Prediction of graft loss in living donor liver transplantation during the early postoperative period
Fig 3
Predicting stratified groups in the validation cohort: (A) Workflows for the multi-class classification and hierarchical classification models are illustrated. For the multi-class classification, the multi-class classifier was constructed and trained to predict whether a given patient belonged to the corresponding group (i.e., early-loss, intermediate-loss, and late/no-loss) by using group-labeled derivation cohort data. The validation data were classified into three groups: early-loss*, intermediate-loss*, and late/no-loss*. In contrast, for hierarchical classification, early-loss and late/no-loss binary classifiers were constructed and trained to predict whether a given patient belonged to the corresponding group (i.e., early-loss and late/no-loss) by using group-labeled derivation cohort data. The validation data were then first classified into early-loss* or non-early-loss* using the early-loss classifier. Second, non-early-loss* patients were classified into late/no-loss* and others using the late/no-loss classifier. The remaining patients (i.e., others) were assigned to intermediate-loss*. This figure was created by hand. (B) Graft survival curves by predicted group in the validation cohort data by multi-class XGBoost classifier are plotted and colored accordingly. Each colored shaded area and short vertical line on the curve indicate the 95% confidence interval and a censored patient, respectively. (C) Graft survival curves by predicted group in the validation cohort data by the hierarchical XGBoost classifier are plotted and colored accordingly. Each colored shaded area and short vertical line on the curve indicate the 95% confidence interval and a censored patient, respectively. (D) Graft survival curves by classified group in the validation cohort data by predictive score are plotted and colored accordingly. PG1: Group with predictive score less than 1.15; PG2: Group with predictive score between 1.15 and 1.30; PG3: Group with predictive score of 1.30 or higher.