Prognostic evaluation of HCC patients undergoing surgical resection: an analysis of 8 different staging systems

All patients who were referred for surgical treatment to our institution underwent a detailed clinical work-up as previously described [2, 4]. Therefore, the number, size, and location of tumor nodules as well as the presence of distant metastases were evaluated by magnetic resonance imaging (MRI) or computed tomography (CT). The preoperative risk assessment was carried out based on the American Society of Anesthesiologists (ASA) and the Eastern Cooperative Oncology Group (ECOG) performance status, calculation of the future liver remnant (FLR), as well as parenchymal liver function as assessed by standard laboratory parameters and the LiMAx test (Humedics® GmbH, Berlin, Germany) [22]. Non-invasive liver function tests were routinely carried out, but no preoperative liver biopsies were obtained to assess the quality of the liver parenchyma. Patients staged BCLC A to BCLC C without any evidence of extrahepatic spread as well as compensated liver function were considered candidates for surgery as primary treatment. The definitive decision for hepatectomy was made by a staff hepatobiliary surgeon and approved by the institutional interdisciplinary tumor board in every patient. Liver resection was carried out in accordance with common clinical standards [2, 4]. In brief, an intraoperative ultrasound was performed to visualize the local tumor spread and other suspicious lesions. The decision for either anatomic resections or non-anatomic atypical wedge resections with an adequate resection margin was based on the surgeon’s preference. Parenchymal transection was carried out using the Cavitron Ultrasonic Surgical Aspirator (CUSA®, Integra LifeSciences®, Plainsboro NJ, USA) with low CVP and intermittent Pringle maneuvers if necessary in open hepatectomy. In laparoscopic hepatectomy, parenchymal transection was commonly performed by Thunderbeat® (Olympus K.K., Tokyo, Japan), Harmonic Ace® (Ethicon Inc. Somerville, NJ, USA), or laparoscopic CUSA (Integra LifeSciences, NJ, USA) in combination with vascular staplers (Echelon, Ethicon, Somerville, NJ, USA) or polymer clips (Teleflex Inc., PA, USA). The anesthesiologic management was based on a restrictive fluid intervention strategy ensuring a low central venous pressure (CVP) during parenchymal dissection.

The primary endpoint of this study was to identify the staging system with the best prognostic ability for OS. Overall performance was defined by homogeneity (small differences in OS among patients within the same stage), discriminatory ability (great differences in OS among patients within different stages), and monotonicity of gradients (longer OS in patients in earlier stages than in more advanced stages within the same system) as previously described [23]. Therefore, Cox regression models of each staging systems were established and subsequently used to calculate the likelihood ratio (LR) χ² to determine homogeneity, linear trend (LT) χ² to assess discriminatory ability, and both LR χ² and LT χ² to measure monotonicity of gradients as well as Akaike Information Criterion (AIC) to describe the explanatory ability of the particular staging system [23]. The degrees of freedom were set to 1 in all calculations to allow the comparison of prognostic systems with a different total number of stages. Higher LR χ² and LT χ² as well as lower AIC indicate a better fitting model to predict OS in this statistical approach. Milan criteria were included in the analysis for reference reasons. Data derived from continuous variables are presented as median and interquartile range. Survival curves were generated by the Kaplan-Meier method and compared with the log-rank test. Median follow-up was accessed with the reverse Kaplan-Meier method. Complications are reported as in-hospital morbidity and in-hospital mortality. Perioperatively deceased patients were included in all survival analyses. The level of significance was set to p < 0.05, and p values are given for two-sided testing. Analyses were performed using SPSS Statistics 24 (IBM Corp., Armonk, NY, USA).

A total of 160 patients with a median age of 68 years and median body mass index (BMI) of 26 kg/m² who underwent curative-intent surgery for HCC at our institution from 2011 to 2019 were included in this study with more than half of the study cohort (60.0%, 96/160) belonging to the performance status category ASA III or higher. The vast majority of the patients were categorized as Child-Pugh A (93.1%, 149/160) with a median Child-Pugh score (CPS) of 5 and a median model for end-stage liver disease (MELD) score of 6. The median nodule count was 1 (range 1–7), and the median largest tumor diameter is 55 mm (range 6–228 mm). A tumor burden > 50% was detected in 5.0% (8/160) of the patients, while an overall invasion to major vessels was observed in 26.3% (42/160). Of all patients, 3.1% (5/160) underwent transarterial chemoembolization (TACE) and 1.3% (2/160) transarterial radioembolization (TARE) prior to surgery. A minority of the individuals treated for HCC (27.5%, 44/160) underwent laparoscopic liver resection, and the median operative time was 207 min. R0 resection was achieved in 95.6% (153/160) of the patients. Median hospital stay was 9 days. No complications were detected in 47% (75/160) of the patients. In contrast, 46 patients (28.8%) experienced major postoperative complications (Clavien-Dindo ≥ 3) and 10 patients (6.3%) deceased perioperatively. More clinicopathological and perioperative characteristics are outlined in Table 2, and a detailed overview of the applied staging systems (Milan criteria, BCLC, HKLC, Okuda, CLIP, ITA.LI.CA staging, ITA.LI.CA score, MESH, and GRETCH) is presented in Table 3.

After a median follow-up of 50 months, the median OS of the cohort was 39 months (95% confidence interval (CI): 32–46 months), and the median RFS was 26 months (95% CI: 16–34 months). Further, we conducted multiple secondary survival analyses within the different staging systems. Patients fulfilling the Milan criteria showed a median OS of 58 (95% CI: 24–92 months) compared to 31 months (95% CI: 24–92 months, 22–40 months) in patients outside the Milan criteria (p = 0.012 log rank). Regarding BCLC staging system, the median OS was 63 months (95% CI: 40–85 months) for BCLC 0, 55 months (95% CI: 32–78 months) for BCLC A, and 23 months (95% CI: 9–37 months) and 13 months (95% CI: 4–22 months) for BCLC C (p = 0.001 log rank). More details regarding OS in different staging systems are shown in Table 3, Figs. 1, and 2.

All nine staging systems demonstrated a significant difference of probability of OS when analyzed using Kaplan-Meier analysis (Table 3, Fig. 2). To determine the “best” fitting model, LR χ², LT χ², and AIC were determined. Due to missing data (alpha-fetoprotein, AFP) which was required for some of the staging systems (CLIP, ITA.LI.CA score, MESH, and GRETCH), the overall cohort (n = 160) was analyzed for Milan criteria, BCLC, HKLC, Okuda, and ITA.LI.CA staging and subset of patients (n = 113) for all staging systems separately (Table 4).

In the overall cohort, BCLC performed best among the analyzed staging system with a LR χ² of 20.48, LT χ² of 13.27, and AIC of 764.21 outranking all other staging systems in each criterion. In contrast, Milan criteria showed the lowest LR χ² (6.21) and highest AIC (775.91), while HKLC was the staging system with the lowest LT χ² (4.35). In the sub cohort, ITA.LI.CA score presented with the highest LR χ² (30.08) and lowest AIC (455.27) of all staging systems. The highest LT χ² (18.95) was determined for CLIP. Similar to the overall cohort, HKLC (LR χ² of 9.23, LT χ² of 2.25, and AIC of 475.78) and Milan criteria (LR χ² of 1.42, LT χ² of 1.27, and AIC of 480.30) were the least fitting models to predict OS. More details regarding the different staging systems are outlined in Table 4.