Development and Validation of a Nomogram Based on Noninvasive Liver Reserve and Fibrosis (PALBI and FIB-4) Model to Predict Posthepatectomy Liver Failure Grade B-C in Patients with Hepatocellular Carcinoma

Hepatectomy is currently one of the most effective treatments for hepatocellular carcinoma (HCC). However, postoperative liver failure (PHLF) is a serious complication and the leading cause of mortality in patients with HCC after hepatectomy. This study attempted to develop a novel nomogram based on noninvasive liver reserve and fibrosis models, platelet-albumin-bilirubin grade (PALBI) and fibrosis-4 index (FIB-4), able to predict PHLF grade B-C. This was a single-centre retrospective study of 574 patients with HCC undergoing hepatectomy between 2014 and 2018. The independent risk factors of PHLF were screened using univariate and multivariate logistic regression analyses. Multivariate logistic regression was performed using the training set, and the nomogram was developed and visualised. The utility of the model was evaluated in a validation set using the receiver operating characteristic (ROC) curve. A total of 574 HCC patients were included (383 in the training set and 191 for the validation set) and included PHLF grade B-C complications of 14.8, 15.4, and 13.6%, respectively. Overall, cirrhosis (P < 0.026, OR = 2.296, 95% confidence interval (CI) 1.1.02–4.786), major hepatectomy (P=0.031, OR = 2.211, 95% CI 1.077–4.542), ascites (P=0.014, OR = 3.588, 95% 1.299–9.913), intraoperative blood loss (P < 0.001, OR = 4.683, 95% CI 2.281–9.616), PALBI score >−2.53 (, OR = 3.609, 95% CI 1.486–8.764), and FIB-4 score ≥1.45 (P < 0.001, OR = 5.267, 95% CI 2.077–13.351) were identified as independent risk factors associated with PHLF grade B-C in the training set. The areas under the ROC curves for the nomogram model in predicting PHLF grade B-C were significant for both the training and validation sets (0.832 vs 0.803). The proposed nomogram predicted PHLF grade B-C among patients with HCC with a better prognostic accuracy than other currently available fibrosis and noninvasive liver reserve models.


Introduction
Hepatocellular carcinoma (HCC) is the sixth highest common malignancy and ranks fourth among cancer-related deaths globally [1]. Hepatectomy is the preferred treatment for liver cancer patients who are diagnosed with resectable and early-stage HCC [2,3]. Although safety and short-term outcomes after surgery have been enhanced as a result of advanced surgical techniques and preoperative management, posthepatectomy liver failure (PHLF) remains the most worrisome complication with high morbidity (0.7-34%) and mortality (50%), which will ultimately prolong hospitalisation, increase hospital costs, and impair quality of life in patients undergoing hepatectomy [4][5][6][7].
Several predicting models of PHLF such as the Child--Pugh grade, the model for end-stage liver disease (MELD) score, albumin-bilirubin (ALBI) score, platelet-albuminbilirubin (PALBI) score, aspartate aminotransferase to platelet ratio index (APRI), and fibrosis-4 index (FIB-4) have been reported [8][9][10][11][12], but the performances of the above models still have drawbacks. e Child-Pugh grade is still widely used to provide an assessment of liver function in clinical work. However, the essential component in this model such as ascites or hepatic encephalopathy lacks objective standards, and the limitations have been suggested in previous studies [13]. MELD is a scoring tool widely used to evaluate the severity of end-stage liver disease and efficacy of transplantation [14] and was original developed to predict survival in patients undergoing a transjugular intrahepatic portosystemic shunt (TIPS) [15]. However, the accuracy of prediction decreases for postoperative morbidity and mortality for patients undergoing hepatectomy, cirrhosis with ascites, or hepatic encephalopathy [16,17]. Furthermore, MELD may not be applicable in patients with PHLF [10,18]. e ALBI grading system was created in 2015 as a convenient and simple tool for assessing liver function in patients with HCC [13], which not only eliminated some subjective variables but also reduced errors due to numerous indexes. Portal hypertension (PH) imposes a high risk of postoperative complication and is closely related to mortality [19]. However, the ALBI grading system does not have any factors associated with PH. e blood platelet count (PLT) was used as an improved marker for PH and added to the PALBI [20], which has shown wonderful power in assessing preoperative liver reserve function and predicting survival [21]. Despite these advantages, the PALBI score still needs to evaluate the accuracy in predicting PHLF by more independent research and with larger amounts of data. e level of hepatic fibrosis is associated with preoperative liver dysfunction and has been discovered to be one of the most important risk factors for the development of liver failure and HCC causing death [22]. e APRI and FIB-4 are noninvasive and reliable fibrosis scores based on laboratory parameters with an impressive performance in assessing significant fibrosis or cirrhosis [12]. However, the APRI score only includes two quantitative variables, serum aspartate transaminase (AST) and PLT counts, which have no ceiling effect and are not optimal factors of liver function. Similarly, the FIB-4 index has obvious limitations on the ability to predict liver fibrosis due to underlying disease, and its predicted performance for late-stage fibrosis is inadequate. erefore, the above conventional models have only focused on preoperative liver reserve function or fibrosis severity without comprehensively considering PHLF-related risk factors, and the accuracy of predicting PHLF is still controversial.
In this study, we retrospectively used a large patient dataset to construct and validate a novel nomogram model combining the PALBI and FIB-4 scores to predict PHLF grade B-C. We then assessed the accuracy of nomogram and conventional model among PHLF patients subjected to hepatectomy.

Study Patients.
is study retrospectively collected data from 574 consecutive patients who underwent hepatectomy for pathologically proven HCC at the Guangdong Provincial People's Hospital from January 2014 to December 2018.
Patients were randomly divided into the training set (n � 383) and validation set (n � 191).
is retrospective study adhered to the provisions of the Declaration of Helsinki. e need for patient written informed consent was waived given the study's retrospective nature. e Research Ethics Committee of our hospital approved the retrospective study.
Patients qualified for inclusion met the following criteria: (1) age 18 to 85 years, (2) preoperative Child-Pugh grade A or B, (3) subjected to R0 resection, and (4) pathologically confirmed as HCC. Patients were excluded for the following reasons: (1) a history of preoperative anticancer therapies such as chemotherapy, radiotherapy, or transcatheter arterial chemoembolisation (TACE); (2) missing important clinical information; and (3) presence of other cancers except for HCC.

Definitions.
PHLF was defined based on the International Study Group of Liver Surgery (ISGLS) definition [23]: serum international normalised ratio (INR) and serum bilirubin concentration increased on or after postoperative day 5. PHLF grade A required no change in the clinical treatment of patients, grade B causing a deviation from the regular clinical management but required noninvasive treatment, and grade C required invasive treatment. In this study, grade B-C was defined as severe PHLF. Clinically significant portal hypertension (CSPH) was diagnosed as the presence of oesophageal/gastric varices by endoscopy and/or a platelet count below 100 × 10 9 /L associated with splenomegaly [24]. Cirrhosis was diagnosed based on histopathological findings. Major hepatectomy was defined as 3 or more liver segmental resections [25]. Ascites was diagnosed by imaging tests such as computed tomography or abdominal ultrasound examination [26].

Clinical Examination Collection and Hepatectomy.
All the preoperative variables based on the results of the latest blood sampling period to surgery including the INR, creatinine, serum albumin, alanine aminotransferase (ALT), AST, alkaline phosphatase (ALP), c-glutamyl transferase (c-GGT), alpha fetoprotein (AFP), and surgical variables (intraoperative blood loss and extent of hepatectomy), tumour characteristics (largest tumour size, number of tumour, and microvascular invasion), cirrhosis, PH, 4 noninvasive liver reserve models (Child-Pugh, ALBI, MELD, APRI, and PALBI), and 1 fibrosis model (FIB-4) were routinely collected. All hepatectomy procedures were performed by three experienced surgeons. e detailed surgical procedures have been illustrated in a previous report [29].

Statistical Analysis.
Categorical parameters are shown as count with percentages (n (%)) and were compared using Fisher's exact test. Continuous parameters with normal distribution are shown as mean ± standard deviation (SD) and compared using Student's t test. In contrast, the Mann-Whitney U test was adopted for continuous parameters with non-normal distributed which were expressed as median (IQR). To determine independent risk predictors of severe PHLF in the training set, univariate analysis and multivariate logistic regression analysis were used. Only parameters with significant difference (P < 0.05) based on univariate regression analysis were selected for multivariate regression analysis. Calibration plots were produced to assess the performance characteristics of the nomograms. Model discrimination accuracy of nomograms, also including Child-Pugh, ALBI, MELD, APRI, PALBI, and FIB-4, for predicting PHLF grade B-C was determined using the area under the receiver operating characteristic (ROC) curve (AUC) and concordance index (C-index). A P value <0.05 was defined as statistically significant in all statistical analyses. All the statistical analyses were performed using Statistical Package for Social Sciences (SPSS) version 26.0 for Windows (SPSS, Chicago, IL, USA). e nomogram and calibration plots were generated using the "rms" package and "boot" method in R software (version 4.0.3).

Patient Characteristics.
e characteristics of 574 HCC cases that met the inclusion and exclusion criteria were recorded and randomly allocated to a training set (n � 383) and an internal validation set (n � 191) at a ratio of 2 : 1. A total of 574 patients were included with a mean age of 53.5 years (SD 11.7), and 507 patients (88.3%) were males. e majority of cases were infected with hepatitis B virus (83.6%), and 32.9% of patients had cirrhosis diagnosed by pathological examination, while CSPH was present in 49 out of 525 cases (8.5%). e baseline clinicopathologic characteristics of the enrolled patients are shown in Table 1. According to the preoperative PALBI score, 55 cases (14.3%) and 34 cases (17.2%) were assigned to the high PALBI group (score >−2.45) in the training and validation sets, respectively. Simultaneously, 49 (12.7%) cases and 26 (13.6%) cases were categorised as high FIB-4 group (score ≥1.45) based on the preoperative FIB-4 score (Table 1). e percentage of PHLF grade B-C was 14.8% in the total set, 15.4% in the training set, and 13.6% in the validation set (Table 1).

Discussion
Preoperative liver functional reserve and fibrosis are critical for surgeons to establish a treatment plan and predicting PHLF grade B-C in HCC patients undergoing hepatectomy. In the present study, we developed and validated a nomogram based on two essential noninvasive liver reserve and fibrosis models (PALBI and FIB-4) in estimating PHLF grade B-C among HCC patients. By utilising AUROC analysis, a graphic and convenient tool demonstrated good capability in terms of prediction of PHLF grade B-C compared with other conventional liver function models including the Child-Pugh grade, MELD, and APRI score and showed a good correlation between prediction by nomogram and actual observation in the training and validation sets. PHLF is a dreadful complication that may cause considerable preoperative death, and its prediction warrants further research and exploration. erefore, the keys to successful outcome include selection of surgical techniques, confirmatory surgery, selection of special perioperative care for patients with possibility of PHLF occurrence, and its severity estimation [4]. e incidence of PHLF grade B-C in our institution was 14.8%, which is consistent with what has been published previously in China in the literature [30] (14.6%, definition criteria: ISGLS). Nevertheless, foreign literature published by Prodeau et al. [4] exhibited a higher PHLF grade B-C incidence in HCC patients with cirrhosis (38%, definition criteria: ISGLS), which may illustrate that cirrhosis is the significant factor to improve the PHLF incidence. is was similar to the results of our study (23.8% in cirrhosis set). e subtle differences of the incidence may be due to the different preoperative evaluations, surgical techniques and postoperative care between home and abroad. Journal of Oncology e function of a remaining healthy liver is strongly associated with developing PHLF [31]. For many years, Child-Pugh grade and ALBI grade have been widely utilised assessment tools for functional liver reserve before liver resection. However, Child-Pugh grade integrated ascites and hepatic encephalopathy, which will cause subjective errors and ultimate leads to a prediction performance decline in PHLF. Although the ALBI grade has exhibited optimal prediction performance in PHLF and was developed using numerous objective indices, its discrimination efficacy does not include important PHLF-related risk factors such as the PLT count [32]. Recently, a novel nomogram based on preoperative and postoperative prediction models did not consider PLT count [33]. e proposed PALBI could compensate for the shortcomings of ALBI grading. In our study, the ROC analysis revealed that the PALBI grade had better accuracy in predicting PHLF in the training set than the Child-Pugh or the ALBI grade, and the results were consistent with a previous study [34]. In contrast, Ye et al. [30] and Xu et al. [35] derived opposite conclusions: the ALBI grade showed a larger AUROC curve for predicting PHLF in HCC patients than PALBI and Child-Pugh grade, and the same results were also observed in the validation set from the present study. erefore, much controversy remains regarding the predictive value between PALBI and ALBI in HCC patients with PHLF, which warrants further research and larger datasets. Our study is consistent with previous findings, which demonstrated that the PLT count is the significant factor for predicting PHLF. Hence, we focused on the analysis of predictive ability of high PALBI grade (score >−2.45) in predicting PHLF. As expected, high PALBI grade is a crucial variable in our predictive nomogram.
Liver fibrosis is correlated with end-stage cirrhosis and HCC [36], and fibrosis staging, as an assessment tool, may   Journal of Oncology effectively estimate prognosis and the treatment efficacy of patients with chronic hepatitis virus infection [37]. erefore, accurate presurgical assessment of HCC patients with liver fibrosis was essential for the patient's long-term prognosis. Although a liver biopsy is officially recommended as the gold standard for staging fibrosis [38], the   invasiveness, cost, and inevitable sampling errors have restricted its clinical application [39,40]. Hence, a variety of noninvasive tools such as FIB-4 and APRI have been approved by the World Health Organization for evaluating significant fibrosis [41]. e FIB-4 and APRI scores all achieved good predictive performances in predicting PHLF grade B-C in both training and validation sets. Moreover, ROC analysis showed the AUROC curves of FIB-4 were larger than that of APRI (training set: 0.758 vs 0.707; validation set: 0.716 vs 0.604). We incorporated the high FIB-4 grade (score ≥1.45) into the multivariate regression analysis, and the results demonstrated that high FIB-4 grade was an independent risk factor for predicting PHLF grade B-C. With the above approach, we developed nomogram and explored its capacity in predicting PHLF grade B-C.
Increasing research has shown that patients presenting major hepatectomy and intraoperative blood loss were factors strongly related to the risk of PHLF [4,33]. e balance of remnant liver parenchyma volume and oncologic characterisation have been maintained in clinical practice. Major hepatectomy caused by anatomical resection can provide patients with a superior prognosis, but it seems to result in an increased risk of PHLF, especially in HCC patients with cirrhosis or hepatic fibrosis [42]. Prodeau et al. [4] collected data from 343 patients and revealed that increased blood loss was associated with a higher risk of PHLF. e formation of ascites is closely related to liver function, which is one of the reasons why it is included in the nomogram. In keeping with previous findings, major hepatectomy, intraoperative blood loss (>400 mL), and ascites were significant risk predictors for the development of PHLF grade B-C in the multivariable logistic regression analysis. Although several previous reports have proposed that ICG clearance can predict the development of PHLF after hepatic resection [43], ICG clearance did not lead to positive results that significantly predicted PHLF [44]. erefore, several previous researches have revealed these three indicators were predictors of PHLF. Rather than adding ICG clearance to the nomogram we developed, we incorporated the major hepatectomy, intraoperative blood loss (>400 mL), and ascites reflecting liver reserve function into our nomogram model. ere were several limitations to the present study. First, as the incidence of hepatitis B virus (HBV) infections in our country has increased, the majority of the patients (83.6%) in this study had experienced chronic HBV infection. As such, the universality of the novel model needs to explore different etiological populations with cirrhosis and HCC. Second, this research was based on data from a single centre and retrospective study, which may cause selection bias. Furthermore, the sample size was relatively small, and the clinical application of our novel nomogram requires independent external and prospective multicentre studies with large dates for optimal validation.
ird, postoperative future liver remnant volume was not considered, which might be taken into further research to promote the accuracy value of predicting PHLF.

Conclusion
In the current study, the results demonstrated that the high PALBI grade (score >−2.45) and high FIB-4 grade (≥1.45) were independently associated with PHLF grade B-C by multivariate regression analysis. e diagnostic nomogram model proposed herein based on four essential clinical variables (cirrhosis, major hepatectomy, ascites, and intraoperative blood loss) and two noninvasive liver reserve and fibrosis model (high PALBI grade and high FIB-4 grade) had good performance for suitably predicting PHLF grade B-C in HCC patients, in contrast to the currently available conventional noninvasive liver reserve and fibrosis (Child--Pugh, MELD, ALBI, and APRI) models in both training and validation sets. is model may contribute to facilitate the preoperative workup of clinicians in predicting the probability of risk of PHLF in HCC patients after hepatectomy.
Data Availability e datasets generated and/or analysed during the current study are available from the corresponding author upon reasonable request.

Conflicts of Interest
e authors declare that they have no conflicts of interest.