Artificial intelligence could transform clinical care, but regulatory hurdles slow adoption. Ensuring healthcare AI has adequate policy safeguards is paramount for safe, widespread integration across the sector.   

AI Improves Care and Solves Dilemmas

AI stands to improve care, cut costs, and solve workforce shortages, but regulatory challenges persist. The tide seems to be turning in favor of healthcare AI adoption, with the recent historic rollout of the FDA AI Scientific Review Pilot. This underscores that AI adoption is no longer a future aspiration, but a present-day reality.

The FDA review pilot fast-tracked new generative AI internally, cutting tedious tasks and freeing scientists and subject-matter experts to expedite processes. FDA Commissioner Dr. Martin Makary said this agency-wide advancement “holds tremendous promise in accelerating the review time for new therapies.” 

AI stands to speed up image analysis, drug development, and personalized medicine. Clinicians utilizing AI to review CT scans, stratify prostate cancer risk, and test for breast cancer recurrence are streamlining processes and maximizing effectiveness and improving patient outcomes. Currently, Cleerly, HeartFlow, and Caristo Diagnostics are exploring AI analysis of coronary CT angiography. 

 The Evolution of AI 

While AI’s ascendance in clinical care has moved at a rapid pace in recent years, it has been utilized as a healthcare tool for half a century. Theadvent of AI in healthcare dates back to the 1970s, with the creation of the world’s first artificial medical consultant, INTERNIST-1. The diagnostic tool demonstrated AI’s potential to advance medicine and alleviate clinical burdens. Soon after, the National Institutes of Health held the first AI in Medicine Conference at Rutgers University. 

These interdisciplinary meetings led to new medical AI systems and applications. From the 1976 launch of MYCIN, which suggested antibiotics for pathogens, to further diagnostic tools and disease-specific platforms, healthcare AI development was steady through the 1980s. The 1991 Pathway Expert Interpretive Reporting System and 2003’s Human Genome Project each moved the innovation needle forward. Neurologists used IBM’s Watson for ALS diagnosis in 2017. The same year, new FDA-approved technologies included an AI product to analyze heart MRIs and the first AI-powered device for operating room use. That was followed in 2019 with AI devices to diagnose cancer and read brain MRIs and the launch of the Cedars-Sinai Division of Artificial Intelligence in Medicine. 

This decade has seen rapid AI momentum in healthcare, starting with Google’s DeepMind, which predicted a protein’s 3D structure from its amino-acid sequence in 2020. In 2022, the FDA authorized 91 AI-powered devices, including the EchoGo Heart Failure tool, which detects heart failure from a single echocardiogram. By 2024, a full 79% of healthcare organizations reported in a Microsoft study that they were using AI technology.  

Real-World Benefits of Health AI

Today, AI diagnostics empower physicians to concentrate more time and energy focused on treatments and patient care. This includes technology for cancer screening in mammography and repurposing CT scans for cardiac screening. AI is enhancing the coronary diagnosis and prognosis prediction through CT angiography and MRI.  AI holds great potential to reduce the need for more invasive imaging for cardiac patients. Mayo Clinic is using AI to advance the diagnostics of neurodegenerative diseases by analyzing routine electroencephalography (EEG) recordings. 

Demonstrating the technology’s broad benefits, oncology researchers are using AI-applied genomic data to guide treatment decisions. The costly, time-intensive Oncotype DX (ODX) test predicts breast cancer risk and recurrence and the benefits of chemotherapy. Researchers seeking an efficient alternative are using AI to predict ODX scores using clinical markers and images, revealing AI’s capacity to reduce healthcare costs and improve accessibility to vital diagnostics. 

Illustrating its life-saving impact, a patient with a rare, deadly autoimmune disease was set to enter hospice care until a group led by researchers at the Perelman School of Medicine at the University of Pennsylvania used an AI tool to find a life-saving medicine. After examining 4,000 existing medications, the AI tool found one to treat the rare idiopathic multicentric Castleman’s disease, which has few treatment options. The patient is now in remission. 

AI could also redefine drug discovery paradigms of trial-and-error experimentation reliant on researcher experience. New large-language models and generative AI enhance both efficiency and effectiveness in the development pipeline, namely in the breast cancer and rare disease arenas. This new reality with AI promises faster, more targeted, and ultimately more affordable drug development.

(IDx) ADVI secured some of the first AI codes in 2021, such as Category I code, 92229, for Imaging of retina for detection or monitoring of disease: point-of-care autonomous analysis and report, unilateral or bilateral. The code was novel in that it was nationally valued on the Medicare Physician Fee Schedule (PFS) to reflect physician work, and in the hospital Outpatient Prospective Payment System (OPPS) setting to reflect hospital payment.

Protecting Patients in the AI Age, Prioritizing Privacy and Equity 

Patient advocates and U.S. leaders remain vigilant against jeopardizing health privacy through AI implementation. While AI models rely on acquiring and processing large amounts of data to train models, the use of patient data must be weighed against HIPAA regulations and privacy risks. Evidencing the Federal government’s cautious approach, in a February 2024 speech, Federal Trade Commission Chair Lina Khan said sensitive patient health data “is simply off limits for model training.” In 2021, the U.S. Department of Health and Human Services established the Office of the Chief Artificial Intelligence Officer to lead AI strategy and governance, coordinating with other agencies and offices. This proactive stance is important but should be balanced with policies that encourage responsible data utilization for progress. 

Physicians, on the other hand, are boldly embracing AI. A 2024 American Medical Association poll found that AI use nearly doubled in a year. Of doctors surveyed, 66% use AI in the clinical setting, up from 38% in 2023. Administrative tasks accounted for 57% of use – including billing, medical charts, visit notes, and discharge instructions. With such rapidly increasing demand, AI developers need clear guardrails to ensure that emerging AI models are compliant with Health Insurance Portability and Accountability Act (HIPAA).

Collaboration across sectors and dynamic regulatory systems are necessary to develop ethical AI, preserve privacy, and mitigate bias. Leaders can look at the regulatory successes and hurdles faced by other emerging healthcare technologies. For example, maintaining electronic medical records compliance poses challenges due to CMS program and HIPAA privacy rule updates. This can potentially offer valuable lessons on navigating the complexities of AI integration. 

While AI brings unquestionable value to clinical practice, it could also inflict harm through bias. It is critical to recognize and mitigate bias based on gender, race, age, socioeconomic status, geography, age, and other factors before it exacerbates existing inequities. Not only must we identify bias, but we must also eliminate it from the AI model lifecycle – and address it in policies. The American Medical Association is addressing the issue of AI bias in Current Procedural Terminology (CPT®) and through new AMA policy. It is a moral and professional obligation to address potential biases to ensure AI serves all patients equitably.  

Current Policy Landscape & AI Regulatory Hurdles

Despite AI’s undeniable clinical benefits, regulatory hurdles delay broader use. A 2024 JAMA review found that the FDA had authorized almost 1,000 AI-enabled medical devices, yet this is just the tip of the iceberg. JAMA recommended regulators advance flexible ways to keep pace with AI development in biomedicine and healthcare, but obstacles are at play. 

Primarily, there’s the issue of reimbursement, which economically inhibits AI advancement, disincentivizing innovation. Under the current CMS Physician Fee Schedule (PFS), software is considered an indirect expense rather than a direct, reimbursable one. This issue must be addressed to encourage greater AI adoption and manage patient costs. Without a clear path to reimbursement, the power of AI in healthcare will remain largely unrealized. 

The current FDA stance on AI is at odds with the developer’s “machine learning” perspective. AI technologies, capable of continuously learning and morphing, present unprecedented regulatory challenges for the FDA. Recognizing this, the Coalition for Health AI created a framework to enhance the safety and effectiveness of healthcare AI. In 2021, the FDA partnered with international agencies to release guiding principles called predetermined change control plans (PCCPs) for machine learning-enabled medical devices. NIH funding is another stream of support for emerging AI and machine learning research. 

In January, the FDA issued draft guidance for using AI to support drug development decisions. The framework, expected to be finalized later this year, provides risk-based AI assessments in cases of regulatory decisions regarding product safety, effectiveness, or quality. The agency has also published draft guidance for AI in software as a medical device.

Topping the list of legislative roadblocks is the CMS refusal to reimburse software as a direct clinical expense. The persistent dilemma of bundled payments for outpatient hospital treatments requires that AI costs must be absorbed, as they are not separately reimbursable as part of a physician’s evaluation and management. These factors dissuade greater AI adoption. 

Policy Recommendations for Enhancing AI in Patient Care 

Given the potential AI holds and the rapidly growing demand for it in the clinical setting, it’s time to embrace new reimbursement models. We believe CMS should recognize software as a direct practice expense. Because generalist AI systems can perform multiple complex tasks, it’s important to assess and reimburse the value of these tools. Potential solutions include new reimbursement structures, continuous performance evaluation, and cost-effectiveness assessments.

It’s also critical that the U.S. balances regulation, prioritizing safety while utilizing AI’s power. Medical technology association AdvaMed recently released an AI policy roadmap with solid recommendations ranging from legislative solutions for budget neutrality constraints to ensuring the FDA examines generative AI using existing risk-based frameworks. This document is worth careful consideration, as it offers a reasonable path forward for realizing AI’s potential. 

An MIT report likened AI to the steam engine, the internal combustion engine, electrification, and computers, “general-purpose technologies” with the power to transform economies and societies. The FDA, healthcare professionals, and technology sector leaders must cautiously pursue broader AI adoption with proper guardrails. Based on the successes of AI as a healthcare administration vehicle, the U.S. should pursue greater integration of AI into healthcare, such as with EMR systems. Need persists for continued NIH funding to support AI research specific to healthcare, as this is doubtless a key innovation vehicle of the future and crucial for keeping the U.S. at the forefront of medical technology.  

To learn more about our work on AI, technology enablement, and ADVI’s work with CMS through clients and the Duke Margolis Institute for Health Policy contact us at innovate@advi.com.

ADVI offers expert advice and influential solutions, informed by data and guided by a clear vision of the complex intersection of life sciences/healthcare innovation, economics, policy, and patient impact. Get in touch to receive insights from our Market Access, Precision Medicine team.