When AI sounds right, but isn't: how to better train your AI for clinical care

The most dangerous thing about health advice from AI isn’t that it can be wrong. It’s that it’s almost always delivered like it’s right. It is confident, step-by-step, empathetic, tuned to how you like to read. Without the proper training, it can be impossible for someone seeking advice from an AI tool to recognize when and where the AI reasoning has gone sideways.

I've seen this in my everyday life as I try out new apps that use AI. When I review another AI chatbot’s answer to a health question, gaps jump out to me, as a physician, immediately. But, ask me to use an AI design for a woodworking DIY project and that instinct disappears — I cut, drill, mortise and assemble until the project is done.  And upon completion I find the furniture piece is too big for the space or the door swings the wrong way.  Whether the problem is my AI prompt or the woodworking knowledge trained into the underlying model, the result is the same - a costly, failed project.

The same confidence that has misled me in furniture-making is what patients could experience, when the cost of being wrong is far greater.

As a trained emergency medicine physician, what keeps me up at night is the patient who acted on plausible-sounding AI generated advice that was incomplete or wrong and never had a reason to doubt it. Or, it’s someone navigating a complex, life-altering treatment decision who needed a comforting hand on their shoulder, not merely an AI-generated polished paragraph.

Physicians, health systems, clinical researchers, have spent years asking whether AI is fit for healthcare: purpose-built models, clinical workflows, safety guardrails. 

I often ask myself: what will it take to feel confident that the people using the AI tools I help develop will be safe? Trust has to be earned through consistently delivering accurate recommendations that are relevant for the user, transparently stating the tool’s limitations, and explaining how safety and performance is watched after launch, including when a human is looped in for oversight.

The answer starts with the same foundation clinicians rely on — access to underlying data and an understanding of the individual’s complete health picture. Human clinicians rarely recommend where to seek care, suggest possible diagnoses, offer treatment advice, or refill prescriptions without first consulting a medical record, asking questions to fill gaps, and accounting for context: social circumstances, health literacy, and what a patient may be leaving unsaid. We also have been trained to recognize when we don’t know, and we should say so.

Patient-facing AI should behave the same way. Recommendations need to be grounded in the highest quality available information and explicit about where that information comes from and what might be missing. The danger isn’t only wrong answers. It’s confident, polished answers built on patient-reported inputs alone, with no acknowledgment that a critical signal may be absent, such as an unreported pregnancy, an escalating mental health crisis, or a misunderstanding of medication instructions. Any of those gaps can turn plausible guidance into unintentional harm.

Equally important is attempting to fill in knowledge gaps before delivering guidance. AI tools that offer health related advice should be trained to recognize missing information in an individual’s health record or from their description of symptoms - similar to how clinicians fill in medical histories and update medication lists upon clinic intake.  

At Verily, the patient-facing AI agents that power Violet in Verily Me are designed with that clinical posture in mind. We’ve 

 that architecture works in practice — from structured assessments to multi-stage safety checks. My focus here is simpler: 

a confident tone is not a substitute for clinical context.

Clinicians don’t stop learning after training. We undergo continuing education, peer review, and performance oversight throughout our careers. AI that delivers health-related guidance needs the same discipline after launch — not only before it.

It is unrealistic to assume every patient interaction can be reviewed by a human in real time. What is realistic is transparency: the goal for any AI tool used for health advice should be to offer explanations of what gets monitored, how often monitoring occurs, who is notified when something goes wrong, how quickly issues are addressed, and how learnings feed back into product improvement. Patients and clinicians should be able to calibrate trust to risk — the way we already approach many aspects of care today.

For the Verily Me app, our clinical quality team has designed a monitoring framework with clear standards for performance oversight. That framework has been designed to connect real-world safety and performance signals back into our formal product development lifecycle — so iteration, escalation, and quality improvement are driven by evidence, not assumption. Models can improve over time; they can also drift. Monitoring has to account for both.

Be clear about monitoring — and about its limits

One of the most important aspects of designing AI in healthcare is ensuring that patients are directed to licensed medical providers when additional clinical evaluation or judgment is needed. Today, AI tools can help people prepare for visits, understand options, and navigate routine questions. But when symptoms are complex, evidence is uncertain, or decisions may significantly impact a patient’s health, the system must reliably recognize its limitations  and direct patients to appropriate medical care immediately. 

While AI can support many aspects of care navigation and decision support, it still cannot replace the patient-clinician relationship when evidence is uncertain and choices are life-altering. As AI continues to advance, defining — and then redefining — clear rules for when to redirect to a telehealth or in-person medical visit will remain as important as any model capability.

At Verily Health, we put these principles into practice by ensuring our AI agents prioritize accuracy and compliance. We achieve this through strict data quality standards and automated guardrails that oversee the AI-generated recommendations. By clearly defining how our AI tools make recommendations using available health data — and how they are monitored — we are laying the foundation to build trust for all users. We owe patients more than confidence. We owe them clarity, context, and care that knows its limits.

*The information contained in this page is intended to outline Verily’s general product direction and is not a commitment or legal obligation to deliver any functionality. Product capabilities, timeframes, and features are subject to change.*

Card image

How to avoid AI that sounds confident, but isn’t right. Verily Health is designing and training models and AI agents with clinical standards and reasoning.

How to avoid the fluent but fallible AI

It’s time to welcome AI to the care team

Agentic AI for enterprise health care remains primarily in the back office rather than at the bedside. According to a 

, 81% of physicians now use AI in their practice, but mostly for administrative workflows, even though 76% see an advantage to using AI tools in patient care. AI is most commonly leveraged for summaries of medical research (39%), creation of discharge instructions and care plans (30%), and documentation of billing and chart summarization (28%); activities where the perceived risk is low and the potential cost savings are quantifiable. Only 17% of physicians reported use of AI for assistive diagnosis and the reported use for patient-facing recommendations was much lower. When it comes to delivering care in clinical settings, AI remains in the background.

This is in stark contrast to the widespread appetite among consumers to engage with AI directly on symptoms, possible 

. According to a recent report published by OpenAI, 40 million+ Americans use ChatGPT 

 for health related guidance. Patients are uploading their medical records, perhaps at risk to their privacy, to LLMs to ask clinical questions about their health. Increasingly, it is evident that consumers want to be able to ask a question, get an answer that is informed by their medical record, and take action according to clinical guidelines and best practice. They don’t want to wait to call, send a portal message, or make an appointment.

At the same time, the policy and regulatory environment shaping the use of AI in care delivery is increasingly favorable and quickly evolving. HHS has 

 on how to accelerate adoption and use of AI as part of clinical care, FDA has been working to clarify regulatory pathways for 

 for AI has been announced to spur AI innovation across industry sectors. 

Amidst all of this momentum, we have to ask ourselves: what is holding us back? 

What will it take for us to embrace the role of AI as part of the patient’s care team?

One way to think about AI is as a digital team member, handling a defined scope of practice and working alongside clinicians within defined boundaries. 

The American Medical Association (AMA) defines 

 as “those activities that a person licensed to practice as a health professional is permitted to perform.” This framework allows for flexibility in how activities are conducted while outlining clear boundaries based on the individual's training, evaluation, and certifications. As an example, the 

 includes monitoring patient vitals, administering medication, and discussing care plans, but it does not include diagnosis or prescribing medication. For physicians, the activities within and boundaries for their scope depends on their specialty. For example, pediatricians can diagnose and treat children for acute and chronic conditions but cannot perform surgery on children unless they have completed both a general surgery residency and pediatric surgery fellowship. 

Under a scope of practice, human clinicians can practice with some degree of flexibility, without needing to give the exact same response to every similar patient, while also clearly understanding the limits as to what they can and cannot do and say to patients under their care. Similarly, AI tools functioning under a defined 

 could produce outputs that still mirror the appropriate variations necessary for standard of care. When a model output reaches the edge of that boundary, AI tools could invoke another role on the care team, either by escalating vertically up the team responsibilities, in the same way that Physician Assistants hand off complex diagnosis cases to MDs, or horizontally across the team, engaging a complementary role, such as when the patient needs medical explanations provided from a different perspective.

Defining the role of AI under the familiar scope of practice framework helps to conceptualize how AI could work as part of a patient’s care team. Yet, to make this a reality, there are many unanswered questions that will first require clarity. These include:

Second, what’s necessary to give AI a role

At Verily, we are thinking intentionally about ways to help bring clarity to some of these unanswered questions. We consider the intended care scope for our AI agents, and how they might be used within a patient’s broader care paradigm, when establishing criteria for evaluation and ongoing monitoring and improvement. 

Verily grounds the use of AI models in a foundation of safety and compliance, while also recognizing that individual AI models differ in risk, complexity, and regulatory requirements. As a result, each AI model is uniquely evaluated for its complexity, risk, and the appropriate regulatory and compliance path.

For example, for evaluation we employ a risk-based testing approach for our AI agents, matching the level of testing to the clinical risk and defined scope. While wellness and education use cases focus on rapid iteration based on user feedback, higher-risk use cases must pass formal validation standards and undergo clinician-led testing. Similar to how clinicians are evaluated and trained by more experienced professionals in apprenticeships or residency programs, the performance of our AI agents is evaluated by qualified healthcare professionals for safety and accuracy. 

For ongoing monitoring and improvement, we are developing a monitoring system that uses automated alerts to flag potential safety concerns, model errors, or uses beyond the AI agent’s intended scope. We are building in direct pathways for automated and user-initiated escalation to human clinicians, similar to how nurse practitioners or medical residents have protocols for escalating to physician attendings when a task exceeds their defined scope or skill set. We also have plans to conduct periodic, formal audits of our AI models where we will re-evaluate the application against established test sets to proactively identify and address any performance degradation or "model drift" that may occur over time. In the similar way to how performance concerns with human clinicians are escalated to review boards and evaluated for disciplinary action, we also plan to escalate model performance issues to a cross-functional governance team for review. When needed, we intend to modify the model's guardrails, roll back to a previous version, or temporarily take the agent offline to ensure user safety.

Ultimately, we believe patient care can be both clinician-led and AI-forward, with appropriate workflows and engagements driven by technology. As the broader industry begins to clarify national standards and frameworks for evaluation, monitoring, reimbursement, workflow integration, and accountability structures, a future where AI functions as part of a patient’s care team is in reach.


*The information contained on this page is intended to outline Verily’s general product direction and is not a commitment or legal obligation to deliver any functionality. Product capabilities, timeframes and features are subject to change and should not be viewed as commitments.

Despite high demand, AI remains in the background for patient care. Learn what we think it will take to embrace the role of AI as part of the care team

An image of the Verily Health logo

Verily was born as an Other Bet at the outset of Alphabet—a ‘Moonshot’ with a shared ambition to impact lives at scale, but specialized with a focus on technology deeply rooted in clinical and scientific excellence. 

Today, we’re moving into our next phase of growth to accelerate precision health and improve patient lives, now as an independent company. 

I’m incredibly proud to share that we’ve closed a $300M investment round led by Series X Capital. This round includes an investment from Alphabet, who will remain a significant minority investor, but will no longer have a controlling stake in Verily. UCHealth and the University of Colorado Anschutz who joined us in a strategic collaboration last October, now become investors as well. They have been deep in the trenches with us on this journey, and seeing them move from customer to strategic investor is the best kind of validation for the path we’re on. Thank you!

To mark this new chapter, we are officially moving forward as Verily Health.

This $300M funding round is an important accelerator for Verily. We’re using it to scale the parts of our ecosystem that are already making an impact in the market.

That means accelerating our AI roadmap—including Violet, our AI companion available through 

, the precision health platform that serves as the data and technical foundation for all Verily solutions today. 

This means further investment in Pre data solutions such as 

, supporting over 21,500 researchers worldwide. It means expanding capabilities for 

, an essential solution for nationwide monitoring of infectious disease, and growing 

 to help more people manage the day-to-day realities of diabetes and hypertension with tailored care and always-on support. On the clinical research side, it means further growing our 

 tools to run real-world studies, using our registry of re-contactable participants, 

 for health data harmonization, and Workbench for multi-modal data analysis.

We’ve always believed that in healthcare there is no path to "doing it alone." Our work is always in service of our customers, partners, and most importantly, the individuals whose lives we aim to improve. We’ve focused on building an ecosystem where our unique blend of scientific, clinical, and technological capabilities meet people where they already are.

 is a great example of this. We’ve integrated their full-stack AI tools and latest GPUs directly into our Pre Workbench, which lets our researchers 

. But we’re also moving closer to the patient—working with 

 to integrate the Galaxy Watch 8 into clinical research, and contributing to 

 Agentforce Health to help providers make sense of messy medical data. And our work with 

 team remains our foundational model for how a health system can actually use AI-ready data to change care on the ground.

The Power of What We’re Building Together

I’ll be the first to say it: this hasn't been a linear journey. Over the last few years, we’ve gone through a deliberate—and honestly, sometimes difficult—transformation. We had to make hard calls about where to double down and where to let go. But that work fueled the focus that was needed to launch innovative solutions like the Verily Pre platform. 

To our Veeps (employees), you have shown incredible patience and a lot of grit. Change at this scale is never easy, and I want to personally thank you for staying the course. This important milestone and 

 that went into achieving it, belongs to you!

Verily was purpose-built for this moment.

 We aren't just a tech company; we’re a team of data scientists, clinicians, and researchers and many other critical functions that come together as One Verily and believe healthcare should be as precise as it is personal.

Our strength is in the intersection. We’ve spent ten years learning how to blend the frontier of data science with the non-negotiable standards of medical rigor. We know that in health, "moving fast" only matters if you’re moving safely. That’s why scientific and regulatory excellence isn't an afterthought for us—it’s in our DNA.

As we embark on this new chapter today, a sincere thank you again to our team, our customers and partners, and our investors. We are Verily Health, and we’re just getting started!

Verily announces $300M in funding led by Series X Capital, and with UCHealth and University of Colorado Anschutz, to accelerate precision health AI.

Our Next Chapter: Verily Health

Powerful AI tools are rapidly reaching millions of consumers  — an exciting but complex trend, especially in the high-stakes world of healthcare. General AI models can be helpful for organizing and making sense of scattered medical information. But when it comes to health, patients need a system that offers high quality, clinically informed guidance, in order to truly make a difference. As these tools evolve from passing exams to interpreting years of personal health history, we must ask: are general purpose AI tools ready to navigate the complexities of human health?

Closing gaps in care requires a purpose-built approach. As both a care and technology company, Verily’s clinical, scientific, data science, and engineering teams partnered together to develop new approaches to symptom assessment and education tools that modeled validated clinical protocols and the ‘in-clinic’ process. Verily recently introduced Violet, a personal, private AI companion accessible through the 

. With Violet at their fingertips, anyone using Verily Me can better understand their health history, prepare for doctor visits, and get questions answered about symptoms.

Behind Violet sit a number of specialized healthcare AI agents. One of these multi-agent architectures was developed to guide users in their symptom assessment, to provide self-care recommendations, and give advice on when to see a healthcare professional. This clinically grounded approach — designed and modeled after the ‘in clinic’ process — preserves patient context, and supports individuals with clear, helpful information.

Within the Verily Me app, specialized healthcare AI agents sit behind Violet’s interface to address complex health tasks like assessing symptom context, based on the following principles:

Active investigation (evaluate, don't just listen).

 Patients know how they feel, but they may not realize which details are medically important. Through active investigation, agents prompt the user with relevant questions to gather deeper context and detail, and other agents actively 

 The model leverages available medical records, not just what a patient verbally mentions.

Extracting more accurate results from messy data.

 Medical data is often messy and inconsistently represented. While standard AI models try to process all this raw information — mistakes and all — Verily believes fine tuning and organizing data can result in more accurate results. Within Verily Me’s secure and private environment, anyone using Verily Me can ask health questions that can be modeled on their fine tuned health dataset.

 The system's process is made more transparent and predictable. Instead of accepting a final answer; models are trained to 

 and conclusions. This minimizes hallucination risks and helps to ensure that the final recommendations are rigorous, reliable, and that we can state to the user the rationale leading to the AI's conclusion and recommendation.

The clinical logic behind the AI companion in Verily Me

Violet addresses the challenges of symptom assessment via a purpose-built design philosophy. Rather than asking a single AI to do everything at once — collect symptoms, check records, and generate recommendations in one conversational turn — we decompose the assessment into specialized components, each with a focused clinical responsibility.

It’s like a relay race instead of a solo sprint. Rather than a single agent trying to manage the whole process, a coordinated team handles specific tasks and passes the 'baton' of information smoothly to the next. 

Our architecture is inspired by how triage actually works in an urgent care setting.

 When you walk into an urgent care clinic, you don't have a single conversation with one person who handles everything, and you don’t have to know what information is important to share. 

Inside Violet: How a team of AI agents work to ensure safety

When a patient asks Violet about a symptom, they progress through a carefully orchestrated architecture:

Violet’s patient journey: A coordinated sequence

Our architecture is built on a safety-first mentality, ensuring checks occur continuously throughout the patient journey rather than at a single point of failure.

 We don't wait until the end to flag danger. Emergencies are screened throughout symptom collection, after collection, and during the final assessment. If any stage detects a high acuity scenario, the system is built to immediately override the workflow to guide the patient to appropriate level of care when a high level of risk is detected. 

 The system automatically screens for complex histories — such as pregnancy or active cancer, organ transplants, or severe autoimmune conditions —  and guides patients to talk to their healthcare provider about their symptoms. 

 Our agents are architected to deliver structured outputs based on clinical practice guidelines. Every assessment follows a strict schema: a clear recommendation, a list of possible conditions, specific self-care steps, guidance on when to see a doctor, and “why we recommend this” (the agentic reasoning). This format focuses on giving users helpful advice right away, while also showing users exactly how the AI reached its conclusion. 

 To ensure the clinical rigor of the design translates into patient comprehension, we conducted user research and used user‑centered design with iterative prototyping. This allowed us to identify high‑value improvement areas, uncover beliefs and barriers to AI use, and measure sentiment around privacy and safety. We focused on reducing cognitive burden and "confusion-related" errors, and prioritized the key principles of transparency and empathy. We developed a linear, easy-to-answer flow that was designed for the needs of the intended user population in its intended environment.

 Verily’s Quality Management System (QMS) ensures strict adherence to regulatory and industry standards. Before release, every AI model must clear rigorous compliance controls; post-launch, we utilize formal change controls to vet and approve every update. This end-to-end lifecycle ensures that our technology remains safe, effective, and high-quality from initial design through retirement.

Built for trust: The core safety systems of Violet

To solve the output variability problem inherent in LLMs, we treat the model less like a chatbot and more like a medical board.

Addressing the output variability of LLMs with a design built for healthcare

Assessing safety: Violet's early performance in beta

This deep dive into our comprehensive approach highlights the critical difference between Verily Me and other AI tools currently available to ask questions about your health. 

While general-purpose LLMs are powerful, Verily Me’s Violet, a coordinated set of healthcare AI agents, embeds individual steps in the clinical process to assess symptoms, much like a coordinated team at an urgent care clinic, and is delivering more personalized, context aware support for users even in its early days as a beta feature.

We’re just getting started — and we begin with an uncompromising commitment to user safety. By building a multi‑agent architecture with multi-stage emergency detection, we’re delivering a system grounded in clinical rigor and focused on patient well‑being.

*Violet’s powerful, but she can’t treat or diagnose. 

She’s not human. Always consult a licensed healthcare provider for diagnosis or treatment. 

Violet is developed in collaboration with medical experts

 and undergoes regular optimization to ensure high standards of accuracy and safety.

 She bases her responses on the info you give her, your available health records, and training data. Errors or gaps in any of these sources could affect Violet’s responses.

Violet can only offer guidance for a limited set of situations, see 

The Verily Me difference: Built for healthcare

Special thanks to the Verily clinical, science, data science and AI, and product teams whose input made this work possible.

Explore Verily’s multi-agent architecture for symptom assessment, designed to prioritize clinical rigor over general-purpose AI.

Is your AI fit for healthcare? 

From delays to delivery: Accelerating study activation for clinical research sites

For academic medical centers (AMCs) and health systems, clinical research is a critical pillar for institutional reputation and financial health. However, while the science of medicine is advancing quickly, the administrative process required to start a trial is struggling to keep up.

The average site initiation time for Phase 1-3 trials at AMCs and hospitals stretches to a lengthy 9.4 months.&sup1; This stands in stark contrast to ambitious guidelines, such as the National Cancer Institute's 90-day target. This six-month activation gap represents more than just lost time; it represents lost revenue, lost sponsor trust, and, most importantly, lost opportunities for patients awaiting life-saving therapies.

Many site leaders attempt to solve startup delays by adding more coordinators or pushing teams to simply work faster. However, the bottleneck is rarely due to a lack of effort, but rather, from the compounding complexity of the modern clinical trial.

Today’s activation process involves a number of stakeholders:

When these departments operate in silos, information is passed like a baton in a relay race — often dropped, frequently delayed, and usually invisible to the leadership team until a deadline is missed. Coordinating between disparate spreadsheets, email threads, and legacy clinical trial management systems (CTMS) can lead to prolonged activation times. 

The impact of the activation gap extends far beyond the research office. To understand the true value of shortening this cycle, leadership must look at three specific areas of risk:

Every month a trial sits in the startup phase is a month of zero revenue for a study that has already incurred costs in the form of administrative labor. For sites operating on thin margins, the inability to move quickly from a signed contract to a billable patient visit can lead to significant budgetary shortfalls.

A research site’s ability to rapidly activate studies is an important selection factor for sponsors and CROs. After all, every day a clinical trial is delayed can cost sponsors an estimated $800,000 in lost revenue opportunities.&sup2; If a research site consistently takes 200+ days to activate, they’re likely being deselected for high-profile trials before they even receive the feasibility questionnaire. 

For patients with late-stage or rare diseases, time is of the essence. When a trial activation is delayed by six months due to administrative friction, it can limit access to potential cures, making operational efficiency critical to patient care.

The hidden costs of delayed study activation

To close the activation gap, sites must move away from status update culture where teams meet weekly to discuss what happened in the past. The evolution should be toward a predictive culture, where teams use data to proactively manage study activation workflows. 

True innovation in study startup involves:

Shifting from reactive to proactive management 

The 9.4-month average is not a law of nature; it’s a symptom of fragmented processes and outdated tools. As clinical trials grow more complex — spanning decentralized models, real-world data, and advanced biomarkers — sites that succeed will be those that make activation speed a core capability, not a bottleneck.

Solving the complexity of study startup requires a blueprint for operational excellence.

 are designed to address these industry challenges head-on, bringing structure to complexity, connecting workflows, and helping sites move faster without sacrificing rigor. Drawing from this approach, Verily has identified the key drivers of activation delays and outlined a practical roadmap for sites targeting the 90-day NCI benchmark.

Built for sites: Accelerated study activation to revolutionize trials

The faster path forward — less time, same quality

. “Overcoming Study Start-Up Delays: Best Practices for Research Sites.” ACRPnet.org, May 7, 2025. 


Smith, Zachary, Joseph A. DiMasi and Kenneth Getz

. Quantifying the Value of a Day of Delay in Drug Development. Tufts Center for the Study of Drug Development, August 2024. 


Whitepaper: Accelerating activation to revolutionize trials 

Download for efficiency strategies specific to study sites

Learn how to accelerate activation and improve patient access for reducing study startup delays at clinical research sites. 

From Delays to Delivery: Faster Study Activation for Sites 

Two decades after precision medicine began, critical insights are still trapped in low-resolution data — a bottleneck that Verily, UCHealth, the University of Colorado Anschutz, and RefinedScience are now overcoming. Terabytes of valuable clinical data, the natural byproduct of patient care, are largely unfit for AI and machine learning because they were never collected as research-grade evidence. This struggle with low-resolution data spans nearly every disease and demands a fundamental re-evaluation of data governance and culture. To achieve high-fidelity evidence at scale, governance must now integrate deep clinical expertise — and new infrastructure must be built to support data transformation. With 

, a powerful curation engine built on Pre, disparate and multimodal data is transformed into a clinically informed, FHIR native model. Following positive results in a test case study with the University of Colorado Anschutz and RefinedScience, the platform is now providing the "resolving power" to make high-resolution clinical data actionable and unlock its latent potential.

Acute myeloid leukemia (AML) is a deadly blood cancer with survival rates significantly lower than common cancers like breast cancer (

) and even lower among older patients. CU's hypothesis is that by leveraging advanced molecular and clinical data, including next-generation sequencing techniques (like CITE-seq), they can discover new, clinically important subclasses of AML. Using the data, they identified the anti-CD70 antibody drug 

, which had been deprioritized by Janssen after a conventional Phase 2 trial. A subsequent analysis by RefinedScience suggests a complex precision biomarker can now identify a specific subclass of AML patients who 

 the drug, creating a major opportunity for drug rescue. Their joint venture, OncoVerity, currently has a new 

 underway to prove this. Ultimately, Verily's physician-scientists and data technologists are building the necessary AI "lenses" to accelerate this new understanding of disease and research, establishing a scalable design pattern for precision health.

University of Colorado Anschutz: Refining biomarkers and precision opportunities

The recent project with UC Health, CU Anschutz and RefinedScience confirms the promise of their approach but also reveals the technical gaps they need to cross. This work leveraged 

, powered by Pre, to take clinical data, predominantly molecular diagnostic reports from the electronic health record, and turn it into structured labels (a matrix of patient rows and abstracted label columns). Historically, this abstraction has been a time-consuming, "artisanal" process. 

Their team then uses this matrix (X) to discover new AML classes by trying to find a machine learning function 

 that reliably predicts a clinical outcome (Y, such as 5-year survival) using the formula 

The results showed that Verily could automate the creation of these label columns using modern, scalable methods.  Verily deployed targeted AI models, including LLMs capable of processing TIFFs while preserving tabular formatting. The results were encouraging, indicating that automated systems could match and even exceed manual curation quality:

However, our work revealed a greater opportunity for Verily: we can greatly enhance the model's resolving power by utilizing a clinical knowledge-augmented approach that injects deep clinical expertise into the data processing pipeline, establishing a clinically informed, FHIR native data model that is AI and analytics-ready.

(a key AML metric), a simple, "zero-shot" AI approach achieved 72% accuracy. 

When clinical knowledge was incorporated into the extraction, the accuracy rose to 95%. Expertise matters.

Here are three key opportunities to extend their accomplishments:

In clinical medicine, some tasks must be deterministic (e.g., treating severe bleeding before a benign mole), and there is little room for AI "hallucinations." A knowledge graph, which represents facts ("entities") connected by their relationships ("semantics"), provides the structure needed to enforce this clinical logic.

Expertise matters: How clinical knowledge boosts AI accuracy to 95%

The results of our recent project with UC Health, CU Anschutz and RefinedScience (see full case study) — demonstrated by high-value clinical use cases (e.g., the potential to rescue cusatuzumab) combined with an evolving tech stack of data refinement and knowledge graphs — establishes a scalable design pattern for precision health. Following the initial program's success, CU Anschutz has committed to a multi-year expansion, scaling the solution to population-level datasets across new therapeutic areas, including cardiovascular disease, neuroscience, and ophthalmology. 

The promise of AI to transform health is historic, harkening back to past technological leaps. We’re now positioned to use data, knowledge, and technology to provide new, high-resolution "lenses" on human disease. Our modern applications help partners resolve new understandings of disease and identify value where it was previously unrecognized. The full benefits of AI for precision health will emerge through an intentional build of our tools that brings together technical capabilities, clinical expertise, with health system data. By focusing on patient safety and impact, this approach creates the structural knowledge network necessary to accelerate precision health for all.

A new understanding of disease to accelerate research and care at scale

&sup1; Data on file
&sup2; Based on Verily estimate - Data on file

To get more details from our case study.

Verily, the University of Colorado, and Refined Sciences are overcoming data bottlenecks in cancer research. Discover how our AI-driven Refinery platform achieved 95%+ accuracy and a 97% reduction in manual data abstraction time for AML.

The 'unseen' value of precision data

Improving real-world results with real-world data for GLP-1s, cardiometabolic care & beyond 

Cardiometabolic disease is an ever-growing public health crisis, spurring the need for better life-changing health insights and innovation at a faster pace. To keep up, biomedical research demands evidence generation that’s designed to represent the real world, providing a continuous perspective that captures the complexity of each individual's real-life health experiences with diabetes, obesity and related cardiometabolic conditions. Verily is dedicated to realizing this vision, building upon the foundation of the Verily-sponsored Baseline Health Study — a research initiative launched in 2017 that generated a deeply phenotyped, multi-dimensional map of health for over 2,500 individuals. The Baseline Health Study results have fueled over 20 publications that have deepened biomedical knowledge in many therapeutic areas, including cardiometabolic health, such as 

 in walking data from wearables that may reveal early heart failure before symptoms appear, and that 

 not merely a risk factor. 

This commitment to leadership in biomedical research and evidence generation continues with the 

 solution, fueled by datasets from the newly-launched and Verily-sponsored Lifelong Health Study and flagship, Cardiometabolic Cohort. This participant-centered research registry integrates multi-modal health data to help researchers answer the most pressing questions in contemporary healthcare and medicine, including the challenges starkly highlighted by the often mystified patient journeys with glucagon-like peptide-1 receptor agonists — or GLP-1s.

The first drug in the GLP-1 class came to market in 2005 for treatment of type 2 diabetes. Today, GLP-1 medications are still transforming treatment for diabetes, but also obesity, weight loss and other chronic conditions, such as chronic kidney disease, metabolic dysfunction-associated steatohepatitis, and obstructive sleep apnea. Yet even with time and use, the medical industry — payers, providers and innovators like pharmaceutical companies — still have unresolved questions, lacking a clear, contextual view as to why patients struggle with, switch, or stop treatment.

Why? Largely because researchers need more comprehensive, real-world data (RWD) from individuals about their real-world experience and behaviors over time, which is lacking from traditional sources like fragmented electronic health records (EHRs) and static claims. With more ongoing, high-resolution biological and behavioral data streams, researchers could reveal the context behind patient outcomes and cardiometabolic disease, and better understand the effectiveness of new treatments. Experiences with GLP-1 discontinuation is a primary case in point. Standard RWD can answer the what, showing that a prescription wasn’t refilled or that therapy was stopped, but this data may not provide clarity on the reason why. Was it due to significant side effects, perceived lack of efficacy, insurance hurdles, or medication unavailability? 

The lack of context severely limits the ability to modify existing drug regimens, develop programs to improve patient outcomes, and uncover new therapeutic opportunities.

A GLP-1 dilemma: Knowing what vs. Understanding 

The Verily Viewpoint Evidence solution empowers end-to-end evidence generation for clinical research, in part, from the newly-launched, 

. The initiative is a Verily-sponsored, observational registry that continuously follows participants on their health journeys to provide a multi-dimensional, realistic view of how health changes over time. This consented, engaged community will help transform RWD from static pulls into a dynamic, ongoing participant conversation.

As Lisa Lehmann, MD, PhD, and Verily's Medical Director for Clinical Research describes, “Through the longitudinality and integration of a vast array of health datasets, the Lifelong Health Study will provide pertinent, missing RWD to accurately portray the continuum from illness to wellness and vice versa.”

Learn more, directly from Verily experts. 

, From data gaps to dynamic dialogue: A new era of real-world evidence. 

Designed for research that reflects real life

Verily’s integrated evidence capabilities are built on the 

 — an AI-native platform for precision health that utilizes a FHIR-native approach to blend complex health data into actionable, reusable evidence. This platform enables Verily’s flagship Cardiometabolic Cohort,  a dataset that is sourced directly from this engaged Lifelong Health Study community. 

This deep, diverse health data — clinical, patient-reported, medication and device/wearables — within the cohort and curated by Pre, will help also bridge fragmented data for GLP-1s and other cardiometabolic research and care, enabling clearer clinical context and greater real-world relevance.  


Focus on Verily’s Cardiometabolic Cohort

Viewpoint Evidence provides continuous insights through easier, ongoing connections with Lifelong Health Study participants for deploying surveys, drawing from a pre-engaged, consented population that is ready for rapid, precise sub-cohort selection. This can advance GLP-1 and cardiometabolic research by reducing the time, cost, and complexity of recruitment and screening, enabling efficiency from study design to real-world insight.

Combined with easier access to integrated clinical, lifestyle/behavioral, and device/wearables data from the Cardiometabolic Cohort, this allows 

researchers and life sciences teams to build a complete picture of cardiometabolic disease progression and use of GLP-1 therapy.

According to Andrew Trister, MD, PhD and Verily Chief Scientific and Medical Officer, “These precision evidence capabilities are necessary to address care challenges in clinical practice and people’s daily lives. For example, let’s consider that 20% of patients on GLP-1 therapies don’t have significant weight loss despite adherence, and genetics offers no clues as to why there is variability in response. Verily can facilitate a study of this left-out cohort, revealing the myriad of reasons for non-response — cost, incorrect usage, side effects, etc. — that are difficult to track using typical health data alone.” 

This high-resolution, continuous data may be valuable to pharmaceutical companies, particularly for GLP-1 research, as it allows monitoring and linking of patient-reported symptoms and side effects, such as specific gastrointestinal side effect profiles, to clinical events and device data. Dr. Trister goes on to explain that this deep understanding enables the refinement of existing GLP-1 treatment, such as through the potential microtitration of GLP-1s to optimize individual patient outcomes and side effect management, such as limiting nausea.

Decoding causes of GLP-1 successes and setbacks

As researchers delve deeper into the mechanisms of action and effects of GLP-1s and related, emergent and next-generation drug classes and formulations, they are expanding investigation on potential applications beyond cardiac and metabolic diseases. A key area focuses on the role of GLP-1s in reducing inflammation. Dr. Lehmann notes, “Since inflammation is believed to drive chronic diseases more generally, and because GLP-1s appear to target inflammatory pathways, they may have very far-reaching uses and health consequences beyond diabetes and weight loss, which is very exciting to me. Verily will help companies unlock pharmacologic applications in these areas, such as connections to longevity research. Basically, the research community believes we’re just at the ‘tip of the iceberg’ in understanding GLP-1 capabilities.”

So as the GLP-1 indications and use expand, so does the need to understand their everyday impact for revealing the mechanisms of disease progression 

 help fill these critical gaps by uniting high-resolution clinical, lifestyle, and behavioral data, in part, from continuous patient input. This approach empowers and equips pharmaceutical and life science teams with research-ready, relevant data to refine therapies, design more effective adherence strategies, and accelerate progress toward precision cardiometabolic care and beyond. The result: a clearer, more complete understanding of how GLP-1s and other future therapies can deliver on their promise for individuals in their real-lives.

Connecting real-world data to real-world results

On demand webinar: A new era of real-world evidence

Explore closing health data gaps with Verily experts on your time.

Stories