By Staff Writer | MedTech Industry Desk

The U.S. Food and Drug Administration (FDA) continues to refine its clearance mechanisms for medical devices, with a pronounced shift toward software-enabled products, artificial intelligence (AI) algorithms, and decentralized clinical evidence. As of the fourth quarter of 2025, the agency has issued over 3,200 510(k) clearances year-to-date, a modest 4% increase from the same period last year, according to internal agency data shared at the recent MedTech Conference in Boston. However, the more telling statistic is the composition: nearly 38% of all clearances now involve products with a software component, up from 22% just three years ago.

This trend underscores a fundamental recalibration of what “FDA clearance” means in an era where hardware is increasingly commoditized and clinical value resides in data interpretation. Industry observers note that the traditional predicate-based pathway is being stretched, and the agency’s response—through guidance documents, pilot programs, and a growing reliance on real-world evidence (RWE)—is shaping the next decade of device innovation.

The Rise of the “Software as a Medical Device” (SaMD) Predicate

One of the most significant developments in recent months is the FDA’s acceptance of SaMD products as valid predicates for future 510(k) submissions. Historically, a predicate device had to be a physical, legally marketed device. But in March 2025, the agency cleared an AI-based cardiac monitoring algorithm using a prior AI-based algorithm as its predicate—marking the first time a purely software product served as the primary comparator for clinical equivalence.

“This is a quiet but seismic change,” says Dr. Elena Marsh, a former FDA reviewer and now regulatory strategy lead at a Silicon Valley diagnostics firm. “The agency is signaling that if you can demonstrate your algorithm produces outputs within an acceptable performance envelope of a previously cleared algorithm, you may not need a full clinical trial. That drastically lowers the barrier for iterative AI improvements, which is exactly what the industry needs.”

The FDA’s Digital Health Center of Excellence has also released a draft guidance titled “Predicate Considerations for Software as a Medical Device,” which explicitly outlines factors such as input data types, intended patient populations, and output decision thresholds. Public comments closed in August 2025, and stakeholders expect final guidance by mid-2026. Legal experts warn, however, that the “software predicate” concept could face litigation from consumer advocacy groups concerned about diminishing clinical oversight. The agency counters that its Total Product Life Cycle (TPLC) framework, including post-market surveillance requirements, provides sufficient safeguards.

Real-World Evidence Moves from Adjunct to Core

Another key trend is the FDA’s growing acceptance of real-world data (RWD) as primary evidence for 510(k) clearance, particularly for modifications to existing devices. In September 2025, the agency cleared a next-generation insulin pump based largely on data drawn from continuous glucose monitor (CGM) records of 1,200 patients who used the predecessor device. No new randomized controlled trial was conducted.

This approach aligns with the FDA’s 2024 “Real-World Evidence for Regulatory Decision-Making” guidance, which encouraged sponsors to use registries, electronic health records, and payer claims data. However, the agency has also tightened its expectations for data quality. A recent refusal-to-file letter, issued to a mid-sized orthopedic company, cited inconsistent data curation across three different health systems, prompting the FDA to require a formal audit before resubmission.

“We are seeing a bifurcation,” notes Marcus Chen, a partner at a Washington D.C.-based regulatory law firm. “For low-risk modifications, RWE is becoming the gold standard. For novel, high-risk devices, the FDA still wants prospective trials. The challenge is that many companies are trying to use RWE for both, and that’s where they run into trouble.”

Chen points to the FDA’s new “Evidence Threshold Framework,” introduced in draft form in July 2025, which asks sponsors to define a minimum clinically meaningful difference for their device before selecting an evidence source. This framework is designed to prevent “data fishing” and to ensure that RWE submissions are scientifically rigorous, not just voluminous.

The 510(k) vs. De Novo Boundary Blurs

A third development is the gradual erosion of the line between 510(k) clearance and De Novo classification. Historically, the De Novo pathway was reserved for novel devices with no predicate. But as the FDA clears more software and combination products, sponsors are increasingly choosing De Novo for products that could theoretically find a predicate—because De Novo grants the manufacturer exclusive rights to the new device type for a period, and it sets a new, modern performance standard.

Data from the FDA’s public database shows that De Novo requests have surged by 45% in 2025 compared to 2023. In response, the agency has hired additional classification officers and initiated a “fast-track De Novo” pilot for devices that address urgent public health needs, such as point-of-care diagnostics for antimicrobial-resistant infections.

“The 510(k) pathway is no longer the default for clever startups,” says Dr. Priya Raghavan, a biomedical engineer and former industry consultant. “If your device is genuinely new in mechanism or intended use, De Novo gives you a cleaner regulatory story and better reimbursement positioning. But the FDA is also stretching its resources. We’ve seen some De Novo reviews drag past the statutory 180-day deadline, which creates uncertainty.”

Expert Outlook: A More Adaptive, But More Complex, System

Looking ahead to 2026, the FDA is expected to finalize its “Predetermined Change Control Plans” (PCCPs) for AI/ML devices—a mechanism that would allow manufacturers to make pre-specified, algorithm-level updates without submitting a new 510(k) each time. The agency has already approved two PCCPs in 2025 for imaging software, and industry groups are lobbying for broader applicability.

However, experts caution that faster clearance does not mean less scrutiny. The FDA is simultaneously expanding its post-market surveillance requirements, including mandatory cybersecurity reporting for connected devices and more rigorous adverse event analysis using automated natural language processing on complaint databases.

Dr. Marsh summarizes the current sentiment: “FDA clearance is no longer a one-time stamp. It’s a continuous dialogue between the sponsor and the agency, with data flowing throughout the product life cycle. That’s harder on companies, but it’s also more honest about how technology actually evolves.”

For manufacturers, the practical takeaway is clear: invest early in data infrastructure, engage with the FDA’s digital health pre-submission programs, and treat RWE as a strategic asset rather than a last resort. The regulatory bar is not dropping—it is changing shape, and those who adapt will find a faster, more predictable path to market. Those who do not will face an increasingly bewildering maze of new guidance, pilot programs, and evidence expectations.