Using Unsupervised Clustering to Characterize Phenotypes Among Older Kidney Transplant Recipients: A Cohort Study.

Abstract

Background: Older kidney transplant recipients have inferior outcomes compared to younger recipients, and this risk may be compounded by donor characteristics.

Objective: We applied an unsupervised machine learning clustering approach to group older recipients into similar phenotypes. We evaluated the association between each cluster and graft failure, and the impact of donor quality on outcomes.

Design: This is a nationally representative retrospective cohort study.

Setting and patients: Kidney transplant recipients aged ≥65 years identified from the Scientific Registry of Transplant Recipients (2000-2017).

Measurements and methods: We used unsupervised clustering to generate phenotypes using 16 recipient factors. Donor quality was evaluated using 2 approaches, including the Kidney Donor Risk Index (KDRI). All-cause graft failure was analyzed using multivariable Cox regression.

Results: Overall, 16 364 patients (mean age 69 years; 38% female) were separated into 3 clusters. Cluster 1 recipients were exclusively female; cluster 2 recipients were exclusively males without diabetes; and cluster 3 recipients were males with a higher burden of comorbidities. Compared to cluster 2, the risk of graft failure was higher for cluster 3 recipients (adjusted hazard ratio [aHR] = 1.25, 95% confidence interval [CI] = 1.19-1.32). Cluster 3 recipients of a lower quality (KDRI ≥1.45) kidney had the highest risk of graft failure (aHR = 1.74, 95% CI = 1.61-1.87) relative to cluster 2 recipients of a higher quality kidney.

Limitations: This study did not include an external validation cohort. The findings should be interpreted as exploratory and should not be used to inform individual risk prediction nor be applied to recipients <65 years of age.

Conclusions: In a national cohort of older kidney transplant recipients, unsupervised clustering generated 3 clinically distinct recipient phenotypes. These phenotypes may aid in complementing allocation decisions, providing prognostic information, and optimizing post-transplant care for older recipients.