Construction and validation of a prognostic prediction model based on albumin-bilirubin grade after resection of hepatocellular carcinoma

Abstract

BACKGROUND

Hepatocellular carcinoma (HCC) is one of the world's major health problems and high recurrence after curative resection limits long-term survival. Existing prognostic tools, such as Child-Pugh score, are limited by subjective aspects. Albumin-bilirubin (ALBI) grade (objective lab parameters) is shown to have an advantage in assessing liver function and prognosis. Given the limitations of existing prognostic tools and the proven prognostic value of ALBI grade in HCC, there is a need to develop an integrated, objective prediction model for postoperative recurrence risk stratification.

AIM

To construct and validate a prognostic prediction model based on ALBI grade after HCC resection.

METHODS

We looked back at clinical data from 596 patients with HCC who had a radical hepatectomy from 2021 to 2023. Using Kaplan-Meier analysis and Cox regression, we identified independent predictors of recurrence-free survival (RFS). We then built a nomogram that combines these predictors to estimate 1- and 2-year RFS.

RESULTS

In the ALBI 2/3 group, the median RFS was 20.0 months. While the median RFS in the ALBI level 1 group was not reached at the end of the study, the differences between the two groups were statistically significant (P < 0.001). Multifactor Cox regression analysis showed that ALBI score, age, tumor size, tumor number, and alpha-fetoprotein levels were independent predictors of the effect on RFS (P < 0.05). We built nomogram with a consistency index (C-index) of 0.647 in the training set and 0.643 in the validation set. From the results of the time dependency analysis, area under the curve was 0.638-0.676 in the 1- and 2-year RFS predictions.

CONCLUSION

ALBI grade independently predicts HCC postoperative recurrence. Our integrated nomogram combining ALBI and clinical factors enables personalized recurrence risk stratification to guide individualized postoperative management.

Key Words: Albumin-bilirubin grade; Hepatocellular carcinoma; Hepatectomy; Prediction model; Nomogram; Recurrence-free survival

Core Tip: We proposed and validated a new predictive tool, the nomogram, to assess whether patients with hepatocellular carcinoma will experience recurrence and how long they will live after radical surgery. This tool uses the level of bilirubin (commonly known as albumin-bilirubin). At the same time, it also combines known tumor indicators such as age, tumor size, number of tumors, and levels of alpha-fetoprotein. In this way, we can provide each patient with a more personalized and clinically relevant risk stratification, helping physicians better develop postoperative plans for patient monitoring and treatment.

INTRODUCTION

Hepatocellular carcinoma (HCC) is one of the most common and lethal malignancies in the world[1]. For patients with early-stage disease, radical hepatectomy is the key method to curate their disease and has shown to improve survival; however, high postoperative recurrence rates still pose a potential risk factor to short-term prognosis[2,3]. Therefore, developing a realistic and effective prognostic model of individualized risk stratifying and recurrence monitoring is a very important clinical objective[4]. Current prognostic assessment is based largely on standard clinical markers such as Child-Pugh score, alpha-fetoprotein (AFP) and tumor size. While the Child-Pugh scoring system includes subjective aspects, such as ascites and hepatic encephalopathy, this may undermine its objectivity[5]. Additionally, their individual values often provide low predictive accuracy and do not capture the entire disease state or prognostic risk profile[6]. Our albumin-bilirubin (ALBI) grade is a recent and objective liver function assessment tool, based solely on two laboratory values, serum albumin and total bilirubin, which is simple and can be easily computed. These two measurements are easy to calculate and demonstrate a good reproducibility. The ALBI grade has been successfully evaluated and used in evaluating hepatic functional reserve and predicting results for patients with HCC[7,8]. In recent years, there has been interest in building recurrence prediction models after hepatectomy by associating the ALBI with different clinicopathology factors to build and validate effective models[9,10].

The rationale for selecting ALBI grade as the core predictor in this study is threefold. First, ALBI grade relies solely on two objective laboratory parameters (albumin and total bilirubin), eliminating subjective assessments such as ascites and hepatic encephalopathy required by the Child-Pugh system, thereby enhancing reproducibility and interobserver consistency. Second, compared with the Model for End-Stage Liver Disease score, which incorporates creatinine and international normalized ratio, ALBI grade is more focused on hepatic synthetic and excretory functions directly relevant to HCC prognosis, while remaining computationally simple. Third, emerging evidence suggests that ALBI grade outperforms Child-Pugh classification in predicting postoperative outcomes for patients with HCC undergoing hepatic resection. Therefore, we hypothesized that integrating ALBI grade with other clinicopathological variables would yield a robust and clinically practical prognostic model. From a retrospective analysis of the clinical data of HCC with curative treatment, our study explores the relationship between the ALBI grade and the recurrence prevention with other important risk factors.

MATERIALS AND METHODS

Research subjects

We carried out this retrospective study using data from patients who underwent curative-intent surgical resection for HCC at Nantong First People’s Hospital from January 2021 to December 2023. To be included, patients had to meet all of the following: They had to be aged ≥ 18 years; have a histopathologically confirmed HCC after surgery; achieve an R0 resection with no residual tumor seen macroscopically or microscopically; have complete clinical records covering demographics, clinicopathological features, lab results, and follow-up data; and have received no preoperative anticancer therapy - meaning no radiotherapy, chemotherapy, molecular targeted agents, or immune checkpoint inhibitors.

This study excluded some cases that did not meet the criteria, such as patients with other cancers besides HCC; patients with major organ dysfunction (such as heart, lung, or kidney failure); patients who die due to complications within 30 days after surgery (as recurrence cannot be assessed for such patients); and patients who already have blood disorders or other systemic diseases that may significantly affect the calculation or results of ALBI scores.

After screening, 596 patients ultimately entered our analysis. Although this was a retrospective study and sample size was determined by the available patient pool during the study period, we performed a post-hoc power analysis to evaluate statistical adequacy. Based on the final sample size of 596 patients, with an expected effect size [hazard ratio (HR)] of 1.5 for the primary predictor (ALBI grade), a two-sided α of 0.05, and an overall event rate of 45.5% (271/596 recurrence events), the achieved statistical power was 92.3% for detecting a significant association between ALBI grade and recurrence-free survival (RFS). This indicates that our sample size was sufficient to yield reliable statistical inferences.

Subsequently, we randomly divided these patients into two groups: 412 were used to construct the model, and the remaining 184 were used to validate the model's effectiveness. The grouping ratio between the two groups was approximately 7:3. Before we officially began data collection and analysis, this research plan had already been approved by the Medical Ethics Committee of the Nantong First People’s Hospital.

Data collection

We extracted the clinical data of patients from the electronic medical record system of the hospital. The information collected is mainly divided into three categories: The first category is demographic information, including gender, age, body mass index (kg/m²), smoking and drinking history, diabetes, family history of related diseases, liver cirrhosis, and hepatitis B virus infection. The second category is clinicopathological features, including tumor size (the maximum diameter measured according to postoperative pathology), tumor number, presence or absence of portal hypertension, presence or absence of vascular or perineural infiltration, and degree of tissue differentiation. The third category is the preoperative laboratory test results and liver function related parameters. The key indicators include AFP, des-gamma-carboxy prothrombin, total bilirubin, albumin, alanine aminotransferase, aspartate aminotransferase, Child-Pugh score, and ALBI classification.

All data have been standardized and strictly controlled. Cases with missing core variables (including age, tumor size, tumor number, preoperative AFP, albumin and total bilirubin for ALBI calculation) were excluded from the cohort. Missing data analysis was performed before screening: Overall, 63 patients had partial missing laboratory or pathological data, accounting for 9.57% of the initially screened population. Missing values were randomly distributed, and no obvious selection bias was observed after comparison of baseline data between patients with and without missing data. To avoid imputation deviation, we directly excluded cases with missing core indicators. When we encounter some ambiguities or problems, we will go back to check the original medical records to ensure that the information is accurate.

The calculation formula of ALBI score is: ALBI = {[log10 total bilirubin (μmol/L) × 0.66] – [0.085 × albumin (g/L)]}[11]. According to the calculated score, we divided the patients into three groups: The corresponding score of level 1 was lower than -2.60, the score of level 2 was between -2.60 and -1.39 (including both ends), and the score of level 3 was higher than -1.39.

Follow-up

We have carried out long-term follow-up of patients from the day of surgery to December 31, 2025, through outpatient, telephone and hospitalization records. To determine tumor recurrence, we adopted the following standardized criteria: (1) Identification of new intrahepatic lesions exhibiting typical HCC imaging features (arterial hyperenhancement with portal venous or delayed washout) on contrast-enhanced computed tomography or magnetic resonance imaging, confirmed independently by at least two experienced radiologists; (2) Extrahepatic metastasis verified by imaging and/or histopathological examination; or (3) Sustained elevation of serum AFP (> 400 ng/mL) with progressive serial increases, in the absence of alternative explanatory conditions. Equivocal imaging findings were further clarified through positron emission tomography-computed tomography or ultrasound-guided biopsy. All recurrence diagnoses were retrospectively reviewed and consensus-validated by a multidisciplinary tumor board comprising hepatobiliary surgeons, radiologists, and oncologists. One of the indicators we are most concerned about is the RFS, which is the time from the end of surgery to the first detection of recurrence or the last contact with the patient. In the first 2 years after operation, we reviewed every 2 months; After 2 years, it was checked every 3 months.

Statistical analysis

SPSS 26.0 software (IBM, Armonk, NY, United States) was used for statistical analysis. For those continuous variables with normal distribution and bell curve characteristics, we describe them in the form of mean ± SD; data that do not conform to the normal distribution are represented by the median and interquartile range. Categorical variables are reported as the n (%).

In terms of survival analysis, Kaplan-Meier method was used to estimate the RFS rate (RFs) of patients, and log-rank test was used to compare the survival curves between different groups. To identify the risk factors that may affect the prognosis, we first performed a single factor Cox proportional hazards regression analysis. For those variables with P < 0.05 in univariate analysis, we will incorporate them into the multivariate Cox regression model to determine which are the truly independent predictors of RFs.

Based on these screened independent predictors, we constructed nomogram with the help of RMS package in R (4.4.1) software (R Foundation for Statistical Computing, Vienna, Austria) and used it to predict the probability of no recurrence within 1 year and 2 years after surgery. We use the consistency index (C-index) to evaluate the discrimination ability of the model. The closer this index is to 1, the stronger the ability of the model to distinguish patients with different risks. In addition, we also drew the time-dependent receiver operating characteristic curve and calculated the area under the curve (AUC) to further test the performance of the model. The calibration curve is used to check whether the probability predicted by the model is consistent with the observed results. For internal verification, we used bootstrap resampling method, repeated sampling 500 times, to test the stability of the model. Finally, all statistical tests were bilateral, and the P < 0.05 were considered statistically significant.

RESULTS

General characteristics of study subjects

In total, 596 patients with HCC who underwent resection were included (Table 1). Out of the entire cohort, 248 patients (41.61%) fell into ALBI grade 1, 340 (57.05%) into grade 2, and only eight (1.34%) into grade 3. Because the grade 3 group was so small, we decided to combine those eight patients with the grade 2 group for all subsequent analyses. That left us with 348 patients (58.39%) in a single ALBI grade 2/3 group.

Table 1 General characteristics of study participants, n (%)/mean ± SD.

Variables	All patients (n = 596)
Sex (male/female)	460 (77.18)/136 (22.82)
Age (years)	56.50 ± 9.53
BMI (kg/m2)	24.61 ± 2.52
Smoking history (no/yes)	398 (66.78)/198 (33.22)
Alcohol use (no/yes)	387 (64.93)/209 (35.07)
History of diabetes (no/yes)	503 (84.40)/93 (15.60)
Family history (no/yes)	527 (88.42)/69 (11.58)
History of liver cirrhosis (no/yes)	363 (60.90)/233 (39.09)
History of hepatitis B (no/yes)	307 (51.51)/289 (48.49)
Tumor size (cm)	4.51 ± 1.02
Tumors number (single/multiple)	429 (71.98)/167 (28.02)
Portal hypertension (no/yes)	500 (83.89)/96 (16.11)
Vascular invasion (no/yes)	458 (76.85)/138 (23.15)
Perineural invasion (no/yes)	540 (90.60)/56 (9.40)
Degree of differentiation (well/moderate/poor)	123 (20.64)/349 (58.56)/124 (20.81)
Child-Pugh class (A/B)	450 (75.50)/146 (24.50)
AFP (μg/L)	299.24 ± 78.89
DCP (mAU/mL)	97.97 ± 30.24
ALT (U/L)	40.34 ± 10.07
AST (U/L)	42.17 ± 9.82
TBIL (μmol/L)	16.73 ± 4.10
Albumin (g/L)	38.96 ± 5.74
ALBI grade (1/2/3)	248 (41.61)/340 (57.05)/8 (1.34)

BMI: Body mass index; AFP: Preoperative alpha-fetoprotein; DCP: Des-gamma-carboxy prothrombin; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; TBIL: Total bilirubin; ALBI: Albumin-bilirubin.

Prognostic analysis of patients with HCC after resection based on ALBI grade

By the end of follow-up, 271 out of 596 patients had developed tumor recurrence. We divided the patients into two groups for prognostic analysis: ALBI grade 1 and ALBI grade 2/3. The median RFS was 20.0 months in the grade 2/3 group, whereas the median had not yet been reached in the grade 1 group. This difference was statistically significant (P < 0.001), as shown in Figure 1.

Figure 1 Cumulative recurrence-free survival in patients with hepatocellular carcinoma after resection based on albumin-bilirubin grade. ALBI: Albumin-bilirubin.

Analysis of factors associated with RFS after HCC resection

Our survival analysis showed that ALBI grade was significantly linked to RFS (P < 0.05). Therefore, we decided to include ALBI grade along with other relevant variables in both univariate and multivariate Cox regression analyses. Table 2 lists how we coded each variable.

Table 2 Variable assignment and coding.

Variables	Assignment of value
Sex	Female = 0, male = 1
Age	Continuous
Body mass index	Continuous
Smoking history	No = 0, yes = 1
Alcohol use	No = 0, yes = 1
History of diabetes	No = 0, yes = 1
Family history	No = 0, yes = 1
History of liver cirrhosis	No = 0, yes = 1
History of hepatitis B	No = 0, yes = 1
Tumor size	Continuous
Tumor number	Single = 0, multiple = 1
Portal hypertension	No = 0, yes = 1
Vascular invasion	No = 0, yes = 1
Perineural invasion	No = 0, yes = 1
Degree of differentiation	Well-differentiated = 0, moderately differentiated = 1, poorly differentiated = 2
Child-Pugh class	Class A = 0, class B = 1
AFP	Continuous
DCP	Continuous
ALT	Continuous
AST	Continuous
TBIL	Continuous
Albumin	Continuous
ALBI grade	Grade 1 = 0, grade 2/3 = 1

AFP: Preoperative alpha-fetoprotein; DCP: Des-gamma-carboxy prothrombin; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; TBIL: Total bilirubin; ALBI: Albumin-bilirubin.

We ran the analysis using the training set. On univariate Cox regression, we found that ALBI grade, sex, age, history of cirrhosis, tumor size, tumor number, vascular invasion, degree of differentiation, AFP level, and albumin level all came out as potential predictors of RFS after HCC resection (all P < 0.05; Table 3).

Table 3 Univariate regression analysis of recurrence-free survival in patients after hepatocellular carcinoma resection.

Variables	β	SE	Wald	P value	HR (95%CI)
Sex	0.452	0.197	5.261	0.022	1.571 (1.068-2.312)
Age	0.023	0.008	7.802	0.005	1.023 (1.007-1.039)
Body mass index	-0.010	0.028	0.118	0.732	0.990 (0.937-1.047)
Smoking history	-0.118	0.160	0.544	0.461	0.888 (0.649-1.216)
Alcohol use	-0.044	0.156	0.179	0.778	0.957 (0.705-1.299)
History of diabetes	0.110	0.208	0.279	0.597	1.116 (0.742-1.679)
Family history	0.180	0.215	0.701	0.402	1.197 (0.785-1.826)
History of liver cirrhosis	0.398	0.148	7.247	0.007	1.489 (1.114-1.989)
History of hepatitis B	-0.013	0.148	0.008	0.929	0.987 (0.739-1.318)
Tumor size	0.226	0.073	9.536	0.002	1.254 (1.086-1.447)
Tumor number	0.373	0.156	5.714	0.017	1.452 (1.069-1.972)
Portal hypertension	0.182	0.197	0.851	0.356	1.199 (0.815-1.764)
Vascular invasion	0.385	0.164	5.519	0.019	1.469 (1.066-2.026)
Perineural invasion	0.029	0.255	0.013	0.910	1.029 (0.625-1.695)
Degree of differentiation	0.282	0.117	5.795	0.016	1.326 (1.054-1.668)
Child-Pugh class	-0.189	0.174	1.172	0.279	0.828 (0.589-1.165)
AFP	0.004	0.001	17.381	< 0.001	1.004 (1.002-1.006)
DCP	-0.001	0.002	0.379	0.538	0.999 (0.994-1.003)
ALT	-0.002	0.008	0.043	0.835	0.998 (0.984-1.013)
AST	-0.002	0.008	0.071	0.790	0.998 (0.983-1.013)
TBIL	-0.011	0.019	0.357	0.55	0.989 (0.953-1.026)
Albumin	-0.025	0.013	3.965	0.046	0.975 (0.951-1.000)
ALBI grade	0.576	0.157	13.514	< 0.001	1.779 (1.309-2.419)

HR: Hazard ratio; CI: Confidence interval; AFP: Preoperative alpha-fetoprotein; DCP: Des-gamma-carboxy prothrombin; ALT: Alanine aminotransferase; AST: Aspartate aminotransferase; TBIL: Total bilirubin; ALBI: Albumin-bilirubin.

Further, we entered those variables into a multivariate Cox regression model. The results indicated that five factors were independent predictors of RFS: ALBI grade [P = 0.004, HR = 2.216, 95% confidence interval (CI): 1.289-3.812], age (P = 0.037, HR = 1.017, 95%CI: 1.001-1.033), tumor size (P = 0.003, HR = 1.261, 95%CI: 1.085-1.466), tumor number (P = 0.009, HR = 1.507, 95%CI: 1.106-2.054), and AFP level (P = 0.001, HR = 1.003, 95%CI: 1.001-1.005). Table 4 summarizes these results.

Table 4 Multivariate regression analysis of recurrence-free survival in patients after hepatocellular carcinoma resection.

Variables	β	SE	Wald	P value	HR (95%CI)
Sex	0.291	0.202	2.080	0.149	1.338 (0.901-1.998)
Age	0.017	0.008	4.364	0.037	1.017 (1.001-1.033)
History of liver cirrhosis	0.197	0.152	1.667	0.197	1.217 (0.903-1.640)
Tumor size	0.232	0.077	9.098	0.003	1.261 (1.085-1.466)
Tumor number	0.410	0.158	6.743	0.009	1.507 (1.106-2.054)
Vascular invasion	0.285	0.168	2.893	0.089	1.330 (0.957-1.849)
Degree of differentiation	0.215	0.121	3.140	0.076	1.240 (0.977-1.572)
AFP	0.003	0.001	10.745	0.001	1.003 (1.001-1.005)
Albumin	0.037	0.022	2.673	0.102	1.037 (0.993-1.084)
ALBI grade	0.796	0.277	8.276	0.004	2.216 (1.289-3.812)

HR: Hazard ratio; CI: Confidence interval; AFP: Preoperative alpha-fetoprotein; ALBI: Albumin-bilirubin.

Construction of a prognostic prediction model for patients after HCC resection

Using the five independent predictors we identified from the multivariate analysis - ALBI grade, age, tumor size, tumor number, and AFP level - we built a prognostic nomogram to predict RFS after liver resection for HCC. We generated this nomogram with the RMS package in R (4.4.1). In the nomogram, we assigned a point value to each variable based on its contribution. Adding up those points gives a total score, which then translates into estimated probabilities of remaining recurrence-free at 1 year and 2 years after surgery (Figure 2).

Figure 2 Nomogram for predicting postoperative recurrence-free survival in patients with hepatocellular carcinoma. AFP: Preoperative alpha-fetoprotein; ALBI: Albumin-bilirubin.

Model validation

In the training set, our model achieved a concordance index of 0.647 for predicting RFS, and in the validation set it came out at 0.643. That suggests the model has acceptable discriminative ability. When we looked at the time-dependent receiver operating characteristic curves (Figure 3), the AUC for predicting 1-year RFS was 0.638 in the training set and 0.663 in the validation set. For 2-year RFS, the AUC values were 0.656 and 0.676, respectively.

Figure 3 Time-dependent receiver operating characteristic curves for recurrence-free survival prediction. A: Training set; B: Validation set. AUC: Area under the curve.

From the calibration curve in Figure 4, the curves in the training set and the verification set are basically coincident with the diagonal line for both 1- and 2-year RFs. This shows that the prediction results of the model based on ALBI grade are in good agreement with the actual situation, and the accuracy is reliable.

Figure 4 Calibration curves for predicting recurrence-free survival. A: Training set for 1- and 2-year recurrence-free survival; B: Validation set for 1- and 2-year recurrence-free survival. RFS: Recurrence-free survival.

DISCUSSION

Prognostic value of ALBI grade in postoperative HCC recurrence

This project retrospectively analyzed the clinical data of 596 patients with HCC who underwent conventional hepatectomy, evaluated the association between ALBI classification and RFs, developed and validated a postoperative prognosis nomogram including ALBI classification. The results showed that there was a significant relationship between ALBI grade and postoperative RFS: The median RFS of patients with ALBI grade 2/3 was 20.0 months, while the median RFS of patients with ALBI grade 1 was much longer. These findings underscore the prognostic value of ALBI classification in the treatment of HCC after operation; i.e., the higher the ALBI classification, the lower the RFS of patients. This trend is also consistent with some previous research reports[12,13].

Mechanistic basis of ALBI grade in predicting recurrence

As an objective indicator for evaluating liver function reserve, ALBI grade depends solely on two laboratory parameters: Serum albumin and total bilirubin. Unlike the Child-Pugh classification, which incorporates subjective clinical assessments such as ascites and hepatic encephalopathy[14,15], ALBI grade offers enhanced reproducibility and objectivity. The biological mechanisms underlying its prognostic efficacy may be explained as follows. Albumin, synthesized by hepatocytes, directly reflects hepatic synthetic capacity. Reduced albumin levels not only indicate compromised liver reserve but may also signify systemic inflammatory response, poor nutritional status, or heavy tumor burden - conditions that collectively create a microenvironment unfavorable for anti-tumor immunity and postoperative tissue repair[16]. Elevated total bilirubin, conversely, often reflects impaired hepatic excretory function, biliary obstruction, or hepatocellular damage, typically accompanied by endotoxemia and oxidative stress[17]. ALBI grade integrates these two objective parameters, and when liver function is impaired (ALBI grade 2 or 3), postoperative recovery may be compromised and immune surveillance attenuated, thereby facilitating tumor growth and metastasis[18].

Independent predictors of RFS

Multivariate Cox regression analysis showed that ALBI grade, age, tumor size, tumor number and AFP level were independent predictors of RFs after hepatectomy. In all these variables, the risk ratio of ALBI classification is the highest, reaching 2.216. In other words, it plays the most important role in our prognosis model. In this regard, Lee et al[19] analyzed the data of 1069 patients after hepatectomy and found that the recurrence risk of patients with ALBI grade 2/3 was 1.94 times that of patients with ALBI grade 1, which was very consistent with our results. In general, the higher the ALBI score, the worse the liver function and the heavier the overall burden of disease. After liver resection, the remaining liver tissue may not be enough to completely restore the normal physiological balance, so the risk of disease recurrence is naturally higher[20].

Age also was independent predictor of recurrence and increased risk increases with advancing age. This is consistent with the observations of Hsiao et al[21], where a larger group of 1701 the study found a significantly higher recurrences risk in older subjects than younger participants, possibly because of age-related physiological decline. The increased age is associated with immunosenescence, a lower immune function that decreases immune surveillance, and less ability to eradicate residual malignant cells. The microscopic residual disease then becomes more proliferative, which increases the risk for intrahepatic recurrence[22]. Tumor size and number are established reliable predictors for overall tumor burden and biological aggressiveness. In this analysis, larger tumor diameter was independently associated with a higher risk of postoperative recurrence, which agrees with other work[23]. A multicenter study also demonstrated that patients with tumors larger than 5 cm were more likely to recur than patients with lesions measuring ≤ 5 cm[24]. Also, multiple tumor size, as opposed to solitary lesions, was independent predictor of less promising outcomes[25]. Large and more diverse tumors are also likely to increase the risk to intrahepatic recurrence and are associated with an adverse prognosis[26].

The AFP is most widely used serum tumor marker for HCC, and in our analysis, it also emerged as an independent predictor of postoperative RFS. Earlier reports have shown that higher preoperative AFP levels are linked to a greater risk of recurrence[27]. The increase of AFP level indicates that tumor proliferation is more active, and the biological behavior of tumor is more invasive. Both mechanisms contribute to tumor invasion and metastasis[28]. Taken together, these findings further illustrate that it is valuable to include AFP in postoperative risk stratification and suggest that for those patients with high AFP levels, closer postoperative follow-up should be arranged[29].

Performance and clinical utility of the nomogram

We combined ALBI classification with other tumor-related factors to construct a nomogram for predicting postoperative RFS, and performed internal validation. From the overall performance, the performance of this model is at a medium level, but relatively stable: In the training group and the validation group, the C-index was 0.647 and 0.643 respectively. AUC values ranged between 0.63 and 0.68. Although it does not reach the high-performance level above 0.8, as a prediction tool, this result is acceptable and relatively consistent[30]. Calibration curve analysis shows that the recurrence probability predicted by nomogram is in good agreement with the recurrence situation observed in the training group and the validation group, indicating that the risk estimation given by this model is reliable, and the prediction result is close to the real outcome of patients[31].

Compared to a single tumor index, this combined model enables a more comprehensive and personalized risk assessment. Clinicians can calculate the total risk score according to the specific clinical situation of each patient, to estimate the probability that the patient will not relapse after 1 year and 2 years. In this way, risk stratification can be done more accurately, and it is also convenient for doctors to formulate more targeted postoperative management plans for high-risk patients; for example, for those patients with high recurrence probability, intensified imaging surveillance may be considered, or adjuvant drug therapy may be administered[32].

Comparison with existing prognostic models

A review of the literature reveals that multiple prognostic scoring systems have been developed for patients with HCC following curative resection. The Barcelona Clinic HCC staging system integrates tumor burden, liver function, and performance status; however, it is primarily designed for treatment allocation rather than precise recurrence prediction[1,33]. The tumor-nodes-metastasis staging system focuses predominantly on the anatomic extent of the tumor while overlooking the functional reserve of the liver[6]. Although the Child-Pugh score is widely applied, its inclusion of subjective clinical parameters (ascites and encephalopathy) may compromise its objectivity[34].

In contrast, our nomogram combines ALBI grade with age and tumor-related characteristics (size, number, AFP) to provide individualized RFS probability estimates. Although the C-index (0.647) did not reach the high-performance threshold (> 0.8), it is comparable to previously reported HCC prognostic models[30,32]. Notably, our model offers the advantage of simplicity: All variables are routinely available preoperatively, facilitating bedside clinical application without complex computations.

Study limitations

This study has some limitations that should be acknowledged. First, as a single-center retrospective study conducted at a tertiary hospital in Eastern China, our findings may be subject to regional bias and limited generalizability. The patient population was predominantly hepatitis B virus-related HCC (48.49%), reflecting the endemic epidemiology of hepatitis B virus in this region, which may differ from Western populations where non-alcoholic steatohepatitis and hepatitis C are more prevalent. Therefore, external validation in multicenter cohorts with diverse ethnic and etiological backgrounds is warranted before widespread clinical application. Second, the median follow-up duration was relatively short (maximum 2 years for patients enrolled in 2023), which limits our ability to assess long-term recurrence patterns and 3-5-year RFS. Given that late recurrence (> 2 years) accounts for a substantial proportion of HCC recurrences and is associated with distinct biological mechanisms (de novo carcinogenesis vs intrahepatic metastasis), longer follow-up is essential to fully evaluate the model’s predictive performance across the entire postoperative timeline. Third, the number of independent predictors included in our nomogram was relatively limited. Future studies should incorporate additional variables such as inflammatory markers (neutrophil-to-lymphocyte ratio, platelet-to-lymphocyte ratio), microvascular invasion status, and molecular biomarkers to enhance predictive accuracy[35]. Finally, although we performed internal validation using bootstrap resampling, external validation in independent cohorts remains necessary to confirm the model’s robustness and generalizability.

CONCLUSION

Overall, ALBI grading can independently predict the RFS of patients with HCC after surgery. On this basis, we further by combining ALBI grading with patient age, tumor characteristics, and AFP levels to create a nomogram. This model is readily applicable, and after verification, its prediction results are also we calibrated.

With this nomogram, we clinicians may consider to identify patients who may experience early postoperative recurrence. For these high-risk populations, we can now consider adopting more intensive follow-up and even consider involving them in some clinical trials of adjuvant therapy. We believe that applying this personalized risk assessment to clinical practice will ultimately help improve the long-term prognosis of patients with HCC.