Donor endothelial grading and specular images did not measurably improve graft-failure prediction over a clinical reference in the Cornea Preservation Time Study

Background Eye banks accept or discard donor corneas on endothelial criteria. Whether they predict postoperative outcomes beyond the clinical record and an independent cell density has not been tested with incremental-value statistics. Methods We analyzed 931 eyes with 69 all-cause graft failures from the public release of the Cornea Preservation Time Study, screened at eye-bank endothelial cell density (ECD) ≥ 2,300 cells/mm 2 . Five arms were compared on identical folds. The first was the eleven ordinal grades eye bank technicians record; the release carries no coefficient of variation or hexagonality. The other four were frozen representations of the donor image, three pretrained outside the modality, one in-domain. Each was appended to a clinical reference of covariates, ECD, diagnosis and site. The criterion was a 95% cluster-bootstrap interval on the paired difference in concordance excluding zero. Results The clinical reference reached C = 0.686. The grading did not measurably improve discrimination in any of six strata, with intervals about half the width of the learned arms’, and no representation improved on the reference. At 69 events only increments above a hazard ratio of roughly 2.7 per standard deviation could be excluded; smaller effects are unresolved. For six-month ECD on 770 eyes, the grading was worse than the reference (Δ R 2 −0.008, −0.012 to −0.004), three of four representations improved on it, up to +0.028 (+0.004 to +0.053), and a general pathology model matched the in-domain one. Three of eleven grades were absent in over 90% of eyes, so part of the null is restricted range. The same eye bank, re-imaging and recounting the same cornea a median of 5 days later, agreed with its first reading at r = 0.73 (limits of agreement about ±400 cells/mm 2 ), varying with method, cells counted and interval. Conclusion In this screened cohort neither the ordinal grading eye banks record nor any learned representation of the same image demonstrated incremental prognostic value over the clinical reference, though the image predicted the endothelium's near-term trajectory. These nulls are bounded by the variability of the deciding measurement, which argues for standardized, reproducible endothelial assessment before its prognostic value is judged.

Authors

Institutions

Publication Details

Journal
Frontiers in Medicine
Published
2026-11-17
DOI
https://doi.org/10.3389/fmed.2026.1993740
Primary Topic
Corneal surgery and disorders
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Donor endothelial grading and specular images did not measurably improve graft-failure prediction over a clinical reference in the Cornea Preservation Time Study

Benton Chuter, Monica M. Jablonski, William G. Gensheimer, Chelsea Joe
1 links
Frontiers in Medicine
Corneal surgery and disorders
article

Donor endothelial grading and specular images did not measurably improve graft-failure prediction over a clinical reference in the Cornea Preservation Time Study

Benton Chuter, Monica M. Jablonski, William G. Gensheimer, Chelsea Joe
article en

Abstract

Background Eye banks accept or discard donor corneas on endothelial criteria. Whether they predict postoperative outcomes beyond the clinical record and an independent cell density has not been tested with incremental-value statistics. Methods We analyzed 931 eyes with 69 all-cause graft failures from the public release of the Cornea Preservation Time Study, screened at eye-bank endothelial cell density (ECD) ≥ 2,300 cells/mm 2 . Five arms were compared on identical folds. The first was the eleven ordinal grades eye bank technicians record; the release carries no coefficient of variation or hexagonality. The other four were frozen representations of the donor image, three pretrained outside the modality, one in-domain. Each was appended to a clinical reference of covariates, ECD, diagnosis and site. The criterion was a 95% cluster-bootstrap interval on the paired difference in concordance excluding zero. Results The clinical reference reached C = 0.686. The grading did not measurably improve discrimination in any of six strata, with intervals about half the width of the learned arms’, and no representation improved on the reference. At 69 events only increments above a hazard ratio of roughly 2.7 per standard deviation could be excluded; smaller effects are unresolved. For six-month ECD on 770 eyes, the grading was worse than the reference (Δ R 2 −0.008, −0.012 to −0.004), three of four representations improved on it, up to +0.028 (+0.004 to +0.053), and a general pathology model matched the in-domain one. Three of eleven grades were absent in over 90% of eyes, so part of the null is restricted range. The same eye bank, re-imaging and recounting the same cornea a median of 5 days later, agreed with its first reading at r = 0.73 (limits of agreement about ±400 cells/mm 2 ), varying with method, cells counted and interval. Conclusion In this screened cohort neither the ordinal grading eye banks record nor any learned representation of the same image demonstrated incremental prognostic value over the clinical reference, though the image predicted the endothelium's near-term trajectory. These nulls are bounded by the variability of the deciding measurement, which argues for standardized, reproducible endothelial assessment before its prognostic value is judged.

Frontiers in MedicineVol. 13
Dartmouth College (US), Dartmouth–Hitchcock Medical Center (US), University of Tennessee Health Science Center (US), White River Junction VA Medical Center (US)
Good health and well-being
Corneal surgery and disorders
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon