Life Sciences / Regulatory Review 🧬

📌 Navigate

📋 Exec Summary

Q4 2022 delivered three structural shifts simultaneously: FDORA signed December 29, codifying PCCP into statute and giving FDA explicit authority for adaptive AI device modification plans. AI/ML device clearances were on the public list in the high hundreds with radiology dominant. And ChatGPT entered healthcare without a regulatory framework -- clinicians began offloading cognitive work to a general-purpose AI with no FDA oversight, widening the gap between regulated and actually-used AI in clinical settings.

📊 What Moved

📈 Trend Arcs

Arc 1: PCCP — From Guidance to Statute

Velocity: Accelerating

The predetermined change control plan concept had a long runway before FDORA. FDA first proposed the framework in a January 2021 discussion paper, inviting public comment on how to handle AI/ML software that needed to adapt post-clearance. The April 2019 discussion paper on AI/ML SaMD had identified "locked" versus "adaptive" algorithms and noted that the existing regulatory framework was better suited to the former. The 2021 PCCP discussion paper represented FDA's attempt to develop a regulatory pathway for adaptive AI that kept pace with the technology; the AI/ML PCCP draft guidance would not arrive until 2023. But guidance documents, even finalized ones, have a different legal status than statute. They can be revised, reinterpreted, or disregarded in enforcement. They do not bind Congress's successors.

The Q4 arc, culminating in the December 29 FDORA signing, moved PCCP from the guidance layer to the statutory layer. Section 515C did not replicate the draft guidance verbatim — there were differences in how the law framed manufacturer obligations — but the core structure was preserved: a manufacturer could submit a PCCP at the time of the original device submission, the plan would describe anticipated modification types and the methodology and performance testing the manufacturer would use to validate those modifications, and upon approval of the PCCP, the manufacturer could implement those changes without a new submission as long as they stayed within the approved plan. The significance was not just legal. It was a signal about where FDA and Congress expected the AI medical device category to go: toward adaptive, continuously learning systems with robust change management infrastructure, not toward static algorithms that required re-clearance every time the underlying model was updated.

Throughout October and November, the regulatory community was tracking the appropriations process — FDORA was embedded in the omnibus spending bill, which went through multiple near-collapses before its December passage. The uncertainty around passage created operational uncertainty for manufacturers who were planning device submissions with PCCP elements. The December 29 signing resolved that uncertainty, but with very little time left in the year for manufacturers to adjust their 2023 planning.

Where it stands at quarter close: PCCP is statute. FDA does not yet have an AI/ML PCCP draft guidance; the law creates the framework and the implementation details are still being worked out.

Arc 2: The Radiology AI Maturity Ceiling

Velocity: Decelerating

The radiology AI market entered Q4 2022 at a moment of visible maturity. The major cleared-device manufacturers — Aidoc, Viz.ai, Nanox (via Zebra Medical), Veracyte, Avicenna.AI, and a significant cohort of others — had FDA clearances, hospital contracts, and in some cases published evidence of clinical impact. The workflow integration problem, which had been the dominant adoption barrier in 2019 and 2020, was increasingly solved: PACS integrations, EHR connections, and radiology information system compatibility had been worked out by enough implementations that new hospital deployments had established templates to follow. In other words, the category was maturing.

Maturity in a device category has specific characteristics: competition intensifies on price rather than differentiation, the evidence bar for clinical adoption rises (a hospital that has already implemented one radiology AI solution wants outcomes data before implementing a second), and the venture premium for early entrants begins to compress. Q4 showed all of these dynamics. Several radiology AI companies that had raised large rounds in 2020 and 2021 — when the category was new and cleared devices were few — were navigating 2022 with slower growth than their investors had underwritten. The cleared device count was high, but the number of hospitals deploying AI in radiology workflows was still a minority of the total market, and the path to majority adoption was going through reimbursement constraints, not regulatory ones.

The dynamics that were decelerating in radiology were simultaneously beginning in adjacent specialties. Pathology AI — computational pathology, whole-slide image analysis — was where radiology had been in 2018: early cleared devices, active FDA engagement, significant clinical interest, but pre-mainstream deployment. Cardiology AI was a similar story, with cleared devices for ECG interpretation, cardiac imaging analysis, and structural heart disease detection but limited commercial scale. The capital and talent that had flooded into radiology AI between 2019 and 2021 was beginning to look for the next radiology.

Where it stands at quarter close: Radiology AI is a mature category — the market has moved from "will hospitals adopt this?" to "on what terms and at what price?" The adjacency plays (pathology, cardiology, dermatology) are where the early radiology AI dynamic is now visible.

Arc 3: The Regulatory Gap Widens — ChatGPT Edition

Velocity: Accelerating (new arc, started November 30)

This arc did not exist on October 1, 2022. It came into being the moment ChatGPT launched and clinicians started using it.

The FDA's Software as a Medical Device framework had been developed over years to handle software that claimed a specific intended medical use. The IMDRF's 2017 SaMD framework — the international standard that FDA's guidance was based on — distinguished SaMD from software that simply supported a medical device or general-purpose software used in a healthcare setting. Under that framework, ChatGPT used by a clinician to draft a clinical note was not a medical device: it had no specific intended medical use built into the product, and the clinical judgment remained with the clinician. But this framing, while legally coherent, did not match the operational reality that was emerging in Q4: clinicians were offloading cognitive work to ChatGPT in ways that affected clinical outputs, without any regulatory visibility into that process.

The gap this created was not a new conceptual problem — FDA had been grappling with the boundaries of clinical decision support software for years, publishing a final guidance on CDS in September 2022 that tried to define when CDS software crossed into SaMD territory. But ChatGPT's Q4 arrival exposed how unprepared the existing framework was for general-purpose AI with emergent clinical uses. The September 2022 CDS guidance was written with purpose-built clinical software in mind. It was not written for a system that could answer medical questions, draft clinical documents, explain diagnoses to patients, and synthesize research literature — all without being marketed as a medical device.

By December 31, the regulatory gap between "the AI we know how to regulate" and "the AI clinicians are actually using" had widened dramatically in a matter of weeks. No enforcement action, no guidance update, and no public FDA statement had addressed it. The arc was new and accelerating, with no indication of when or how the regulatory response would come.

Where it stands at quarter close: ChatGPT is in clinical use with no applicable FDA oversight. The regulatory framework does not have a clear mechanism for addressing general-purpose AI used in clinical settings. FDA has not publicly acknowledged the gap. The arc is wide open going into 2023.

🗺️ Landscape Shift

🧭 Regulatory Direction of Travel

FDA's trajectory in Q4 2022 is best understood as consolidation and codification. The agency had spent 2019–2022 developing the conceptual framework for AI/ML SaMD: the 2019 proposed action plan, the 2021 PCCP discussion paper, and the September 2022 CDS final guidance. By Q4, the conceptual work was largely done. FDORA's passage codified the most important piece — PCCP authorization — into statute. The agency's near-term work was shifting from framework development to implementation: what does an approvable PCCP actually look like? What evidence does FDA expect manufacturers to include? How will the agency review PCCP submissions in practice?

The answers to those questions were not published in Q4. Manufacturers who wanted to use PCCP in 2023 were working from a statutory authority without final FDA implementation guidance. That ambiguity was the dominant regulatory operational challenge going into Q1 2023.

Clearance velocity and mix:

• Public-list AI/ML devices cleared through year-end 2022 remained on a snapshot-dependent basis
• Q4 clearance pace roughly consistent with the 2022 trend of ~3-4 AI/ML devices per week
• 510(k) remained the dominant pathway, accounting for the vast majority of AI/ML clearances
• De Novo grants for AI/ML devices were rare but consequential — each De Novo that established a new predicate opened a 510(k) pathway for future entrants in that clinical area
• PMA for AI/ML devices remained essentially unused in practice; the evidentiary bar for PMA was inconsistent with the evidence available for most AI diagnostic algorithms

Evidence standards:

• FDA's expectations for AI/ML device validation were evolving but not yet published in binding guidance
• The 2021 AI/ML Action Plan had committed FDA to issuing good machine learning practice (GMLP) guidance; this had not yet been finalized by Q4 2022
• Reader studies and retrospective validation on curated datasets were still the norm; prospective clinical trial evidence was the exception, not the requirement
• Subgroup analysis — ensuring AI device performance was equitable across demographic groups — was an increasing area of FDA reviewer scrutiny, driven in part by published literature documenting performance disparities in deployed AI medical devices

International alignment:

• IMDRF's AI/ML SaMD working group continued to publish guidance documents intended to harmonize regulatory approaches across FDA, EMA, Health Canada, MHRA, TGA, and PMDA
• The degree of actual harmonization remained limited in practice; different jurisdictions were at different stages of framework development
• EU MDR/IVDR, which had fully entered force in 2022, created a separate compliance pathway for devices sold in the EU that was not aligned with FDA's SaMD framework in several important respects
• UK MHRA was developing its own AI/ML medical device guidance post-Brexit, creating a third distinct regulatory pathway for companies seeking global market access

💰 Funding & Deal Pattern

Q4 2022 life sciences AI funding reflected the same bifurcation visible in the broader AI sector, with sector-specific dynamics layered on top.

🔍 The Counter-Narrative

• The consensus: PCCP is the key unlock for AI medical device adoption. The reality: PCCP addresses re-clearance friction for adaptive AI, but the real bottleneck is reimbursement. A cleared device with an approved PCCP can update its model post-clearance — but if there's no standalone CMS coverage determination, the hospital still can't separately bill for the AI's contribution. The regulatory pathway is smoother; the commercial pathway is no different.

• The consensus: A high cleared-device count proves a mature, broad-based sector. The reality: The public list was still heavily concentrated in radiology. Psychiatry, neurology, gastroenterology, dermatology — near-zero cleared AI devices. The aggregate count suggested breadth; the distribution revealed deep concentration in imaging, with most of clinical medicine still in pre-clearance stage.

📐 Builder's Benchmark

Median clearance timelines (Q4 2022 estimates):

• 510(k) for AI/ML-enabled devices: approximately 90–180 days from submission to clearance, for well-prepared submissions in established predicate areas (radiology)
• 510(k) for AI/ML-enabled devices in novel clinical areas: 12–18 months or more, often requiring multiple Q-subs (pre-submission meetings) to align with FDA on the regulatory strategy before filing
• De Novo: 12–24 months for AI/ML applications; limited precedent makes estimates uncertain
• PMA with AI/ML components: multiple years; essentially unused for AI-primary device applications

Reimbursement coverage rates for AI-enabled devices:

• Standalone CMS coverage decisions for AI medical devices: a small minority of the public-list devices had specific coverage determinations
• Most AI medical device revenue in 2022 was generated through indirect billing — the device was deployed at hospitals, the underlying clinical service (imaging interpretation) was billed at standard rates, and the AI was captured through volume-based vendor agreements
• CMS's transitional coverage for emerging technologies (TCET) framework was not yet public in Q4 2022; it would be proposed in 2023 and finalized later

Clinical validation study designs being accepted:

• Reader studies comparing AI-assisted vs unassisted radiologist performance were the standard for radiology AI clearance
• Retrospective validation on curated, annotated datasets was acceptable for 510(k); FDA was beginning to raise questions about dataset representativeness and demographic subgroup performance
• Prospective clinical trials were not required for 510(k) clearances but were increasingly requested by health system procurement committees even when not required for clearance
• The evidence package required by payers was substantially more demanding than what FDA required for clearance — hospital procurement and payer contracting had become a higher evidence bar than the regulatory process itself

PCCP adoption rate:

• Near zero in Q4 2022 — the mechanism was just entering statute. No manufacturer had gone through a full PCCP review and approval cycle
• Expectation: the first PCCP approvals would come in 2023 once FDA published implementation guidance and manufacturers had time to develop compliant submissions

👀 What to Watch

Q1 2023 catalysts:

• FDA final PCCP guidance publication (no date, expected H1 2023): The statutory framework is in place, but the operational guidance is still pending. Final guidance will answer critical questions about what an approvable PCCP looks like, how FDA will review them, and what the performance monitoring obligations are post-approval. Watch for: scope of modifications covered, evidence requirements, post-market surveillance obligations.

• First generative AI clinical workflow products seeking regulatory clarity (Q1–Q2 2023): Companies building ChatGPT-powered tools for clinical documentation, prior authorization, clinical decision support, and patient communication will need to determine whether their products are SaMD. Watch for: first FDA pre-submission (Q-sub) meetings involving generative AI tools in clinical contexts, and FDA's response to those inquiries.

• CMS TCET framework development (H1 2023): The transitional coverage for emerging technologies pathway was not yet public in Q4 2022. Watch for: TCET finalization and how AI medical devices are positioned within the framework. This is the most direct potential intervention on the reimbursement gap.

• EU AI Act foundation model provisions (ongoing negotiation): The debate over whether foundation models like ChatGPT require their own regulatory category — and what obligations that category would impose — directly affects healthcare AI developers building on foundation models. Watch for: Parliament position on general-purpose AI, how medical device AI interacts with both the AI Act and the MDR/IVDR, and any indication of a provisional agreement timeline.

• Health system AI governance policies (Q1 2023): The gap between ChatGPT's clinical adoption and institutional policy will drive health system compliance departments to act. Watch for: published AI use policies from major academic medical centers or health systems, and whether those policies treat generative AI as categorically different from purpose-built clinical AI.

Covers: October 1 – December 31, 2022. Primary events: FDA CDS final guidance (September 2022, carry-forward), FDORA signed December 29 (PCCP codified into statute), ChatGPT launch November 30 (created regulatory gap in healthcare), ~170 AI/ML devices cleared. Next quarterly: 2023-Q1.

📎 Sources

Key references for this quarter. Links provided where available; historical entries may reference publications by title and date.