By Editorial TeamAutonomous AI that makes clinical decisions on its own is a completely different beast from assistive AI that just helps a doctor. For a primary care clinic, the promise of the former is a new revenue stream with lower overhead. But for VCs looking at pre-IPO health AI companies, the commercial models for these two approaches couldn’t be more different. Anyone kicking the tires on these companies needs to understand the divergent paths in market adoption, liability, and especially reimbursement to get a real sense of a company’s floor valuation.
The Regulatory Chasm: De Novo for Autonomy, 510(k) for Assistance
The first big split between autonomous and assistive AI happens at the FDA. Assistive AI, the kind that acts as a co-pilot for a clinician, usually gets to take the FDA’s 510(k) clearance pathway. This just requires showing it’s substantially equivalent to a device already out there, which means a lower regulatory lift and a faster path to market. Autonomous AI, on the other hand, is almost always pushed down the much harder De Novo classification path because it does something entirely new with no existing predicate. Digital Diagnostics set the precedent here, getting the first ever FDA De Novo clearance for its autonomous system, IDx-DR, which detects diabetic retinopathy without a specialist. That 2018 approval was a big deal, showing the FDA was open to the idea. In contrast, Eyenuk’s EyeArt system, which also offers autonomous AI for diabetic retinopathy, managed to get a 510(k) clearance in 2020. This regulatory choice dictates everything, development timelines, the scope of clinical trials, and how a company can even position itself to sell.
Reimbursement Realities: CPT Codes and Economic Viability
Getting an FDA clearance is just the first checkpoint. You don’t have a business until you can get paid. For autonomous diabetic retinopathy screening, everything revolves around CPT code 92229. The problem is the language in the code itself: “remote data analysis and interpretation by a physician or other qualified health care professional.” That phrase has been a major hang-up for truly autonomous systems that are designed to eliminate the interpreting physician. For 2024, the Centers for Medicare & Medicaid Services (CMS) set the national payment for CPT 92229 at about $37.05, but how physician involvement is defined is everything. The whole point of a system like IDx-DR is to let a technician in a primary care office run the test and get a result on the spot, cutting out the need for a specialist. The economic argument is simple: you break the specialist bottleneck and lower costs. But the CPT code’s emphasis on “interpretation by a physician” creates real payment friction. VCs have to ask pre-IPO companies how they’re getting around this, are they lobbying for new CPT codes that specifically cover autonomous interpretation, or can they prove a cost-savings argument so compelling that health systems will pay for it anyway? CMS Physician Fee Schedule for CPT 92229 Assistive AI, like Eyenuk’s products, fits much more cleanly into existing billing since the human doctor is still the one doing the official interpretation and submitting the bill. This gives them a more direct, if less revolutionary, path to near-term revenue.
Market Penetration and Primary Care Enablement
The real value for both AI types is getting them into primary care offices, where most patients are. The American Diabetes Association recommends annual diabetic retinopathy screenings, but millions of people don’t get them simply because there aren’t enough ophthalmologists, especially in rural areas. Autonomous AI is built to solve this access problem. For example, Digital Diagnostics’ IDx-DR is designed for a primary care tech to operate, with the AI giving a yes/no answer right away. This could massively expand screening, improve guideline adherence, and in the end save the healthcare system a fortune by catching disease earlier. Though it’s still early, primary care adoption rates are the key metric for future growth. Investors should be demanding to see proof of easy integration into existing clinic workflows and evidence that staff are actually following the protocol. Assistive AI also helps get screening into primary care, usually by making existing cameras smarter or flagging suspicious cases for a general practitioner to look at before making a referral. This approach still boosts efficiency and accuracy in a primary care office, cutting down on bad referrals and improving early detection.
Liability and the Autonomous Frontier
The liability question for autonomous AI is a huge, and frankly, unsettled issue for any investor. When an algorithm makes the final diagnosis, who gets sued for a bad call? Unlike assistive AI where the doctor holds the bag (and the liability), autonomous systems are a new legal frontier. By getting its De Novo clearance, Digital Diagnostics has already had to make its case to the FDA on this front. But the long-term legal precedents for AI malpractice are still being written in real-time. Companies in this space need airtight quality management systems (QMS), solid post-market surveillance, and clear plans for what happens when the AI gets it wrong. Algorithmic drift, where the model’s accuracy degrades as it sees new real-world data, is a major liability that needs constant monitoring and a solid Predetermined Change Control Plan (PCCP). FDA guidance on AI/ML change control With assistive AI, the liability stays with the doctor, though the software vendor still has product liability. This straightforward chain of responsibility can make assistive tech a safer bet for some investors, at least until the courts figure out who to sue when a fully autonomous AI makes a mistake.
Conclusion
The go-to-market models for autonomous and assistive AI in diabetic retinopathy screening create very different investment risks and rewards. Autonomous AI, like what Digital Diagnostics is doing, is the “holy grail” that promises complete automation and massive market access, but it comes with the massive headaches of a De Novo application, unsolved reimbursement puzzles, and novel liability questions. Assistive AI, seen in Eyenuk’s offerings, fits neatly into existing clinical and billing workflows, offering a much cleaner (if less disruptive) path to making money. For VCs focused on primary care and med devices, the decision comes down to risk appetite. You can bet on the long-term, high-risk disruption of autonomous AI, or you can go with the more incremental, but more immediate, commercial path of assistive tech. The right choice depends on a deep read of the regulatory pathways, the payment realities, and the very real risks of letting an algorithm make the final call. American Diabetes Association screening guidelines
Frequently Asked Questions
What are the key regulatory differences between autonomous and assistive AI in diagnostics?
Assistive AI typically follows the FDA’s 510(k) clearance pathway, demonstrating substantial equivalence to existing devices, leading to a lower regulatory burden and faster market entry. Autonomous AI, performing novel functions, usually requires the more rigorous De Novo classification pathway, as seen with Digital Diagnostics’ IDx-DR.
How do current reimbursement structures impact autonomous AI solutions, particularly in primary care enablement?
Current CPT codes, like 92229 for diabetic retinopathy screening, often include clauses requiring ‘interpretation by a physician,’ which can create friction for truly autonomous systems designed to remove specialist involvement. This necessitates companies to advocate for new CPT codes or demonstrate significant cost savings to overcome these reimbursement complexities, despite the potential for reduced specialist bottlenecks and improved access in primary care.
What is the primary market penetration strategy for autonomous AI in primary care, and what metrics should investors consider?
Autonomous AI aims to democratize screening by enabling primary care technicians to operate devices and receive immediate, definitive diagnostic results, expanding screening capacity. Investors should look for evidence of seamless integration into existing primary care workflows and high fidelity to protocol as key indicators of future growth and adoption.
