Retrospective analysis of risk factors for early recurrence after hepatocellular carcinoma resection

Population selection

The Institutional Review Board of the Second Affiliated Hospital of Anhui Medical University approved the protocol (YX2024-155). The requirement for informed consent was waived as the study was retrospective. The study obtained medical data from cases undergoing HCC resection at the Second Affiliated Hospital of Anhui Medical University between October 2018 and May 2021. This period was selected to provide an adequate length of postoperative follow-up. The study excluded patients with mixed HCC and cholangiocarcinoma as well as those who were treated preoperatively with transarterial chemoembolization, percutaneous ethanol injection, and radiofrequency ablation. Moreover, patients without follow-up data or with isolated extrahepatic recurrence were also excluded. Of the 141 patients in the original group, nine were excluded due to insufficient baseline data for continuous variables.

Collection of clinicopathologic and imaging data

The demographic information gathered from patients included their age and sex. All patients had traditional laboratory tests, including haematological evaluations, serum biochemistry tests, and tumor marker testing, before curative liver resection. Specific laboratory indicators evaluated in this investigation were AFP, aspartate aminotransferase (AST), alanine aminotransferase (ALT), bilirubin, albumin, gamma-glutamyltransferase (GGT), total bilirubin, direct bilirubin, and indirect bilirubin.

Two experienced pathologists independently evaluated all pathological specimens. The tissue sample was processed according to the standard recommendations for handling hepatic surgical specimens [8], including specific locations, distances from the tumor edge, and tissue volume to ensure diagnostic accuracy. Histopathological features assessed included tumor size and number, degree of cellular differentiation, capsular invasion, and status of the surgical margin.

Follow-up

Tumor recurrence within two years after liver resection was defined as early HCC recurrence. Following tumor resection, all patients underwent systematic follow-up in accordance with clinical guidelines [9]. Follow-up was performed every three months using imaging (ultrasound, computed tomography [CT], and magnetic resonance imaging [MRI]) and laboratory analyses (such as AFP level and tests for liver function). Two or more diagnostic techniques, such as hepatic artery angiography, contrast-enhanced CT, ultrasound, or MRI, supported by elevated AFP levels when present, were required to diagnose HCC recurrence. For those patients who remained recurrence-free, the follow-up concluded on December 31, 2023.

Statistical analysis

In this study, the overall rate of missing data was low (<5 % for all variables) and was addressed separately for each variable before model development. Missing values in continuous variables were imputed using the median of non-missing observations, while those in categorical variables were imputed using the mode. Continuous variables were dichotomized or categorized using pre-specified cut-offs derived from clinical guidelines, established literature, or conventional reference ranges. When multiple potential thresholds existed, we prioritized those most widely cited in hepatocellular carcinoma research to enhance interpretability and clinical utility. Chi-squared tests were applied for comparison of categorical data. p-values <0.05 represented statistical significance in risk factor identification. Univariable logistic regression was performed to identify potential predictors of early recurrence. To account for multiple testing in the univariable screening process, we applied the Benjamini-Hochberg procedure to control the false discovery rate (FDR). Variables with an FDR-adjusted p<0.05 were considered statistically significant and were included in the subsequent multivariable analysis. Analysis outputs included odds ratios, 95 % confidence intervals (CI), and p-values. Moreover, a predictive model for early HCC recurrence was created by randomly dividing the data into training and validation cohorts in a 6:4 ratio while employing stratified sampling. The development of the model was carried out using the “rms” package in R (version 3.6.1), based on statistically significant predictors from the multivariate regression. The receiver operating characteristic (ROC) analysis, calibration plots, and decision curve analysis (DCA) were used to assess the model’s discrimination, calibration, and clinical utility. Model performance was then evaluated in the validation cohort.

Participant characteristics

Between October 2018 and May 2021, 132 patients who underwent curative HCC resection at the Second Affiliated Hospital of Anhui Medical University were enrolled. Of these, 109 (82.58 %) were males and 23 (17.42 %) were females. Regarding the etiology of HCC, 101 (76.52 %) cases were associated with hepatitis B virus (HBV). Portal hypertension was found in 24 (18.18 %) patients, while cirrhosis was present in 104 (78.79 %). Splenomegaly and ascites were observed in 38 (28.79 %) and 23 (17.42 %) patients, respectively. Laboratory findings showed that 50 (37.88 %) patients had AFP levels above 400 ng/mL, 37 (28.03 %) had albumin levels below 35 g/L, and 72 (54.55 %) had higher GGT levels equal to or exceeding 50 U/L. Furthermore, 13 (9.85 %) were classified as Child-Pugh class B, while 119 (90.15 %) were categorized as Child-Pugh class A.

Factors linked to early HCC recurrence following after resection

The cases were divided into two groups based on postoperative tumor recurrence status: those exhibiting early recurrence (within 2 years post-surgery) and those without early recurrence (no recurrence or metastasis observed over at least a 2-year follow-up period). A range of variables were analyzed, including demographics (sex, age, and body mass index [BMI]), clinical parameters (albumin-bilirubin [ALBI] grade, surgical approach, margin status, operative time, hepatic inflow occlusion, intraoperative transfusion), laboratory findings (AFP, hepatitis B status, presence of portal hypertension, cirrhosis, splenomegaly, ascites), surgical specifics (extent of resection, anatomical liver resection), pathological characteristics (maximum tumor diameter, tumor count, intravascular tumor thrombus, satellite nodules, TNM classification, Child-Pugh score), and biochemical markers (total protein, total bilirubin, direct bilirubin, indirect bilirubin, albumin, GGT, ALT, AST, and AST/ALT ratio). Table S1 presents the distribution and statistical significance (p-values) of these variables between the two groups. Univariate logistic regression analysis revealed significant associations between early recurrence and AFP≥400 ng/mL (p=0.004), portal hypertension (p=0.005), tumor size>10 cm (p=0.002), albumin<35 g/L (p<0.001), AST≥40 U/L (p=0.010), and GGT levels≥50 U/L (p=0.002), representing potential risk factors (Table 1). Multivariate logistic regression was then performed to identify independent predictors while adjusting for confounding effects. AFP≥400 ng/mL (OR=2.603, 95 % CI: 1.056–6.587), portal hypertension (OR=3.556, 95 % CI: 1.195–11.049), tumor size>10 cm (OR=5.036, 95 % CI: 1.264–23.237), albumin<35 g/L (OR=0.127, 95 % CI: 0.041–0.353), and GGT≥50 U/L (OR=2.172, 95 % CI: 0.863–5.612) were identified as independent predictors of early recurrence. Comprehensive statistical details are provided in Table 2.

Establishment of the predictive nomogram

To verify the correlation between these factors and early postoperative HCC recurrence, a model for the prediction of early recurrence was established. Table S2 presents the distribution of the relevant variables among HCC patients comparing groups with and without early recurrence in both the training and validation datasets. Five predictors identified in the multivariate regression were incorporated into a nomogram to estimate the risk of early recurrence in HCC cases after resection. Figure 1 illustrates this nomogram, with the top bar indicating the score for each variable, and the bottom bar showing the early HCC recurrence rates following resection.

Figure 1:

Novel nomogram for predicting the risk of early recurrence after resection of hepatocellular carcinoma (HCC) patients. The early recurrence after resection of HCC patients’ nomogram was developed in the cohort, with the use of alpha-fetoprotein (AFP), gamma-glutamyltransferase (GGT), portal hypertension, albumin, tumor maximum diameter. *p<0.05, ***p<0.001.

Assessment of nomogram performance

To determine how the nomogram performed in terms of calibration and discriminatory ability, calibration and ROC curves were used. The calibration curve indicated good agreement between predicted and observed outcomes (Figure 2A) in the training cohort. Compared to the other factors, the nomogram’s AUC was higher at 0.844 (95 % CI: 0.774–0.913) for identifying early HCC recurrence following resection (Figure 2B). The comparative AUV values of individual variables were as follows: AFP (0.628), portal hypertension (0.603), maximum tumor size (0.641), albumin (0.689), and GGT (0.664). Furthermore, the clinical applicability of the nomogram was analyzed through DCA (Figure 2C).

Figure 2:

Apparent performance of the predictive nomogram in the cohort. (A, D) Calibration curves of the early recurrence nomogram prediction in (A) the training cohort and (D) the validation cohort: The x-axis represents the predicted early recurrence after resection of HCC patients. The y-axis represents the actual early recurrence after resection of HCC patients. (B, E) Receiver operating characteristic (ROC) curves of the nomogram in (B) the training cohort and (E) the validation cohort: The ROC curve is displayed in a solid line, and the reference is displayed in a dotted line. The ROCs of the predictive nomogram in the training and validation cohorts with the AUC of 0.882 and 0.923, respectively. (C, F) Decision curve analysis (DCA) of the nomogram in (C) the training cohort and (F) the validation cohort. Using this predictive nomogram to predict early recurrence after resection of HCC patients adds more benefit than the intervention-all-patients scheme or the intervention-none scheme.

To confirm the robustness of the nomogram, its calibration, discriminative, and clinical utility were evaluated in the validation cohort. The calibration curve (Figure 2D), AUC analysis (Figure 2E), and DCA (Figure 2F) yielded results in close alignment with those of the training cohort. The AUC of the nomogram was 0.874 (95 % CI: 0.794–0.954), exceeding the predictive accuracy of other independent variables that contribute to early HCC recurrence. The AUC for individual predictors were lower than the nomogram: 0.658 for AFP, 0.574 for portal hypertension, 0.717 for tumor maximum diameter, 0.687 for albumin, and 0.663 for GGT. This indicates that the nomogram offers a consistent and enhanced predictive capacity for early HCC recurrence following resection.