The allocation score has a survival model inside it. It works — modestly, and we can only see three years of it.
Asking whether the Composite Allocation Score predicts survival is a category error: it is a priority ranking, and a sicker candidate ranks higher. But one component exists to predict post-transplant benefit. Tested on its own, it does.
Source: OPTN/UNOS STAR, Dec 2025 release · 9,323 lung and heart-lung transplants with a CAS, 2023–2025. 4,379 have a full year of follow-up; none has five.
25.2% → 12.0%1-year mortality, lowest to highest post-transplant subscore
A 2.1-fold difference across quartiles of the one component built to predict post-transplant outcome. C-index 0.579 on graft survival.
01 · The caveat that comes first
Three years of follow-up, testing a five-year claim
Continuous distribution for lung began in March 2023. That is the whole reason this is an interim read rather than a verdict.
The component under test targets five-year benefit; we can observe one year. Of 9,323 transplants with a CAS, 4,379 have completed a year and none has reached five. What follows measures the early portion of what the subscore claims to predict, on the earliest cohort to receive it. It is informative about direction and ordering. It is not a verdict on the instrument, and it cannot be — the data to deliver one will not exist until roughly 2028.
02 · The component that works
Post-transplant subscore: modest, and in the right direction
One-year mortality falls monotonically across its quartiles.
1-year mortality by post-transplant subscore quartile (%)
Quartile
n
Median subscore
1-year mortality
Severe PGD
Q1 (lowest)
2,340
17.73
25.2%
37.5%
Q2
2,335
19.11
17.7%
31.0%
Q3
2,339
19.92
17.0%
30.3%
Q4 (highest)
2,309
20.77
12.0%
32.0%
C-index 0.579 on graft survival across 8,683 transplants; AUROC 0.597 for surviving one year. Modest by the standards of a clinical prediction model — but this is a subscore contributing part of a priority ranking, not a prognostic tool, and the gradient it produces is clinically legible.
03 · Why the total runs backwards
A C-index below 0.5 is the score working, not failing
C-index for graft survival, by component
0.5 is no discrimination. Below 0.5 means a higher score predicts worse survival.
Total CAS scores 0.455 and waitlist urgency 0.443. Both sit below 0.5, and both should. Urgency exists to predict death without a transplant; a candidate who scores high on it is sicker, and sicker recipients do worse afterwards. That is sickest-first allocation functioning as designed.
Reporting either number as evidence that "CAS fails to predict survival" would misrepresent what the score is for. Only the post-transplant subscore makes a prognostic claim, and only it should be judged on one.
04 · What none of it predicts
No component predicts primary graft dysfunction
Across all five components, discrimination for severe PGD at 72 hours falls between 0.470 and 0.553 — no better than chance, on 5,263 assessable recipients.
Coherent with the mechanism, not disappointing. CAS describes the recipient and the offer: urgency, blood group, height, sensitisation, distance. Severe PGD is ischemia-reperfusion injury of the donor organ. Companion briefs in this series reach the same place from other directions — HLA antigen mismatch does not predict PGD either, and neither does preservation solution. The consistent implication is that PGD risk lives in donor and preservation variables rather than recipient priority, which is where an imaging-based donor assessment would have to earn its value.
One instance of a pattern. Across this series of briefs, five clinically decisive quantities turned out not to be collected at all: implant sequence in multi-organ procedures; HLA typing at allele resolution; the lung perfusion device; and molecular rejection surveillance. Each is recorded somewhere in the transplant system — by a centre, a laboratory, a manufacturer, or OPTN itself — and none reaches the research release. The registry records the organ’s disposition but not its journey, the graft’s dysfunction grade but not the operative sequence that produced it. Distinguishing a question that is underpowered from one that is unanswerable in principle is the discipline that separates the two.