A week or so ago, Bloomberg Law published (behind a pay wall) an article entitled “Medical Device Makers May Risk Liability Following FDA Effort.”  The Bloomberg article relied upon the same FDA initiative regarding artificial intelligence/machine learning (“AI/ML”) in medical devices that we covered here, in a guest post.  Our guest post raised preemption as one issue to consider:

As product liability litigators, concerns about the proposed framework’s effects on preemption and duty to warn/learned intermediary doctrine are on our radar screens.  For example, will regulation of SaMD self-modification be “rigorous” enough to support preemption?

The Bloomberg article, on the other hand, pushed the view that the FDA’s regulation was creating sort of a preemption “grey area,” and that it’s “unclear” whether the FDA regulation would be “rigorous enough.”  We even read, “if algorithmic changes expand a device’s scope beyond that for which it was approved, preemption becomes an ‘open question.’”

That’s about as far as “fair use” lets us sample the Bloomberg piece, so we’ve decided to examine the subject in a bit more detail.  First of all, as long as the FDA chooses to regulate a medical device featuring AI/ML as requiring pre-market approval (“PMA”) – and the Supreme Court continues to follow Riegel v. Medtronic, Inc., 552 U.S. 312 (2008) – there will be preemption.  We suppose that, should the still rather vague concept of “focused review” mutate into an entirely new form of FDA pre-market device review (one not currently provided by the Medical Device Amendments to the FDCA), either the Agency in the first instance or the courts would have to decide whether that new, neither-fish-nor-fowl process was preemptive.  But as long as the FDA includes its approach to AI/ML medical devices within the rubric of the PMA framework, Riegel preemption will apply.

Assuming the FDA continues to use the framework, for AI/ML medical devices, tort claimants relying on state law are not permitted to second-guess the FDA’s decision to regulate a device as requiring PMA versus any other regulatory path to marketing.  Plaintiffs may not relitigate the sufficiency of FDA attention given to particular PMA devices:

The thrust of [plaintiffs’] argument is that the FDA could not have granted premarket approval to the [device] because it did not have sufficient data to do so.

However, the sufficiency of the data submitted to the FDA with respect to the safety and efficacy of a device does not govern the scope of the premarket approval.  Whether a product is FDA-approved is determined by the language in the approval letter, not by the application documents submitted to it for review.  The FDA, not litigants, is entrusted with the responsibility to police the sufficiency of the evidence to support a PMA approval.

Duggan v. Medtronic, Inc., 840 F. Supp. 2d 466, 471-72 (D. Mass. 2012).  State-law claims are “preempted [where they] require[] a fact finder to make a determination under state law that federal law leaves exclusively to the FDA.”  Zimmerman v. Novartis Pharmaceuticals Corp., 889 F. Supp.2d 757, 777 (D. Md. 2012).  At least since Buckman Co. v. Plaintiffs Legal Committee, 531 U.S. 341 (2001), was decided not quite 20 years ago, and up until, like, now, we know of no case in which a plaintiff was allowed to second-guess an FDA decision whether to require PMA for a medical device, as opposed to §510(k) “substantial equivalence” – or vice versa.  See McNeil-Williams v. Depuy Orthopaedics, Inc., 2019 WL 2179217, at *6 (E.D.N.C. May 20, 2019) (denying, as preempted by Buckman, “discovery regarding whether defendants, ‘submissions to the FDA were timely, truthful and complete’”) (this is the “now”).

Thus the idea floated in the Bloomberg article that the FDA not requiring a separate PMA for every algorithm alteration created by AI/ML could affect preemption doesn’t hold water.  Assuming the FDA chooses to regulate such changes through a PMA approach evaluating rates and direction of change, and that decision holds up under federal administrative law as within the Agency’s discretionary authority, preemption will bar state tort claimants from second-guessing the FDA’s supposed lack of “rigor” in the context of the Agency’s PMA evaluation.

[I]t is inappropriate for a jury to second-guess FDA’s scientific judgment on such a matter that is within FDA’s particular expertise.  FDA determines the scope of a device, including . . . the appropriate regulatory pathway for the device.  FDA subsequently determines whether the device meets the PMA approval standard.  The agency makes a reasoned and deliberate decision as to the correct pathway of regulation and whether to approve the device.  Juries lack the scientific knowledge and technical expertise necessary to make such judgments.  [T]he prospect of hundreds of individual juries determining the propriety of particular device approvals, or the appropriate standards to apply to those approvals, is the antithesis of the orderly scheme Congress put in place and charged the FDA with implementing.

Blunt v. Medtronic, Inc., 738 N.W.2d 143, 151 (Wis. App. 2007) (quoting Horn v. Thoratec Corp., 376 F.3d 163, 178 (3d Cir. 2004)), aff’d, 760 N.W.2d 396 (Wis. 2009).  See In re Orthopedic Bone Screw Products Liability Litigation, 264 F.3d 344, 364 (3d Cir. 2001) (“[b]ecause substantial equivalence determinations as well as the manner in which those decisions get made are functions committed to the discretion of the FDA, we will not second guess their outcomes.”); Schering Corp. v. FDA, 51 F.3d 390, 399 (3rd Cir. 1995) (“[FDA’s] judgments as to what is required to ascertain the safety and efficacy of drugs fall squarely within the ambit of the FDA’s expertise and merit deference from us.”); Center for Science in the Public Interest v. FDA, 74 F. Supp. 3d 295, 305 (D.D.C. 2014) (refusing to reevaluate “FDA’s unique − and authoritative − position to view its projects as a whole, estimate the prospects for each, and allocate its resources in the optimal way”).  See also Our recent post on Merck Sharp & Dohme Corp. v. Albrecht, ___ S. Ct. ___, 2019 WL 2166393 (U.S. May 20, 2019), about who should be deciding FDA-related issues that affect preemption.

The other approach suggested in the Bloomberg article questioned preemption in a situation where “algorithmic changes expand[ed] a device’s scope beyond that for which it was approved.”  That argument is belied by the plethora of off-label use cases that we’ve covered on the Blog over the years, many of which we summarized here.  To summarize our summary, the teaching of the great majority of the off-label use cases is that, preemption depends on pre-market approval of the “device” itself,” not a device’s potentially many “intended uses,” a regulatory term that can change over time.  The FDA approves “devices” not “uses.”  So once PMA is obtained, preemption exists under Riegel, regardless of which particular “use” or “intended use” a device is put to in any given case.

That’s the same concept identified in the Bloomberg piece, only instead of a manufacturer’s alleged “off-label promotion” changing a device’s “intended use,” the purported change is a consequence of incremental changes causes by AI/ML.  Again, that is for the FDA, not private state-law litigants to decide.  The FDA’s ultimate regulation of AI/ML medical devices will no doubt superintend the evolutionary process of machine learning by establishing boundaries beyond which substantial evolution of a self-teaching device will require an additional PMA.  Indeed, that’s exactly what we read the FDA’s discussion paper as doing:

For example, an app that was used to flag skin moles as potentially cancerous (and recommend a visit to a doctor) could morph into a program that could definitively diagnose melanoma (without a trip to the doctor).  Such a change in the possible use for a life-threatening condition would necessitate pre-market approval (“PMA”).

Thus, the FDA will decide when evolutionary changes to a medical device caused by AI/ML will – and equally importantly, will not – require a new PMA.  As far as private plaintiffs attempting to second-guess an FDA decision not to require another PMA at any particular time, that takes us back to the first argument, since these type of decisions may not be collaterally attacked via state-law litigation.

Thus, upon further reflection, we don’t think that anything in the FDA’s currently envisioned approach to AI/ML medical devices portends any weakening of PMA preemption.  What the FDA says is a PMA device will be treated as a PMA device in product liability litigation, particularly including preemption.  If the FDA ever clears an AI/ML device under §510(k) as “substantially equivalent,” then the device will have whatever preemption protection that process is interpreted as providing.  Plaintiffs cannot now, and will not in the future, be able to second the FDA’s decision how, or when, to require PMA.