Tumor size and location affect the treatment selection for solitary small recurrent hepatocellular carcinoma after initial hepatectomy

Objective: Currently, liver resection is the most effectivecurative treatment for patients with resectable hepatocellularcarcinoma (HCC). Severe post-hepatectomy liver failure (PHLF)is a serious complication for HCC patients undergoing liverresection. Models have been used to predict severe PHLF inpatients with HCC, such as the Child-Pugh score, model forend-stage liver disease (MELD) score, albumin-bilirubin (ALBI)score, and aspartate aminotransferase to platelet ratio index(APRI) score. However, the ability of these models to predictsevere PHLF is limited. Accurate preoperative individualizedprediction of severe PHLF is very important for surgeons toprecisely plan and perform liver resection procedures. Thisstudy was conducted to develop a novel and stable nomogrammodel for predicting the risk of severe PHLF for patients withHCC undergoing liver resection. Methods: Data on 1,044 consecutive patients who underwentcurative liver resection at Affiliated Tumor Hospital of GuangxiMedical University between September 2013 and December2016 were included in this retrospective study. All patientswere divided into a derivation set (n = 696) and validationset (n = 348) according to the 2:1 matching principle. In thederivation set, univariate and multivariate logistic analyseswere performed to investigate risk factors associated withsevere PHLF. The β coefficients from multivariate logistic analyze were used to construct a novel nomogram model forestimating severe PHLF risk. Next, a validation set was usedto externally validate the model’s performance. The predictivediscrimination and calibration ability of the novel nomogrammodel were assessed in terms of its area under the receiveroperating characteristic curve (AUC) and calibration curve, andcompared with four currently conventionally used predictionmodels for severe PHLF. Results: Severe PHLF occurred in 75 of 696 patients (10.8%)and 37 of 348 patients (10.6%) in the derivation and validationsets, respectively. The baseline characteristics between thetwo sets were not significantly different. In the derivationset, univariate and multivariate analyses revealed that the riskfactors associated with sever PHLF were preoperative plateletcount, total bilirubin, prealbumin, aspartate aminotransferase,tumor size, major resection, and intraoperative bloodloss ≥ 400 mL. Incorporating these seven factors into thenovel nomogram model resulted in an AUC of 0.827 (95%confidence interval 0.780–0.874; P < 0.001) and 0.799 (95%confidence interval 0.726–0.872; P < 0.001) for predictingsevere PHLF in the derivation and validation sets, respectively,and showed satisfactory goodness-of-fit calibration curves.Compared with current conventionally used prediction models,the AUC of the novel nomogram model (0.827) for predictingsevere PHLF was significantly greater than that of the Child–Pugh score (0.584), MELD score (0.658), ALBI score (0.677),and APRI score (0.765) for the derivation set, as well assuperior in the validation set (corresponding AUC: 0.799 vs.0.593–0.772). Conclusions: The novel nomogram model predicted the riskof severe PHLF in HCC patients undergoing liver resectionmore accurately than the Child-Pugh, MELD, ALBI, andAPRI scores, which is very important for surgery plan andperformance. DOI: 10.20892/j.issn.2095-3941.2018.S082