For patients
Long periods of uncertainty can mean missed care, avoidable cost, and the exhaustion of having real symptoms repeatedly dismissed.
The Invisible Patient Finder
Make invisible illness visible.
A patient-centered research concept designed to learn from the years before diagnosis—so care teams can recognize overlooked patterns earlier, measure uncertainty, and choose a better next question.
Concept-stage decision support. Not a diagnostic tool or medical advice.
Longitudinal signal
Calibrated uncertainty
Patient partnership
Open standards
The gap
The system sees codes.
Patients live the years between them.
Many diagnostic models learn only from people who already have a diagnosis code. That leaves out the population that matters most here: people still moving through repeat visits, changing symptoms, contradictory explanations, and inconclusive tests.
The opportunity is to study the whole trajectory—not ask one visit to tell the entire story.
For clinicians
A busy encounter rarely reveals a multi-year pattern. Care teams need an early signal that is explainable and appropriately uncertain.
For health systems
Fragmented histories can lead to repeated, unfocused workups when the more useful evidence lives across time and settings.
“The question is not only what is happening today. It is what trajectory this person has been on.”
The proposed approach
Follow the signal across time.
The Invisible Patient Finder is conceived as a clinician-facing risk flag—not an autonomous diagnosis. It combines longitudinal pattern learning with calibrated uncertainty and a clear next step.
-
01
Read the journey
Organize symptom sequences, repeat visits, laboratory trends, and changing diagnoses into a longitudinal view.
-
02
Learn before the label
Use positive-unlabeled learning and anomaly detection to study what happened before known diagnoses—and look for related patterns in people without a label.
-
03
Calibrate the signal
Return an interpretable risk flag, a possible condition family, and an estimate of where the patient may be in the diagnostic journey.
-
04
Support the next question
When the record is sparse, surface high-information questions and possible next steps for a qualified care team to consider.
Initial areas of exploration
Long COVID
ME/CFS
Fibromyalgia
POTS
Lyme disease
Data foundations
Built to meet care where it already happens.
The proposed pipeline uses established health-data models and exchange standards so a useful signal can move into clinical workflows rather than remain in a research notebook.
OMOP
HL7 FHIR
USCDI+
Blue Button 2.0
Development is planned to begin with synthetic and public sources, including CMS DE-SynPUF and federal survey and surveillance data, before validation in governed research environments under the appropriate approvals.
What good looks like
Earlier recognition without false certainty.
Calibrated, not certain
Make uncertainty visible. Offer a signal for review, never a verdict.
Built with patients
Bring lived experience into product decisions, evaluation, and governance from the start.
Interoperable by design
Use reusable standards so the work can generalize across conditions and care settings.
Measured in the real world
Evaluate earlier recognition, false alerts, equity, workflow fit, and avoidable pre-diagnosis burden.
Built with patients,
not only for them.
About the initiative
Undiagnosed No More
The concept is led by Amir Asiaee, a computational and statistical researcher with experience in machine learning, biostatistics, health-data modeling, electronic health records, and claims data, affiliated with Vanderbilt University Medical Center.
The proposed next phase pairs technical development with a patient advocate who brings lived experience and a clinician collaborator who understands the realities of care delivery.
Current status
Early-stage research and MVP concept