Can AI decide when a donor kidney should leave the waitlist sequence? A US preprint models the trade-off
An interpretable system resolved 93.5% of historical routing decisions and flagged many discarded kidneys for expedited placement. It was a retrospective simulation—not a transplant trial.
The short version
- What happened
- Researchers trained and evaluated models on 143,821 US deceased-donor kidneys from 2016–2019 and 2023–June 2025, keeping kidneys from the same donor in one data split.
- Who it affects
- For people waiting for a kidney, faster placement could expand opportunities—but out-of-sequence routing can also bypass candidates with higher priority. This study models that tension without showing who would actually receive an organ or whether outcomes would improve.
- What remains unproven
- Unreviewed preprint
Keep in mind
Unreviewed preprint
Retrospective US registry modelling
143,821 kidneys
21,574-organ held-out test
Acceptance within 100 offers as proxy outcome
No prospective deployment
No transplant or graft-survival outcome
No observed equity outcome
Policy-sensitive cost setting
Exploratory AI literature audit without pre-specified protocol
Source analysis · 1 linked source. Primary analysed source date: 28 October 2026.
Keep following the evidence
Join The Impact Brief founding list for selected AI reports, their effect on people and the evidence limits. Read a sample first.
Read the full report · 4 min The evidence, methods, implications and limitations
The model tries to protect both scarce organs and waitlist order
A new US preprint proposes an AI system that recommends keeping a deceased-donor kidney in the standard waitlist sequence, moving it to expedited out-of-sequence placement, or deferring the decision when its evidence is uncertain. The problem is real and ethically difficult: moving too slowly can leave an organ unused, while bypassing the sequence can deny higher-priority candidates a fair opportunity.
The researchers linked Organ Procurement and Transplantation Network match-run records to donor and waitlist registries for 2016–2019 and 2023 through June 2025, excluding pandemic-disrupted years and several non-standard allocation pathways. The final dataset contained 143,821 kidneys: 100,674 for training, 21,573 for validation and 21,574 for held-out testing. Kidneys from the same donor stayed in the same split. The prediction target was acceptance within the first 100 offers—not transplantation success.[1]
Interpretable rules covered most test cases, with a measurable cost
Two interpretable models produced clinical-feature rules and 99% uncertainty intervals. The Arbitrator acted only when at least one model certified the same preferred route and neither contradicted it. On the test set, both models agreed for 17,121 kidneys; one model certified another 3,060. That resolved 20,181 decisions, or 93.5%, while 1,393 were deferred. In the experiment, deferred cases went to a black-box XGBoost model; real use would require an accountable human process.
Interpretability did not come free. At the authors' main cost setting, the combined Arbitrator-plus-XGBoost pipeline had 0.802 accuracy and a 0.143 average decision cost, versus 0.815 and 0.131 for XGBoost alone. The cost setting valued preventing nonuse twice as strongly as preserving the standard sequence in an individual trade-off. That is a policy choice, not a clinical fact, and different values change which kidneys are routed out of sequence.[1]
The distinction that matters
What the held-out test actually measured
The system predicted whether a kidney would be accepted within 100 standard offers. It did not predict graft survival or prove that expedited placement would lead to transplantation.
Held-out kidneys
21,574
14,127 accepted within 100 offers; 7,447 not accepted
Rule-backed decision
20,181 kidneys
93.5% received an interpretable on- or out-of-sequence recommendation
Deferred
1,393 kidneys
6.5% lacked a non-conflicting certified rule
Discarded lower-risk subset
1,920 of 2,738 flagged
Retrospective candidate count—not observed transplants
The headline number is a projection, not kidneys saved
Among 2,738 historically discarded test kidneys with Kidney Donor Profile Index scores below 85%, the system retrospectively recommended expedited placement for 1,920, or 70.1%. The authors extrapolate that result to suggest 43.9% of all 8,418 kidneys recovered but unused in 2023—about 3,700—might have been candidates for faster placement. The model did not show that those kidneys would be accepted, transplanted successfully or reach the patients who would benefit most.
The work is unreviewed and retrospective. Historical acceptance can encode centre behaviour and inequity; acceptance within 100 offers is only a proxy for difficult placement. The agentic audit of 113 rules was exploratory: an AI selected literature without pre-specified search terms or study criteria, and clinical validation remains future work. No patient outcomes, graft survival, implementation delays or distributional fairness effects were measured. Prospective evaluation should compare routing decisions, transplant outcomes and who gains or loses before any automation enters allocation practice. The authors declared no competing interests and reported MIT, NIH and NSF support.[1]
Watch the background, then check the evidence
Illustrative photo · not a video frame
The Royal Society · 44 min 28 sec · Background explainer
Is this the AI future we want? | The Royal Society
YouTube loads only when you choose; playback starts when you press play.
A Royal Society discussion with Madhumita Murgia and biomedical engineer Alison Noble on AI’s potential in healthcare, science and culture, recorded at the 30 June 2026 event.
Evidence check: A discussion of opportunities and risks, including future possibilities. It is not evidence that every application described is available or clinically proven.
Published 2026-07-01. Source checked 14 October 2026. Check captions on the publisher’s video page.
Check the healthcare evidencePhoto: Daniel Sone / National Cancer Institute / Unsplash · Licence
Continue the story
Read next
Continue exploring clinical decision support: compare the finding and its evidence limits.
Can an interpretable AI spot small liver cancers on CT? External testing is promising—but incomplete
A peer-reviewed multicentre study tested dLI-RADS on 133 external patients. Sensitivity reached 93%, versus 78% for radiologist-assigned LI-RADS, but only sensitivity could be compared externally and every lesion was manually segmented.
Read the next reportExplore the full source and evidence record
What this means for people
- For people waiting for a kidney, faster placement could expand opportunities—but out-of-sequence routing can also bypass candidates with higher priority. This study models that tension without showing who would actually receive an organ or whether outcomes would improve.
Global context
Organ-allocation rules, waiting lists and expedited-placement authority differ between countries. A model trained on US historical offers could reproduce local acceptance patterns and should not be transferred to another allocation system without policy, clinical and equity validation.
What the evidence does not yet show
- Unreviewed preprint
- Retrospective US registry modelling
- 143,821 kidneys
- 21,574-organ held-out test
- Acceptance within 100 offers as proxy outcome
- No prospective deployment
- No transplant or graft-survival outcome
- No observed equity outcome
- Policy-sensitive cost setting
- Exploratory AI literature audit without pre-specified protocol
What to watch next
- Independent peer review
- Prospective silent evaluation
- Observed transplant and graft-survival outcomes
- Equity by patient and centre characteristics
- Sensitivity to the policy cost setting
- Human review of deferred decisions
- Pre-specified clinical audit protocol
- External validation in later years and other allocation systems
Living evidence record
Impact record IAI-0Z5C49V
Evidence stage
Studied
Confidence
Supported
Reporting basis
Source analysis
Independent or research support
Present
Record status
Monitoring
Last checked
31 October 2026
Source trail
1 direct source across 1 source type.
People impact
Documented in this record.
Uncertainty
Limits and next checks are explicit.
Stages describe the evidence available—not whether a technology is good or bad. See the public method.
Single-source reporting disclosure
This record analyses one direct source. It can establish what arXiv published or reported, but it is not independent corroboration of every performance claim or predicted outcome. The confidence label will change only when broader evidence is added.
Evidence trail
Sources used for this report
Links checked 31 October 2026
Research question, topics and regions
Uncertainty-aware AI for deceased-donor kidney allocation
Topics: Kidney transplantation · Organ allocation · Clinical decision support · Uncertainty · Fairness · Interpretable AI
Regions: United States · North America
This report is labelled source analysis. We summarise and analyse source material in our own words; company statements remain attributed claims until independently supported. Translated summaries preserve the meaning of the original source and link back to it. Read our editorial standards.
Reader commentary
What does this mean where you are?
Tell us what you have seen at work, in your community or in everyday life. A question is welcome too. No account is needed; we review notes before publishing them.
Explore commentary across the portal →Published reader notes
0No published reader notes yet. You can start the evidence-led discussion above.
Prefer a private correction or response? Contact the newsroom.