External validation of prediction models for time to death in potential donors after circulatory death.

Predicting time to death in controlled donation after circulatory death (cDCD) donors following withdrawal of life-sustaining treatment (WLST) is important but poses a major challenge. The aim of this study is to determine factors predicting time to circulatory death within 60 minutes after WSLT and validate previously developed prediction models. In a single-center retrospective study, we used the data of 92 potential cDCD donors. Multivariable regression analysis demonstrated that absent cough-, corneal reflex, lower morphine dosage, and midazolam use were significantly associated with death within 60 minutes (area under the curve [AUC] 0.89; 95% confidenence interval [CI] 0.87-0.91). External validation of the logistic regression models of de Groot et al (AUC 0.86; 95% CI 0.77-0.95), Wind et al (AUC 0.62; 95% CI 0.49-0.76), Davila et al (AUC 0.80; 95% CI 0.708-0.901) and the Cox regression model by Suntharalingam et al (Harrell's c-index 0.63), exhibited good discrimination and could fairly identify which patients died within 60 minutes. Previous prediction models did not incorporate the process of WLST. We believe that future studies should also include the process of WLST as an important predictor.