|
Carta, Hinge Crushes Q1, and Why New LLMs Love to Hallucinate May 8, 2025
|
|
|
|
Together with
|
|
|
“We shouldn’t forget that today’s ambient AI products are doing maybe 10% of what they’ll be able to do in two or three years. We’re at the very beginning of a long journey, so it’s important to pick a partner that understands your needs and can take that journey with you.”
|
Nabla CEO Alex LeBrun
|
|
Clinical documentation might be the “killer app” for AI in healthcare, but its real impact still lies ahead. In the latest Digital Health Wire Show, Nabla CEO Alex LeBrun unpacks the ongoing evolution of ambient AI, the long journey to come, and what it takes to build a product clinicians actually want to use.
|
|
|
Better reasoning apparently doesn’t prevent LLMs from spewing out false facts.
Independent testing from AI firm Vectara showed that the latest advanced reasoning models from OpenAI and DeepSeek hallucinate even more than previous models.
- OpenAI’s o3 reasoning model scored a 6.8% hallucination rate on Vectara’s test, which asks the AI to summarize various news articles.
- DeepSeek’s R1 fared even worse with a 14.3% hallucination rate, an especially poor performance considering that its older non-reasoning DeepSeek-V2.5 model clocked in at 2.4%.
- On OpenAI’s more difficult SimpleQA tests, o3 and o4-mini hallucinated between 51-79% of the time, versus just 37% for its GPT-4.5 non-reasoning model.
OpenAI positions o3 as its most powerful model because it’s a “reasoning” model that takes more time to “think” and work out its answers step-by-step.
- This process produces better answers for many use cases, but these reasoning models can also hallucinate at each step of their “thinking,” giving them even more chances for incorrect responses.
The Takeaway
Even though the general purpose models studied weren’t fine-tuned for healthcare, the results raise concerns about their safety in clinical settings – especially given how many physicians report using them in day-to-day practice.
We’re testing a new format today – let us know if you prefer two shorter Top Stories or one longer Top Story with this quick survey!
|
|
2025 State of Payer Enrollment & Credentialing
With increasing costs and shifting regulations, outdated processes won’t cut it. Medallion’s new report breaks down the biggest challenges, emerging trends, and how automation is transforming the landscape. Get the insights you need – read the full report today.
|
|
Tia Turns to Nabla for Patient-Centered Care
Tia’s mission is to drive the highest standard of care for women, which is why it partnered with Nabla to strengthen the connection between its patients and providers. See why Tia’s clinicians trust Nabla’s ambient AI assistant to keep care centered on patients by streamlining medical documentation.
|
|
|
Manual clinical data abstraction is a prime target for AI automation, and Carta Healthcare is looking to hit the center of the bullseye after raising $18.25M of Series B1 funding.
Carta’s data abstraction platform leverages AI and a team of expert abstractors to automate the collection of clinical data and surface actionable insights.
- The platform converts both structured and unstructured data into standardized datasets, giving healthcare orgs a foundation of high-quality data to build on.
Although AI’s potential in the data abstraction arena has attracted an army of new competitors, Carta’s been duking it out against manual workflows since 2017.
- That gave it a big head start to refine its tech and build trust with partners, many of which participated in the round, including MemorialCare, Memorial Hermann, and MGB.
- Carta plans to put the new capital toward expanding its customer footprint, particularly in the life sciences market where it can help match patients to clinical trials using AI it acquired from Realyze.
The Takeaway
Traditional methods of data abstraction are about as labor-intensive and time-consuming as you can imagine, which makes them ideal targets for AI solutions and startups like Carta that are bringing them to market.
We’re testing a new format today – let us know if you prefer two shorter Top Stories or one longer Top Story with this quick survey!
|
|
BPM Pro 2: Unparalleled Reliability
BPM Pro 2 increases data reliability with two new features that ensure patients are taking their reading properly and prompting them to rest and retake it if their first reading was abnormally high.
|
|
Future-Proof Your Health System with Scalable, Accessible Primary Care
The reality of modern health system primary care: more patients, fewer resources, constant pressure to deliver. K Health offers a strategic advantage, with a clinical AI and Virtualist care model that provides the scalable solution you need to meet patient demand anytime, anywhere. Join the ranks of forward-thinking systems like Cedars-Sinai, Hackensack Meridian Health, and Hartford HealthCare. Explore the future of primary care with K Health.
|
|
- Function Acquires Ezra: Function Health is leveling up its preventative health platform with the acquisition of Ezra, which adds AI-powered full-body MRI screening to its current lab testing offerings. The announcement also included the launch of a new 22-minute full-body MRI service for $499 (a typical MRI is an hour long and $1,500), with Ezra’s AI reportedly allowing it to accelerate scans and pass the savings on to customers. The ultimate vision is to combine the imaging with over 160 lab tests to support earlier disease detection, while simultaneously “creating a dataset that will allow AI to become truly predictive.”
- Hinge Post Strong Q1: Hinge just amended its S-1 filing with its Q1 numbers, and all signs seem to point to it pulling off its IPO on schedule. The virtual MSK platform raked in $124M in revenue during the first quarter (up 50% year-over-year), hitting an eye-popping 81% gross margin in the process. Although market conditions still aren’t necessarily ideal, there’s no telling when a better IPO window might come along, and pocketing $100M in Q1 seems like a decent indicator that Hinge’s business is strong enough for a public debut.
- AI For Clinician Wellbeing: New research out of Sutter Health highlighted the potential of ambient AI to improve clinician wellbeing. Within three months of implementing its AI documentation platform, Sutter clinicians reported a significant decrease in time spent on notes per appointment (6.2 to 5.3 minutes), as well as “a marked reduction in mental demand” like feeling rushed or weighed down by administrative work. Better yet, the percentage of physicians reporting burnout dropped from 42.1% to 35.1%.
- Cedars-Sinai + Redesign: Cedars-Sinai is teaming up with Redesign Health to build health tech startups from the ground up. The LA-based health system already has a venture arm and a startup accelerator, but apparently decided to work with Redesign to co-develop a new Digital Innovation Platform that will allow it to be more hands-on. The partnership focuses on four areas: tools to cut down on specialist wait times, diagnostics that better leverage its genomics data, hospital workflow tech, and RCM solutions.
- AltaMed Partners With Abridge: Abridge is rolling out across AltaMed Health Services Corporation, the largest federally qualified health center in the country. AltaMed serves over 500k patients in southern California, and opted for Abridge in part because of its support for diverse and multilingual patients. The ambient AI platform supports 28+ languages across a wide range of care settings, transforming medical conversations into clinically useful and billable documentation at the point of care.
- Prescription Spending Nears $100B: Americans forked over $98B out of their own pockets for prescription drugs in 2024, marking a 25% increase over the last five years. IQVIA’s annual U.S. prescription breakdown found that total spending on medications rose 11.4% to $487B in 2024 alone, despite average prescription costs remaining flat year-over-year. IQVIA attributed the increase to a “greater use of medicines with significant clinical benefits,” particularly for obesity and oncology. It was also interesting to see that over a quarter of new prescriptions weren’t filled last year, mostly because they aren’t covered by payors.
- AMIE Gains Vision: Google’s AMIE diagnostic AI agent just went multimodal through an update that allows it to incorporate visual information like skin photos into diagnostic dialogues. Not only is this a massive upgrade given how often real-world medicine involves sharing images alongside text, but Google’s fantastic blog post on the launch also painted a picture of a not-too-distant future where LLMs will be able to incorporate real-time audio/visual feedback to help address patients’ non-verbal cues, perform visual assessments, and help conduct guided examinations.
- Oura Integrates With Dexcom: Oura debuted new metabolic health features that allow its signature smart ring to track meals and blood sugar levels through an integration with Dexcom’s Stelo over-the-counter continuous glucose monitor. The integration doesn’t come as a surprise given that Dexcom pitched in $75M toward Oura’s latest funding round as it looks to find new ways to get its CGMs in front of a non-diabetic audience.
- Relatient Launches Dash Voice AI: Relatient hopped on the voice AI agent train with the launch of Dash Voice AI, a conversational AI for automating patient scheduling calls. Similar to Zocdoc’s Zo agent that debuted earlier this week, Dash handles everything that goes into appointment-related phone calls, including confirmations, rescheduling, and cancellations. Dash reportedly offloads a quarter of scheduling call volume instantly, allowing human representatives to focus on high-impact work.
- GLP-1 Head-to-Head: A study presented at last week’s SCAI cardiology conference found that Eli Lilly’s tirzepatide (Mounjaro) leads to a significantly lower all-cause mortality rate than semaglutide (Ozempic) in patients with type 2 diabetes, chronic kidney disease, and heart failure. Among 9k randomized patients, semaglutide was linked to a higher one-year risk of all-cause mortality, heart attack, ischemic stroke, and hospital readmission than tirzepatide. However, HbA1c levels were more likely to be under 7% for patients treated with semaglutide.
|
|
Humans of Healthcare: AltaMed’s Multilingual Community
AltaMed, the nation’s largest Federally Qualified Healthcare Center (FQHC), is partnering with Abridge to serve its majority-multilingual population. Everyone deserves access to cutting-edge healthcare technology, no matter what language they speak. Learn more here.
|
|
Navina Reduces Burnout and Improves Value-Based Outcomes
An independent study from The American Academy of Family Physicians found that Navina’s AI assistant significantly reduced physicians’ administrative burden and improved value-based outcomes. Download the full report to learn why family physicians trust Navina, and how they’re leveraging AI to streamline their workflows.
|
|
The First 30 Days: What to Expect With AI
Implementing AI documentation tools promises significant benefits, but how do you ensure a smooth transition? Playback Health has you covered with this comprehensive 30-day roadmap outlining what to expect, industry best practices, and its own proven implementation approach.
|
|
|
|
|