Anticipating the unexpected: a predictive model for early hospital readmissions in kidney transplant recipients using machine learning
4 affiliations
- 1Nephrology Department, Colombiana de Trasplantes
- 2Research Department, Colombiana de Trasplantes
- 3Transplant Surgery Department, Colombiana de Trasplantes
- 4Information Technology Department, Colombiana de Trasplantes
Abstract
Background Early hospital readmissions (EHR) within 30 days after kidney transplantation are common and add clinical and financial burden. Early identification of high-risk recipients can guide post-discharge care. We aimed to develop a machine learning model to predict 30-day hospital readmission in adult kidney transplant recipients. Methods We conducted a retrospective cohort study of 2,110 adult kidney transplant recipients treated at Colombiana de Trasplantes between July 2008 and December 2023. Patients with allograft thrombosis at the time of transplantation were excluded. Clinical and demographic data were extracted from electronic records. Missing values were imputed using multivariate imputation by chained equations with classification and regression trees (MICE-CART). We then implemented a 10-fold cross-validation framework, restricting oversampling to the training folds to prevent data leakage. Two machine learning models (Random Forest and XGBoost) were trained and compared against a baseline logistic regression model on both discrimination and calibration metrics. Results Overall, 14.5% of patients were readmitted within 30 days. Readmitted recipients more often had diabetes mellitus, longer pre-transplant dialysis, delayed graft function, and surgical complications. Evaluated via 10-fold cross-validation, XGBoost yielded the highest discrimination (AUC 0.750; 95% CI 0.722–0.778). Random Forest (AUC 0.744) and baseline logistic regression (AUC 0.747) performed comparably. Both algorithms identified initial hospitalization length, dialysis duration, GFR at day 7, recipient age, and surgical reintervention as the most important predictors. Conclusions A machine learning model that combines pre- and early post-transplant variables can identify kidney transplant recipients at higher risk of 30-day readmission. Based on this model, we developed a web-based risk calculator that returns individualized risk estimates to support discharge planning and follow-up. External validation in independent cohorts is needed before clinical adoption.
Related
- Association of eplet mismatch load and allograft injury in kidney transplant recipients
- The Tell-Tale Heart ― Persistent Immune Activation in Histologically Rejection-Free Heart Transplant Recipients ―
- Retrospective evaluation of predictors and clinical impact of postoperative delirium in liver transplant recipients
- Virologic and Exploratory Immunologic Characterization of Relapsing Human Cytomegalovirus (HCMV) DNAemia in Solid Organ Transplant Recipients: A Prospective Observational Cohort Study
- Clinicopathologic Features and Diagnostic Challenges of Mycophenolate Mofetil-Associated Enterocolitis in Lung Transplant Recipients
BibTeX
@article{pintoramirez2026anticipating,
title = {{Anticipating the unexpected: a predictive model for early hospital readmissions in kidney transplant recipients using machine learning}},
author = {Jessica Pinto-Ramírez and Andrea Garcia-Lopez and Santiago Cabas and Yenny Báez-Suarez and Néstor Pedraza and Andrea Gómez-Montero and Juan García and Fernando Girón-Luque},
journal = {BMC Medical Informatics and Decision Making},
year = {2026},
month = jul,
doi = {10.1186/s12911-026-03683-y},
url = {https://doi.org/10.1186/s12911-026-03683-y},
note = {CC BY},
}
RIS
TY - JOUR
TI - Anticipating the unexpected: a predictive model for early hospital readmissions in kidney transplant recipients using machine learning
AU - Jessica Pinto-Ramírez
AU - Andrea Garcia-Lopez
AU - Santiago Cabas
AU - Yenny Báez-Suarez
AU - Néstor Pedraza
AU - Andrea Gómez-Montero
AU - Juan García
AU - Fernando Girón-Luque
PY - 2026
DA - 2026/07/01
JO - BMC Medical Informatics and Decision Making
DO - 10.1186/s12911-026-03683-y
UR - https://doi.org/10.1186/s12911-026-03683-y
AB - Background Early hospital readmissions (EHR) within 30 days after kidney transplantation are common and add clinical and financial burden. Early identification of high-risk recipients can guide post-discharge care. We aimed to develop a machine learning model to predict 30-day hospital readmission in adult kidney transplant recipients. Methods We conducted a retrospective cohort study of 2,110 adult kidney transplant recipients treated at Colombiana de Trasplantes between July 2008 and December 2023. Patients with allograft thrombosis at the time of transplantation were excluded. Clinical and demographic data were extracted from electronic records. Missing values were imputed using multivariate imputation by chained equations with classification and regression trees (MICE-CART). We then implemented a 10-fold cross-validation framework, restricting oversampling to the training folds to prevent data leakage. Two machine learning models (Random Forest and XGBoost) were trained and compared against a baseline logistic regression model on both discrimination and calibration metrics. Results Overall, 14.5% of patients were readmitted within 30 days. Readmitted recipients more often had diabetes mellitus, longer pre-transplant dialysis, delayed graft function, and surgical complications. Evaluated via 10-fold cross-validation, XGBoost yielded the highest discrimination (AUC 0.750; 95% CI 0.722–0.778). Random Forest (AUC 0.744) and baseline logistic regression (AUC 0.747) performed comparably. Both algorithms identified initial hospitalization length, dialysis duration, GFR at day 7, recipient age, and surgical reintervention as the most important predictors. Conclusions A machine learning model that combines pre- and early post-transplant variables can identify kidney transplant recipients at higher risk of 30-day readmission. Based on this model, we developed a web-based risk calculator that returns individualized risk estimates to support discharge planning and follow-up. External validation in independent cohorts is needed before clinical adoption.
SN - 1472-6947
ER -