The biggest challenges for AI-enabled care are delivering real-time insights into the clinical workflow and gaining acceptance from caregivers as well as patients. Innovative solutions continue to emerge.
Since the beginning of the year, there has been a significant uptick across health plans, healthcare providers, and analytics firms utilizing AI to change how healthcare is delivered and how patients can be more engaged in their care.
AI applications in healthcare are all over the map today. Data from my firm’s digital health intelligence database, DamoIntelTM, has identified a significant rise in the launch of AI use cases across clinical and administrative areas in 2020. An analysis of AI/ML applications deployed by the top 50 health systems across the United States identifies that AI-enabled solutions fall into multiple technology categories: machine learning, natural language processing (NLP), conversational interfaces such as chatbots, and robotic process automation (RPA). COVID-related use cases in clinical and administrative areas contributed to the growth in the adoption of newer technologies such as chatbots in healthcare.
A focus on real-time interventions with AI-enabled solutions at the point of care
The biggest challenge for AI-enabled care is to deliver insights in real-time into the clinical workflow at the point of care. For example, voice recognition technologies are effective with lower-level tasks such as scribing doctor-patient encounters. However, they are yet to evolve into decision-support systems that deliver additional insights at the point of care for diagnostic and treatment decisions.
On the other hand, solutions that can deliver real-time insights are yet to reach scale and broad-based adoption. An example is Stanford University’s smartwatch-based COVID diagnosis app, in partnership with Amazon, that analyzes elevated heart rates and other abnormalities for pushing real-time alerts to patients suspected of COVID infection. Dr. Michael Snyder, Professor, and Chair of Genetics is working to scale the solution with the aim of creating a continuous monitoring framework for health indicators at an individual level. His goal is to cover anyone, anywhere who has a smartwatch. Amazon has offered millions in cloud computing credits for similar diagnostic solutions for digital health innovators across the world.
Data collaborations to drive advanced real-time analytics
If there’s one new trend this year, it’s data collaboratives. Truveta, a consortium of 14 health systems launched in February, aims to pool patient data from all the member systems to drive advanced analytics for improved healthcare outcomes. Google has announced a series of partnerships with healthcare enterprises, including Mayo Clinic, Ascension Health, and Highmark. Use cases include but are not limited to data analysis for quality measures, benchmarking, and administrative reporting. In addition to its partnership with Google, Mayo Clinic has launched new data collaboration initiatives with AI startups, targeting data from remote monitoring devices. Highmark, a major health plan based in Pennsylvania has formed a 10-year partnership with Christiana Care in Delaware to pool medical and claims data to drive better outcomes. Expect to see more consortiums as large payers and providers pool their datasets to drive efficiencies through advanced analytical insights.
Other trends that will drive the AI-enabled future of health care
Increased innovation in AI-enabled applications following the CMS final rule that allows patients to access their medical information and share it with developers looking to build new digital health products and services.
Hospital rooms of the future that will incorporate superior experiences driven by AI-enabled digital interactions among caregivers, patients, and their families. An example is the $1.5 Bn investment by Penn Medicine in Philadelphia. Titled The Pavilion, this 500-bed facility comes with patient rooms featuring interactive 75-foot monitors on the walls. John Donohue, VP of Entity Services for Medicine, has been closely involved in the technology enablement aspects of the patient room of the future. He references Disney-inspired user experience design as a part of the 6-year project in the making.
Analytics from remote monitoring devices. AI-enabled applications that ingest and analyze vast amounts of data from home monitoring devices and sensors will drive the next stage of the evolution of health care. As healthcare moves from hospital to home, expect heavy investments in analyzing data from remote sensors and monitoring devices. Amazon’s recently launched Amazon Care offering includes home-based care in addition to virtual care services as part of the overall package. Large health systems such as Kaiser Permanente and Mayo Clinic have also gotten into the game. They have announced investments in Medically Home, a tech firm that caters primarily to home-based care.
The patient is ready now – or is she?
While technologies and the computing infrastructure for AI-enabled care have matured, the adoption of AI-enabled care is driven by the varying readiness levels for the incumbents in the current healthcare ecosystem and concerns around the safety of AI-enabled care, specifically for complex clinical conditions.
Patients are not sure about AI-enabled care either: a recent study points out that patients find chatbots intrusive and are hesitant to take medical advice from a bot. Administrative use cases for AI-enabled applications may deliver better ROI in the short term. Sachin Patel, CEO of Apixio, a healthcare analytics company acquired by Centene in 2020, attests to a 4x to 7x return on AI applications in financial operations such as risk adjustments.
My firm’s research indicates that over half of all hospitals in the country continue to use electronic health record (EHR) systems as the primary tool at the point of care. New, cloud-based, AI-enabled solutions continue to face challenges in integrating seamlessly into the clinical workflow at the point of care. Interoperability concerns and the challenges with standardization and normalization of healthcare data will continue to remain a significant challenge for AI-enabled applications. Further, standards such as ICD, SNOMED, and FHIR continue to evolve, representing a continuing demand for authoritative code change management and data normalization solutions validated by subject matter experts. New and emerging data sources, such as genomic data, will require additional guardrails around ethics and privacy prior to use in AI applications.
A final concern around AI in healthcare relates to the lack of visibility to how the algorithms are trained to work in healthcare, further exacerbated by systemic bias inherent in many AI applications. Despite advances in AI techniques, algorithms trained on a data set cannot be easily transferred to another data set, especially as the role of operational data and social determinants of health in population health risk assessment increases. As cloud platforms become the dominant data repositories for developing AI-enabled solutions, concerns around the privacy protections for the data will drive trust and consent required for advancing the adoption of AI tools.
A bright spot for AI in healthcare is the brisk pace of AI adoption in administrative functions. Health systems executives must expand the scope of these applications to cover new operational areas, including access and patient engagement, to drive efficiencies and improved quality of experience. Clinical leaders must continue to expand the use of AI applications cautiously and focusing on operational areas that don’t necessarily look to replace human intuition and judgment. An example of this is the use of AI to optimize schedules for chemotherapy at Penn Medicine.
As healthcare leaders look to accelerate the adoption of AI, they must also carefully weigh the costs and benefits of the efforts involved in developing and deploying AI solutions. The question always comes back to what we can do with insights we get back from AI applications. If we cannot move the needle based on the insights and information, clinical leaders must question the value of the program and the energy that goes into producing the insights in the first place. The key is to invest in areas where we can see demonstrated results and build it out from there. We are still several years away from the pervasive use of AI in core clinical aspects of healthcare. Until then, we simply continue to push the frontiers.