Machine learning methods to identify risk factors for corneal graft rejection in keratoconus

Abstract

Abstract Machine learning can be used to identify risk factors associated with graft rejection after corneal transplantation for keratoconus. The study included all keratoconus eyes that underwent primary corneal transplantation from 1994 to 2021. Data relating to the recipient, donor, surgery, and postoperative course that might be associated with the occurrence of a graft rejection reaction were compiled. This study used five supervised learning algorithms including artificial neural network, support vector machine, gradient boosting, extra trees classifier, and random survival forests to select the most predictive factors for graft rejection. A total of 1214 consecutive eyes of 985 keratoconus patients were included in the study, and the technique of keratoplasty included penetrating keratoplasty in 574 eyes (47.3%) and deep anterior lamellar keratoplasty in 640 eyes (52.7%). The overall prevalence of first graft rejection was 28.1%. All five models had similar ability in identifying predictive factors for corneal graft rejection. Technique of keratoplasty was associated with an increased risk of graft rejection in all models. Other identified risk factors included patient age, keratoplasty in the fellow eye, donor age, graft endothelial cell density, duration of corticosteroid application, time from keratoplasty to complete suture removal, and suture-associated complications. It is advisable that in the absence of any contraindication, post-transplant keratoconus eyes receive a low dose topical corticosteroid until all sutures are removed.

Keywords

• Keratoconus
• Corneal transplantation
• Graft rejection
• Prediction
• Machine learning
• Artificial neural network