Severe graft dysfunction is a cliff, not a slope — and the fall happens in year one.
Grading primary graft dysfunction by severity rather than presence changes what it predicts. Moderate PGD looks like no PGD at every horizon. Severe PGD costs about eleven points of survival, all of it lost in the first year.
Source: OPTN/UNOS STAR, Dec 2025 release · 39,821 bilateral lung recipients, 13,856 with a derivable 72-hour grade. Follow-up landmarked at day 3.
2.51×hazard in year one, severe PGD
Adjusted, severe PGD carries a 2.51× hazard of death through the first year — then 1.12× from one to five years, and 1.12× beyond five, no longer significant. Moderate PGD is null throughout.
01 · The threshold
Moderate PGD is indistinguishable from none
ISHLT grades PGD 0–3 on gas exchange and radiographic infiltrates. The registry records no radiographs, so grades 0 and 1 are pooled; grade 2 is a P/F ratio of 200–300, grade 3 is below 200 or ECMO at 72 hours.
Survival by grade (%)
no PGDmoderatesevere
Horizon
No PGD (0–1)
Moderate (2)
Severe (3)
Severe − no PGD
30-day
98.9%
98.7%
92.7%
-6.2 pp
1-year
90.4%
89.8%
78.2%
-12.3 pp
5-year
63.1%
62.7%
52.2%
-10.9 pp
10-year
37.0%
38.8%
25.2%
-11.7 pp
Grade 2 carries no survival penalty. The gap against no PGD never exceeds one percentage point at any horizon, and at ten years moderate is nominally higher. Whatever the middle grade measures, it does not predict death. The clinically meaningful boundary is grade 3.
02 · Timing
One hazard ratio would be the wrong answer
Proportional hazards is decisively violated for severe PGD (χ² = 145). A model forced to report a single ratio averages a large early effect against a null late one and describes neither.
Hazard ratio by follow-up window (1.0 = no difference)
This resolves an apparent contradiction. The absolute survival gap is ~11 points at one year and still ~11 points at ten — yet the hazard ratio collapses after year one. Both are true: the deficit opens early, and thereafter the two groups die at nearly the same rate, so the gap persists without widening. Severe PGD is a one-time loss, not an ongoing penalty.
03 · Against the literature
We do not reproduce the published null
A 2024 analysis of the same registry, stratified the same way, reports no independent association after adjustment (HR 0.972). Adjusting for baseline characteristics alone, we get 1.522; adding their post-transplant covariates attenuates it to 1.215 — still elevated.
Adjusted hazard ratio, severe PGD
Two reasons to distrust the adjusted comparison. Treated rejection and post-transplant dialysis are measured after the 72-hour assessment, so conditioning on them blocks the causal path rather than removing confounding — attenuation follows by construction.
Worse, their missing categories behave like death indicators: unknown rejection status carries a hazard ratio of 11.4 and unknown dialysis status 4.5, because status goes unrecorded largely when the patient dies before it can be recorded. A covariate that encodes the outcome will absorb it.
04 · Is the graded subset representative?
Missingness here is era, not severity
Only 34.8% of recipients carry a derivable grade, which invites the suspicion that grading is available preferentially for sicker patients. It is not — the 72-hour fields are simply not collected before 2015.
Transplant era
Share with a derivable grade
1987-2004
0.0%
2005-2014
1.6%
2015-2019
57.6%
2020-2025
62.1%
The comparison itself is era-balanced. Graded and ungraded recipients have near-identical survival (86.3% vs 87.6% at one year), and all three grades share a median transplant year of 2021. That check is not routine here — era confounding has already dissolved two other findings in this series.
Related.Severe PGD@72h Landscape sizes the endpoint and its offer-time drivers; The Long-Horizon Survival Landscape covers the label layer this analysis draws on. Both predate the grading and the day-3 landmark used here, so their survival figures are superseded by this brief rather than in conflict with it.