Performance validation of the ALPPS risk model

Abstract

BACKGROUND Based on the International ALPPS registry, we have recently proposed two easily applicable risk models (pre-stage1 and 2) for predicting 90-day mortality in ALPPS but a validation of both models has not been performed yet.
METHODS The validation cohort (VC) was composed of subsequent cases of the ALPPS registry and cases of centers outside the ALPPS registry.
RESULTS The VC was composed of a total of 258 patients including 70 patients outside the ALPPS registry with 32 cases of early mortalities (12%). Development cohort (DC) and VC were comparable in terms of patient and surgery characteristics. The VC validated both models with an acceptable prediction for the pre-stage 1 (c-statistic 0.64, P = 0.009 vs. 0.77, P < 0.001) and a good prediction for the pre-stage 2 model (c-statistic 0.77, P < 0.001 vs. 0.85, P < 0.001) as compared to the DC. Overall model performance measured by Brier score was comparable between VC and DC for the pre-stage 1 (0.089 vs. 0.081) and pre-stage 2 model (0.079 vs. 0087).
CONCLUSION The ALPPS risk score is a fully validated model to estimate the individual risk of patients undergoing ALPPS and to assist clinical decision making to avoid procedure-related early mortality after ALPPS.