Photo: Eva Katalin Kondoros/Getty Images
Grace Chang, CEO and cofounder of Kintsugi, said the company is shutting down and moving its research and artificial intelligence-driven mental health acoustics technology into the public domain in order to empower behavioral health providers.
Making a gift of free access to the technology was the best course of action for voice biomarker developers who believe "the biology of the human voice holds the key to making mental healthcare objective, accessible and universal," she said in the company's announcement.
While the startup successfully developed a platform to detect depression and anxiety through acoustics, the high cost and lengthy timeline of U.S. Food and Drug Administration regulatory hurdles ultimately made the venture-backed business model unsustainable, Chang explained to Healthcare IT News.
Clinical markers of distress in voices
Kintsugi's voice biomarker models – the foundational AI architectures built to identify clinical markers of distress – and the scientific methodologies to ensure objectivity and reduce bias developed during the company's regulatory journey are now available open source to all.
"I am sharing the difficult news that Kintsugi is shutting down its commercial operations," Chang said in a social media post. "However, the mission itself is not over."
The company's platform uses novel machine learning and deep learning to attribute clinical depression and anxiety from the sound of patients' voices, not what patients say during their medical visits.
It's based on years of research into the intersection of acoustics and mental health.
Designed to empower mental healthcare practitioners, the advanced AI detects depression and anxiety from 20 seconds of free-form speech to help them better identify and triage patients. The API-based platform was aimed at improving mental health services and outcomes by analyzing acoustic signals in human speech, Chang explained on HIMSSTV four years ago.
The technology also helps protect the privacy of patients because the models do not analyze what they are saying. The models are also language-agnostic.
Ultimately, with the AI focusing on how patients were speaking, Kintsugi, which launched in 2019, developed "a simpler and more elegant architecture where we could produce real-time results," she said.
As a startup, Kintsugi raised $8 million in seed funding in 2021 and then $20 million in Series A funding in 2022.
In 2023, the company reported having supported mental health screenings by one of the country’s largest health payers. The health plan initially relied on standard patient health questionnaires to indicate that 3% of its members surveyed were dealing with some form of depression. The payer then screened recently discharged emergency department and maternal health patients using Kintsugi's voice biomarker AI, which detected moderate-to-high depression in 33% and severe depression in 14% of members screened.
Despite telehealth and application-based deployments and established commercial relationships with health providers and payers to assess depression and anxiety in patients and also clinicians experiencing burnout, it was the cost of regulatory approvals that proved too much for the venture capital-funded startup to bear.
"We navigated the rigorous path of FDA De Novo Clearance and pushed the boundaries of what voice biomarker AI could achieve in detecting depression and anxiety," Chang said in the company's statement.
"By open-sourcing Kintsugi, we are removing the paywalls and proprietary barriers, allowing a global community of scientists to unlock the potential of voice-based identification, triage and monitoring for everyone, everywhere."
Costly regulatory hurdles
"I would say the challenge in this intersection of AI and regulatory can be a really big hurdle for startups," Chang told Healthcare IT News on Wednesday.
While healthcare AI requires validation, that is going to come from hospitals, not expensive regulatory processes, according to one healthcare AI governance expert.
"Right now, the FDA treats AI like a traditional medical device – once it's approved, it's essentially locked in place," Pelu Tran, CEO and cofounder of Ferrum Health, told Healthcare IT News last year. "Any updates, even small ones, can trigger a whole new approval cycle. That model works for devices like pacemakers, but it doesn't make sense for AI, which is supposed to evolve and get smarter over time."
Scaling up a health or medtech company is a long and costly process, not only in the U.S., but also in Europe. Industry experts on both sides of the Atlantic want to see government bodies like the FDA move faster on health AI.
Meanwhile, for Kintsugi, the window of opportunity has run out – for now.
"Timing and luck are often the silent partners in business, and while they did not align for us commercially, the global need for objective mental health measurement has never been more urgent," Chang said.
After exploring the use of synthetic voice data from Sesame and ElevenLabs to augment training for its mental health models, Kintsugi found it would be of no value. Because synthetic voices do not carry the same biological and cognitive attributes of real human speech, the company's models can detect the difference every time.
Despite expanding into deepfake detection last year with Kintsugi Signal, it was still time for the company to cease commercial operations, she said.
"The time horizon, for example, we've spent probably $16 million or so on four years of FDA work for presubmissions, and we're very close to the finish line," explained Chang. "And yet, there was still another, probably another $1 to $4 million left of submission, getting it stood up and working to align with different associations to ensure that the current CPT code of 96127 would align with some of our standard of care practices that have been shown to demonstrate performance in smaller behavioral health and primary care examples.
"That's asking a lot for venture investors who want to see $100 million in [annual recurring revenue] before you're 10."
Andrea Fox is senior editor of Healthcare IT News.
Email: afox@himss.org
Healthcare IT News is a HIMSS Media publication.
