A donor with a linked chest CT was transplanted 91.5% of the time — versus 15.2% without one. The imaging set we train on is overwhelmingly a record of lungs that were used. The discarded arm — the whole basis for asking “could this lung have been used?” — is just 790 donors.
Source: OPTN/UNOS STAR file, Dec 2025 release · DECEASED_DONOR_DATA lung dispositions (DONDISP) intersected with the DICOM→STAR donor crosswalk (9,258 donors).
91.5%of CT-linked donors → transplanted
CT presence is nearly a proxy for use. That makes the imaging cohort excellent for modeling accepted lungs — but it means the accept-vs-decline frontier is represented by a thin, selected sliver. Read this brief alongside The Discarded-Lung CT Cohort, which studies that sliver up close.
01 · The selection
Having a CT and getting transplanted are almost the same event
Among the 9,258 donors in the DICOM→STAR crosswalk, 91.5% had at least one lung transplanted. Among the ~307,000 deceased donors without a linked CT, only 15.2% did. A chest CT enters this crosswalk because the donor was worked up for lung donation and proceeded — so the imaging set inherits that survivorship.
Lung-transplant rate by CT linkage (% of donors)
Donor counted “transplanted” if any lung disposition = 6 (Transplanted). The 6× gap is selection, not a lung-quality difference the CT reveals.
02 · The study set that remains
From 9,258 imaged donors to a few hundred adjudicable declines
9,258
CT-linked donors · 51,238 studies
8,468
Transplanted — ≥1 lung used (91.5%)
790
No lung transplanted (8.5%) — the discard arm
~182
Image-assessable declines within it
The 790-donor arm is smaller still once you subtract lungs that were not clinical failures: a share were recovered for research or exported and transplanted outside the U.S. (see the companion brief). What’s left for the “could it have been used?” question is a few hundred donors — and only a fraction were declined for a reason a CT can adjudicate.
03 · Who the declines are
The unused lungs differ mostly on things a CT doesn’t see
Compared with the transplanted arm, the not-used donors are older and far more often DCD — axes driven by physiology and procurement type, not chest-CT parenchyma. Age and smoking do leave CT traces (emphysema); DCD status, the largest gap, does not.
Donor profile — not-used vs transplanted Not used Transplanted
Median donor age: 41.5 (not used) vs 36.0 (transplanted). DCD = donation after circulatory death (NON_HRT_DON).
04 · Why they were declined
The largest single reason isn’t a picture — it’s a number
Discard reasons for the 790-donor arm (LUB_DISCARD_CD; 663 coded). Distinctly image-assessable findings — diseased organ, anatomical abnormality, trauma, infection — are highlighted. But the plurality, “poor organ function,” is a gas-exchange judgment (donor P/F) that a CT does not directly measure.
Discard reason — not-used CT donors (count)
■ image-assessable · “poor organ function” and “other” shown neutral — physiologic or uncoded, not a CT finding.
Net scope. Of ~663 coded declines, roughly 182 turn on a finding a CT could adjudicate; ~234 more turn on gas exchange the donor ABG already captures. So the imaging cohort can, today, validate the accept/decline call on only a few hundred lungs — and those are older, DCD-heavy, marginal donors, not the broad discard pool. This is a real study, but a narrow one: it answers “can CT add at the margin for fully-worked-up declines?”, not “can CT unlock the discard pool?”
05 · What would widen it
The bottleneck is linkage, not modeling
The discard arm is thin because the crosswalk over-represents transplanted donors, not because discarded lungs lack CTs. Nationally, most declined lungs never enter this crosswalk. Growing the answerable cohort for pool-expansion is therefore a data-linkage task — pulling and crosswalking the chest CTs of donors whose lungs were declined — before it is a modeling task. Until then, models train on accepted lungs and can be honestly validated against only the ~182 image-assessable declines above.