Investment in medical note-taking apps using AI will double in 2024, as big tech companies including Microsoft, Amazon and startups race to grab a share of the $26 billion AI healthcare market.
AI startups focused on creating digital “scribes” for health workers raised $800 million in 2024, compared to $390 million in 2023, according to data from PitchBook.
Startups like Nabla, Heidi, Corti and Tortos raised money last year Supporters Including Khosla Ventures, Entrepreneur First and French tech billionaire Xavier Niel.
The surge in funding comes as the groups rush to get off the ground Artificial intelligence-powered products It aims to speed up doctors' taking of medical notes and improve patient interactions, as health has become a major growth area in the artificial intelligence boom.
Microsoft, which owns AI speech recognition company Nuance, as well as Amazon and Oracle, has launched so-called AI assistant pilots for doctors that use large language models and speech recognition to automatically generate transcripts of patient visits, highlight medically relevant details and generate… Clinical summaries.
“I don't think I've seen anything more transformative in 15 years of healthcare than this,” said Harpreet Sood, a primary care doctor in south London, who has been trialling French startup Nabla's app for the past 15 months. .
In a full-day clinic with about 40 patients, traditional note-taking can take at least two hours of writing time, said Sood, a former technology and innovation adviser to the chief executive of NHS England.
He added: “It was great, easily saved 3-4 minutes from each consultation (10 minutes) and really helped capture the consultation and what it was about.”
Nabla's note-taking app uses Whisper, a transcription tool from OpenAI, which makes ChatGPT, and was used to record about 7 million medical visits as of October last year.
Hospitals and GPs across the UK's National Health Service are trialling note-taking using AI as a way to save time and improve doctor-patient interactions. According to a Mayo Clinic study, doctors spend an average of one-third of their workday on administrative work, such as paperwork.
Meanwhile, Nuance's DAX Copilot tool, which launched just over a year ago, now documents more than 1.3 million doctor-patient encounters each month in more than 500 U.S. healthcare groups, Microsoft said.
Nuance, which Microsoft acquired for nearly $20 billion in 2022, said the AI tool reduces the amount of time doctors spend on clinical documentation by 50 percent.
At Stanford University School of Medicine, more than 50 primary care doctors have tried Nuance's AI note-taking device in 2024, with two-thirds of users saying it saves time.
The AI-generated notes were closely scrutinized by doctors for accuracy, and the vast majority of them, about 90%, had to be manually edited to correct errors, a person familiar with the trial said.
Despite this, the results have prompted Stanford to plan to roll out the DAX Copilot system to all providers.
Sood said that while he examines every report generated by Nabla, the cognitive load of writing and listening at the same time during a consultation is “significantly reduced, if not completely eliminated” by the tool.
“You can focus more on the patient, listen to them, be more present, understand their body language. I enjoy my consultations more now,” he added.
However, the rise in medical note-taking has drawn criticism from researchers about the dangers of artificial intelligence-generated fabrications, known as “hallucinations,” which can be particularly harmful in the medical context, as well as the issue of patient data privacy.
Researchers at Cornell University and the University of Virginia Analyzed Thousands of text snippets generated by Whisper from 2023 found that approximately 1 percent of audio transcriptions contained “entire hallucinatory phrases or sentences that were not present in any way in the underlying audio.”
The study said that about 40% of the hallucinations included harmful fabricated content, such as perpetuating violence, or making up inaccurate associations.
“I won't just rely on the tool, I'll read every note to check it and go back to the text,” Sood said. “There is work to be done there but . . . for me personally, this has been a huge transformation.”
