Artificial Intelligence in Healthcare vs the Brick Wall of Reimbursement

Why the U.S. reimbursement system is not ready for AI, or software at all, really.

I too have been living in the artificial intelligence (AI) hype. In healthcare, especially, visions of AI physicians replacing doctors, chat-bots for chronic disease management, risk prediction models, opportunities to discover new drugs, and better medical decision making have all been circulating the healthcare sphere. In particular, here, in Washington, D.C., there have been hundreds of events to discuss AI and health.

I do not want to rain on anyone's parade (because health AI is cool and will solve big problems), but there are two major issues that must be solved before either the highly debated value or harm can be realized.

Introduction to the Problem

The U.S. healthcare system stands at a critical juncture, with artificial intelligence (AI) poised to revolutionize patient care, improve outcomes, and reduce costs. However, widespread adoption of AI in healthcare faces a significant hurdle: outdated Medicare reimbursement processes that fail to account for AI-driven care models and technologies. To fully realize the transformative potential of AI in medicine, Medicare must modernize its reimbursement framework to incentivize and support AI integration across the healthcare ecosystem.

This problem can be broken down into two overarching problems: first, medical coding is controlled by the American Medical Association (AMA). In other words, the physicians control the majority of the language of healthcare payment. The AMA has a vested interest in protecting the economic moat of physicians and they are already changing the narrative around AI by adopting the term "augmented intelligence." Basically, the physicians reject your vision of an AI doctor and subscribe to the concept that AI will serve as a tool to be used by physicians.

Second, CMS has built decades of reimbursement policy on the use of codes that describe the amount of time spent by a human clinician. The basis for the rate of payment and the construction of the reimbursement system is still largely based on time and material cost. This system is not built for software-based medical technologies.

These two problems are politically difficult to solve, technically challenging, and constrained by the constant cost-pressures faced by policymakers. If you are a venture capital firm investing in health AI, this issue may significantly constrain your returns. If you are a hospital or physician seeking to adopt AI technologies, these constraints mean you will be funding out of your internal budget (i.e., it's a cost center). If you are a developer of health AI, this issue will affect your business model.

All hope is not lost as a system that was constructed by humans can be deconstructed and rebuilt in a new form given sufficient need and will and with the growth of value-based payment models there is an opportunity to develop business models less reliant on reimbursement. However, there is a higher evidence standard required to gain adoption for value-based payment purposes.

The Promise of AI in Healthcare

Artificial intelligence has already demonstrated remarkable capabilities in healthcare, from early disease detection to personalized treatment planning. AI algorithms can analyze medical images with superhuman accuracy, predict patient risks based on vast datasets, and even assist in robotic surgeries. The potential benefits are:

1. Improved diagnoses: AI can detect subtle patterns in medical data that human clinicians may miss, leading to earlier and more accurate diagnoses.

2. Personalized treatments: By analyzing genetic information and treatment outcomes across large patient populations, AI can help tailor therapies to individual patients.

3. Enhanced efficiency: AI-powered tools can automate routine tasks, allowing healthcare providers to focus more time on direct patient care.

4. Reduced medical errors: AI systems can flag potential medication interactions, dosing errors, and other safety concerns in real-time.

5. Cost savings: By improving preventive care, reducing unnecessary tests, and optimizing resource allocation, AI has the potential to significantly lower healthcare costs.

The Reimbursement Roadblock

Despite these promising applications, the adoption of AI in healthcare remains limited. A key factor holding back widespread implementation is the current Medicare reimbursement structure, which was designed for traditional care models and fails to adequately account for AI-driven approaches. This misalignment creates several barriers:

1. Lack of specific AI reimbursement codes: Medicare's current procedural terminology (CPT) codes do not include categories for many AI-assisted diagnostic and treatment processes, making it difficult for providers to bill for these services and the AMA controls the editorial process for new CPT codes.

2. Value-based care misalignment: While Medicare has been shifting towards value-based reimbursement models, the metrics used often fail to capture the full value that AI can provide in improving long-term outcomes and reducing overall costs. Similarly, for healthcare providers participating in VBC arrangements, it is difficult to attribute cost-savings to any one technology or program.

3. Upfront investment challenges: Healthcare organizations face significant costs in implementing AI systems, but current reimbursement models offer limited ways to recoup these investments. AI development and deployment represent significant expense derived from specialized labor, computing, and validation work.

4. Regulatory uncertainty: The lack of clear guidelines for AI reimbursement creates uncertainty for healthcare providers and technology developers, slowing innovation and adoption. This issue will be solved and the current discourse in Washington, D.C. is close to implementing policies, however, most people will tell you that the FDA has a relatively good system in place for companies seeking to market AI-based medical devices or software.

5. Bias towards traditional care models: Existing reimbursement structures often incentivize traditional, volume-based care approaches rather than innovative, AI-driven solutions that may ultimately prove more effective and efficient.

An Example: Digital Therapeutics

By now you are probably familiar with the fate of Pear Therapeutics and Akili Interactive. Pear brought the first digital therapeutic mobile application, reSET, through the FDA process to treat substance use disorder. Akili brought a video-game-like interface to treat ADHD in pediatric patients. Despite positive clinical evidence and FDA approval, these companies did not succeed in the same manner as pharmaceutical companies as digital therapeutics are not a covered benefit for Government Health Programs and therefore reimbursement was a major headwind. While reimbursement has improved as of 2024, it was too late for Pear and Akili.

This same outcome is likely for many AI-based technologies unless there is significant reform to reimbursement systems.

The Path Forward: Reimagining Medicare Reimbursement

To accelerate the adoption of AI in healthcare and unlock its full potential, Medicare must evolve its reimbursement processes. Key changes should include:

1. Develop AI-specific reimbursement codes: Create new CPT codes that specifically cover AI-assisted diagnostics, treatment planning, and other AI-driven healthcare services. In many circumstances, AI technology could save money through creating more efficient care processes and through better care that mitigates downstream costs. Similarly, it is possible to use coding modifiers to provide an add-on payment for certain existing procedure codes when a validated AI technology is used and approved for that procedure type.

2. Expand value-based care metrics: Incorporate metrics that capture the long-term value of AI in improving patient outcomes, reducing readmissions, and lowering overall healthcare costs.

3. Implement AI investment incentives: Develop programs that help offset the initial costs of AI implementation for healthcare providers, similar to existing incentives for electronic health record adoption. Once validation standards are set, this may be an effective means to promote the use of provider-developed AI programs that can demonstrate real clinical and cost effectiveness.

4. Establish clear regulatory guidelines: Work with the FDA and other relevant agencies to create a clear regulatory framework for AI in healthcare, including standards for reimbursement eligibility. The FDA is working to develop processes and pathways for AI, but will argue that over 800 medical devices and algorithms have been approved already.

5. Create AI innovation pathways: Develop fast-track processes for evaluating and approving reimbursement for novel AI applications in healthcare. A new system that decouples the language of reimbursement from the AMA may be necessary to promote the use of novel technologies.

6. Support AI validation studies: Fund large-scale studies to validate the effectiveness and cost-efficiency of AI in various healthcare applications, providing data to support reimbursement decisions. More prospective and randomized clinical trials are necessary. This is a great read to learn more about this topic.

7. Encourage data sharing: Develop incentives and infrastructure for secure, privacy-compliant sharing of healthcare data to train and improve AI systems.

Conclusion

Pharmaceutical and medical device companies have the budgets to fund significant evidence generation efforts for their products because the reimbursement system and market rewards the investment. This same set of market incentives does not exist in software-as-a-medical-device and AI tools in health. This is a fundamental problem to be solved.

The potential for AI to transform healthcare is immense, promising better patient outcomes, improved efficiency, and reduced costs. However, realizing this potential requires a fundamental shift in how Medicare approaches reimbursement. By modernizing its processes to account for AI-driven care models, Medicare can play a pivotal role in accelerating the adoption of these transformative technologies.

As healthcare continues to evolve, it is crucial that our reimbursement systems keep pace with innovation. Modernizing Medicare's approach to AI is not just about adopting new technologies – it's about creating a healthcare system that is more effective, efficient, and equitable for all Americans. The time for change is now, and the benefits of acting decisively will resonate throughout the healthcare system for generations to come.