How doctors are using Google’s new AI models for health care

Google On Wednesday, MedLM announced a new healthcare-specific artificial intelligence model suite designed to help clinicians and researchers conduct complex studies and summarize doctor-patient interactions. announced.

The move marks Google’s latest attempt to monetize AI tools in the healthcare industry, where competition for market share remains fierce. Amazon and microsoft. CNBC spoke with companies testing Google’s technology, including HCA Healthcare, and experts said that while they are taking cautious steps toward implementation, the potential for impact is real. Says.

The MedLM suite includes large and medium-scale AI models, both built on Med-PaLM 2. Med-PaLM 2 is a large-scale language model trained on medical data that Google first announced in his March. The AI ​​suite will be generally available to eligible Google Cloud customers in the U.S. starting Wednesday, and Google says the cost of the AI ​​suite will vary depending on how companies use the various models, but that the mid-range model will be more expensive to operate. was said to be low.

Google also said it plans to bring a healthcare-specific version of Gemini, its latest and “most capable” AI model, to MedLM in the future.

Aasima Gupta, Google Cloud’s global director of healthcare strategy and solutions, said the company has discovered that various medically tuned AI models can perform certain tasks better than others. So instead of building a “one-size-fits-all” solution, Google decided to introduce a set of models.

For example, Google says that larger MedLM models are better suited to perform complex tasks that require deep knowledge and a lot of computational power, such as conducting research using data from a healthcare organization’s entire patient population. He said that But if a company needs a more agile model that can be optimized for specific or real-time functionality, such as summarizing doctor-patient interactions, a medium-sized model should work better. says Mr.Gupta.

Actual usage example

When Google announced Med-PaLM 2 in March, the company initially said it could be used to answer questions such as “What are the first warning signs of pneumonia?” and “Can it cure incontinence?” But Greg Corrado, head of health AI at Google, said the use case changed as the company tested the technology with customers.

Collard said Google hasn’t seen much demand from customers for such features because clinicians don’t often need help with “accessible” questions about the nature of a disease. Rather, he hopes AI will help healthcare organizations solve back-office and logistical problems, such as managing paperwork.

“They want something that solves a real pain point or slowdown in their workflow that only they know about,” Collard told CNBC.

for example, HCA HealthcareGoogle, one of the largest health systems in the United States, has been testing Google’s AI technology since the spring. The company announced the following: official collaboration The company partnered with Google Cloud in August to use generative AI to “improve workflows for time-consuming tasks.”

Dr. Michael Schlosser, HCA’s senior vice president of care transformation and innovation, said the company uses MedLM to enable emergency physicians to automatically record patient interactions. For example, HCA uses an ambient audio recording system from a company called Augmedix to transcribe doctor-patient meetings. Google’s MedLM suite can take these transcripts and split them into components for ER provider notes.

Schlosser said HCA uses MedLM in four hospital emergency rooms and the company hopes to expand its use over the next year. Schlosser added that by January, Google’s technology will be able to successfully generate more than half of the notes without the help of a provider. Schlosser said saving time and effort can make a meaningful difference for doctors, who can spend up to four hours a day processing paperwork.

“This was a huge step forward for us,” Schlosser told CNBC. “We believe we will reach a point where AI alone can accurately create more than 60 percent of notes before humans review and edit them.”

Schlosser said HCA is also working on using MedLM to develop handoff tools for nurses. The tool reads electronic medical records and identifies relevant information for nurses to pass on to the next shift.

Schlosser said handoffs are “labor intensive” and a major headache for nurses, so automating this process would be “powerful.” Nurses across HCA’s hospitals perform approximately 400,000 handoffs a week, and HCA’s two hospitals are testing nurse handoff tools. Schlosser said nurses will compare traditional handoffs and AI-generated handoffs side-by-side and provide feedback.

However, in both use cases, HCA found that MedLM is not foolproof.

Schlosser said the fact that AI models can spew out false information is a major challenge, and HCA is working with Google to devise best practices to minimize such fabrications. . He added that token restrictions that limit the amount of data that can be fed into models, and managing AI over time, he said, pose further challenges for HCA.

“What I’m saying right now is that the hype around the current use of these AI models in healthcare has exceeded the reality,” Schlosser said. “Everyone is fighting this problem, but no one has left these models alone in the health care system at scale because of it.”

Still, Schlosser said providers’ initial response to MedLM has been positive, recognizing that they are not yet dealing with a finished product. He said HCA is working hard to implement this technology in a responsible manner that does not put patients at risk.

“We’re being very cautious about how we approach these AI models,” he said. “We’re not using any use cases where the model’s output has any impact on someone’s diagnosis or treatment.”

Google plans to bring a healthcare-specific version of Gemini to MedLM in the future. The company’s stock soared 5% after Gemini’s launch earlier this month, even as the company acknowledged that Google faced scrutiny over a demonstration video that was not conducted in real time. bloomberg.

Google told CNBC in a statement: “This video is an illustrative depiction of possible interactions with Gemini, based on real-world multimodal prompts and output from tests. December 13. “Gemini Pro will be available for access on Sunday and we look forward to seeing what people create.” . ”

Google’s Corrado and Gupta said Gemini is still in its early stages and the model needs to be tested and evaluated with customers in managed care environments before being rolled out more broadly through MedLM. .

“We have been testing Med-PaLM 2 with our customers for many months and now feel comfortable accepting it as part of MedLM,” said Mr. Gupta. Gemini will continue to do the same.

Schlosser said HCA is “very excited” about Gemini, noting that the company already has plans in place to test the technology, adding: “If it comes to fruition, we’ll get another level of performance. I think it’s a possibility.”

Another company using MedLM is BenchSci, which aims to use AI to solve drug discovery problems. Google is Investor Published in BenchSci, the company has been testing MedLM technology for several months.

Riran Berenzon, co-founder and CEO of BenchSci, said the company will integrate MedLM’s AI with BenchSci’s proprietary technology, which will be key to helping scientists understand disease progression and how to cure it. He said he was able to identify biomarkers.

Berenzon said the company spent a lot of time testing and validating the model, including providing feedback to Google about needed improvements. Now, Belenzon said BenchSci is in the process of bringing the technology to market more broadly.

”[MedLM] “It won’t work on its own, but it will help accelerate concrete efforts,” he said in an interview with CNBC.

Corrado said research on MedLM is ongoing and believes Google Cloud healthcare customers will be able to tailor the model to multiple different use cases within their organizations. He added that Google will continue to develop domain-specific models that are “smaller, cheaper, faster, and better.”

Like BenchSci, Deloitte tested MedLM “many times” before implementing the technology with medical customers, said Dr. Kureni Ghebreyesus, Deloitte’s U.S. life sciences and healthcare consulting leader.

Deloitte uses Google technology to help health systems and health plans answer members’ questions about access to care. For example, if a patient needs a colonoscopy, they can use MedLM to find a provider based on gender, location, benefits, and other criteria.

Gebreyes said clients find MedLM to be accurate and efficient, but like other models, AI is not always good at deciphering user intent. It can be difficult if patients don’t know the correct words or spellings or use other colloquial terms related to colonoscopies, she said.

“At the end of the day, this is not a substitute for diagnosis by a trained professional,” Ghebreyesus told CNBC. “This brings expertise closer and more accessible.”