Medicina (May 2023)

Differences between Very Highly Sensitized Kidney Transplant Recipients as Identified by Machine Learning Consensus Clustering

  • Charat Thongprayoon,
  • Jing Miao,
  • Caroline C. Jadlowiec,
  • Shennen A. Mao,
  • Michael A. Mao,
  • Pradeep Vaitla,
  • Napat Leeaphorn,
  • Wisit Kaewput,
  • Pattharawin Pattharanitima,
  • Supawit Tangpanithandee,
  • Pajaree Krisanapan,
  • Pitchaphon Nissaisorakarn,
  • Matthew Cooper,
  • Wisit Cheungpasitporn

DOI
https://doi.org/10.3390/medicina59050977
Journal volume & issue
Vol. 59, no. 5
p. 977

Abstract

Read online

Background and Objectives: The aim of our study was to categorize very highly sensitized kidney transplant recipients with pre-transplant panel reactive antibody (PRA) ≥ 98% using an unsupervised machine learning approach as clinical outcomes for this population are inferior, despite receiving increased allocation priority. Identifying subgroups with higher risks for inferior outcomes is essential to guide individualized management strategies for these vulnerable recipients. Materials and Methods: To achieve this, we analyzed the Organ Procurement and Transplantation Network (OPTN)/United Network for Organ Sharing (UNOS) database from 2010 to 2019 and performed consensus cluster analysis based on the recipient-, donor-, and transplant-related characteristics in 7458 kidney transplant patients with pre-transplant PRA ≥ 98%. The key characteristics of each cluster were identified by calculating the standardized mean difference. The post-transplant outcomes were compared between the assigned clusters. Results: We identified two distinct clusters and compared the post-transplant outcomes among the assigned clusters of very highly sensitized kidney transplant patients. Cluster 1 patients were younger (median age 45 years), male predominant, and more likely to have previously undergone a kidney transplant, but had less diabetic kidney disease. Cluster 2 recipients were older (median 54 years), female predominant, and more likely to be undergoing a first-time transplant. While patient survival was comparable between the two clusters, cluster 1 had lower death-censored graft survival and higher acute rejection compared to cluster 2. Conclusions: The unsupervised machine learning approach categorized very highly sensitized kidney transplant patients into two clinically distinct clusters with differing post-transplant outcomes. A better understanding of these clinically distinct subgroups may assist the transplant community in developing individualized care strategies and improving the outcomes for very highly sensitized kidney transplant patients.

Keywords