Short answer: A digital aromatherapy idea can combine an oil, delivery device, app, sensor, questionnaire, or algorithm, but those parts do not become a validated therapy or olfactory biomarker merely by being connected. Identify the product, intended use, measured signal, reference standard, population, comparator, privacy path, and decision made from the result. A scent preference is not a diagnostic measurement.

Validation belongs to a defined use

The FDA-NIH BEST framework separates analytical validation—whether the test measures the intended quantity—from clinical validation—whether the measured biomarker represents the relevant clinical concept. An exploratory correlation or an algorithm trained on one dataset is not sufficient for every proposed use. A clinical-utility question goes further: does acting on the result improve the intended decision or outcome?

For an olfactory system, identify the scent set, delivery conditions, response measurement, algorithm version and intended population. Preserve the validation sample and reference assessment, including errors and missing tests. A change in cartridge, scoring model or population can change the applicable evidence even if the screen still displays the same score.

Name the system before judging it

“Digital therapy” may mean a reminder app, a device that releases a scent, a breathing exercise, a research platform, or software that makes a clinical recommendation. “Olfactory biomarker” may mean an odor-identification score, threshold test, reaction time, sensor signal, chemical profile, or model output. Write the exact workflow: what the person receives, what they do, what is measured, how the data are processed, and who acts on the output.

FDA’s digital-health overview spans general-wellness software, sensors, telehealth, personalized medicine, and medical-device contexts. That range is why a product’s intended use matters. A pleasant aroma delivered by a connected device is not the same as software intended to diagnose disease or guide care. Do not infer the category from a product name, dashboard, or claim such as “personalized.”

Separate signal, endpoint, and decision

LayerQuestionsUnproven leap
SignalOdor identity, threshold, reaction time, sensor output, questionnaire, or chemical reading?That the signal is stable, specific, or clinically meaningful.
EndpointWhat outcome is measured, over what time, against what reference or comparator?That detecting an odor measures memory, anxiety, inflammation, or disease stage.
DecisionDoes the system display information, recommend a routine, change treatment, or alert a clinician?That an algorithm’s recommendation is safe, effective, or regulated as intended.

Control the olfactory test conditions

NIDCD explains that smell disorders include reduced detection and altered perception such as anosmia, hyposmia, parosmia, and phantosmia, and that evaluation may involve an ear, nose, and throat specialist. A person’s score can be affected by congestion, infection, age, medicines, smoking, environment, language, attention, fatigue, and the odor materials themselves. Preserve the test version, odor identity, concentration or presentation method, order, room, timing, instructions, exclusions, and missing data.

Do not treat one missed odor as a diagnosis or one improved score as a treatment response. A useful validation record includes a reference test, repeatability, sensitivity and specificity where appropriate, known confounders, prespecified endpoint, missing-data handling, independent test data, and a population that resembles the intended users. If the input is an aroma preference or self-report, name it as such.

Keep software claims proportional

FDA’s clinical decision-support guidance discusses software functions, including functions used by patients or caregivers, and the distinction between information and decision support. It does not classify this particular app or clear a scent protocol. If software interprets a person’s smell result, selects an exposure, or recommends a health action, preserve the exact version, model, training data, output, explanation, and human review. A black-box score should not silently direct a person to change medicine, activity, or treatment.

Record product and data safety

  1. Identify the oil or finished product, delivery route, label, lot, concentration, device model, firmware, app version, and intended user.
  2. State what data are collected, where they go, who can see them, how long they remain, and how a person can withdraw.
  3. Log calibration, room conditions, odor presentation, device failures, adverse symptoms, and deviations.
  4. Keep a human stop rule for headache, nausea, cough, breathing difficulty, distress, or a worsening symptom.
  5. Send diagnostic, treatment, or urgent safety questions to the responsible qualified authority instead of the aroma algorithm.

What would count as evidence?

For a therapy claim, look for a controlled comparison of the exact intervention, population, outcome, duration, adherence, harms, and clinically meaningful result. For a biomarker claim, look for a prespecified reference condition, analytic validity, clinical validity, and clinical utility. A prototype demonstration, association, sensor graph, or user testimonial can motivate research without proving benefit.

Where this page stops

This page does not diagnose disease, validate a biomarker, classify a device, prescribe an aroma, or provide a digital treatment protocol. The answer depends on the exact system, signal, endpoint, user, data practice, intended decision, and independent validation.