Transplant Trial Watch

Machine learning in predicting graft failure following kidney transplantation: A systematic review of published predictive models.

Senanayake, S., White, N. et al. (2019).

International Journal of Medical Informatics [record in progress].

The aim of this systematic review was to assess the usefulness of different machine learning methods used to predict graft outcomes among kidney transplant patients in decision-making.

The Medline, EMBASE, Cumulative Index of Nursing and Allied Health Literature (CINAHL), PsycINFO and Cochrane databases were searched for clinical prediction models, based on an adult patient population in graft survival following kidney transplantation.

18 studies met the inclusion criteria for this review (population size varied from 80 to 92844).

Graft outcomes, including graft survival and failure following kidney transplantation were explored.


CET Conclusions
This systematic review looks at machine-learning derived models for predicting graft failure following renal transplantation. The authors identify 18 studies, using a mixture of neural networks, decision trees and Bayesian belief models. In general, models performed well, although in those that compared machine learning techniques to traditional regression modelling the benefits were uncertain. Given the theoretical benefits of the machine learning approach (fewer assumptions about underling relationships, distributions and interactions) it is perhaps surprising that this approach did not demonstrate greater accuracy over regression modelling. This may be a reflection of the complexity of the determinants of clinical outcomes, due to missing data, or due to uncaptured variables that influence outcomes. It may be that machine learning has a role to play in outcome prediction and allocation, but further studies are required to determine the best parameters and models to use.

Quality notes
SR - QA not necessary.

Trial registration

Funding source
Non-industry funded