Using nomogram of the Barcelona Clinic Liver Cancer system for treatment selection in patients with stage C hepatocellular carcinoma

During a 14-year period between 2002 and 2016, 1317 newly diagnosed BCLC stage C HCC patients in our hospital were identified and retrospectively analyzed in this study. Etiology of underlying liver disease, serum biochemistry, number and size of tumor(s), PS, severity of liver cirrhosis and cancer stage were comprehensively recorded at the time of diagnosis. The survival status of all patients was checked every 2–3 months after enrollment and was confirmed by using the database of National Cancer Registry, Taiwan. This study has been approved by the institutional review board (IRB) of Taipei Veterans General Hospital. Before analysis, waiver of consent form from each patient was obtained as justified by the IRB, and patient information was blinded and de-identified.

Findings of typical radiological features in at least two imaging modalities including magnetic resonance imaging (MRI), contrast-enhanced dynamic computed tomography (CT), ultrasound and hepatic arterial angiography, or by a single positive imaging study associated with serum α-fetoprotein (AFP) level ≥ 400 ng/mL or histological confirmation were used to diagnose HCC [11]. Daily consumption of at least 40 g of alcohol for 5 years or more was considered alcoholic liver disease [12]. Patients who were seropositive for anti-hepatitis C virus (HCV) antibody were classified as HCV-related HCC. Hepatitis B virus (HBV)-related HCC was defined as seropositive for HBsAg. Vascular invasion was diagnosed by the presence of thrombus adjacent to the tumor in portal system by at least two imaging modalities. Total tumor volume was calculated based on tumor diameter of every HCC nodules [13]. The Eastern Cooperative Oncology Group (ECOG) criteria were used to evaluate the overall physical status of each patient [14]. Patients with single tumor smaller than 2 cm in size were coded as tumor burden grade 0. Patients with tumor burden beyond grade 0 and within the Milan criteria (one nodule < 5 cm, up to 3 nodules < 3 cm, no vascular invasion or extrahepatic involvement) were classified as tumor burden grade 1 [10, 15]. Patients were recorded as tumor burden grade 3 if lymph node involvement, vascular invasion, or distant metastasis were confirmed at the time of diagnosis. All remaining patients were coded as tumor burden grade 2 [10]. Chest CT scan was performed to detect metastatic lesion(s) and lymph node involvement. Bone metastasis of HCC was surveyed by bone scan and confirmed by MRI if indicated. Surgical resection, ablation, liver transplantation and TACE were collectively defined as aggressive treatments, and targeted therapy and supportive treatment were classified together as the reference group based on the original design of BCLC system. All clinical data were recorded at the time of diagnosis before specific anti-cancer therapy was performed.

General criteria of surgical resection were (1) patients with tumor involving no more than 3 Healey’s segment, (2) Child-Turcotte-Pugh (CTP) class A, and (3) no main portal vein trunk involvement or distant metastasis. Liver transplantation was considered in patients fulfilling the Milan criteria [15]. TACE was performed depending on the size and number of tumor nodules as previously reported [5, 7]. Seldinger’s technique of arterial embolization was performed as the standard TACE procedure. After tumor stain was identified, infusion of a mixture of 20–30 mg adriamycin (Carlo Erba, Milan, Italy) and 5–10 mL Lipiodol (Laboratoire Guerbet, Villepinte, France) was performed after the artery supplying the tumor was catheterized superselectively. Sufficient amount of emulsion and 2–3 mm strips of Gelfoam (Upjohn, Kalamazoo, MI) were delivered to the tumor area until complete flow stagnation was achieved.

The propensity score was generated by using a logistic regression, which calculated the possibility of each patient to receive aggressive or targeted/supportive treatments. Possible variables associated with long-term survival and treatments allocation, including age, gender, severity of cirrhosis, PS, vascular invasion, tumor burden, renal function, serum AFP level, diabetes mellitus were included comprehensively for propensity score generation. To compare the association between experimental variable (anti-cancer treatment) and response (survival), one-to-one pairs were selected by using the propensity score and greedy algorithm to reduce potential biases in subsequent survival analysis [16, 17]. The greedy algorithm used in this study performed 5 to 1 digit matching mechanism; p-value and standardized difference were used to evaluate the goodness of this propensity score with one-to-one matching model. All statistical analyses were conducted with the SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

The baseline patient characteristics are provided in Table 1. Their mean age was 64 years, and 78% of them were male. Hepatitis B (55%) was the most common etiology of chronic liver disease. There were 41% patients who had multiple tumors, and 64% of patients had a main tumor diameter of more than 5 cm. Sixty-nine percent of patients were classified as CTP class A, and 24%, 53% and 23% of patients had PS 0, 1 and 2, respectively. Vascular invasion was documented in 45% of patients, and 27% of patients were diabetic. A total of 15% of patients were confirmed with distant metastasis at study enrollment.

Of all patients, 21% of patients received partial hepatectomy, and 0.3%, 11%, 33%, 8% and 26% of patients underwent transplantation, ablation, TACE, targeted therapy and best supportive care, respectively.

As shown in Fig. 1, nomogram points were assigned to all enrolled patients by using PS, severity of cirrhosis and tumor burden according to our recent study [10]. The range of total nomogram points was between 3 (PS 1–2, CTP class A and tumor burden 0) to 18.2 (PS 1–2, CTP class B and tumor burden 3) for stage C patients (Fig. 2). Ninety-five percent of patients with 3 points underwent aggressive treatments, and 82% of patients with 18.2 points received targeted/supportive therapies. Patients with more advanced HCC had a significantly higher tendency to receive targeted/supportive treatments (p <  0.001).

Patients were divided by nomogram points into four groups (total points: 3–4.2, 6.7–10, 11.9–13, 15.2–18.2). Comparisons of baseline characteristics between patients receiving aggressive and targeted/supportive treatments within these four groups are given in Table 2. Within each group, patients receiving different treatments had similar distributions of age, gender, etiology of chronic liver disease and prevalence of diabetes mellitus (all p > 0.05), but significant differences in tumor burden, PS, and severity of cirrhosis were found among patients undergoing different forms of treatment within these four groups (p < 0.05).

Within all four groups, patients receiving aggressive treatments had significantly better overall survival compared to patients receiving targeted/supportive therapies (all p < 0.05; Fig. 3).

By using the propensity model, 15, 87, 112 and 45 pairs of matched HCC patients were selected for the four groups with absolute differences in propensity score of matched pairs of < 0.0001, 0.0007, 0.0007 and < 0.0001, respectively (Table 3). Among matched patients stratified by the nomogram points, there were no significant differences in gender, etiology of chronic liver disease, severity of cirrhosis, number and size of tumor, PS, and prevalence of diabetes mellitus (all p > 0.05). The majority of standardized differences of significant prognostic variables (e.g., tumor burden, cirrhosis, and PS) were controlled under 0.2; some standardized differences in group 1 (nomogram points: 3–4.2) and 4 (nomogram points 15.2–18.2) were larger than 0.2.

For patients with nomogram points 3–4.2, 6.7–10 and 11.9–13, aggressive treatments were associated with significantly better survival (all p < 0.05). For patients with nomogram points 15.2–18, there were no significant differences in long-term survival after all variables were controlled by the propensity model (p > 0.05; Fig. 4).