A simpler way to rank heart transplant patients could save lives

by · News-Medical

By analyzing data from more than 33,000 heart transplant candidates, researchers reveal how a new risk score could reshape the US transplant waiting list, prioritize the sickest patients, and potentially save lives while reducing opportunities to game the system.

A recent paper published in the journal JAMA reports on the second version of the US Candidate Risk Score (US-CRS 2.0), which could improve the prioritization of heart transplant candidates without increasing donor heart travel distances when incorporated into the existing allocation framework.

Why the current heart allocation system needs reform

About one in five adult heart transplant candidates waitlisted in the US in 2020 died or were removed from the list without a transplant within three years.

Since 2018, donor hearts have been allocated by a 6-status system, with status 1 the most urgent. The statuses depend mostly on the treatments a patient is receiving, making the system susceptible to manipulation through treatment decisions. For instance, the use of the intra-aortic balloon pump (IABP) has risen sharply, and it can qualify a patient for status 2. The authors also cite a growing number of "exception" requests for patients who do not meet standard criteria.

Federal rules require graft allocation by "objective and measurable medical criteria." The original US Candidate Risk Score (here termed US-CRS 1.0) was based on five laboratory values and two device variables, and was developed to demonstrate that it was feasible to replace the categories with a continuous score. It distinguished between candidates at higher and lower risk of death more accurately than the existing 6-status system.

This second version arises from a collaboration between the US-CRS 1.0 authors and the Scientific Registry of Transplant Recipients (SRTR) and the Organ Procurement and Transplantation Network (OPTN) Heart Committee. The aim is a "policy-ready" US-CRS 2.0.

Tracking over 33,000 US heart transplant candidates

The data came from a retrospective registry study of all adults on the US heart waitlist between January 2019 and August 2025. The January date marks the start of data collection under the 2018 policy.

The cohort of 33,097 adult candidates was split by calendar time into training, validation, and test cohorts. The training cohort included all adults on the waitlist between January 2019 and August 2023, comprising 21,859 candidates. The validation cohort comprised 7,897 candidates on the waitlist between September 2023 and August 2024, while the test cohort included 8,256 candidates between September 2024 and August 2025.

Some candidates contributed nonoverlapping follow-up to more than one set. In the training cohort, the mean age was 53 years, and 74% of the candidates were male.

Mortality risk predictors

The authors studied the primary outcome of six-week mortality without transplant, defined as death or removal from the list for deterioration. The six-week period was chosen to capture short-term risk, similar to the Model for End-Stage Liver Disease (MELD) score for liver allocation.

The potential risk predictors included laboratory results (albumin, total bilirubin, estimated GFR, BNP or NT-proBNP, sodium), the primary diagnosis, four device categories (durable left ventricular assist device (LVAD), IABP, microaxial flow pump, and advanced mechanical circulatory support (extracorporeal membrane oxygenation (ECMO) or temporary surgical ventricular assist device (VAD)), and hemodynamic indices (cardiac power output, aortic pulsatility index, pulmonary artery pulsatility index).

The authors used a discrete-time survival model with 14-day landmarks to align with the frequency of status data updates. However, transplant recipients are often sicker than those still waiting, which can bias mortality estimates. To adjust for this, they fitted a marginal structural model with inverse probability weights. As a result, the score aims to estimate mortality risk in a hypothetical world without transplant.

Testing two alternative approaches to heart allocation

Following this, the authors ran a policy simulation. They first converted the score to a 1 to 50 scale, with 50 representing the highest mortality risk. They then reordered 45,530 actual donor-heart match runs from 2019 to 2024 under two counterfactual systems:

  • 6-Status US-CRS: this keeps the current allocation structure but replaces the adult statuses with score-based tiers, ranking candidates by their individual scores within each tier.
  • National US-CRS: this ranks the entire national waitlist by score, with blood type, distance, and waiting time as tiebreakers.

Five laboratory measures and four devices predict mortality

The authors chose two US-CRS models based on their ability to distinguish between higher- and lower-risk candidates and the accuracy of their mortality estimates in the validation set, one being simple and one complex. The complex US-CRS offered no meaningful advantage over the simpler one in the test set, which was therefore selected as the final model. This was based on five laboratory values (albumin, bilirubin, eGFR, BNP/NT-proBNP, sodium) and four devices.

The adjusted hazard ratios for mortality during the six-week follow-up, compared with candidates without the respective device, were as follows. A durable LVAD was associated with lower risk, at 0.65. An IABP was associated with higher risk, at 2.13, and a microaxial flow pump with similar risk, at 2.12. Advanced mechanical circulatory support (ECMO or a temporary surgical VAD) carried the highest risk, at 5.34.

Predicted mortality varied considerably among candidates in the same urgency status, even among those using the same device. This highlights the limitations of relying primarily on treatment type to determine transplant priority.

Simpler model improves prediction of waitlist mortality

In the test set, the simple model had a mean area under the curve (AUC), a measure of how well the model distinguishes between candidates at higher and lower risk of death, of 0.807, against 0.693 for the current 6-status system and 0.680 when waiting-time tiebreakers were added. The more complex model scored 0.795, which the authors take as a sign of overfitting. The simple model was significantly better than status plus waiting time in 3 of 8 calendar cross-sections.

Accurate risk estimates could reduce allocation exception requests

Calibration, or how closely predicted mortality matched observed mortality, was acceptable by status, diagnosis, and exception status. Notably, the model predicted mortality well for exception candidates. This suggests an exception pathway might not be needed for them under a score-based system.

Observed mortality in exception patients at the top statuses was lower on average than in standard-criteria patients. Both models overestimated risk in the very highest-risk candidates, and more so for the complex model.

The results were robust to alternative methods of handling missing data and to excluding candidates who had also contributed to the training set from the test set.

New scoring system reduces opportunities for allocation manipulation

With this score, adding a device could not, by itself, lead to a high US-CRS status, unlike the current system. For example, a current status 6 patient would typically jump to status 2 upon starting an IABP. Under US-CRS, adding an IABP or microaxial flow pump to a candidate without a device usually resulted in no change or a one-tier increase when laboratory values were held constant.

Moreover, treatment may improve a patient's laboratory values, potentially offsetting the priority increase associated with starting a device.

Score-based allocation prioritizes sicker heart transplant candidates

Median predicted six-week mortality rose from 0.32% among candidates in the lowest US-CRS urgency tier to 22.57% among those in the highest.

Of 47,442 match runs, 45,530 had complete data, with 21,787 of these hearts accepted by adult candidates. The effects on the waitlist were large, with 67% of actual transplant recipients moving down in priority with the 6-Status US-CRS allocation. In the 6-Status US-CRS reordering, 57% of recipients moved 10 or more positions. In comparison, 80% of recipients moved 10 or more positions when reallocated by the National US-CRS.

The observed six-week mortality, censoring at transplant, among candidates ranked first on the match run increased from 11.2% under the current system to 15.7% under the 6-Status US-CRS and 23.0% under the National US-CRS. This indicates that both proposed systems gave greater priority to candidates at risk of dying without a transplant.

Donor-to-candidate distance was shorter at early positions under 6-Status US-CRS and longer under National US-CRS. The authors read the shorter distances as a sign that the current system lets lower-risk candidates displace higher-risk ones located closer to the donor.

The score-based systems prioritized more candidates from racial and ethnic minority groups. They gave lower priority to candidates with blood type O and to those with durable LVADs. Priority changed little by sex or stature.

Balancing medical urgency with equitable transplant access

The authors acknowledge several limitations. Regarding calibration, both models overestimated risk in the highest-risk patients. They consider discrimination more important than calibration for allocation, and propose periodic refitting to keep up with changes in clinical practice.

The lower priority for durable LVAD and blood type O patients requires separate policy structures, such as a time-based pathway for LVAD patients. Patients with durable LVADs generally face a lower short-term risk of death but still need a pathway to transplantation before serious complications develop. A June 2026 policy already awards high status to long-term LVAD patients without requiring complications.

Post-transplant outcomes are not reflected in match-run analysis. That is, it cannot show how directing hearts to sicker candidates would affect post-transplant survival. Nor does it establish which candidates would ultimately receive donor hearts or how many waitlist deaths would be prevented. A full simulation is needed, and if prioritization by US-CRS is projected to adversely affect post-transplant survival, an expected post-transplant survival component could be added to a composite score.

The authors argue that device-based score manipulation is harder because of the multivariable nature of the score. Excluding durable LVAD use would bias the coefficients and reduce predictive accuracy, while relying solely on laboratory values might incentivize undertreatment.

New urgency score could reduce heart transplant waitlist deaths

The authors conclude that US-CRS 2.0 ranks candidates more accurately than the 6-status system, would prioritize the sickest patients, and could potentially reduce waitlist deaths while making heart allocation less susceptible to manipulation. Incorporating the score into the existing allocation framework could also shorten donor heart travel distances. They emphasize the need for a full simulation, with fair access for LVAD and type O candidates, before adopting any policy.

Journal reference: