Gemini LLMs score 100% accuracy in summary generation tasks
Gemini summaries up to 32% more accurate than other LLMs
Gemini summary generation 98% faster than its closest competitor
Gemini outperforms other models in accuracy and speed when generating clinical studies and summaries.
Gemini outperforms other models in accuracy and speed when generating clinical studies and summaries.
Doctor, what's your evidence?
Your doctor makes treatment decisions based on clinical evidence. But not all evidence is equal.
There's evidence established through double-blind randomized clinical trials and published in medical journals. There are opinions based on a clinician's years of experience. And somewhere in between lie observational research studies: information based on raw, highly specific data compiled from millions of anonymized health records, insurance claims, genetic testing results, data from wearable devices, and other artifacts of day-to-day patient care.
Amassing, analyzing, documenting, and summarizing this "real-world evidence," or RWE, can be a herculean undertaking involving months of work for a team of healthcare professionals and data scientists. That's where Atropos Health comes in.
"To manifest the value of real-world evidence, the data needs to be aggregated and converted from a wall of numbers in a database into something healthcare practitioners can easily understand," explains Neil Sanghavi, president and head of product at the five-year-old Series B startup. "Atropos Health uses Gemini models to generate and summarize real-world evidence studies from structured and unstructured data in a fraction of the time that humans can," he continues. "What used to take months to compile and publish is now available on demand in days or even minutes with our software."
Gemini models help us generate and summarize real-world evidence studies from structured and unstructured data in a fraction of the time that humans can.
Neil Sanghavi
President and Health of Product, Atropos Health
While not a replacement for well-designed clinical trials, RWE can be far more specific to a situation a physician faces. It's therefore a key resource in clinical decision-making. "In healthcare and life sciences, saving time not only saves costs, it can also mean extending people's lives," Sanghavi adds.
That's an appropriate mission for a company named for the Greek goddess who decided how long mere mortals would live.
Users from hundreds of health systems, academic medical centers, and life sciences companies have taken notice. And so have investors: Atropos Health has attracted over $50 million in venture capital to date.
The fastest, most accurate AI yields tangible, real-world benefits
Nowhere is the need for accuracy in AI-driven content summarization more crucial than in healthcare, so Atropos Health used the utmost care in selecting a large language model (LLM) for its platform by conducting a trial of its own. Sanghavi's team pitted three different Gemini LLMs — Flash 2.0, Pro 2.5, and Flash 2.5 — against five other leading models.
Actual data was fed into all eight LLMs. Each model was then benchmarked for three types of accuracy: "direction of effect," which evaluated whether the model reported a study's outcome correctly; "numerical accuracy," which evaluated whether the model chose the correct numerical data for its summary; and "completeness," which evaluated whether the summary included all statistically significant data. The experiment also measured "inference time," or the speed at which each model completed its task.
"Gemini models outperformed the other models across the board," reports Sanghavi, "and a Gemini model scored a perfect 100% in each of the accuracy tasks."
Gemini's scores in each accuracy category bested the other models' up to 32%, and it performed those tasks 98% faster than its closest competitor. "We've since made Gemini a foundational part of the Atropos technology stack," Sanghavi says.
These performance metrics have wide-ranging implications. By accelerating their work, Atropos Health's evidence-generation tools enable medical researchers to validate their hypotheses more quickly, increasing the amount of research they can conduct and helping them publish those results sooner.
Pharmaceutical companies use Atropos Health to conduct drug trials more cost-effectively and determine efficacy faster. They can also use the solution to justify the cost of approved medications to insurers so more patients have access to them, and to refine guidelines for their use with granular information that would be otherwise difficult to collect.
And of course, Atropos Health helps hospitals and healthcare providers improve patient care. "Automation with Gemini means we can put more information in the hands of more clinicians," says Sanghavi. "As a result, doctors can answer patients' questions more readily and find better treatment options for them more easily."
Gemini vastly outperformed models by Anthropic and OpenAI in tasks that measured both accuracy and speed. We've made it a foundational part of our technology stack.
Neil Sanghavi
President and Health of Product, Atropos Health
Expert guidance and new ideas for building life-saving technologies
Sanghavi credits the Google Cloud team with helping Atropos Health over the learning curve that comes with adopting any new technology. "Google Cloud offers a plethora of tools, and our Google team advised us on how to get the most out of the platform," he recalls. "Having direct access to solution architects who listen to our feedback and find answers to the challenges we encounter has truly boosted productivity."
Informal brainstorming sessions with members of DeepMind's research team have also generated ideas that now appear on Atropos Health's roadmap. This includes an agentic AI guardrail that monitors platform users' actions in real time to ensure that the parameters they set for a real-world evidence study will yield useful results.
Gemini is helping healthcare providers get the right treatments to the right patients — treatments that have the potential to save lives.
Neil Sanghavi
President and head of Product, Atropos Health
Atropos Health's roadmap also includes support for expedited medical journal articles: Gemini-generated summaries of RWE studies. Journals will be able to evaluate and review these summaries before a traditional journal publication, further accelerating the publishing cycle and getting important research into the hands of healthcare professionals more quickly.
"Across all use cases, Gemini is accelerating our customers' ability to apply real-world evidence to clinical decisions, and faster decision-making can change the trajectory of illness," Sanghavi concludes. "Gemini is helping healthcare providers get the right treatments to the right patients — treatments that have the potential to save lives."
Atropos Health's platform and AI-generated real-world evidence studies and summaries accelerate clinical decision-making while lowering research costs for healthcare and life sciences organizations.
Industries: Healthcare and Life Sciences, Startups
Location: United States
Products: Google Cloud, Gemini
