External Validation of Six Liver Functional Reserve Models to predict Posthepatectomy Liver Failure after Major Resection for Hepatocellular Carcinoma

Objective: To validate and compare the predictive ability of albumin-bilirubin model (ALBI) with other 5 liver functional reserve models (APRI, FIB4, MELD, PALBI, King's score) for posthepatectomy liver failure (PHLF) in patients with hepatocellular carcinoma (HCC) who underwent major hepatectomy. Methods: Data of patients undergoing major hepatectomy for HCC from 4 hospitals between January 01, 2008 and December 31, 2019 were retrospectively analyzed. PHLF was evaluated according to the definition of the 50-50 criteria. Performances of six liver functional reserve models were determined by the area under the receiver operating characteristic curve (AUC), calibration plot and decision curve analysis. Results: A total of 745 patients with 103 (13.8%) experienced PHLF were finally included in this study. Among six liver functional reserve models, ALBI showed the highest AUC (0.64, 95% CI: 0.58-0.69) for PHLF. All models showed good calibration and greater net benefit than treating all patients at a limit range of threshold probabilities, but the ALBI demonstrated net benefit across the largest range of threshold probabilities. Subgroup analysis also showed ALBI had good predictive performance in cirrhotic (AUC=0.63) or non-cirrhotic (AUC=0.62) patients. Conclusion: Among the six models, the ALBI model shows more accurate predictive ability for PHLF in HCC patients undergoing major hepatectomy, regardless of having cirrhosis or not.


Introduction
Hepatocellular carcinoma (HCC) is the fifth most common malignancy in the world. Partial hepatectomy (PH) is still the mainstay of curativeintent treatment for HCC [1]. With improvements in surgical techniques and perioperative management, increasing number of HCC patients are able to undergo PH. However, postoperative morbidity and mortality still exists, and posthepatectomy liver failure (PHLF) remains the most common cause of mortality particularly in patients who underwent major hepatectomy [2]. The reported incidence of PHLF is 0.7%-9.1%, especially in patients underwent major hepatectomy as high as 58.22%, and PHLF is shown as the major cause (between 18% and 75%) of postoperative mortality [3][4]. In consideration of such a high incidence, accurate prediction of PHLF is very important for patient selection and perioperative management in HCC patients following major hepatectomy.

Ivyspring International Publisher
Previously, some liver functional reserve models have been created to assess liver function and predict posthepatectomy outcomes in patients with HCC. MELD model was used to predict survival after hepatectomy of HCC patients [5][6][7], APRI, FIB4 and King's score models were created to assess liver fibrosis and cirrhosis of HBV and HCV patients [8][9][10][11][12]. These models were gradually adopted to predict PHLF and exhibited a certain predictive ability [13][14][15][16]. Recently, ALBI and PALBI model were developed to evaluate liver function and used to predict PHLF after hepatectomy for HCC [17][18][19][20][21][22][23]. Although various studies demonstrated the usefulness of these liver functional reserve models in prediction of PHLF, few studies compared their accuracy in HCC patients undergoing major hepatectomy.
The aim of this study is to validate and compare the predictive ability of the ALBI model with other 5 existing liver functional reserve models (PALBI, APRI, MELD, FIB4, and King's score) for PHLF in patients with hepatocellular carcinoma after major hepatectomy.

Study design
A multicentric retrospective analysis of patients with hepatocellular carcinoma who underwent major hepatectomy (≥ 3 segments) from the Eastern Hepatobiliary Surgery Hospital in Shanghai, Zhongda Hospital, Southeast university in Nanjing, Qin Huai Medical District of Eastern Theater General Hospital in Nanjing and The First Affiliated Hospital of Xi'an Jiao Tong University in Xi'an between January 01, 2008 and December 31, 2019 were recorded. Inclusion criteria are obeyed strictly as follows: (1) pathologically confirmed as hepatocellular carcinoma; (2) underwent major hepatectomy, major hepatectomy is defined as resection of Couinaud's segmentation 3 and above [4]; (3) no anti-tumor treatment before surgery, such as ablation, TACE; (4) no major blood vessel invasion, bile duct cancer thrombus and distant metastasis; (5) age ≥18. Exclusion criteria: (1) intraoperative radiofrequency ablation, particle placement and other treatments; (2) data were missing on the fifth day after surgery, and PHLF could not be defined.

Surgical technique
The technique of hepatectomy has been previously described [24]. Parenchymal transection was usually achieved under intermittent pedicle clamping (15-minute occlusion and 5-minute reperfusion). Under anesthesia, the hemodynamic management aimed to maintain low central venous pressure to minimize blood loss with reduced volume perfusion.

Variables of interest
Data were collected by direct extraction from electronic health records, complemented by manual curation. Variables of interest in the dataset included: demographics (age, gender, BMI), laboratory measurements (TBIL, ALB, PLT, PT, INR, ALT, AST and Cr), underlying liver disease (HBsAg, HBeAg, HBV-DNA, anti-HCV and antiviral therapy), radiology reports (ascites and cirrhosis), surgical factors (clamping time, blood loss and intraoperative transfusion) and pathological reports (tumor size, tumor number, microvascular invasion and tumor differentiation).

Liver functional reserve models
A total of 6 liver functional reserve models were investigated: ALBI, PALBI, MELD, APRI, FIB4, King's score which were depicted in detail in Table S1. Comparison of these models was performed through discrimination.

Outcomes and definitions
The main outcome of this study was PHLF. The "50-50 criteria" was used as the definition of PHLF in this study: prothrombin activity < 50% and posthepatectomy serum bilirubin > 50 μmol/L, fifth day after operation [25]. To provide a better overview of the predictive abilities on distinct patient populations, this study performed subgroup analyses based on liver cirrhosis according to imaging findings.

Statistical analysis
Continuous variables were expressed as median with the interquartile range (IQR). Categorical variables were presented as numbers and percentages. The X 2 test or Fisher's exact test was used to analyze categorical variables and the Mann-Whitney ranked sum test for continuous variables. The discrimination ability of models was evaluated by the Receiver Operating Characteristic Curve (ROC) and area under the curve (AUC). Comparisons between the ROC curves were performed using the corresponding DeLong's tests. We assessed calibration by visualizing calibration of predicted vs. observed risk using loess-smoothed plots. Decision curve analyses were done using the rmda package in R [26]. A two-tailed P values <0.05 was considered statistically significant. Statistical analysis was performed using Stata/SE 15.1 (College Station, TX77845, USA) and R (Version 4.0.2, https://www. r-project.org).

Decision curve analyses to assess clinical utility
This study compared net benefit for each model to the strategies of treating all patients, treating no patients for PHLF. Although all models showed greater net benefit than treating all patients at a limit range of threshold probabilities, the ALBI demonstrated net benefit across the largest range of threshold probabilities (Figure 4).

Subgroup analysis
Finally, subpopulation analyses were done in the patients with or without cirrhosis to assess the overall performance of ALBI compared with the other 5 models. Subgroup analysis was presented in a forest plot ( Figure S1). In cirrhotic patients, ALBI and PALBI had the better predictive accuracy (AUC = 0.63 and 0.60, respectively) compared with other 4 models. In subgroup patients without cirrhosis, ALBI, MELD, FIB4 and King's score had significant predictive accuracy (P = 0.012, 0.030, 0.011, 0.007 and AUC = 0.62, 0.59, 0.61 and 0.63, respectively).

Discussion
Liver failure is one of the leading causes of perioperative death in patients with hepatocellular carcinoma after major hepatectomy. This study applied the "50-50 criteria" proposed by Silvio Balzan et al. in 2005 as a diagnostic criterion for liver failure.
In our study, the incidence of liver failure after major liver resection is 13.8%, the 90-day mortality of patients with PHLF was significantly higher than the patients who did not experience liver failure (38.6% vs. 26.2%, P = 0.011) in our multi-institutional cohort. Selection criteria for major hepatectomy and prediction of the individual PHLF risk are debatable. Accurate assessment of preoperative liver reserve function is one of the keys to reducing the occurrence and death of liver failure after major liver resection [27]. Clinicians have been preferring to use new classification of objective indicators to assess liver function more accurately. Therefore, MELD, APRI, FIB4, King's score, ALBI and PALBI models have been proposed successively. This study aimed at evaluating these six common models of PHLF in HCC patients undergoing major hepatectomy.
In this article, 745 patients with hepatocellular carcinoma who underwent major hepatectomy were retrospectively analyzed. As for univariable risk factor, BMI, HBV-DNA, TBIL, ALB, PLT, prothrombin time, INR, and blood loss were predictors for PHLF. Thus, the factors which were related to patients' demographic, HBV-related, liver related and surgical factors have been shown the associations with PHLF. The result was partially consistent with Kauffmann et al. and Shoup et al. who validate TBIL and coagulation as risk factors of PHLF [28][29].
As an integrated variable, ALBI indicated the best predictive accuracy, regardless of having cirrhosis or not. Although these models are calculated using objective indicators, the numerical results and the accuracy of predictions vary a lot. The ALBI model proposed in 2015 was simply calculated from two indicators, ALB and TBIL, and many literatures have confirmed its accuracy in assessing liver function in patients with hepatocellular carcinoma [30][31]. At the same time, ALBI and King's score models were equally statistically significant in predicting PHLF after major liver resection (P = 0.395). As verified by this article, for HCC patients undergoing major hepatectomy, the ALBI model has a better ability to predict PHLF then PALBI (P = 0.005) and MELD (P = 0.007). This result was in accordance with the study reported by Alexander M. that ALBI can predict PHLF better than MELD after hepatectomy [21]. Subgroup analysis was then performed based on patients with or without cirrhosis. For cirrhotic patients, except ALBI, PALBI may be also a good choice for predicting PHLF. Moreover, for patients without cirrhosis, ALBI, MELD, FIB4 and King's score may be more suitable.
The strengths of this study lie in the large patient population. However, the present study has some limitations. First, it is a retrospective study with inherent shortcoming. Second, although our data are from four well-known hospitals, they are all from Chinese people, whether our conclusion is suitable in patients from the West remains uncertain. Third, there is no data on remnant liver volume and ICG-15, which may be useful in predicting PHLF. Further external validation of our results is needed.

Conclusion
In conclusion, among the six liver functional reserve models (ALBI, APRI, FIB4, MELD, PALBI, King's score), the ALBI exhibits more accurate predictive ability for PHLF in HCC patients undergoing major hepatectomy, regardless of having cirrhosis or not. data in the research.

Data availability statement
All data generated or analyzed during this study are included in this article.
• Collection and assembly of data: all authors.
• Data analysis and interpretation: all authors.