To construct a nomogram model for predicting post-hepatectomy liver failure (PHLF) based on machine learning (ML), and evaluate its application value in PHLF prediction.

Clinical data of 235 patients who underwent hepatectomy in Beijing Tiantan Hospital affiliated to Capital Medical University from June 2021 to June 2023 were retrospectively analyzed. The informed consents of all patients were obtained and the local ethical committee approval was received. Among them, 118 patients were male and 117 female, aged from 27 to 83 years, with a median age of 57 years. All patients were randomly divided into the training and test sets at a ratio of 8∶2. ALT, TB, ALB, Plt, PT, INR, AST to Plt ratio index (APRI), total liver volume (TLV), remnant liver volume (RLV), spleen volume (SV), standard remnant liver volume (SRLV), remnant liver volume/weight ratio (RLV/Weight, RLV/W) and spleen volume/liver volume ratio (SV/LV) and other imaging indexes were observed. In the training set, LASSO regression and Logistic regression analyses were used to select features and screen key risk factors related to PHLF. ML-nomogram and Log-nomogram models were constructed, respectively. The predictive value of these two models was analyzed by the ROC curve. AUC between two models was compared by Delong test.

The incidence of PHLF was 16.2%(38/235), including 30 cases of grade A and 8 cases of grade B. In the training set, 6 variables including liver cirrhosis, Plt, APRI, SV, RLV/TLV and SV/LV were screened by LASSO regression as the key features of PHLF prediction. The AUC of the constructed ML-nomogram in the training and test sets was 0.937 and 0.839, respectively. Logistic regression analysis screened 3 risk factors including Plt, SV and RLV/TLV. The AUC of the constructed Log-nomogram in the training and test sets was 0.934 and 0.813, respectively. Both ML-nomogram and Log-nomogram had high predictive ability for the incidence of PHLF (Z=1.21, 1.19; both P>0.05).

In this study, ML-nomogram is successfully constructed for predicting PHLF. Both ML-nomogram and Log-nomogram are convenient and have high prediction efficiency and accuracy.