Transplant Trial Watch

Application of machine learning models for predicting acute kidney injury following donation after cardiac death liver transplantation.

He, Z. L., et al.

Hepatobiliary & Pancreatic Diseases International 2021 [record in progress].


Aims
This study aimed to compare the performance of machine learning models versus a logistic regression model for predicting acute kidney injury (AKI) in patients undergoing donation after circulatory death liver transplantation (DCDLT).

Interventions
The study sample was randomly split into a training set and a test set.

Participants
493 adult (>18 years) DCDLT patients.

Outcomes
The main outcome was the assessment of the predictive power of four machine learning models (random forest, support vector machine, classical decision tree, and conditional inference tree) versus a logistic regression model for predicting the incidence of postoperative AKI in DCDLT patients.

Follow-up
20.4 (11.4–35.5) months [median (interquartile range)]

CET Conclusions
This interesting study looks at the role of machine learning models in predicting the incidence of acute kidney injury (AKI) following DCD liver transplantation. The authors split a single-centre dataset into training and validation cohorts and compare the performance of different regression and ML-based models. The incidence of AKI was high, and AKI was associated with inferior survival. Machine learning models, especially random forest models, outperformed regression-based models in predicting AKI. Overall, the study is well designed and reported. The fact that the models were derived from single-centre data means that generalizability may be limited and new models are likely to be required for other settings. It is not entirely clear how this information could be used in clinical practice, and what interventions may help to reduce risk of AKI in these patients – these aspects should be the focus of future studies.

Trial registration
N/A

Funding source
Non-industry funded