TY - JOUR
T1 - Machine learning for post-liver transplant survival
T2 - Bridging the gap for long-term outcomes through temporal variation features
AU - Balakrishnan, Kiruthika
AU - Olson, Sawyer
AU - Simon, Gyorgy
AU - Pruinelli, Lisiane
N1 - Publisher Copyright:
© 2024 Elsevier B.V.
PY - 2024/12
Y1 - 2024/12
N2 - Background: The long-term survival of liver transplant (LT) recipients is essential for optimizing organ allocation and estimating mortality outcomes. While models like the Model-for-End-Stage-Liver-Disease (MELD) predict 90-day mortality on the waiting list, they do not predict post-LT survival accurately. There is a need for predictive models that can forecast post-LT survival beyond the immediate period after transplantation. Method: This study introduces new temporal variation features for predicting post-LT survival during the waiting list period. Cox Proportional-Hazards regression (CoxPH), Random Survival Forest (RSF), and Extreme Gradient Boosting (XGB) models are utilized, along with patient demographics and waiting list duration. Data from 716 LT patients from the University of Minnesota CTSI (2011–2021) are used to develop, evaluate, and compare post-LT survival prediction models. Results: The temporal variation features, particularly when combined with the RSF model, proved most effective in predicting post-LT survival, with a C-index of 0.71 and an IBS of 0.151. This outperformed the predictive capability of the most recent MELD score, which had a C-index of <0.51 in the same cohort. Conclusions: Incorporating temporal variation features with the RSF model enhances long-term post-LT survival predictions. These insights can assist clinicians and patients in making more informed decisions about organ allocation and understanding the utility of LT, ultimately leading to improved patient outcomes.
AB - Background: The long-term survival of liver transplant (LT) recipients is essential for optimizing organ allocation and estimating mortality outcomes. While models like the Model-for-End-Stage-Liver-Disease (MELD) predict 90-day mortality on the waiting list, they do not predict post-LT survival accurately. There is a need for predictive models that can forecast post-LT survival beyond the immediate period after transplantation. Method: This study introduces new temporal variation features for predicting post-LT survival during the waiting list period. Cox Proportional-Hazards regression (CoxPH), Random Survival Forest (RSF), and Extreme Gradient Boosting (XGB) models are utilized, along with patient demographics and waiting list duration. Data from 716 LT patients from the University of Minnesota CTSI (2011–2021) are used to develop, evaluate, and compare post-LT survival prediction models. Results: The temporal variation features, particularly when combined with the RSF model, proved most effective in predicting post-LT survival, with a C-index of 0.71 and an IBS of 0.151. This outperformed the predictive capability of the most recent MELD score, which had a C-index of <0.51 in the same cohort. Conclusions: Incorporating temporal variation features with the RSF model enhances long-term post-LT survival predictions. These insights can assist clinicians and patients in making more informed decisions about organ allocation and understanding the utility of LT, ultimately leading to improved patient outcomes.
KW - Liver transplantation
KW - Machine learning
KW - Post-transplant survival prediction
KW - Survival analysis
KW - Temporal variation features
UR - http://www.scopus.com/inward/record.url?scp=85205443536&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85205443536&partnerID=8YFLogxK
U2 - 10.1016/j.cmpb.2024.108442
DO - 10.1016/j.cmpb.2024.108442
M3 - Article
C2 - 39368442
AN - SCOPUS:85205443536
SN - 0169-2607
VL - 257
JO - Computer Methods and Programs in Biomedicine
JF - Computer Methods and Programs in Biomedicine
M1 - 108442
ER -