Position: Machine Learning for Heart Transplant Allocation Policy Optimization Should Account for Incentives

The allocation of scarce donor organs constitutes one of the most consequential algorithmic challenges in healthcare. While the field is rapidly transitioning from rigid, rule-based systems to machine learning and data-driven optimization, we argue that current approaches often overlook a fundamental barrier: incentives. In this position paper, we highlight that organ allocation is not merely an optimization problem, but rather a complex game involving organ procurement organizations, transplant centers, clinicians, patients, and regulators. Focusing on US adult heart transplant allocation, we identify critical incentive misalignments across the decision-making pipeline, and present data showing that they are having adverse consequences today. Our main position is that the next generation of allocation policies should be incentive aware. We outline a research agenda for the machine learning community, calling for the integration of mechanism design, strategic classification, causal inference, and social choice to ensure robustness, efficiency, fairness, and trust in the face of strategic behavior from the various constituent groups.

💡 Research Summary

The paper “Machine Learning for Heart Transplant Allocation Policy Optimization Should Account for Incentives” presents a position that the next generation of heart‑transplant allocation policies must be incentive‑aware. The authors first describe the current U.S. system, which classifies adult heart‑transplant candidates into six urgency tiers. They argue that this tiered system, together with the use of device status (e.g., intra‑aortic balloon pump versus durable left‑ventricular assist device), creates a “device game” in which clinicians manipulate patient features to move patients into higher‑priority tiers. Empirical data show a three‑fold increase in IABP use after the 2018 policy change, suggesting strategic behavior rather than purely medical necessity.

The paper then examines the “out‑of‑sequence” or “open‑offer” mechanism that allows organ procurement organizations (OPOs) to bypass the official priority queue when organ ischemic time threatens viability. Although intended to reduce waste, the authors document a sharp rise in such discretionary allocations (from ~1 % to >10 % in recent years) and highlight the opacity of the trigger criteria, the potential for collusion between OPOs and transplant centers, and demographic biases favoring affluent patients.

A third source of incentive distortion is the performance‑monitoring regime administered by the Scientific Registry of Transplant Recipients (SRTR). Centers are publicly rated on waiting‑list mortality, transplant rate, and one‑year graft survival. Because these metrics affect reputation and reimbursement, centers may list patients early to accrue “wait‑time credit” or reject viable offers to protect reported outcomes, thereby worsening overall efficiency.

To address these problems, the authors propose a research agenda that integrates several machine‑learning sub‑fields:

1. Strategic classification – model the cost of feature manipulation (e.g., device implantation) and train classifiers that are robust to such gaming, drawing on work in strategic classification and repeated risk minimization.
2. Causal inference – identify non‑manipulable, causally relevant clinical variables (e.g., biomarkers) to replace or down‑weight manipulable features in urgency scores.
3. Mechanism design & selective verification – introduce randomized audits (selective verification) to raise the expected cost of misreporting, and design allocation mechanisms that are incentive‑compatible (e.g., continuous distribution scores rather than discrete tiers).
4. Social choice & fairness – develop transparent, community‑driven preference elicitation procedures and ensure that any continuous scoring respects equity constraints across demographics.

The authors also discuss concrete policy changes: moving from tiered to continuous priority scores, embedding hard, objective thresholds (e.g., ischemic‑time limits) for triggering open offers, and using computer‑vision‑based viability assessment of ex‑vivo perfused organs to inform real‑time allocation decisions.

Overall, the paper makes a compelling case that without accounting for strategic behavior by clinicians, OPOs, and transplant centers, any machine‑learning‑driven optimization will be undermined by gaming, leading to inefficiency, inequity, and potential harm to patients. The proposed interdisciplinary agenda—combining mechanism design, strategic classification, causal inference, and social choice—offers a roadmap for the ML community to develop truly robust, fair, and trustworthy organ‑allocation policies.