This study was a nationwide multicenter registry-based comparative analysis of low (LP) vs. high (HP) preoperative PIVKA-II in patients with hepatitis B-related HCC who were initially diagnosed between January 1, 2008 and December 31, 2014 in Korea. All eligible patients with hepatitis B-related HCC were assigned to 1 of 2 groups (LP or HP) at a 1:1 ratio using propensity score (PS) matching.

The Korea Central Cancer Registry is a governmental organization with a statutory nationwide cancer registry. This population-based registry began in 1999. Registry completeness accounted for more than 95% of all cancers. The present study was based on data from the Korean Primary Liver Cancer Registry, a joint project between the Korean Liver Cancer Association and the Korea Central Cancer Registry. The Korean Primary Liver Cancer Registry is a retrospective randomly selected nationwide database from the Korea Central Cancer Registry using a systematic random sampling. It has a random sample consisting of approximately 15% of patients newly diagnosed with primary HCC in Korea. The present study was focused on those with newly diagnosed HCC between 2008 and 2014 from the Korean Primary Liver Cancer Registry.

The database included information such as age, sex, date of diagnosis, etiology (hepatitis B or C virus-related, alcoholic liver disease, etc.), Child-Pugh class, Model for End-Stage Liver Disease (MELD) score, performance status, laboratory results such as albumin or bilirubin, tumor markers (serum α-FP and PIVKA-II levels), portal vein invasion, tumor number, tumor size (defined as the diameter of the largest tumor), American Joint Committee on Cancer/International Union Against Cancer TNM stage, Barcelona Clinic Liver Cancer staging, data on first treatment modality and timing, and survival outcomes until December 31, 2016. Individuals with missing data for any of the above variables were excluded from the analysis. Individuals with Child-Pugh classifications B and C were excluded from the analysis. Only individuals with pathologically confirmed HCC were included in this study.

PS matching is increasingly used for reducing the effects of confounding in observational studies. A matching was performed for 956 primarily selected patients using the PS matching method. The PS for probability assigned to each group was estimated using logistic regression with baseline characteristics including age, sex, histologic tumor number, largest tumor size based on diagnostic imaging, preoperative serum α-FP level, and underlying liver disease (combined hepatitis C virus-related liver disease, or alcoholic liver disease). After PS estimation, we matched patients using a 1:1 nearest neighbor matching. The 2 matched groups were compared to examine covariate balance [1920] and to determine whether there were statistically significant differences in baseline covariates between groups.

Outcomes included long-term disease (HCC)-specific survival in matched groups and comparative prognostic factors using univariable and multivariable analyses. HCC-specific survival was measured from the date of liver resection of HCC until the date of HCC-related death or the last follow-up. HCC-specific death was based on data obtained from the registry. Variables such as age, sex, obesity (body mass index of ≥27 kg/m²), smoking, diabetes, hypertension, MELD score, hepatitis C virus-related liver disease, alcoholic liver disease, pathologic tumor number, pathologic largest tumor size, preoperative tumor markers (α-FP, PIVKA-II), ascites on preoperative imaging, preoperative laboratory test results (indocyanine green-R₁₅, serum total bilirubin, serum albumin, serum ALT, PT/international normalized ratio, serum sodium, platelets, and serum creatinine levels), and pathologic results (microvascular invasion, bile duct invasion, and lymph node positivity) were analyzed to determine prognostic factors of HCC-specific survival.

Receiver-operating characteristic (ROC) curve analysis was used to identify the optimal cutoff for baseline α-FP and PIVKA-II that had the highest sensitivity and specificity in discriminating between HCC-related survivors and nonsurvivors. Survival rates were estimated using the Kaplan-Meier method. They were compared using the log-rank test for proportional hazard. PS matching [21] and other statistical analyses were performed with IBM SPSS Statistics ver. 28.0 (IBM Corp., Armonk, NY, USA). In the process of PS matching, Student t-tests or Wilcoxon rank sum test was used for continuous variables, while the chi-square test or Fisher exact test was used for categorical variables for analysis and comparison of baseline covariates. After PS matching, baseline characteristics and survival outcomes of matched groups were compared using paired t-test or Wilcoxon signed-rank test for continuous variables and McNemar test for categorical variables. All categorical data are expressed as numbers or frequencies with percentages. All continuous data are presented as mean ± standard deviation. Comparative prognostic factors were assessed by univariable and multivariable analyses of Cox proportional-hazards regression models. Multivariable analysis was performed using all variables that showed P-values less than 0.05 in univariable analyses. All reported P-values are 2-sided. Statistical significance was defined as a P-value of <0.05.

Patient selection is shown in Fig 1. A total of 10,578 patients initially diagnosed with HCC between 2008 and 2014 in Korea were randomly selected by systematic random sampling from the Korea Central Cancer Registry. Among these, 6,770 patients had HCC related to hepatitis B. Excluding liver transplantations, 1,367 patients underwent liver resection as the first treatment after initial diagnosis of hepatitis B-related HCC. Excluding patients with Child-Pugh B/C and without preoperative PIVKA-II values, a total of 956 patients with preoperative PIVKA-II values were finally enrolled in this study.

Fig. 2 shows ROC curve of α-FP and PIVKA-II in this study population. The cutoff value for baseline α-FP was 40.5 ng/mL, with a sensitivity of 61.7% and a specificity of 56.4% for HCC-specific survival after liver resection. The area under the ROC curve was 0.632 (P < 0.001). The cutoff value for baseline PIVKA-II was 106.0 mAU/mL, with a sensitivity of 64.6% and a specificity of 62.9% for HCC-specific survival after liver resection. The area under the ROC curve was 0.662 (P < 0.001).

After PS matching, LP and HP groups each contained 245 patients (44.8% and 59.9%, respectively). Baseline characteristics of the 2 groups before and after PS matching are summarized in Table 1. Before PS matching, there were significant differences in several baseline variables between the 2 groups (Table 1). After PS matching, there were no statistically significant differences in baseline variables between the 2 groups, except baseline PIVKA-II (Table 1). Hence, these 2 groups were well-matched.

Overall survival time and HCC-specific survival time of patients in the 2 matched groups are shown in Fig. 3. In overall and HCC-specific survival rates, there were no significant differences between the LP group and the HP group (P = 0.460 and P = 0.605, respectively). The 1-, 2-, 3-, and 5-year overall survival rates were 95.6%, 92.2%, 89.5%, and 83.7% in the LP group and 95.6%, 90.7%, 86.4%, and 80.0% in the HP group, respectively (Fig. 3A). The 1-, 2-, 3-, and 5-year HCC-specific survival rates were 94.7%, 90.7%, 88.0%, and 80.9% in the LP group and 95.9%, 90.7%, 85.5%, and 78.7% in the HP group, respectively (Fig. 3B).

Prognostic factors were analyzed for variables obtained from the database using a Cox proportional-hazards model. Results of univariable and multivariable analyses are summarized in Table 2. In univariable analysis, lymph node positivity (hazard ratio [HR], 35.474; 95% confidence interval [CI], 10.36–121.49; P < 0.001), microvascular invasion (HR, 4.245; 95% CI, 2.48–7.28; P < 0.001), hyponatremia of <135 mEq/L (HR, 4.134; 95% CI, 1.51–11.33; P = 0.006), preoperative ascites (HR, 4.077; 95% CI, 1.76–9.46; P = 0.001), largest tumor size of ≥5.0 cm (HR, 2.939; 95% CI, 1.86–4.65; P < 0.001), multiple HCC (HR, 2.306; 95% CI, 1.37–3.87; P = 0.002), and thrombocytopenia (<100 × 10³/µL) (HR, 1.839; 95% CI, 1.01–3.34; P = 0.045) were significant prognostic factors for worse HCC-specific survival (Table 2). Preoperative high α-FP (>40.5 ng/mL) (HR, 1.402; 95% CI, 0.90–2.19; P = 0.139), and preoperative high PIVKA-II (>106.0 mAU/mL) (HR, 1.183; 95% CI, 0.76–1.85; P = 0.461) were not significant prognostic factors for worse HCC-specific survival (Table 2).

For multivariable analysis, Cox regression analysis was performed in a backward manner. In multivariable analysis, hyponatremia of <135 mEq/L (HR, 4.855; 95% CI, 1.67–14.12; P = 0.004), preoperative ascites (HR, 4.072; 95% CI, 1.59–10.43; P = 0.003), microvascular invasion (HR, 3.112; 95% CI, 1.69–5.74; P < 0.001), and largest tumor size of ≥5.0 cm (HR, 2.665; 95% CI, 1.65–4.31; P < 0.001) were independent predictors for worse HCC-specific survival (Table 2).