Phenotypic clustering to identify high-risk adults undergoing heart transplantation
W Baker, C I Coleman, S Ohira, A JaiswalAbstract
Background
While risk scores have been developed to predict 1-year survival after heart transplant (HT), clinicians continue to seek a simplified approach to identify patients at highest risk.
Purpose
To use cluster analysis of national registry data to identify recipient risk profiles and evaluate their associations with 1-year mortality among adults undergoing HT.
Methods
Adult recipients who underwent HT between January 2010 and December 2023 were identified from the United Network for Organ Sharing (UNOS) database, with follow-up through December 2024. Machine-learning-based clustering was used to identify recipient phenotypes based on readily available clinical characteristics: sex, race, body mass index (BMI), diabetes mellitus (DM), and ischemic cardiomyopathy (ICM). Given the strong association between recipient age and post-HT outcomes, clustering was performed within 3 age strata: <50 years, 50-64 years, and ≥65 years. Multivariable Cox proportional hazards regression, adjusted for recipient age, ischemic time, donor-recipient predicted heart mass ratio, and average annual transplant center volume, was used to compare 1-year mortality risk across clusters.
Results
The study cohort included 34,734 HT recipients (median age 56 years, 73% male), of whom 3,024 (8.7%) died within 1 year of HT. Within each age stratum, 5 distinct clusters were identified. Across all age groups, clusters with the highest 1-year mortality were predominantly men with higher BMI, frequent DM and/or ICM (Figure 1). In contrast, clusters with the lowest risk generally consisted of women with lower bodyweights or those without a history of DM and/or ICM. Among the younger (<50 years) group, there was no significant differences in adjusted 1-year mortality risk across clusters relative to the highest-risk cluster (Figure 2). In the 50-64 years age group, compared with the highest-risk cluster (heavy White men with ICM), significantly lower adjusted mortality was observed in clusters of lighter women without DM (HR 0.81, 95% CI 0.68-0.98) and White men without DM or ICM (HR 0.68, 95% CI 0.58-0.81). Similarly, among ≥65 years of age group, compared with the highest-risk cluster (heavier White men with DM/ICM), lower mortality risk was observed in clusters of lighter White men (HR 0.75, 95% CI 0.63-0.90) and lighter White women with no DM history (HR 0.678, 95% CI 0.48-0.93).
Conclusion
Cluster analysis identified clinical phenotypes of HT recipients with differing risks of 1-year mortality. While risk did not vary by cluster among younger recipients, older age groups demonstrated higher mortality in clusters with higher BMI, in men, and those with DM/ICM. These findings suggest that simplified, phenotype-based risk stratification may complement existing prediction tools, particularly in older recipients.For image description, please refer to the figure legend and surrounding text.For image description, please refer to the figure legend and surrounding text.