Editors: Raymond D. Aller, MD & Dennis Winsten
Stanford clinicians use AI to draft lab test result comments
June 2025—Whether it’s easier to edit a document created by another source or create more or less the same document from scratch is open to debate. But clinical informaticians at Stanford Health Care, Palo Alto, Calif., are banking on clinicians preferring to assess and adjust versus starting with a blank page. Therefore, the health care system has been conducting trials on the use of generative artificial intelligence to draft patient-centric interpretations of pathology test results that these care providers can review and edit and then share with patients.
The tool was built at Stanford using Anthropic’s Claude 3.5 Sonnet general-purpose large language model through the Amazon Bedrock Web service for building, deploying, and scaling generative AI applications. It interprets laboratory test results and offers a plain-language explanation of those results that clinicians can edit and then share via the health care system’s patient portal. “The intellectual property that drives this capability is embedded in the generative AI components,” says Christopher Sharp, MD, chief information officer at Stanford Health Care and a clinical professor of medicine at Stanford Medical Center. Stanford has, in turn, embedded the AI’s functionality for drafting reports and explanations into its workflows via the health system’s Epic EHR.
“Since the 21st Century Cures Act, we have been challenged to ensure that we provide results in a transparent way to patients, but also in way that is empathetic and supportive,” Dr. Sharp says. To that end, those involved in the Stanford trial are evaluating not only whether the use of generative AI makes it easier for clinicians to create result comments, but also whether it increases the proportion of patient test results released with a comment and how those comments influence patients or are received by them.
Stanford’s tool provides a concise explanation that is personalized for the patient based on a set of prompts. However, clinicians must review those result comments since AI can generate incorrect or misleading information, or hallucinations. The accuracy of the final documents is solely the responsibility of the clinicians creating them, and use of the technology is optional, Dr. Sharp adds.
“We think it’s very important that the clinicians themselves understand the source that this is being created from and the potential hazards,” Dr. Sharp says. “First of all, it’s very transparent that this was not drafted by a human; this was not written by the pathologist; this was drafted by a generative AI model. And we talk about the types of errors that may be seen and how to be vigilant around this.”
In the initial pilot test of the technology, conducted with a group of approximately 40 Stanford primary care physicians, about half of the participants consistently used the tool as intended—the other half drafted their own comments from scratch or did not write comments. “We think that this is success,” says Dr. Sharp, who noted that about 300 primary care providers are now using the tool. “We see, in even partial utilization, that this provides a lot of value to those who use it. When we survey them, they tell us that.”
Stanford initially pilot tested its AI tool with primary care physicians because they tend to order less complex laboratory tests and fewer tests than specialists. It has since expanded the trial into areas more heavily reliant on clinical pathology, such as dermatology and gastroenterology.
“There are many high-volume simpler reports that can have relatively repetitive interactions—colon polyps or simple skin excisions that might have relatively repetitive but higher volume communications,” explains Dr. Sharp. “In those settings, we think these tools might be particularly valuable. In settings where the pathology is more complex—in an oncology setting or a hematology setting—we have not yet gone into those areas with this technology.”
“In the coming months,” Dr. Sharp adds, “we will be working with pathology explicitly to hone, understand, and assess the technology around pathology specimen result comments.” Working with the pathology department, information technology staff and clinical informaticians will adjust the AI’s prompts as necessary to “bring forth the best interpretation as a draft result comment generated by AI.”
—Manny Frishberg
Pitt and Vizzhy using AI and precision medicine to address health care on a global scale
The University of Pittsburgh School of Medicine and the artificial intelligence provider Vizzhy are collaborating to leverage AI technology to scale disease-management efforts and reduce costs via a new platform.
The AI-based platform, called Gainmed (for generative and agentic intelligence navigated multiomics medicine), is designed to advance comprehensive patient-centered medical care. The first phase of the project involves recruiting 20,000 participants and 200 to 300 primary care providers worldwide, with the latter focused on sequencing samples and building the Gainmed knowledge base.
The initiative will also establish the Pitt-Vizzhy Longevity Labs, in partnership with Illumina. The labs will provide multiomics laboratory services to expand precision medicine.
“By combining data analyzed at Pitt-Vizzhy Longevity Labs with generative AI, Gainmed delivers more comprehensive care plans, empowering physicians and nurse practitioners to make more informed and personalized treatment decisions,” according to a press release from the University of Pittsburgh.
The Pitt-Vizzhy labs will eventually extend across Europe and Asia to collect and analyze data, with the long-term goal of delivering “top-tier health care” to underserved populations worldwide, according to the press announcement.
The initiative, led by Pitt, will use the health care institution’s cloud and data centers in lieu of those provided by technology companies. The partners have established a five-year objective of sequencing samples from 1 million people and setting up regional laboratories.
Also participating in the partnership are Thermo Fisher Scientific, Amazon Web Services, L&T-Cloudfiniti, and LTIMindtree, among others.
Applied Spectral Imaging updates cytogenetics software
Applied Spectral Imaging has launched GenASIs version 8.4.1, which offers integrated workflows and other capabilities for pathology and cytogenetics laboratories.
“By cutting in half the time needed to prepare G-banded or Q-banded karyograms, ASI’s new software version [allows] cytogenetics labs to manage their workload more efficiently, freeing up precious time for higher-value tasks,” says ASI CEO Limor Shiposh, in a company press statement. Pathology lab users can seamlessly align H&E or IHC whole slide images from third-party scanners with FISH slides scanned on the ASI system.
The software, which combines digital pathology and cytogenetics workflows in one platform, is compatible with Windows 10 and Windows 11 operating systems.
The new software version has been incorporated into ASI’s brightfield, fluorescence, and spectral imaging and analysis solutions, including HiBand, HiFISH, CytoPower, HiPath Pro, and PathFusion.
Applied Spectral Imaging, 760-929-2840
Illumina and Tempus AI join forces to train genomic algorithms
Illumina and Tempus AI are extending their long-standing partnership by combining artificial intelligence technologies from Illumina with Tempus’ multimodal data platform to train genomic algorithms, with the intent of accelerating clinical adoption of molecular testing.
“By expanding our collaboration with Illumina, we are combining our strengths in technology and data analytics with their strengths in developing new sequencing technologies to drive forward innovation and advance precision medicine,” said Terron Bruner, chief commercial officer for Tempus, in a company press statement.
The vendors have previously jointly focused on developing assays and other tools for patient care, health economics, and bioinformatics pipelines.
Illumina, 858-202-4500
Cleveland Clinic and Akasa collaborate on AI tools for coding and documentation
Cleveland Clinic has announced that it is employing an artificial intelligence solution for medical coding from Akasa and pilot testing a clinical documentation integrity tool from the vendor.
“The AI coding assistant [Akasa Coding Optimizer] can read a clinical document in less than two seconds and process more than 100 documents in 1.5 minutes,” according to a joint press release from Akasa and Cleveland Clinic. “In addition, the technology is designed to understand clinical context, beyond keywords, and adapt to a patient’s complexity.”
Revenue cycle staff at Cleveland Clinic typically review more than 100 clinical documents per case, including pathology reports, progress notes, and discharge summaries, according to the health care institution.
“We chose to pilot this technology with Cleveland Clinic because we wanted to test our AI against some of the most complex patient encounters in the world,” said Malinka Walaliyadde, CEO and cofounder of Akasa, in the press announcement. “We are proud to now be rolling it out, as well as collaborating with Cleveland Clinic’s coders and CDI [clinical documentation improvement] specialists in developing additional products to make the revenue cycle process easier and more efficient.”
Cleveland Clinic reported that it will apply multiple Akasa AI-powered tools during the mid-revenue cycle at all of its U.S. locations.
Akasa, 650-209-0358
Precision for Medicine partners with PathAI
Precision for Medicine and PathAI are jointly developing artificial intelligence-based technologies and integrating PathAI’s digital pathology and analysis capabilities across Precision for Medicine’s clinical trial and biospecimen operations.
Under the agreement, Precision for Medicine will deploy select tools from PathAI, including the latter’s AISight digital pathology image-management system and AI-powered algorithms, to enhance biospecimen and clinical trial services.
“Together, we’re deploying tools that add critical quality control steps to tumor biopsy workflows and apply machine-based learning and unsupervised algorithms early in the development cycle to help identify which biomarkers are most relevant for clinical efficacy,” said Darren Davis, PhD, Precision for Medicine’s senior vice president of global digital pathology, genomics, and liquid biopsy solutions, in a press announcement from the com-
panies.
The collaboration will specifically focus on offering tools and analytical services in the areas of biomarker discovery, spatial biology, and tissue-based clinical research. The tools generated through the partnership will be available through Precision for Medicine.
PathAI, 617-500-8457
Dr. Aller practices clinical informatics in Southern California. He can be reached at rayaller@gmail.com. Dennis Winsten is founder of Dennis Winsten & Associates, Healthcare Systems Consultants. He can be reached at dennis.winsten@gmail.com.