A Look Inside Claude’s New AI-Assisted Patient Overview

At the core of every positive veterinary interaction (and its ultimate outcome) there is one constant: a trusting relationship between the care provider and the pet parent.

While trust in veterinary care can be built (or broken) in many ways—great medical quality being table stakes—we believe that one of its most important foundations is a doctor’s ability to enter an exam room and genuinely connect with their clients around a shared understanding of a pet’s needs.

In great experiences, people feel like their veterinarian has somehow been able to match their own intimate understanding of their pet to the finest details and provide sound advice that feels contextual and personalized to them. In poor experiences, they may feel like their interaction was transactional, or that the advice they’ve been given lacks connection to their pet’s history or their care preferences.

This should go without saying, but at Modern Animal, we want to empower our doctors to provide great experiences.

Of course, for veterinarians, this is easier said than done. At scale, doctors need to maintain hundreds, if not thousands, of client relationships. To do this while also making consistently good medical decisions, they must take on the daily challenge of quickly absorbing context on their patients before each appointment—a process that involves sifting through sometimes complex medical histories consisting of notes on past appointments, diagnoses, prescriptions, vaccinations, external records, and more.

Information should be easy to find, not hard.

Understanding the critical importance of pre-appointment preparation and knowing we could find ways to streamline it, we undertook an in-depth exploration to understand the answers to two questions:

1. What, and how much, information does a doctor need to connect with their clients, make great medical decisions, and provide an exceptional experience?
2. How can they access this information as quickly as possible?
   

Through our time with the team, we realized that it’s universally difficult for veterinarians to quickly synthesize lengthy medical history (especially for complex cases or patients with years of history across multiple care providers) and fully understand everything they need to know to present confidently in the exam room. When doctors nail this preparation they feel more equipped to deliver high quality medicine, provide a continuity of care, and make their patient’s parents feel heard (i.e. they don’t have to rehash their pet’s history every time they visit). 

For Modern Animal doctor’s using Claude (Modern Animal's EMR), these pre-appointment preparations were taking at least 2 to 5 minutes per case and often involved upwards of 15 to 20 clicks to gather all the necessary information.

Put simply, information was too hard to find, and it was taking too long to find it.

We conceptualized two paths forward: On one end we considered compiling static data pulled from various parts of our system (think copy-pasted) into one summary screen. On the other end, a dynamic solution that could synthesize available information and generate useful insights and recommendations to our doctors. We learned that while static data aggregation might reduce clicks, it wouldn't significantly enhance the quality of care or decision-making.

Ultimately, we decided that our solution should lean towards the 'insights' end of the spectrum.

Enter: AI-Assisted Patient Overview

Key Design Considerations

By leveraging the power of Generative AI, this feature ingests a patient’s entire available medical history and aggregates all pertinent medical details into an easily accessible snapshot that is available in just one click from anywhere in a patient’s chart or in our messaging console. It’s a major step-forward in how we apply new technology to streamline our doctor’s days, and honestly, it feels like magic.

Some the things that make this feature special are:

• Comprehensive Aggregation: The AI-Assisted Patient Overview pulls together signalment, recent chat logs, diagnoses, medications, vaccinations, diagnostics, and more from our in-house EMR system.
• Continuity of Care: One of the features our doctors love the most is that the Patient Overview tells them how many times they personally have seen that pet or that owner so that they can simply pick up the conversation where they left off last time. This is especially helpful when you haven’t seen a pet in a while.
• Connecting Dots: The overview can point out historical patterns in the pets health history that other EMR’s don't. For example, it can note that the pet has had a history of allergies and what treatments have or have not worked over the past few years.
• Dynamic Updates: Content refreshes automatically before and after appointments, when prescriptions are added, after diagnostic tests, and following chat interactions with members.

Pointing AI in the Right Direction

To harness the power of GPT-4o to execute the feature, we began by framing the AI as a “medical document analyzer” tasked with generating comprehensive overviews for veterinary review. This prompt included specific instructions detailing the problem context, the significance of each data point, and guidelines for ensuring relevance and clarity in generated summaries.

We meticulously outlined the data inputs provided to the AI, emphasizing their relevance to immediate veterinary tasks and how each piece of information should be interpreted within the given context. Clear instructions were essential to prevent the AI from overlooking pertinent details or including extraneous information.

Next, we structured specific directives for the AI model, emphasizing the need for well-written, concise summaries across designated sections like “Patient Identity and Reason for Visit”, “Medical History Highlights”, and “Last Known Medications”. Each section’s content requirements were clearly defined, ensuring that the generated overview effectively encapsulated key patient details for efficient veterinary decision-making.

Finally, we formatted patient-specific data consistently with the outlined context and instructions provided, ensuring seamless integration into the AI’s summarization process.

Feedback and Reflections

Of course, no feature is perfect upon release.

After launching the first version of the Patient Overview, we were eager to gather feedback from our team on how they were leveraging the tool and what improvements could be made to increase its utility.

For example, we quickly began receiving feedback that the format of the Medical History Highlights in the overview was just too long. While the paragraph structure we launched the feature with felt compelling and cohesive, it didn’t lend itself to easy reading on the fly. In response to this feedback, we were quickly able to re-engineer our prompt to generate this information in a bulleted format that simplified the cognitive load of our doctors’ appointment preparation.

Reflecting on the impact of this work, The DTLA Clinic’s Lead Doctor, Dr. Lauren De Silva, sums things up nicely:

“The Patient overview feature is such a unique tool that I utilize multiple times a day. It has helped simplify the normal ritual I perform to prepare myself for an appointment medically and professionally, significantly reducing complexity and increasing my efficiency. I loved partnering with the engineering team to collaborate on little tweaks to maximize its impact on the care team experience!”

— Dr. Lauren De Silva, Lead Doctor at DTLA

For more insight on how we are building with AI at Modern Animal, Follow us on Linkedin and watch AI at Modern Animal: The Early Innings.

Thanks for reading! 👋

Project Team

Technology Team Contributors

Scott Roth, Product Manager
Trevor Sullivan, Product Designer
Eisah Jones, Engineer
Micah Arnson-Serotta, Quality Engineer
Nawaaz Janmohamed, Quality Engineer 

Care Team Partners

Dr. Jennifer Bonnell, Dr. Poorna Chowdry, Dr. Lauren De Silva, Dr. Jessica Friedman, Dr. Nina Han, Dr. Mallory Mathews, Dr. Jessie Parks, Dr. Claire Samuelson, Dr. Janette Shim, Dr. Julie Storm, Dr. Kristin Yang