Skip to Content

What’s the Most Effective Use of AI in Healthcare? 30 leaders share

[Originally published by Randi Haseman at Becker’s Hospital Review – Scroll for Tony Das, MD for an Aspen Forge review!]

AI is beginning to appear more and more in healthcare settings. So much so that the Department of Veterans Affairs created an Artificial Intelligence Institutional Review Board and an AI Oversight Committee to review its usage of AI tools in clinical settings.

Thirty industry leaders explain the most effective use of AI at their health system.

The executives featured in this article are all speaking at the Becker’s Health IT, Digital Health + RCM Annual Meeting: The Future of Business and Clinical Technologies which will take place Oct. 3-6, 2023, at the Navy Pier in Chicago.

To learn more about this event, click here.

As part of an ongoing series, Becker’s is talking to healthcare leaders who will speak at our conference. The following are answers from our speakers at the event.

Question: What has been the most effective use of artificial intelligence at your hospital or health system?

Tina Esposito. System vice president and chief health information officer at Advocate Health (Charlotte, N.C.): We have had a number of use cases around AI and fundamentally what makes it effective is defining success far beyond its implementation status.

It is important to ensure that success is defined in all three of the following ways:

1) validity of the AI model (does the model work well?)

2) efficiency gains (Have we made our clinician’s lives easier?)

3) improved outcomes (Have we moved the needle on the issue we are trying to improve?).

To that end, Advocate Health has seen AI effective in reducing readmissions, early identification of hospitalized patient deterioration, and improved stroke outcomes. Ultimately, it is not AI alone that will solve problems and the technology must be partnered with clinical/operational expertise to be effective.

Salim Saiyed, MD. Vice president and chief medical officer of diplomate clinical informatics at UPMC Central Pa (Pittsburgh): Two important areas AI has been most effective in clinical decision support in radiology and generative AI for clinical documentation. The health system has been testing and developing technology from Abridge that can transcribe patient interactions with physicians. We are in the midst of integrating it within our EHRs in a move intended to relieve documentation burden for physicians.

Additionally, UPMC is testing technology that can give patients more personalized care. The process is called “digital twinning.” I strongly believe AI will revolutionize healthcare now and in the future in dramatic ways.

Richard Zane, MD. Chief innovation officer at UCHealth (Aurora, Colo.): The most impactful use of AI has been with our chat bot “Livy” which has helped millions of patients navigate everything from making appointments, refilling prescriptions, getting results and referrals to finding hiking trails, restaurants and hotels.

We have also had great success in deploying AI with our partners Eon Health in recognizing and following up on incidental findings, like nodules on chest X-rays, making cancer and other potentially life-saving diagnoses earlier.

Michael Pfeffer. CIO at Stanford Health Care (Palo Alto, Calif.): Stanford Health Care uses various levels of AI support across approximately 30 different applications to improve the work experience of our care providers. In all cases, the AI tool remains in an advisory capacity with the final decision resting with the human care provider.

The uses range from automated contouring for radiation therapy planning, to automating analysis of cardiac MRI images, to scoring systems that alert teams about patients at risk, to powering text-messaging for care navigation, to generating draft encounter notes based on physician patient encounters. One of the most fruitful areas has been improving the quantity, timing, and quality of serious illness conversations, where machine learning algorithms are used to identify eligible patients. We improved the care of over 5,300 unique patients, engaging with clinicians from different disciplines and using non-traditional workflows suggested by stakeholders to improve buy-in.

Tony Das, MD. Founder and CEO at Connected Cardiovascular Care Associates (Dallas) and Chief of Cardiovascular Strategic Planning and Development at Baylor Scott and White (Dallas): We have explored the use of AI and NLP tools to enhance several service lines at The Heart Hospital Baylor and in my practice, Connected Cardiovascular Care Associates. We believe that remote monitoring devices to help follow patients are one of the best tools to employ AI in the care of patients between office visits. These physiologic parameters are helpful in early detection of critical changes that allow proactive interventions to avoid costly hospital visits. We believe this will be a helpful program as we transition from fee for service to fee for value. We have partnered with IntelliH, Livmor, Medify and others to aggregate this data.

Additionally, we have looked at machine learning, AI and natural language processing with applications like Aspen Forge which has been able to query our EHR to identify patients with specific medical conditions after reviewing data in multiple disparate formats such as structured, unstructured and even .pdf formats. This has improved our enrollment in clinical research and growing new service lines such as structural heart, peripheral vascular and heart failure.

Jason Stopyra, MD. Regional medical director of safety and security at Atrium Health Wake Forest Baptist Health (Winston-Salem, N.C.): My interaction with AI in workplace violence prevention and emergency medicine is just beginning. We have engaged our nature language processing and machine learning experts to create ways to improve identification of in-facility violent incidents to improve data quality. As we continue to invest in and improve our real-time workplace violence outcomes, AI will likely be at the core.

Cory Ferrier. Vice president at Adventist Health (Roseville, Calif.): We are currently using AI in a few select areas, but one of the most beneficial has been in our neuro (stroke) service line. Utilizing Viz.ai for this service line has helped our physicians and clinical staff better communicate and identify large vessel occlusions sooner, allowing the care team to decide more quickly the path of treatment for these patients. Seconds count when dealing with a stroke, so anything that helps speed up the treatment process translates to better patient care.

Melissa Shipp. Vice president of digital experience at OSF HealthCare (Peoria, Ill.) and OSF OnCall Digital Health: The most effective use of AI our health system has implemented, so far, has been our virtual assistant Clare and her triage symptom checker. We contracted with the vendor (Gyant) in the fall of 2019, went live in December and then COVID-19 hit our first hospital in February 2020. In March 2020 alone, we had 67,000 triage screenings.

Clare acts as a single point of contact, allowing patients to navigate to many self-service care options and find information on their timeline. Clare is available 24/7/365 to help patients both during and outside of business hours. We have found that 48 percent of our patients are accessing Clare outside of the traditional business hours. This allows patients to directly check their symptoms, find a provider, pay their bill, schedule appointments, including asynchronous and telehealth appointments, and understand the best online resources for their clinical and non-clinical needs.

By seamlessly navigating patients to what they need, Clare diverts calls from the contact center.  Since the first year of implementation and changing consumer behavior, having Clare has resulted in $1.2 million in new patient annual net revenue.

Brent Lamm. CIO at UNC Health (Chapel Hill, N.C.): The solution that stands out is our AI-enabled model that helps our care teams match patients to our hospital-at-home service. This solution has proven to effectively identify qualifying patients who are a good fit for this service, streamlining the workflow for our care teams and, most importantly, providing our patients with the best care for their needs.

Tony Ambrozie. Senior vice president and chief digital information officer at Baptist Health South Florida (Coral Gables, Fla.): AI, specifically machine learning, has been a focus for me even prior to me coming to Baptist and continues to be so. The great thing is that the organization is keen to support and adopt these efforts.

We have been working for a while with a variety of vendors to embed and enhance algorithms inside a variety of devices and systems such as imaging (anomaly detection), dictation (voice recognition), physician workload reduction (summarization) and more.

But we have also built and trained our own models based on our own vast clinical and operational data, focused on predictions on clinical (length of stay, readmissions factors) and patient wellbeing, patient services demand and utilization and financial as well as operations and capacity optimizations. We have more than 20 models in various production deployments as well as another 20 more in the works.

I think with AI like ChatGPT (and the many variants, some large language models, some of other types), we are on the precipice of AI becoming a lot more embedded in the clinical activities, supporting clinicians with curated information, smart assisting and even providing suggestions and recommendations. We are not quite there on the clinical side, the technology is very good but still suffers from hallucinations (being wrong with supreme confidence, less so in GPT-4 than GPT-3) but we absolutely cannot underestimate the speed of improvement and learning of these technologies.

D. Matthew Sullivan, MD. Chief medical information officer at Advocate Health (Charlotte, N.C.): The most effective use of AI has been in the imaging space where the dedication of the radiologists and neurologist have helped formulate the right workflows and responses to the outputs to improve patient care.

Amber Fencl. Senior vice president and chief digital health officer at Novant Health (Winston-Salem, N.C.): For me, our use of AI and machine learning for stroke detection is incredibly impactful. My mother suffered a life-altering ischaemic stroke two years ago, hence, I personally know every second counts when it comes to detection and recovery. I am proud Novant Health was the first health system in the state of North Carolina to integrate AI technology, Viz.ai, into its stroke care.

It uses machine learning for early detection, and AI provides simultaneous mobile alerts to all relevant clinicians, eliminating bottlenecks that occur between the time of notification and transfer. While it was already ahead of the national average for time between initial brain scan and treatment, Viz.ai allowed us to further reduce time by an additional 10 minutes. That equates to 19 million brain cells saved per patient- a life changing impact for patients and families.

Onyeka Nchege. Senior vice president and CIO at Novant Health (Winston-Salem, N.C.): We deployed an AI-based scheduling technology to support our new cancer center in Charlotte, N.C. which is built without waiting rooms for the infusion centers. Our patients who come in for cancer treatment often see many providers in back-to-back appointments with the potential for a lot of waiting. Instead, this capability removes some of that downtime or gives patients an idea of exactly when their infusion chair will be ready so they can visit the lounge or the restroom and come back the minute their chair is available.

Erik Smith. Chief digital officer at Memorial Hermann Health System (Houston): One of the more interesting uses of AI that we are exploring today is leveraging virtual voice capabilities to engage with patients. While we already leverage AI messaging extensively, we’ve found that for follow-up appointments, general reminders or when more explanation is needed about a topic (i.e., the importance of annual wellness visits for Medicare patients), a virtual voice experience brings a stronger level of engagement from patients.

Our voice bot utilizes AI to intelligently respond to patients and tailor the conversation based on responses or feedback. We are currently utilizing the voice bot for ER visit follow-ups, chronic condition check-ins and annual wellness visits outreach with future uses in development.

While the virtual voice experience in most cases can complete the interaction with the patient, it also has the ability to offer a warm hand-off to a staff member should the patient make that request or if the complexity goes beyond what the voice bot can do. Already, we are seeing positive early adoption by our patients and efficiency gains by our staff.

Shekar Ramanathan. Director of digital transformation at Atlantic Health System (Morristown, N.J.): We’ve had a lot of success with utilizing AI to improve operational processes such as automating prior authorizations, monitoring the health of our technology systems and improving the accuracy of coding.

Clinically, imaging really stands out as an area where we’ve been able to most effectively impact care delivery. We have deployed technology that empowers our radiologists to detect urgent cases faster and reduce the overall report turnaround time directly within their workflow by utilizing technology that analyzes medical imaging and flags acute abnormalities across the body. Incidental findings that may otherwise have gone without closed loop follow up are identified by AI that reviews radiologists reports to ensure the appropriate care coordination happens and AI allows for personalized 3D models of patients coronary arteries for our cardiologist to create treatment plans for their patients without the need for invasive cardiac CTs.

Alan Weiss. Chief medical information officer at BayCare Health (Clearwater, Fla.): Nuance Dax. It has a broad based support of one of the most important aspects of patient care and a sticking point for providers – creating the patient note.

Nicole Harper. Vice president of revenue cycle services, CFO and chief strategy officer at Eskenazi Health (Indianapolis): To date – one of our most effective uses of AI has been with our inpt/obv level of care process. We’ve partnered with a vendor that utilizes AI to assist with determining appropriate placement of our patients, and it has certainly had a tremendously positive impact for our organization!

Rick Jiggens. Principal cybersecurity engineer at Kinwell Physician Network (Seattle): Artificial intelligence has been an empowering tool that I, as a cybersecurity engineer, use often. From a strategic or governance perspective, AI helps me with development of policies and procedures. From a defense and vulnerability perspective, AI helps me identify potential vulnerabilities and can provide the best recommendations for remediation.

Tracy A. Elmer. CIO at TrueCare (San Marcos, Calif.): TrueCare is actively exploring innovation opportunities in the AI space and have prioritized focus on ways we can enhance patient experience through augmenting our call center services. We are currently planning to implement a conversational AI tool after having great success with a virtual chatbot assistant,  which is integrated on our website. The chatbot was instrumental in helping our organization manage a large surge of COVID-19 related calls in early 2022.

The AI technology is powered by Hyro and was very quickly deployed and proved to be effective in supporting access to accurate COVID-19 info and connection to vaccine appointment information. This tool gave our patients and community convenient, immediate access to support while helping to reduce call volume and hold times.

Peter Chang, MD. Vice president of healthcare design at Tampa (Fla.) General Hospital: The most effective use of AI at TGH is centered around staffing and patient placement predictions based on volume models. Revenue cycle management is also another area where AI is starting to gain some traction at TGH. On the clinical side, we’ve deployed some amazing machine learning capabilities to establish an earlier diagnosis. We’re really excited to work on novel ways of presenting pertinent information to healthcare professionals at the right time to aid in more accurate, timely diagnosis in addition to automating care pathways.

Phillip L. Coule, MD. Vice president and chief medical officer at Augusta (Ga.) University Health System: We are using several different AI systems successfully, most notably in the imaging area but also with a pilot of a documentation AI platform.

HeartFlow is CT-based, non-invasive personalized coronary artery testing. The AI-powered analysis improves the diagnosis of coronary artery disease by evaluating both anatomy and physiology. That is, it helps us look at both blockages and how they affect blood flow to critical areas of the heart.

New technology PixelShine is based on deep-learning to reduce image noise in CT to allow lower radiation doses and improve 3D images.

Rapid AI is a machine learning software that enhances the recognition of perfusion defects for strokes and pulmonary embolism. By more rapidly identifying these defects, the team is able to make clinical decisions to mobilize specialized interventions and thereby improve door to clot retrieval times.

We are also piloting a program from Nuance called Dragon Ambient Experience. This program seeks to reduce the clerical work of the physicians to compose notes by leveraging AI to automatically generate documentation of the encounter though “listening” to the interaction to compose the visit note.

John Finkbeiner. Senior clinical informaticist at South Shore Hospital (Chicago): I would say speech-to-text technology. The most frequent use has been in the ED. I have seen very effective use in the OR setting as well. My experience in smaller hospitals is that there seems to be less use of features such as the use of key or command word. I believe many providers feel like they can work almost as fast using the base capabilities of the technology. A word here about technology acceptance: while user preference is a key, most important is the easy availability of resources as well as situational. For instance, an environment can be made awkward by a provider dictating notes in front of a patient.

David Flannery, MD. Director of telegenetics and digital genetics at Cleveland Clinic: For us in the Center for Personalized Genetic Healthcare at Cleveland Clinic, the most effective use of AI has been in chatbots for genetic risk stratification.

One project used a chatbot which was offered to every patient scheduled for a colonoscopy. If they interacted with the chat, their answers regarding their personal and family history were assessed by an algorithm, and it determined if they had an increased risk of having a genetic predisposition for colon cancer. If they did, it offered the patient the opportunity to set up an encounter with a cancer genetic counselor before or after their colonoscopy to decide whether or not to have genetic testing.

Nasim Eftekhari. Executive director of applied AI and data science at City of Hope (Duarte, Calif.): At City of Hope, our success in applying AI technology across a range of areas has enabled us to continuously improve our care and research practices as well as maximize efficiency in our operations. We have been able to effectively apply AI in many areas such as real-time clinical decision support, research and precision medicine, operations, business and finance.

We have a number of real-time AI-enabled clinical decision support tools that are fully integrated with our EMR system (Epic) and our clinicians’ workflows. These tools are being used by our physicians, nurses and the rest of the care team at point of care to make more data-driven decisions and are continuously learning and being optimized based on new data and user feedback.

In precision medicine, we apply deep learning and generative AI on genomics, imaging, clinical, free text and wearables data for a wide range of use cases. Some examples are understanding disease trajectory and progression, understanding treatment efficacy, automatic tumor segmentation, digital pathology, cancer early detection and prevention, drug discovery, and promoting diversity, equity, and inclusion in cancer research and care.

In business and operations, we have many use cases where we have successfully deployed AI solutions such as optimizing patient throughput, forecasting cash, improving revenue cycle and much more.

Ihuoma Emenuga, MD. Chief medical officer at FHCB Health System (Baltimore): Using AI to improve clinical productivity and documentation,while truly enhancing provider wellbeing is always going to be a win-win because this impacts patient safety. For us, that would be AI scribing. Why is this better than human scribes? It is affordable which makes it accessible to those who need it.

Lisa Grisim, RN, MSN. Vice president and associate CIO at Stanford Children’s Health (Palo Alto, Calif.): The radiologists at Stanford Medicine Children’s Health are leveraging a cloud-based AI algorithm to calculate a patient’s bone age. Based on an imaging study such as a wrist/hand radiograph, the AI is able to calculate the patient’s estimated skeletal maturity and automatically feed it into the radiologist’s report.

Previously, the radiologists would have to manually calculate the bone age and type and dictate it into their report. By implementing this technology, it allows the radiologist to more efficiently and accurately interpret these types of studies.

Geeta Nastasi, RN. Chief nursing informatics officer at NewYork-Presbyterian (Albuquerque, N.M.): At NewYork-Presbyterian, the power of AI is being felt from nursing to obstetrics to cardiology. Perigen has been deployed as an early warning system and clinical decision support tool for obstetrics.

According to Dr. Ashley Beecy, our medical director for AI Operations, “using artificial intelligence and other analytical techniques, Perigen continuously analyzes maternal vital signs, fetal heart rate, contractions, and labor progression to enhance clinical efficiency, timely intervention, and standardization of care.”

Azizi Seixas, PhD. Associate professor of psychiatry and behavioral sciences, director of the media and innovation lab, associate director for the center for translational sleep and circadian sciences, interim chair for the department of informatics and health data science at the University of Miami Miller School of Medicine: The school has implemented a “productivity-as-a-service” model that leverages cloud-based technology to streamline workflow and increase collaboration among researchers and healthcare professionals.

It has also invested in computer vision technology to optimize its 20,000-square-foot university distribution center, where cameras are used to gather data and track insights in real-time, helping to identify inefficiencies and improve processes.

It utilizes AI in its research and clinical operations. For instance, the school has a collaboration with IBM Watson Health to develop an AI-powered clinical decision support system to improve patient care. The system uses natural language processing and machine learning algorithms to analyze patient data and provide clinicians with real-time, evidence-based treatment recommendations.

In the field of radiology, the school is also using AI to improve diagnostic accuracy and speed up the interpretation of medical images. The radiology department has implemented an AI-powered tool that can automatically detect and flag potential abnormalities in medical images, enabling radiologists to focus on reviewing cases that require their expertise.

Another initiative is the Oneness program, which is a collaboration between the University of Miami Health System and Jackson Health System.The program aims to improve patient outcomes by enabling clinicians and researchers to access patient data across both health systems seamlessly, with the highest level of data security and data governance.

Additionally, it has created Hi-RiDE which is a platform with a web-based interface that empowers researchers with tools to aggregate, analyze and visualize de-identified EMR data.

Lastly, the creation of The Media and Innovation Lab and the Department of Informatics and Health Data Science signifies the university’s major commitment to digital health technology and innovation.

Steve Davis, MD. President and CEO at Cincinnati Children’s Hospital Medical Center: Dr. John Pestian and others are working with Oak Ridge National Laboratories to create a mental health early warning system. The project, which is supported by a $10 million investment from Cincinnati Children’s, draws on de-identified data from the EMRs of nine million pediatric visits. That data is merged with other large data sets, including reports on neighborhood pollution and weather patterns, as well as an area’s income, education levels and green space and whether or not it sits in a food desert, to create a predictive algorithm.

Pestian says that in his team’s newest work, which has not yet been fully peer-reviewed, the algorithm can use these findings to determine whether a child is likely to be diagnosed with clinical anxiety in the near future and then apply early interventions. This first-of-its-kind mental health early warning system is part of Cincinnati Children’s national initiative to move mental health care from reactive to proactive.

We are also using a multidisciplinary AI imaging core which focuses on building an enterprise approach to AI adoption in medical imaging, videos and pathology to enhance diagnostic capabilities and clinical decision-making. Ultimately, our goal at Cincinnati Children’s is to improve patient safety, outcomes, and operational efficiency by accelerating AI implementation in all areas while addressing potential inherent biases or inequities in data models.

Erik J. Blutinger, MD. Medical director for community paramedicine at Mount Sinai Health Partners and assistant professor of emergency medicine at Mount Sinai Health System (New York City): One word: accessibility. AI has been a useful tool for identifying patients who could really use population health-focused interventions. For example, AI is being used to identify those patients who may require care management and social work engagement or those who may benefit from a dietary consult while suffering from malnutrition.

Samir Courdy. Senior vice president of informatics at City of Hope (Duarte, Calif.): The best is software as a medical device. I can’t say that we are currently relying on AI to make treatment decisions. We do use ML to predict adverse events and intervene in a timely manner.

James Forrester. Vice president of IT and chief technology officer at the University of Rochester (N.Y.) Medical Center: In my opinion, the most effective use of AI to date at Univ. of Rochester Medical Center and UR Medicine is in diagnostic imaging. Our Radiologists have been heavily invested in this area for several years and the data has lent itself to the AI algorithms.

We are currently using 10 diagnostic imaging detection algorithms providing critical clinical value to our diagnostic radiologists and improving outcomes for our patients with notable successes in pulmonary embolism and intracranial hemorrhage  detection.

AI processing of image data is a rapidly growing area of AI in healthcare, and we can expect to see more exciting advancements in this field in the coming years. At the same time, it is reasonable to expect an unprecedented acceleration of AI tools across many aspects of healthcare. This acceleration will result from a few key factors that include staffing shortages, financial pressures as well as advancements in generative AI. Here at URMC I have started to socialize the term “triple A” as we think about automation and staff augmentation enabled by AI.

Share On: