How a Single eGFR Change Is Exposing the Clinical Trial Industry’s Embedded Patient Bias Problem

A single algorithmic correction — removing one variable from one equation — is now demonstrably responsible for thousands of Black patients gaining access to kidney transplants they were previously denied. If that doesn’t reframe how the clinical trial industry thinks about the tools embedded in its own infrastructure, nothing will.

According to STAT News, removing race as a variable from the estimated glomerular filtration rate (eGFR) calculation — the standard measure used to assess kidney function — produced a 5.3 increase in transplants per 1,000 Black candidates. That is a structural correction to a system that was actively and measurably disadvantaging patients based on a biological assumption with no sound scientific foundation. The race-adjusted eGFR had been artificially inflating kidney function estimates for Black patients, making them appear healthier than they were, and therefore less urgently in need of a transplant. The algorithm wasn’t neutral. It was a gatekeeping mechanism dressed in clinical clothing.

The implications for clinical trials are the story. Because the same class of algorithmic thinking — race-adjusted, historically derived, assumption-embedded — is woven throughout the trial ecosystem. It lives in the eligibility criteria. It lives in biomarker stratification. It lives in the composite endpoints that define whether a drug works. And almost nobody is auditing it.

 

The Problem Is Structural, Not Incidental

The eGFR case is not an isolated failure. The Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) equation, which incorporated race as a correction factor, was developed from population-level data that conflated socioeconomic, environmental, and historical health disparities into a single biological variable labeled “race.” The logic was circular: Black patients in the dataset had measurably different outcomes, therefore the model assumed a biological difference, therefore the model encoded that difference into future clinical decisions. What looked like precision medicine was actually a feedback loop built on structural inequality.

This same logic pattern appears in clinical trial design with troubling regularity. Consider the spirometry equations used in pulmonary function testing, which also apply race-based correction factors and which influence enrollment eligibility in respiratory trials. Or consider cardiovascular risk calculators — the Pooled Cohort Equations, for instance — which use race as a direct input and which are embedded in protocol-level screening criteria for trials evaluating statins, antihypertensives, and novel lipid-lowering agents. 

That sequencing is the problem. The FDA’s guidance on enhancing diversity in clinical trials, most recently formalized in the 2024 Diversity Action Plan requirements under the omnibus legislation, focuses almost entirely on recruitment strategy and demographic representation targets. What it does not do is require sponsors to audit the clinical algorithms embedded in their eligibility criteria for race-based assumptions before those criteria are locked into an IND. The FDA is asking sponsors to recruit more diverse populations into trials that may still be screening those populations out through biased tools. It is not a diversity strategy, but a cosmetic fix applied to a structural wound.

What the Trial Record Actually Shows

The kidney transplant finding isn’t the first signal. It’s the loudest one so far. Research published through the National Lung, Heart, and Blood Institute’s spirometry normalization work has documented how race-based lung function equations systematically exclude Black patients from COPD and asthma trials by setting higher normal thresholds. The practical consequence: a Black patient with clinically significant airflow obstruction may not meet the inclusion FEV1 criteria in a trial that would directly benefit them, because the reference equation assumes their lungs work differently.

Biogen’s aducanumab program (Aduhelm) is instructive from a different angle. The EMERGE and ENGAGE trials (NCT02484547 and NCT02477800) enrolled populations that were over 75% white — a widely criticized disparity in a disease area with documented differences in Alzheimer’s risk and presentation across racial groups. The enrollment failure wasn’t just a recruitment problem. It reflected eligibility criteria built around biomarker thresholds derived from predominantly white research cohorts. When the tools you use to confirm eligibility are calibrated on one population, you will systematically enroll that population. The algorithm is upstream of the recruiter.

The Reframe the Industry Is Avoiding

The conventional framing of algorithmic bias in clinical trials treats it as a diversity and inclusion issue — a downstream problem to be addressed by community outreach, site placement near minority-serving institutions, and culturally competent recruitment materials. That framing, while not wrong, is dangerously incomplete. Algorithmic bias in trial infrastructure is a data integrity issue. It is a scientific validity issue. And it is, increasingly, a regulatory risk issue.

Here is the strategic reality that sponsors are not yet internalizing: if a pivotal trial’s eligibility criteria are built on race-adjusted algorithms, and those algorithms are later demonstrated to be scientifically invalid — as the eGFR correction has now been — the FDA has a legitimate basis to question whether the enrolled population accurately represents the intended use population. The FDA’s own guidance, Considerations for the Use of Real-World Data and Real World Evidence to Support Regulatory Decision-Making for Drug and Biological Products, explicitly requires that data inputs into clinical decision frameworks be evaluated for systematic bias before those frameworks are used in regulatory submissions. The same logic should — and eventually will — apply to the embedded algorithms that define who gets into a trial and what counts as an outcome.

It’s about the fact that a drug approved on data from a biased enrollment process may not work the way the label says it works in the populations that will actually receive it. Post-market surprises of this type are expensive, sometimes fatal, and entirely preventable if the bias is caught upstream. The eGFR story proves that catching it is possible. The kidney transplant data is real-world evidence in the most literal sense: a natural experiment showing what happens when you remove a biased assumption from a clinical decision algorithm. Trials should be generating that kind of corrective evidence, not perpetuating the errors.

What Must Actually Change

The FDA needs to close the gap between its diversity action plan framework and its protocol review process. Specifically, the agency should require that any eligibility criterion, stratification variable, or primary endpoint calculation that incorporates a race-based correction factor be accompanied by a scientific justification in the IND submission — not a diversity statement, but a mechanistic defense of why the biological assumption embedded in that algorithm is valid for the intended study population. If sponsors cannot provide that justification, the criterion should be replaced with a race-neutral equivalent before the trial begins. The tools exist. CKD-EPI 2021, the race-free spirometry reference equations published by the Global Lung Function Initiative, and race-neutral cardiovascular risk frameworks are all available and validated.

Sponsors and CROs need to treat algorithmic audit as a standard component of protocol development — not a regulatory checkbox, but a genuine scientific review of every quantitative tool embedded in eligibility, stratification, and endpoint logic. This means asking, for every algorithm: what population was it derived from, what assumptions does it encode, and what is the documented performance differential across demographic groups? A senior biostatistician or clinical pharmacologist should be signing off on those answers before a protocol goes to IRB. The cost of doing this rigorously is measured in weeks. The cost of not doing it is measured in decades of skewed evidence and populations denied access to treatments that could help them.

The kidney transplant finding is a proof of concept that algorithmic correction works, that its effects are measurable, and that the harm caused by biased algorithms is not theoretical — it is countable in lives. The clinical trial industry has spent years building sophisticated infrastructure for patient recruitment, site monitoring, and data capture. It is past time to build equivalent infrastructure for questioning the assumptions encoded in the tools that infrastructure depends on. The question is whether the industry and regulators are finally ready to stop pretending otherwise.

References

1. STAT News: Removing race from kidney function algorithm helped more Black patients access transplants (March 10, 2026)
2. Inker LA et al. New Creatinine- and Cystatin C–Based Equations to Estimate GFR without Race. NEJM, 2021.
3. Bhakta NR et al. Race-neutral reference equations for pulmonary function testing. NEJM, 2023.
4. ClinicalTrials.gov: EMERGE Trial — Aducanumab in Early Alzheimer’s Disease (NCT02484547)
5. FDA Guidance Documents: Diversity Action Plans and Real-World Evidence Considerations (FDA.gov)