SaMD Misclassification: When Your Software Is a Medical Device and You Didn't Know

← Back to All Insights

Software as a Medical Device (SaMD) is FDA's term for software intended for medical purposes performed without being part of a hardware medical device. The category sounds narrow. In practice, it covers an expanding range of digital health applications—clinical decision support tools, AI-driven diagnostic aids, treatment-planning algorithms, even some patient-monitoring dashboards. Digital health companies that built around "we're just software" are increasingly finding that FDA disagrees.

Misclassification is not a hypothetical. FDA has issued enforcement actions against multiple companies in the past 24 months for marketing SaMD without appropriate clearance.

What Counts as a Medical Device

Under 21 USC §321(h), a medical device includes "an instrument, apparatus, implement, machine, contrivance, implant, in vitro reagent, or other similar or related article" intended for use in diagnosis, cure, mitigation, treatment, or prevention of disease. Software meets the definition when it is intended for any of those uses—even when it does not directly touch the patient.

The 21st Century Cures Act Carve-Out

The 21st Century Cures Act excluded certain clinical decision support (CDS) software from device regulation, but only when the software meets four criteria: (1) it is not intended to acquire/process medical images; (2) it is intended for display, analysis, or printing of medical information; (3) it is intended to support recommendations to healthcare professionals; (4) it allows the professional to independently review the basis of the recommendation. AI-driven tools often fail criterion 4 because the underlying model rationale is not transparent enough for independent review.

Common Misclassification Patterns

Vantage frequently sees these patterns in digital health diligence: (a) a wellness app that has crossed into diagnostic claims; (b) a population health analytics platform that provides treatment-specific recommendations; (c) a clinical decision support tool whose AI model's outputs cannot be independently verified by the clinician; (d) a remote patient monitoring platform that triggers clinical alerts based on proprietary algorithms; (e) a pharmacovigilance tool that recommends dose adjustments.

FDA Enforcement Trajectory

FDA has shifted from primarily reactive enforcement (warning letters after consumer complaints) to more proactive enforcement (warning letters following routine market surveillance). The 2024 FDA AI/ML Software as a Medical Device Action Plan signaled heightened attention to AI-driven SaMD. Companies that marketed AI clinical tools without 510(k) clearance face increasing enforcement risk.

Remediation Pathways

For misclassified products: stop the violating marketing claims; pursue 510(k) clearance (typically 8-15 months and $500K-$2M); or restructure the product to fit within the Cures Act carve-out (which may require feature removal). Each option has different revenue and timeline implications. Investors should examine whether the target has a remediation plan and budget aligned to the regulatory exposure.

Diligence Questions

Does the product make claims related to diagnosis, treatment, or prevention of disease? Are clinicians able to independently verify the recommendation rationale? Has the company conducted a SaMD risk classification analysis (FDA's IMDRF framework)? Is there documented FDA engagement (Q-Sub, Pre-Sub) for any product where SaMD classification is plausible? What is the regulatory liability if a competitor reports the product to FDA?

SaMD misclassification is one of the most common—and one of the most expensive—digital health diligence findings. The remediation path is straightforward when the company catches the issue early. It becomes substantially more expensive (and sometimes existential) when discovered after an FDA enforcement action or during a transaction. Disciplined diligence examines SaMD classification explicitly rather than accepting the company's regulatory characterization at face value.