Improving Pediatric Donor Heart Utilization with Predictive Analytics

Pediatric heart transplant candidates have the highest waitlist mortality rate among all organ transplants, with an alarming 14 percent mortality rate. Despite this, nearly 90 percent of pediatric donor heart offers are declined, and 40 percent of these hearts are never used by pediatric transplant centers in the United States, even when they are in excellent condition. This high discard rate is largely driven by the complex and time-sensitive nature of the decision making process, during which clinicians must evaluate over 100 variables in a very short period. Sometimes biases, such as risk aversion and reliance on institutional practices, can further complicate decisions. For instance, the researchers’ recently found that the primary factor influencing whether a pediatric transplant center accepts a donor heart offer is how many institutions have previously rejected it, regardless of the offer's merit. These unnecessary refusals contribute to the high number of deaths among children on the waitlist without improving transplant outcomes. Additionally, the absence of standardized guidelines or predictive tools for navigating these complex decisions has resulted in significant variability in donor acceptance practices.

This research aims to address these challenges by developing predictive risk models and decision aids to assess the likelihood of successful pediatric heart transplants, thereby enhancing clinician confidence and improving donor heart use.

The specific aims of the research are as follows:

• Construct and validate predictive models to support transplant offer decision making.
• Develop a custom SimUNet interface for pediatric heart offer decision making.
• Develop novel visualizations to support and improve decision making.
• Evaluate the impact of interface design and predictive model scores on acceptance practices.

Researchers will develop predictive models using advanced machine learning techniques to assess transplant success, predict the time to the next donor offer, and estimate survival likelihood for pediatric candidates if an offer is declined. In collaboration with the United Network for Organ Sharing (UNOS), they will create a custom SimUNet interface for pediatric heart transplant decision making, designed for both mobile and desktop use, which mirrors DonorNet®, the actual end-user interface for offer decision making. The researchers will translate donor data and model outputs into intuitive visualizations on the SimUNet interface to assist clinicians in evaluating donor offers and will evaluate the interface's clinical utility by studying its impact on decision making among experienced pediatric transplant cardiologists.

The researchers aim to improve pediatric heart transplant decision making by reducing clinicians' cognitive burden and increasing their confidence in their decisions through the development of the first predictive model to assess post-transplant survival at the time of a donor offer, using machine learning and novel data sources. This model will provide real-time, data-driven guidance to clinicians, addressing critical gaps in current decision making practices. By incorporating detailed donor data from sources such as echocardiograms and indicators of other organ health, the model seeks to optimize donor-recipient matching, reduce unnecessary heart refusals, and improve waitlist mortality rates for pediatric patients. Collaboration with UNOS further enhances the model’s potential for wide-scale adoption, significantly advancing transplant efficiency and outcomes across pediatric and adult populations.