OrganAlert · Lung Donor Evaluation · Decision Support
Clinical decision support · Lung
The donor read,
with the work shown.
Upload a DonorNet packet. OrganAlert turns it into a structured record and runs four published risk scores in code. Every finding shows the source line it came from, and a clinician signs the final read.
OrganAlert
Lung donor evaluation
P/F trend · 18 h
↑ 312 → 512
Extracted record
Oto suitability
1 / 18
Abstract
Extraction is AI-assisted. Scoring is not. The four risk scores run in code, so the same packet always produces the same numbers. The final read belongs to a clinician, who signs it before it informs any decision.
It is decision support, not a diagnostic device.
§ Methods
Four steps, one auditable path
Upload the DonorNet packet
Encrypted on arrival; it stays inside the BAA-covered backend.
AI extraction + vision verification
Every field is cross-checked against the source pages before it enters the record.
Deterministic scoring & flags
The four published models run in code. No model output is allowed to produce a score.
Reviewed, co-signed report
A named clinician verifies and signs before it informs any decision.
§ Evidence · Four published scores
Established models, computed in code
Every score in a report comes from a published, peer-reviewed donor-lung model. We do not invent a risk number of our own. Open any score below to see the arithmetic for the specimen donor.
Computed in code — Oto Score
Computed in code — Eurotransplant Donor Score
Computed in code — ODSS
Computed in code — LUNDON
LUNDON — Not scored · not validated for DCD donors
When a model is asked to score a donor outside its validated range, the report says so instead of printing a number. Point tables come straight from the cited papers. Thresholds are configured per program and await faculty sign-off.
References
- 1.Oto T, et al. Ann Thorac Surg 2007;83:257.
- 2.Smits JM, et al. Transpl Int 2011;24:393.
- 3.Whited WM, et al. Ann Thorac Surg 2019;107:425.
- 4.Heiden BT, et al. Am J Transplant 2023;23:540.
§ Transparency
Every finding cites its source
Findings appear as a ledger. Each row names the rule that fired, what it found, how serious it is, and where in the packet it came from. If a finding cannot cite its source, it does not go in the report.
Table 1. Illustrative flag ledger — PHI-free specimen donor.
§14 · Flag ledger
IllustrativeRule
Finding
Severity
Source
- GAS-02Best P/F 512 on FiO₂ 1.0ReassuringSerial ABG
- IMG-01Serial CXR clear — no infiltrateReassuringChest X-ray
- LAB-01ABG trajectory improving over 18 hReassuringGas exchange
- SERO-01HIV / HBV / HCV non-reactiveReassuringSerology
- HX-0124 pack-year smoking historyReviewSocial hx
Example rows for illustration only — not real donor data. This donor reads reassuring; thresholds and labels are configured per program.
§ Security by design
PHI protection in the foundation
This marketing site never touches PHI. Clinical data lives only in the encrypted, BAA-covered backend, and the constraints below shaped how that backend was built.
PHI encryption at rest
Application-layer AES-256-GCM with per-tenant keys. The database never stores plaintext.
Append-only audit
Access and generation events recorded; IP addresses pseudonymized.
Tenant isolation
Per-request org scoping with Postgres row-level security.
Shift-bound sessions
Idle and absolute session limits enforced on every request.
BAA-covered AI path
PHI reaches a model only through AWS Bedrock, under our signed BAA. No third-party AI APIs.
Partner with us
Bring OrganAlert to your program
We work directly with each lung transplant program through onboarding. Tell us how your team evaluates offers today, and we will walk through the report on a packet of yours.
AI-assisted decision support · not a diagnostic device · verify all findings