Human Versus Computer-based Predictions of Long Allograft Survival

Overview

The clinical decision-making after kidney transplantation is mainly driven by patient individual assessment. However, this task remains difficult and uncertain due to the integration of complex and numerous parameters. We aim to evaluate and compare the ability of transplant physicians to predict long term allograft survival compared with a computer-based survival prediction algorithm (iBox system).

400 kidney transplant recipients among the cohort of 4,000 patients from the Paris Transplant Group prospective kidney transplant cohort (NCT03474003) were randomly selected. We generated an anonymized electronic health record for each included patient including a total of 60 classical kidney transplant prognostic parameters comprising baseline transplant and recipient characteristics, together with post-transplant parameters including allograft function, proteinuria, histology, diagnoses, and immunological profile collected during the first-year post-transplant. The time of risk evaluation for the human and the iBox system were at 1-year post transplant and the death censored allograft survival predictions made at 7 years after risk assessment. We enrolled transplant physicians at various stages of their careers (residents, fellows and seniors) to assign death censored graft survival probabilities at 7 years post risk assessment. The physicians were blinded to the actual patient outcome (allograft failure) and the iBox predictions. The physicians-based predictions will then be compared with the iBox system, a validated computer-based kidney survival prediction system.

Conditions

Interventions

Eligibility

Locations (1)

Outcomes