Survival is the abundant, long-horizon outcome — where PGD@72h is scarce.
Nearly every recipient in the registry has survival follow-up, out to a decade. That makes 1-, 3-, 5-, and 10-year survival the outcomes with enough labels to train on — the long-horizon complement to the sparse, recent PGD@72h window.
Source: SRTR-linked 2024 lung-transplant extract (adult + pediatric, 1987–2024, cleaned) · survival via SSA/CMS + OPTN death linkage. Companion to the STAR-based PGD@72h landscape.
50,247recipients with survival follow-up (99.8%)
Follow-up is near-complete and long. 29,307 deaths are observed, median follow-up 3.4 years, and 32,237 recipients have a full 10 years of potential follow-up — a deep outcome layer no other endpoint matches.
01 · Label coverage
How many recipients have assessable survival at each horizon
"Assessable" means followed at least that long, or died before it — the honest denominator for a survival model at that horizon. A 2023 transplant simply cannot yet have 5-year data, so coverage tapers with distance.
46,254
1-year assessable (92%)
41,275
3-year assessable (82%)
37,575
5-year assessable (75%)
32,237
10-year assessable (64%)
Recipients with assessable survival, by horizon
02 · Crude survival
Survival by horizon — pooled and unadjusted
Crude survival among the assessable at each horizon. This is a complete-follow-up proportion, not Kaplan–Meier, and it pools 37 years of transplants — the long horizons are dominated by older, worse-surviving eras. Read the trend, not the exact 10-year figure.
Crude survival (% of assessable)
1-yr 84% · 3-yr 66% · 5-yr 50% · 10-yr 20%. The 10-year value is depressed by era pooling — see the era split below.
03 · The honest signal
Survival improved sharply across allocation eras
Splitting by era removes the pooling artifact. One-year survival rose from 77% (pre-LAS) to 87% (LAS); five-year is roughly flat (48%→51%), consistent with early gains that don't fully carry to the long term. The CAS era (2023+) is excluded — too few recipients have reached even one year.
Crude survival by allocation era pre-LAS (<2005) LAS (2005–22)
04 · Why these labels are trustworthy
SRTR death linkage catches deaths OPTN follow-up misses
988 recorded deaths have a Social Security / CMS death date but no OPTN follow-up death record — patients the transplant program lost track of, recovered by SRTR's death-registry linkage. Because loss-to-follow-up is not random, a survival model trained on OPTN follow-up alone would be biased; these labels are the more complete ones. This is the outcome layer we should anchor the models to.
05 · What survival reveals about PGD@72h
Severe PGD@72h front-loads death — and the gap persists to 5 years
Linking the 10,455 bilateral-lung recipients who have both a PGD@72h assessment (STAR) and SRTR-linked survival: severe PGD@72h — 32% of the cohort — roughly sextuples 30-day mortality (7.1% vs 1.1%), and its survivors stay behind through five years. SRTR death linkage catches deaths OPTN follow-up missed, sharpening the 3–5-year gap by a few points.
Survival by PGD@72h status (% of assessable) Severe PGD@72h No severe PGD
Perioperative (≤30-day) death: 7.1% with severe PGD vs 1.1% without. 10-year is omitted — the PGD@72h label only begins ~2015, so no reliable decade of follow-up exists yet, even with SRTR linkage.
This is why PGD@72h is the model target. The mortality it carries is front-loaded into the first 30 days — exactly the window a calibrated, offer-time risk estimate could act on. Beyond ~1 year the conditional death timing of the two groups converges: PGD raises mortality early, not uniformly.