Advancements around large language models and generative AI in healthcare are ramping up quickly, and it's challenging to keep up with the evolving news.
In an effort to help keep you informed about the latest initiatives in this fast-paced sector, Fierce Healthcare has launched this generative AI tracker to provide updated coverage of noteworthy projects. But, we'll still profile larger trends and exciting new technologies that catch our eye in depth.
Check back here often to read more about the latest efforts to integrate technology like OpenAI's GPT-4 into the medical field.
Incredible Health implements gen AI to speed up hiring process
Updated Wednesday, December 13
Incredible Health, a career marketplace for healthcare workers, implemented generative AI across its platform to speed up the hiring process for nurses.
The company says it's the first career marketplace to deploy generative AI. The tech has driven a 20% increase in interview acceptance by nurses, dramatically increasing speed to hire, according to the company.
Logistical challenges within nurse hiring limit opportunities for nurses to find permanent positions that best suit them and contribute to the growing U.S. healthcare staffing crisis.
Advanced generative AI offerings, which rolled out this year and are now live for all Incredible Health users. The features include a "resume wizard" for automatic resume creation. This feature enables nurses to generate an impactful resume via the Incredible Health mobile app in less than 5 minutes at no cost, the company said.
Incredible Health also leverages generative AI to help health systems instantly create customized messages to nurse candidates, highlighting important details about hospital benefits, perks and other differentiating factors for why nurses should engage with their organization.
And the company uses generative AI to instantly process resumes and applications, including verifying key specialties and skills, matching open positions within specialties to relevant applications immediately.
The company boasts that 800,000 nurses and more than 750 hospitals nationwide use its platform.
Study on ChatGPT finds greatest accuracy in tasks of final diagnosis
Updated Wednesday, August 23
A new study published in the Journal of Medical Internet Research details the usability ChatGPT in ongoing clinical decision support.
The capacity of large language model-based bots like ChatGPT to assist in the full scope of iterative clinical reasoning, in effect acting as artificial physicians, has not yet been evaluated, researchers wrote in the study.
To do so, the researchers fed ChatGPT 36 published clinical vignettes from the global standard in medical reference, known as the Merck Sharpe & Dohme Clinical Manual.
They then compared its accuracy on differential diagnoses, diagnostic testing, final diagnosis and management based on patient age, gender, and case acuity. The study measured accuracy by what portion of answers the AI got correct, as calculated by human scores.
Overall, ChatGPT was accurate 72% of the time. It demonstrated the highest performance in making a final diagnosis, with an accuracy of 77%. Its lowest performance was in generating an initial differential diagnosis, with an accuracy of 60%. It generated inferior performance on differential diagnosis and clinical management questions as opposed to questions about general medical knowledge.
"ChatGPT achieves impressive accuracy in clinical decision-making, with increasing strength as it gains more clinical information at its disposal," the study's authors concluded.
Laudio bringing AI workflow solution to Northwell Health, Nebraska Medicine
Updated Monday, July 17
Northwell Health and Nebraska Medicine are the latest health systems to employ Laudio’s AI workflow solution. The rollout will add over 15,000 new frontline employees to the platform in order to improve retention, quality care and safety.
Laudio’s AI solution automates repetitive work along with providing daily recommendations and best practices to help clinical and operational staff. The platform uses AI to offer managers recommendations on how to engage team members in order to reduce turnover and burnout. Northwell Health manages 21 hospitals, 890 outpatient facilities and 83,000 employees while Nebraska Medicine consists of two hospitals and 70 specialty and primary care clinics.
“Labor productivity and reducing costs are ongoing issues within health systems today. There is a clear need for health systems to reimagine the future of work and embrace new technologies that bolster their workforce,” said Russ Richmond, CEO and co-founder of Laudio, in a press release. “We’re thrilled to partner with Northwell Health and Nebraska Medicine, who are now using Laudio to automate administrative work and allow managers and leaders to focus on what’s truly important.”
The Boston-based company scored $13 million in series B funding this June in a round led by Define Ventures. Laudio’s total raise to date has reached $25 million. New funds will be used to invest in its AI and analytics platform, Richmond said.
Novant Health, Tufts Medical Center and UNC Health already utilize Laudio’s platform in their own management systems. Through cross-client analysis, Laudio found that sustained interactions in the platform reduced risk of registered nurse turnover by 26%.
The health tech company states that it is on track to partner with over 17 health systems and 250,000 clinical and non-clinical employees in order to achieve goals in human resources, operations, quality and safety and patient experience.
Array Insights launches data federation platform to provide non-profits with AI insights
Updated Monday, July 17
Array Insights, formerly known as Secure AI Labs, will now help non-profits and patient advocacy groups accelerate research through its new data federation platform.
By applying AI to previously siloed datasets from healthcare organizations, including hospitals, Array claims to offer smaller non-profit healthcare organizations insights from patient data. The organization’s agnostic software uses machine learning and analytics on federated data while ensuring that patient data is protected.
“Our new name — Array Insights — is really reflective of the larger vision of our company,” said Anne Kim, co-founder and CEO of Array, in a press release. “When patient advocacy groups tap into our clinical data federation platform, they’re able to access an array of data sets that had previously been inaccessible. Researchers are able to glean tremendous insights from this data, in a way that puts privacy first for patients, hospitals and everyone involved.”
Clinical researchers can use the platform to run queries on patient data sets to fuel healthcare advancements. Array’s technology does not allow for patient data to be copy and pasted, emailed, forwarded, downloaded or otherwise extracted from its platform.
The health tech company touts its platform’s ability to protect patient security more robustly in comparison to data brokers who sell de-identified data that has the potential to be re-identified.
Patient advocacy groups, such as Kidney Cancer Association, Pancreatic Cancer Action Network and Fatty Liver Foundation, are already working with Array’s technology to gain medical insights through AI.
Array moved to change its name from Secure AI Labs following the raising of $4.5 million in seed funding in late 2022. The funding round brought total fundraising to $9 million for the health tech organization.
AI model beats traditional approaches for detecting heart attacks
Updated Monday, July 17
A novel machine learning model using electrocardiogram (ECG) readings outperformed standard methodology when it came to identifying heart attacks, a study published in Nature Medicine found.
The study was led by researchers at the University of Pittsburgh and found that by employing a novel AI model alongside medical professional judgment, identification of occlusion myocardial infarction was faster and more accurate. By measuring 7,313 patients at multiple clinical sites, one in three patients were able to be reclassified into a more accurate diagnosis.
“When a patient comes into the hospital with chest pain, the first question we ask is whether the patient is having a heart attack or not. It seems like that should be straightforward, but when it’s not clear from the ECG, it can take up to 24 hours to complete additional tests,” said lead author Salah Al-Zaiti, Ph.D., associate professor in the Pitt School of Nursing and of emergency medicine and cardiology in the School of Medicine, in a press release. “Our model helps address this major challenge by improving risk assessment so that patients can get appropriate care without delay.”
The model was developed by co-author Ervin Sejdić, Ph.D., associate professor at The Edward S. Rogers Department of Electrical and Computer Engineering at the University of Toronto and the research chair in artificial intelligence for health outcomes at North York General Hospital in Toronto.
Through observing an ECG reading clinicians can often find indications of heart attacks caused by total blockages of arteries. However, according to the study authors, nearly two-thirds of heart attacks caused by severe blockages do not replicate the most standard ECG pattern.
In comparison to experienced clinician interpretation of ECG, commercial ECG algorithms and the HEART score, which considers history at presentation and blood levels of a protein called troponin, the new model showed superior detection abilities.
“In our wildest dreams, we hoped to match the accuracy of HEART, but we were surprised to find that our machine learning model based solely on ECG exceeded this score,” Al-Zaiti said.
Study co-authors suggest that the new tool can be used by emergency medical services and emergency department staff to provide faster care. They stated that the next stage of research would involve connecting the technology to a hospital command center to offer providers guided medical decisions in real-time.
Johns Hopkins University, Emeritus aim to train healthcare leaders in AI utilization
Updated Monday, July 17
Johns Hopkins Carey Business School has partnered with online education platform Emeritus Healthcare to launch an immersive business program educating healthcare professionals on how to utilize evidence-back technology and methodology to improve health outcomes.
The course, titled “Business of Health Care: Driving Impact and Transformation,” covers subject matter including building leadership skills, understanding business functions and emerging trends in healthcare. The primary trend highlighted in the course is in applying AI, automation and technology to improve health equity across patient populations.
“There is a growing sense of urgency for health care leadership teams to assess and adapt to new economic circumstances quickly – all while continuing to improve access, quality and health outcomes,” said Ranil Herath, president of Emeritus Healthcare, in a press release.
“As expectations rise and workplace challenges evolve – leadership training is no longer considered a nice-to-have but is an essential need to best meet the demands of today’s global health system. Through our partnership with Johns Hopkins Carey Business School, we’re continuing to democratize access to high quality business of health education and help the next generation of health care executives achieve sustainable transformation across their organizations.”
Emeritus Healthcare boasts other partnerships with employers designed to continually develop the skills of healthcare professionals, including U.S. Department of Health and Human Services, Aveanna Healthcare and SCL Health.
Generative AI tool added to Wellen’s osteoporosis and osteopenia platform
Updated Friday, July 7
Wellen is aiming to bridge the digital divide and create a fracture-free future for women with osteoporosis and osteopenia through its platform now equipped with a generative AI tool.
The new AI chatbot uses advanced language learning models to help users find resources on bone health. Wellen announced in May that it would be tapping OpenAI’s LLM ChatGPT for the task. Wellen’s chatbot was trained on data from the company’s “Well Guide,” which is written and peer-reviewed by experts in the field, according to CEO and founder Priya Patel.
New York-based Wellen is a bone health-focused fitness startup. The company states that it uses science-based, personalized strength training programs designed to increase bone health and help patients avoid the worst ramifications of bone loss.
“We are using OpenAI’s API to create embeddings that produce a vector store of our content,” Patel told TechCrunch. “We are leveraging a popular open-source framework called LangChain to facilitate the search and retrieval of information within our embeddings.”
Patel said the chatbot provides a step forward for patients by increasing ease of use and access to information. Additionally, with the bot’s ability to “interpret intent, remember history and provide quick, accurate responses” specific questions can be answered in addition to offering lifestyle and nutritional guidance.
Wellen’s Well Guide is continually updated to include expert interviews, information on the interaction of health conditions such as menopause and osteoporosis and “5 foods to eat this summer for healthy bones.”
Nearly one-fifth of females over the age of 50 experience osteoporosis, in contrast to 4.4% of men, according to a 2018 study performed by the Centers for Disease Control. Low bone mass, a precursor to the disease, was found in half of all women aged 50 and older. That fraction fell to one-third in men. Women also tend to experience bone loss earlier in life and see faster declines in density.
While losing bone density is normal for an aging body, women experience the decrease more noticeably due to physiological differences like smaller bones and the decrease of bone-protecting estrogen following menopause.
Navina unveils generative AI assistant for primary care providers
Updated Friday, July 7
A new platform from Navina uses a generative AI assistant to offer primary care providers access to actionable insights, such as care recommendations.
Primary care providers will be offered insights into patient health status, support in managing administrative requests and gain recommendations for care using “natural language of primary care,” according to Navina. The AI assistant can also generate progress notes and referral documents which can be inputted into electronic health record fields.
Following the waning of the COVID-19 pandemic, provider burnout has continued to be a challenge across the country. Navina works with value-based care and multi-state physician groups to leverage AI with the goal of decreasing tedious work.
The new model answers provider questions by pulling insights from individual patient data, clinical guidelines, health information exchange, claims data and scanned documents.
“Our cutting-edge technology is trained to deliver actionable insights by processing large amounts of structured and unstructured data from the individual patient record, and correlating that with clinical standards of care,” said Shay Perera, co-founder and CTO of Navina, in a press release. “By providing physicians with instant answers to all their queries about their patient, Navina’s AI assistant can not only drastically improve workflow efficiency, but also facilitate more meaningful patient-provider interactions.”
Navina scored $22 million in a series B round last October that was partially used to fund its new offering along with advancing its algorithms, expanding integration of emerging data sources and spreading out into new markets. The new pot brought the startup’s total raised to $44 million.
Botco.ai GenAI Chat Cloud designed to bring access and security to generative AI for businesses
Updated Friday, June 30
Botco.ai launched its suite of generative AI products dubbed GenAI Chat Cloud. The offering will allow for users to ask questions of enterprise applications, such as electronic medical records, websites and content management systems, and receive a conversational response.
GenAI Chat Cloud is built from large language models to sort, analyze and contextualize information from a company’s data. The no-code platform comes with SOC2 compliance, data governance capabilities and HIPAA validation. Users in healthcare, wellness and government have seen a six-times return on investment in the tool, according to Botco.ai.
“Generative AI is a proven game changer as enterprises seek new ways of engaging with customers, prospects and other stakeholders,” said Rebecca Clyde, CEO of Botco.ai, in a press release. “But the solutions on the market today are limited in what they can do for an enterprise because they aren’t trained on the data and information that matters most—the enterprise’s own content. Our GenAI Chat Cloud makes it easy for any business to query its own documents and deliver timely, accurate responses to just about any question a prospect or customer might have.”
Three product modules are included in the chat cloud. InstaStack queries documents across entire companies while extracting meaning and context of found text. Botco.ai’s models are able to adapt to changing data sources to reflect the most current information.
InstaChat produces questions and answer pairs from uploaded documents. The company says this module is best suited for smaller chatbots to be deployed rapidly when the underlying information doesn’t change frequently.
InstaGraph interrogates structured data to gain insight into customer behaviors by examining customer questions, structure of customer questions and efficacy of campaigns.
NVIDIA-accelerated AI model used in EEG data to detect delirium in the ICU
Updated Friday, June 30
Through the use of a rapid-response electroencephalogram, or EEG, device boosted with NVIDIA graphic processing AI capabilities, delirium has been shown to be better detected in intensive care units.
The efficacy of the technique was presented in a recent article published in Nature. In the article, researchers explained how they used a deep learning model called vision transformer (ViT) alongside an EEG device to detect delirium in critically ill older adults. The approached showed a 97% success rate. This compares to the roughly 40% success rate of clinician ICU assessments, according to the researchers.
“This method has strong potential for improving accuracy of delirium detection, providing greater opportunity to implement and evaluate individualized interventions,” the researchers wrote. “Once early detection of brain dysfunction associated with poor outcomes such as the need for institutionalization and higher mortality are readily available, it will be possible to identify reversible causes, followed by early intervention and close monitoring to avoid preventable complications.”
The results were “due in part” to the use of NVIDIA’s graphic processing units, GPUs, compared to traditional CPUs, according to the company. While the use of handheld EEG devices could make the screenings more accessible and affordable, a shortage of skilled technicians and neurologists dampens the good news.
The AI model used, called ViT, was initially created for natural language processing, but proved successful in the new application.
Clinical delirium is presented, oftentimes in elderly patients, through alteration of attention, consciousness and cognition. Certain factors make an individual more susceptible to delirium such as brain injury or changes in the individuals environment.
Total costs to the healthcare system annually due to delirium as estimated to be between $16,000 and $64,000 per patient. Nationally, the financial burden therefore reaches between $38 billion and $152 billion each year.
UNC Health pilots internal AI-powered chatbot
Updated Tuesday, June 27
Chapel Hill-based UNC Health is piloting a generative AI internal chatbot, powered by Azure OpenAI Service. The chatbot aims to streamline administrative day-to-day work for care team members by quickly accessing reference materials and documents.
UNC Health will begin the rollout of its internal chatbot tool in June with a small group of clinicians and administrators and is finalizing plans to offer the software more broadly to teammates later this year, the health system said.
Earlier this year, electronic health record giant Epic, in partnership with Microsoft, rolled out a pilot project using generative AI embedded in its medical records software. UNC Health is one of a handful of health systems tapped to be an early adopter to test out the generative AI tools.
The health system developed a conversational bot hosted in a secure, governed internal environment. The bot will respond to questions specific to UNC Health and provide real-time recommendations or directions to help save time and provide more efficient, patient-focused care in administrative use cases, the health system said.
Instead of spending time searching through training libraries, which include hundreds of ‘how-to’ documents, UNC Health teammates will be able to ask questions to leverage the chatbot to quickly access reference materials and documents, health system executives said.
“This is just one example of an innovative way to use this technology so that teammates can spend more time with patients and less time in front of a computer,” said Dr. David McSwain, UNC Health’s Chief Medical Informatics Officer and a pediatric critical care physician at UNC Children’s Hospital, in a statement.
The team anticipates identifying multiple other use cases during the testing and pilot program, and especially after broader release to teammates. “
NYU, NVIDIA predicting patient readmission with large language model
Updated Tuesday, June 20
Researchers at NYU Langone collaborated with NVIDIA to develop a large language model (LLM) that predicts a patient's risk of readmittance within 30 days, along with other clinical outcomes.
Dubbed NYUTton, the model seeks to identify patients in need of clinical intervention. The tool has been rolled out in the NYU health system's six inpatient facilities and applied to over 50,000 patients. Through email notifications, physicians are notified of patient readmission risk.
Researchers used 10 years of health records at NYU Langone Health to train NYUTron with 4 billion words of clinical notes representing 400,000 patients. Through the technique, the model achieved an accuracy improvement of over 10% percent when compared to other machine learning models, according to the health system.
Upon completing the LLM training, the team spun out four other predictive algorithms which predict the patient's hospital stay, the likelihood of mortality and the chances of insurance claims being denied.
“While there have been computational models to predict patient readmission since the 1980s, we’re treating this as a natural language processing task that requires a health system-scale corpus of clinical text,” Eric Oermann, M.D., assistant professor of radiology and neurosurgery at NYU Grossman School of Medicine and a lead collaborator on NYUTron, said in a press release. “We trained our LLM on the unstructured data of electronic health records to see if it could capture insights that people haven’t considered before.”
The U.S. government fines institutions with a high rate of readmissions in order to protect patient safety and ensure quality care. Readmission is one of the costliest episodes in treatment in the country, racking up $41.3 billion in medical costs annually.
In a Nature article titled "Healthsystem-scale language models are all-purpose prediction engines," the authors found that NYUTron showed notable improvements from previous models. The authors also warned that physicians and administrators should not over-rely on the tool.
"An ethical consideration in deployment is that physicians and administrators could over-rely on NYUTron’s predictions owing to its seamless integration with existing medical workflows, thereby leading to undesirable outcomes," the authors wrote. "Further research is needed to optimize human–AI interactions, as well as development of standardized assessments for sources of bias or other unexpected failure points."
NYUTron was trained using the NVIDIA NeMo Megatron framework on a large cluster of NVIDIA A 100 Tensor Core GPUs. To optimize the model for in-hospital use, researchers developed a modified version of the NVIDIA Triton open-source software.
By first pretraining the LLM and then fine-tuning it onsite with the hospital in question's data, the models could boost accuracy. The team believes other institutions can adopt similar models.
“Not all hospitals have the resources to train a large language model from scratch in-house, but they can adopt a pretrained model like NYUTron and then fine-tune it with a small sample of local data using GPUs in the cloud,” Oermann said. “That’s within reach of almost everyone in healthcare.”
Carbon Health rolls out hands-free charting
Updated Tuesday, June 6
Hybrid primary care company Carbon Health debuted hands-free charting—an AI-enabled notes assistant—in its proprietary electronic health record software across each of its clinics and providers.
The company says it's the first to deploy native AI-assisted charting at scale.
Built directly into the Carbon Health EHR and designed to fit into a provider’s existing workflow, the integration saves time and enhances the patient-doctor connection, executives said. Without a screen between them and their patients, doctors can focus more time on care and less on typing.
“The rapid development and deployment of hands-free charting advances Carbon Health's AI-enabled EHR and shows how quickly AI can have a real-world impact on care delivery,” said Carbon Health CEO Eren Bali in a statement. “Our vertically integrated model of providers, clinics, and software makes our system uniquely positioned to leverage AI technologies like GPT-4.”
The aim is to reduce clinicians' workload and allow clinics to see more patients without adding more strain to physicians' already packed schedules.
The company is hopeful that the technology could cut its own costs internally and also become a product Carbon can sell to other provider groups, executives told Stat.
Early data show immediate benefits for providers, the company says. On average, the system generates a complete medical chart in less than four minutes compared to 16 minutes for a manual chart. And 88% of the AI-generated text is accepted by the provider without edits. The AI-generated charts are 2.5x more detailed than manual entry.
Carbon Health operates 125 clinics in 23 states.
The company uses ambient voice technology to capture audio of a patient-doctor visit audio, which is transcribed using AWS Transcribe Medical. The transcript is then combined with other patient information such as demographics, vitals, lab results and manually added notes from the provider. Within minutes, a visit summary is auto-generated in the patient’s chart, allowing a provider to review, adjust and finalize it all within the EHR.
“It’s about eye contact over iPads,” said Bali. “For too long, providers have had to choose between connecting with their patients and taking notes. That will no longer be the case at Carbon Health clinics. We’re unlocking the ability for clinics to meet the needs of the community without trade-offs like burnout or exhaustion.”
Tempus launches AI assistant for oncologists
Updated Monday, June 5
Artificial intelligence and precision medicine company Tempus rolled out an AI-enabled clinical assistant that leverages generative AI to give clinicians easier access to patient data.
Tempus One is a voice and text assistant—available via the Tempus Hub desktop and mobile app—that is designed to provide clinicians with quick access to their patient’s full clinical and molecular profile to help inform clinical decisions in real time. Through Tempus One, clinicians can, among other things: access new clinical test reports and receive status updates of a patient’s report; rapidly filter patient incidence by alteration, gene, or diagnosis; quickly access summarized patient information; review report information on actionable biomarkers; and easily query clinical guidelines for up-to-date standard of care insights, according to the company.
Tempus says it originally unveiled the first prototype of Tempus One in early 2021 and subsequently enrolled a cohort of clinicians into a pilot program as provisional beta users. The company spent the past two years refining the Tempus One technology through informed iterations based on physician feedback and user testing, evolving the technology to incorporate generative AI—built on top of large language models—directly into clinical care settings.
“Tempus One leverages our vast healthcare dataset, which now exceeds 100 petabytes of curated data, and uses pioneering advancements in generative AI, enabling us to place our precision medicine platform at the fingertips of every clinician we serve, ultimately improving their ability to make data-driven treatment decisions on behalf of their patients,” said Eric Lefkofsky, founder and CEO of Tempus, in a statement.