Long-term outcomes of and a risk score model for moderate-risk to high-risk penetrating keratoplasty

BACKGROUND/AIMS: To develop a risk score for moderate-risk to high-risk penetrating keratoplasty (PK) based on various risk factors and use it to stratify grafts. METHODS: We collected data on clinical features and survival status from the electronic health records of patients who underwent moderate-risk to high-risk PK between January 2012 and December 2021. Feature selection was performed using multivariate Cox regression analyses. The risk score was derived from the coefficient of variables obtained through the Cox proportional hazards model. The performance of the risk score was assessed using receiver operating characteristic curve analysis. Risk stratification was performed based on the risk scores. RESULTS: The study cohort comprised 2297 eyes (2166 patients) subjected to moderate-risk to high-risk PK. The risk score was derived based on six variables. Each variable was assigned a score of 0, 1, 2 or 3 based on its coefficient. The discriminative ability of this score was robust, achieving an area under the receiver operating characteristic curve of 0.747 in the external validation cohort, which surpassed the predictive performance of any single variable alone. According to the established risk score, patients were stratified into three distinct prognostic subgroups: moderate-risk (0-2 points), high-risk (3-5 points) and extremely high-risk groups (≥6 points). The survival curves across these stratified groups differed significantly (log-rank test, p