Regulatory Position
1. Why this document exists
Software that informs clinical care sits near a regulatory line. A vendor that has not thought about which side it is on is a risk to its customers, because the customer inherits the problem. This sets out our reasoning so a compliance reviewer can check it rather than take our word for it.
We are not asserting an exemption granted to us. We are describing design constraints adopted deliberately so the product stays on the non-device side, and naming the conditions under which that would stop being true.
2. What Vivantal does, stated for this purpose
Vivantal reads a de-identified patient record and applies deterministic, named rules to identify process gaps: a critical result never acknowledged, a monitoring interval that has lapsed, a referral ordered and never closed. It cites the published guideline or explicit statistic behind each finding, and ranks findings so the more consequential surface first.
It does not diagnose. It does not recommend a treatment or a dose. It does not analyse a signal or image from a medical device. It does not compute a patient-specific risk score for display to a clinician.
3. The four criteria
Section 520(o)(1)(E) excludes software from the device definition when it meets all four of the following. Our position on each:
(i) Not intended to acquire, process or analyse a medical image, signal, or pattern from a signal acquisition system
Met. Vivantal ingests structured record exports and, where present, the text of reports already authored by a clinician. It never touches waveforms, images, or the output of a monitoring device. There is no DICOM path, no signal processing, no image analysis.
(ii) Intended to display, analyse or print medical information about a patient
Met. The inputs are ordinary record contents: labs, medications, encounters, referrals, and report text. This criterion is about the kind of information involved and is satisfied straightforwardly.
(iii) Intended to support or provide recommendations to a health care professional about prevention, diagnosis or treatment
Met, with the emphasis on professional. Vivantal is intended for clinicians, researchers, and quality and safety staff, not for patients or caregivers. Output is framed as a worklist for professional review, and the Terms of Service restrict use accordingly.
This criterion is the one most easily lost by a product decision rather than a code change. A patient-facing view of the same findings would fall outside it.
(iv) Intended to enable the professional to independently review the basis of the recommendation, so they need not rely primarily on it
This is the criterion the product is built around, and where most clinical decision support fails. Four design commitments carry it:
- The engine is deterministic. Findings come from named rules, not from a model. The same record produces the same output every time, and the rule can be read.
- Every finding cites its basis. A published guideline, with the clause quoted, or an explicit statistic with its arithmetic shown. The clinician can check the source independently of us.
- The source data is shown. Each finding points to the record rows that produced it, so the input can be inspected rather than assumed.
- Suppression stays visible. When a guideline covers a finding and Vivantal sets it aside, that finding remains reviewable in the interface, with the rule that cleared it. A clinician can always see what was filtered and why. Hiding it would defeat independent review, which is why the Terms prohibit configuring a workflow where suppression is invisible.
Where a language model is available, it only restates a finding the deterministic engine has already produced. It cannot create, rank, or suppress a finding. That boundary exists precisely so the reviewable basis is never a model output.
4. Where the position is weaker
Stated deliberately, because a memo that finds no weakness is not an analysis:
- The Oracle ranking. Ranking who is most likely to fall through next sits closer to the line than gap detection does. It is mitigated by showing every contribution to the ordering and by presenting a band rather than a numeric risk score, but a reviewer could reasonably scrutinise it hardest, and we would expect them to.
- Time-critical framing. Criterion (iv) is harder to satisfy where a clinician has no practical opportunity to review the basis. Vivantal is built for retrospective panel review, not bedside urgency, and the Terms prohibit emergency use. A deployment that put it in an acute pathway would weaken this.
- Automation. Autopilot drafts outreach for human approval and sends nothing on its own. If it ever acted without review, criterion (iv) would be in real doubt.
5. What would change this position
We would re-examine classification, and seek regulatory advice before shipping, if Vivantal were to: present output directly to patients; issue a treatment or dosing recommendation; analyse an image or device signal; act on a finding without human approval; hide the basis of a finding; or use a model to generate, rank or suppress findings rather than explain them.
None of these are on the roadmap. They are listed so you can hold us to their absence.
6. Other regimes
- Quality improvement versus research. Vivantal is a QI and research tool. Where its use constitutes human-subjects research, your IRB governs, and we make no determination about your protocol.
- Outside the United States. Other jurisdictions classify software differently. In particular, the EU Medical Device Regulation applies a broader test than the US non-device CDS provision, and this analysis does not extend to it.
- Clinical validation. Regulatory classification and clinical validation are separate questions. Vivantal has not been validated against real patient records: no clinician has adjudicated its output and no positive predictive value has been established. We say so in the Terms and repeat it here, because a favourable classification argument is not evidence that the tool works.
7. Status of this analysis
This is our reasoning, prepared internally. It has not been reviewed by regulatory counsel and is not a legal opinion. FDA has not evaluated Vivantal. If your compliance office reaches a different conclusion, we want to hear it: write to partnerships@vivantal.com. We would rather revise this document than defend it.
← Back to Vivantal · Terms of Service · Security Overview · DPA
Research & quality-improvement tool — not a diagnostic device. Vivantal surfaces process gaps for human review. It does not diagnose.