid in cancer diagnosis, research, and treatment. One year after its release, Prov-GigaPath was downloaded over 2 million times by clinical and research teams around the world. Research: Nature “A whole-slide foundation model for digital pathology from real-world data” The model achieved approximately 90% accuracy for detecting actionable EGFR mutations in lung cancer, reducing the need for molecular testing by 43%.
Prov-GigaTime: An open-source model that predicts how patients will respond to immunotherapy. The model is trained to identify “exceptional responders” by reviewing pathology images to identify immune content in tissues. Research: Researchers trained themodel on a Providence dataset of 40 million cells, pairing pathology slideswith mIF data examining 21 different proteins. Prov-GigaTime was applied tosamples from 14,256 cancer patients across the Providence system which includes51 hospitals and more than 1,000 clinics. The work produced a virtualpopulation of approximately 300,000 mIF images that cover 24 cancer types and306 cancer subtypes.
Prov-AI Task Models
Providence is leveraging its early, highly successful foundation model work to rapidly develop a suite of AI tools, or task models, to aid caregivers by thoughtfully integrating AI tools and supports into their daily tasks. Examples include:
Prov-AI Trial Simulator: This clinical simulation tool is powered by Prov-TrialScope and uses AI to extract and analyze real-world evidence (RWE) from electronic medical records (EMRs) to estimate treatment effects and trial designs for large populations. This model could revolutionize the way we conduct clinical trials and generate RWE. Research: New England Journal of Medicine/AI. This work shows that by automating the curation of EMR data and leveraging advanced AI tools, researchers can generate robust, reliable, and cost-effective RWE to help identify the most effective treatments for specific patient populations.
Prov-AI ChartMiner: This AI model transforms free-text clinical documentation—one of the richest but least structured sources of real-world data—into accurate, scalable, research-ready information for oncology. By training on patient-level supervision from existing cancer registries, the model eliminates much of the cost and delay associated with manual chart abstraction. Research: Patterns. In a study of more than 135,000 patients, the deep-learning approach achieved exceptionally high accuracy (AUROC 94–99%) in extracting key tumor attributes and maintained strong performance across external health systems.
An Ethical AI Approach to Transformation
We’re committed to an ethical AI approach that keeps patients, doctors, and their relationships at the center of health care operations. Some of our key principles include:
Patient-centered. AI models should be trained to help optimize outcomes that matter most to patients.
Assist, not replace. AI tools should never replace clinical discernment. Instead, they should support clinical decision making by helping doctors gather and curate the complex data they need to make the right call.
Shared decision making. An AI tool shouldn’t try to tell patients what to do. Instead, it should provide information about the options in ways that help doctors and patients decide together what to do.
Our AI research includes a portfolio of different ProvAI tools to help transform care. Follow our research papers here.
“As research within whole-person data and genomics expands, we are getting better at predicting risks and finding new treatments. This is creating a significant and promising shift in healthcare.”
Anton J. Bilchik, MD, PhD, MBA, FACS
Professor of Surgery and Director, Gastrointestinal Research
Program, St. John’s Cancer Institute, Providence
Learn more
Stay up to date on our research by following some of our primary researchers:
