DISCOVER - Development and valIdation of an Integrated riSk prediction model for COrneal graft surviVal following Endothelial keRatoplasty: an artificial intelligence based approach

Abstract:

Corneal disease is the second leading cause of blindness affecting up to 12 million people worldwide.1,2 With greater surgicalsuccess and improved clinical outcomes, health care systems currently struggle to bridge the increasing gap between supply anddemand. In fact, only one sight restoring donor cornea is currently available for every 70 individuals requiring corneal transplant andapproximately 12.7 million people worldwide are actively awaiting corneal transplantation. As the demand for donor corneascontinues to exceed the supply, development of a valid and reliable risk prediction model for corneal graft failure can improvedecision making on allocation of scarce resources of donor corneas.In order to directly address this need, the DISCOVER project aims to develop and validate a dynamic integrated prediction modelthat systemically incorporates baseline demographic data, longitudinal clinical and patient-reported outcomes of corneal transplantrecipients. With access to high quality patient registry data from the clinical teams of Università degli Studi di Ferrara (UNIFE) andUniversità degli Studi “Magna Graecia” di Catanzaro (UNICZ) and the state-of-the-art artificial intelligence algorithms developed bythe AI teams of UNIFE and Università degli Studi di Napoli Federico II (UNINA), the current project shall build on our previousworks.3-17 The project will comprise two main phases: a system development phase and clinical validation phase. The systemdevelopment phase shall involve data exploration and clustering, feature extraction and development of machine learningalgorithms based on baseline demographic data and deep learning models for image processing of longitudinal clinical outcomes.Through multiple iterative steps of model training and testing and optimization of model hyperparameters, top performing algorithms shall be identified and incorporated as a single integrated prediction model. The clinical validation phase shall involve apreliminary evaluation using retrospective data to further refine and evaluate the credibility of the computational algorithm toensure seamless integration to routine patient care. Finally, the prediction model shall be implemented within the real-world settingthrough a prospective multicenter prospective real-time study on a diverse population of corneal transplant recipients. A strongfocus on the targeted users of the system, i.e. eye surgeons in all phases of the project to assess the true clinical, technological, andeconomic utility of the prediction model.The envisioned output from the system development and clinical validation phases is a dynamic integrated prediction model ofcorneal graft survival following endothelial keratoplasty that can be readily translated into routine clinical practice