Domain Atlas / Clinical decision support & deterioration alerting
OPTN eGFR Waiting-Time Correction
Explore this deployment in the PAN Lab ↗
In the PAN Lab, the readouts of this case's model organization carry a shaded evidence band whose width follows the least-established class among the modeling inputs the readings rest on.
The least-established input behind this case's model organization's readings is an assumption, not a measurement. Evidence base: 2 assumed · 15 published baseline.
The object corrected here is a clinical equation, not a learned system. The standard formulas for estimating glomerular filtration rate from a serum creatinine result — the 1999 MDRD equation and then the 2009 CKD-EPI creatinine equation — applied a coefficient that raised the estimated kidney function of any patient identified as Black; in the 2009 equation that coefficient was 1.159 (95 percent CI 1.144 to 1.170). The stated biological rationale, higher average muscle mass, treated race as a biological rather than a social category, and the MDRD equation behind it was derived from roughly 1,400 White and fewer than 200 Black patients. Inker and colleagues reported in the New England Journal of Medicine on 23 September 2021 that the race-including equation overestimated measured GFR in Black patients by a median of 3.7 mL/min/1.73m2, and concluded that race in eGFR equations is a social and not a biologic construct; OPTN's own patient materials state in a different register that the race variable automatically increased all Black patients' eGFR values, by as much as 16 percent. A kidney transplant candidate begins accruing waiting time when the estimate reaches 20 mL/min/1.73m2 or lower, so an estimate raised above that line delayed the clock; two studies cited in the nephrology commentary put the lost time at 1.3 and 1.9 years for affected Black candidates.[3]
What happened
The object at the centre of this case is not an AI system. It is an equation. For decades the standard formulas for estimating glomerular filtration rate from a serum creatinine result — the 1999 Modification of Diet in Renal Disease (MDRD) equation, then the 2009 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) creatinine equation — applied a coefficient that raised the estimated kidney function of any patient identified as Black. In the 2009 equation that coefficient was 1.159 (95% CI 1.144-1.170). The stated biological rationale, higher average muscle mass, treated race as a biological rather than a social category; the MDRD equation behind it had been derived from roughly 1,400 White and fewer than 200 Black patients. In 2021 a joint National Kidney Foundation and American Society of Nephrology task force recommended immediate adoption of a race-free equation, and Inker and colleagues published new race-free 2021 CKD-EPI equations in the New England Journal of Medicine, reporting that the race-including equation overestimated measured GFR in Black patients by a median of 3.7 mL/min/1.73m2 and concluding that race in these equations is a social and not a biologic construct. OPTN's own patient materials put the same point in a different register: the race variable automatically increased all Black patients' eGFR values, by as much as 16 percent. The two magnitudes come from different registers and are cited separately throughout this file.
What turned a measurement question into a governance one is a threshold. A kidney transplant candidate begins accruing waiting time when their estimated GFR reaches 20 mL/min/1.73m2 or lower. An estimate raised by a coefficient could sit above that line when the same creatinine value without it would have sat below — so the clock started late. Two studies cited in the nephrology literature put the lost time at 1.3 and 1.9 years for affected Black candidates. Accrued waiting time is a term in the ranking donor kidneys are offered down, so this was not an abstraction about a number in a chart; it was a position in an allocation queue.
The correction came in two layers and then a third. After a public comment period, the OPTN Board of Directors unanimously approved "Establish OPTN Requirement for Race-Neutral eGFR Calculations" on 27 June 2022, requiring every kidney transplant program to use a formula without a Black-race variable from 27 July 2022. The National Kidney Foundation called it an important first step, saying there is no place for race-based variables in evaluating organs offered through the allocation system. A prohibition, though, only runs forward. So on 5 December 2022 the Board unanimously approved the retroactive remedy, effective 5 January 2023: every kidney program had to assess its waiting list, identify Black candidates disadvantaged by race-inclusive eGFR, establish whether a race-neutral calculation would have qualified them sooner — documentation had to show the estimate was over 20 mL/min with the race coefficient and 20 or lower without it — and apply to OPTN to backdate that waiting time. Programs had until 3 January 2024 to complete their assessments, file every qualifying modification through UNet, notify every kidney candidate before and after assessment regardless of race, and attest to OPTN that they had done so. Programs that did not comply would be referred to the Membership and Professional Standards Committee. In practice, as the trade guidance written for the programs describes, this meant pulling a candidate list from OPTN's custom reporting tool, retrieving historical laboratory results from hospital records, external and reference laboratories, dialysis centers and record-retrieval systems, comparing the race-inclusive and race-neutral figures, and filing an eGFR Waiting Time Modification Form with supporting documentation.
The numbers came with snapshot dates, and they grew. The early monitoring report (published 4 December 2023, data through 5 July 2023) recorded more than 6,100 Black candidates with modified waiting times at a median of 1.7 years, 491 of whom had received a deceased-donor transplant and 15 a living-donor transplant; at that date only 12 of 232 active programs had attested. The one-year report (published 8 May 2024, data through 4 January 2024) recorded 14,701 waiting-time modifications processed, a median of 1.7 years with roughly half receiving between one and three years, 2,709 modified candidates transplanted from deceased donors and 158 from living donors — and all 230 active kidney programs attested. These are counts of modifications and registrations rather than of distinct people.
Then the outside measurement. Schold and colleagues published a national study in the Journal of the American Society of Nephrology on 6 May 2025: of 44,912 Black candidate kidney waitlist registrations assessed, 32 percent (14,419) received a modification, worth a median of about 610 priority days, and modified candidates had an adjusted hazard ratio of 2.85 (95% CI 2.7-3.02) for deceased-donor transplantation against non-modified candidates. The study also found that modification rates varied significantly by candidate characteristics and by transplant center. The investigators' reading of that variability was that there was still mixed use of the policies and that the requirement had not been articulated clearly at the beginning. The Membership and Professional Standards Committee, having observed programs implementing the requirements in various ways, referred a follow-on project to the Minority Affairs Committee. The Board approved "Monitor Ongoing eGFR Modification Policy Requirements" in June 2025, effective 10 September 2025: programs must now maintain written protocols and document compliance for confirming a candidate's race, for fulfilling notification requirements, and for seeking supporting documentation, naming at minimum which sources will be reviewed; notification duties apply to candidates registered on or after 4 January 2024; and every registered kidney candidate must be assessed for eligibility. Programs have until 11 September 2026 to complete the strengthened requirements. The peer-reviewed commentary on the programme is warm and unillusioned at once: it uses the language of restorative justice, and it records the critique that the policy has been called both unfairly too broad and too narrow, addressing one input and one population while leaving open the question of whether all patients should accrue predialysis waiting time.
The sociotechnical reading
Nearly every deployment in this atlas is read at the moment a correction channel fails. This one is carried because a correction channel worked, and reading it honestly means holding two things at once: what the record establishes was repaired, and what the same record establishes about the repair's shape.
Start with why the defect was invisible from inside. The estimating equation is not a learned system that could drift, be retrained, or be audited against a held-out sample. It is a published formula with fixed coefficients, running identically in thousands of independent laboratories. Every laboratory's quality process checks whether the arithmetic was performed correctly on the specimen in front of it, and the arithmetic was correct every time. A bias living inside the formula itself is invisible to every check that trusts the formula — which is why no local fix was available and why the correction had to arrive as a professional standard converted into an allocation rule. That is also the reason the monoculture on this diagram is drawn at full strength: one formula, one direction of adjustment, every patient it touched, nationwide, for over a decade.
The remedy the record documents is genuinely unusual, and its unusual part is the backward-looking half. Removing the coefficient was the recommendation of a professional task force and the subject of a board vote; that step, on its own, would have left every candidate whose clock had already run late holding a late qualifying date. What the OPTN did next was order its own registry state recomputed: retrieve the history, apply a formula without the coefficient, and write the corrected qualifying date into UNet, where it re-ranks the candidate in the live deceased-donor offer sequence. The measured allocation effect — an adjusted hazard ratio of 2.85 for deceased-donor transplantation among modified candidates — is what makes the point that the correction was not advisory. The map draws this as structure: the registry drives the offer sequence, and the corrected value has no path into the registry except through a person.
That last absence is where the honest complications live, and the record supplies them rather than the atlas inventing them. First, the undo ran by hand. Roughly 230 autonomous programs each executed a uniform rule against messy historical laboratory data spread across hospitals, reference laboratories and dialysis centers, with wide discretion over which sources to pull and how to confirm a candidate's recorded race — and the independent national evaluation measured the result: modification rates that varied significantly by transplant center, which the investigators read as mixed use of a requirement that was not clearly articulated at the outset. A programme's own monitoring reports could count how much restitution had been made; they could not see that it was being made unevenly, and it took an outside study with a denominator to establish that. Second, the correction outran its own checking at the far end: the trade guidance for the audit describes candidates receiving a lot of time back, moving to the top of the list and beginning to receive offers, while centers worked to ensure those candidates had been re-evaluated recently. Nothing settles a changed rank against the record before offers issue on it, and that is drawn on the diagram rather than argued in prose. Third, the correction's scope is a scope, not a completeness. It applies to registered kidney candidates on programs' waiting lists; it can only advance a qualifying date, never delay one; and the peer-reviewed commentary records the standing critique that it addresses one input and one population, leaving open whether all patients should accrue predialysis waiting time. And fourth, the strengthening that answered the variance finding is in force but not finished: programs have until 11 September 2026 to complete the written-protocol requirements, so the record as it stands describes a rhythm that has started.
What the Lab reads here is institutional, and only institutional. Transplant candidates are the subjects of this correction and appear nowhere in the dynamics: no waiting time, qualifying date, offer or transplant outcome for any person is computed on the diagram. The published figures about candidates are recorded external observations, each carrying its own denominator and data-snapshot date — more than 6,100 modifications at six months, 14,701 at one year, 14,419 of 44,912 assessed registrations in the peer-reviewed count, the transplant subtotals, the hazard ratio — and they belong in this file, not on the network. What the network carries is the shape of the thing: a governance body that ordered its own registry to recompute an input's historical effect, measured the result publicly, and then tightened on what the measurement showed.
The concepts used in this reading are defined in the Field Guide; the governance responses live in the Practice Library. The model organization for this case can be stress-tested in the PAN Lab.