Provision of Transplant Education for Patients Starting Dialysis: Disparities Persist
Heliyon · Volume 10, Issue 17, e36542 · August 2024
Federal rules require every patient beginning dialysis in the United States to be told that a kidney transplant is an alternative. A form filed within 45 days records whether that conversation happened, and if it did not, why. Using 944,015 such records covering dialysis initiations between January 2008 and the end of 2018, we document who is left out, what reasons providers give, and how being left out tracks with a patient’s later access to the transplant waitlist. 14.52 percent of patients were not informed. Fewer than two percent of patients in any group declined the information themselves.
Key findings
- 1 One in seven patients is never informed, and more often than not no clinical reason is recorded. Of the 133,414 patients not informed, 52 percent were recorded as “not assessed,” a category that names no clinical barrier at all. 27 percent were recorded as medically unfit.
- 2 The shortfall is not patient refusal. In every group studied, fewer than two percent of patients declined the information. Whatever is producing the gap operates on the provider side of the conversation.
- 3 Black, Hispanic and American Indian patients are disproportionately recorded as “not assessed.” Controlling for insurance, employment, comorbidity, functional status, facility staffing and region, the odds of landing in the unexplained category are 18 to 30 percent higher for these patients than for white patients.
- 4 Chain-owned facilities file the paperwork and lose the outcome. When an independent dialysis facility is acquired by a chain, its patients become far more likely to be recorded as informed and meaningfully less likely to reach the transplant waitlist.
Where the measurement comes from
The Centers for Medicare and Medicaid Services require a Medical Evidence Form (CMS-2728) for every patient who develops end-stage renal disease, filed within 45 days of starting dialysis. Since 2005 the form has carried a question asking whether the patient was informed about kidney transplant options, and a follow-up asking for the reason if they were not.
That question is the paper’s measuring instrument. It is a census rather than a survey: because the form is mandatory and tied to reimbursement, it covers essentially the entire incident dialysis population. The cost of that coverage is that the form records what a provider reported, not what a patient heard.
Figure 1. The reporting checkpoint. Everything in this paper hangs on a single checkbox filed within 45 days of a patient’s first dialysis session. Counts are from Table 2 of the paper; the 52 percent figure is from Section 3.2.
Why patients are not informed
The form offers a menu of reasons: medically unfit, declined information, unsuitable due to age, psychologically unfit, not assessed, or other. The distribution across that menu is the paper’s most uncomfortable result. The modal answer is the one that explains nothing.
Figure 2. The two reasons the paper quantifies. “Not assessed” records that no determination was made, rather than a clinical barrier to transplant. The form permits more than one reason to be selected, so the categories do not sum to 100 percent, and the paper does not report percentages for the remaining categories. Source: paper, Section 3.2 and Figure 2.
A telling internal check: among patients who were unable to ambulate, a group with a genuine physical barrier, only 30.1 percent were recorded as not informed because they were medically unfit. Even where a clinical rationale plausibly exists, the form is frequently not being used to record one.
Who ends up in the unexplained category
If “not assessed” simply reflected clinical severity, it should be predicted by comorbidity and functional status. It is predicted by those things, but it is also predicted by race, ethnicity, and who owns the dialysis facility, in models that already control for insurance, employment status, cause of renal failure, vascular access type, alcohol and drug dependence, ambulation, transfer ability, assistance with daily living, staffing ratios, and region.
Figure 3. Adjusted odds of being recorded as “not assessed” rather than informed, among patients who were not informed. Race and ethnicity coefficients are relative to white patients; facility ownership is relative to independent facilities. Bars are 95 percent confidence intervals. The Asian estimate, shown in grey, is not distinguishable from no difference. Source: paper, Table 3, “Unassessed” column, N = 91,193.
Chain patients are more likely to be informed overall, and also more likely to be placed in the unexplained category when they are not informed. Both are true: chains file more complete paperwork on the information question and give less specific answers when they report a patient was not informed.
Does being told predict getting listed?
Patients recorded as informed reach the transplant waitlist faster. In a Cox model controlling for the same set of characteristics, the hazard ratio on being informed is 1.62 (95% CI 1.58 to 1.65). Translated to levels using the model’s adjusted survival curves at average covariate values, 16 percent of informed patients have reached the waitlist two years after the form is signed, against 10 percent of patients who were not informed.
The survival analyses run on a narrower sample than the descriptive results. So that every patient has a realistic chance of being listed within the observation window, they include only patients whose form was filed by 31 December 2015, and follow-up ends on 8 October 2018, the last waitlist addition recorded in the data. That restriction is why the Cox sample is 478,746 rather than 944,015.
Figure 4. Adjusted two-year waitlisting rates, derived from the Cox model at mean covariate values. These are model-implied quantities, not raw sample means, and they describe an association rather than an estimated causal effect of information. Sample restricted to forms filed by 31 December 2015 and followed to 8 October 2018. Source: paper, Section 3.3 and Figure 3, Cox model N = 478,746.
The corresponding result for the step after listing is a null, and the page would be dishonest to bury it. Conditional on being waitlisted, the hazard ratio on being informed for time to transplant is 1.032, with a 95 percent confidence interval of 0.999 to 1.067 (p = 0.057). Information appears to help patients get onto the list. Once they are on it, the data do not show that it does anything further.
There is also a statistical wrinkle the paper reports directly: a test on scaled Schoenfeld residuals rejects the proportional hazards assumption for the informed indicator, and the residual plot slopes downward early. The authors read this as information mattering most soon after it is received. It also means the single hazard ratio is a summary of something that changes over time, and should not be read as a constant effect.
Disparities that survive getting on the list
Whatever information does, it does not close the racial gap in transplantation. Among patients who reach the waitlist, minority patients wait substantially longer to be transplanted, in the same fully adjusted model.
Figure 5. Adjusted hazard of receiving a transplant, conditional on having been added to the waitlist. A hazard ratio of 0.62 corresponds to a 38 percent lower rate of transplant at any given time relative to white patients with the same recorded characteristics. Same sample restriction as Figure 4, and conditional on reaching the waitlist. Source: paper, Table 4, time-to-transplant column, N = 109,048.
Compliance without outcomes: what happens when a chain buys a clinic
The cleanest comparison in the paper narrows to independent facilities and asks what changes when one is acquired by a chain. Because the same facility is observed before and after, and because the models are run with facility and time fixed effects without the coefficients moving much, this comes closer than anything else here to isolating a change in practice rather than a difference in patients.
Two things happen at once, and they point in opposite directions.
Figure 6. Logit estimates on the subsample of independent facilities, where “acquired” switches on after a chain takes over. After acquisition the odds of a patient going uninformed fall to about a third of their previous level, and the odds of being added to the waitlist fall by roughly 29 percent. Source: paper, Table 5, N = 133,888 and 133,891.
The economics here is about what a measured indicator does when it becomes a target. Chain ownership brings standardized protocols, and one of those protocols evidently covers filling in the transplant question on the CMS-2728. The patient outcome the question exists to promote moves the other way. The observation is consistent with prior work finding that chain acquisition in dialysis raises reimbursement-linked measures while lowering waitlist and transplant rates.
What this design does and does not establish
This is an observational study built on an administrative registry. It is well powered and broadly representative, and it is not a causal design.
The information result is an association. Patients recorded as informed differ from patients recorded as not informed in ways the form does not capture, including how sick a nephrologist judged them to be. The controls are extensive but they are controls, not an instrument or a natural experiment. Nothing here identifies the effect of telling a patient about transplant.
The outcome is a report, not a conversation. The CMS-2728 is frequently completed by ancillary staff. What was said, how well it was explained, and whether the patient retained it are all outside the data. One cited study finds that a meaningful share of patients whose physicians reported informing them did not recall being informed.
The acquisition analysis is a before-and-after comparison. It is the closest thing to a research design in the paper and it survives facility and time fixed effects, but acquisition is not random. Chains may buy facilities on trajectories that already differ.
Two further limitations the authors flag. Patients with a prior kidney transplant were not excluded, and pre-emptive listing before dialysis was not studied, so the paper speaks to the dialysis-initiation checkpoint rather than to the full path into transplantation.
Why it matters for policy
The 45-day checkpoint is the last point in the system at which every patient is supposed to learn that transplant exists. By the time it arrives, a patient has already started dialysis, which is later than clinical guidance suggests: the Organ Procurement and Transplantation Network recommends referral when GFR approaches 30.
Three implications follow from the pattern documented here. Education should happen earlier in the progression of kidney disease rather than at dialysis initiation. Providing information about an option should be separated from judging whether a patient is a suitable candidate for it, since candidacy assessment belongs to transplant centers and the dialysis chair is not the setting for a specialized screening. And a form that lets “not assessed” absorb half of all non-disclosures is not producing usable accountability, so the reporting instrument itself is a policy lever.
The ownership result adds a caution to all of this. Any intervention that raises measured information rates may raise measured information rates and nothing else. Chain facilities already demonstrate what full compliance with the reporting requirement looks like without a corresponding gain in access.
Paper and data
The article is published open access in Heliyon under a CC BY-NC 4.0 licence and is available from the publisher at doi.org/10.1016/j.heliyon.2024.e36542, with free full text at PubMed Central.
The analysis uses restricted data from the United States Renal Data System. These records are not publicly available: access requires an approved Data Use Agreement, and the data cannot be redistributed, so there is no public replication archive for this paper.
My own contribution, as recorded in the article’s CRediT statement, covers conceptualization, data curation, formal analysis, methodology, and writing of the original draft and revisions. The work was carried out while I was a research analyst at the Federal Reserve Bank of Philadelphia. The research received no specific grant funding, and the authors declare no competing interests.
Citation
Villani, V., Bertuzzi, L., Butler, G., Eliason, P., Roberts, J. W., DePasquale, N., Park, C., McElroy, L. M., & McDevitt, R. C. (2024). Provision of transplant education for patients starting dialysis: Disparities persist. Heliyon, 10(17), e36542.
@article{villani2024transplant,
title = {Provision of transplant education for patients starting dialysis:
Disparities persist},
author = {Villani, Vincenzo and Bertuzzi, Luca and Butler, Gabriel and
Eliason, Paul and Roberts, James W. and DePasquale, Nicole and
Park, Christine and McElroy, Lisa M. and McDevitt, Ryan C.},
journal = {Heliyon},
volume = {10},
number = {17},
pages = {e36542},
year = {2024},
doi = {10.1016/j.heliyon.2024.e36542}
}
Figures on this page were built from the statistics reported in the published article. They are not reproductions of the article’s own figures.