Introduction
Hepatocellular carcinoma (HCC) is the fourth most common cause of cancer-related deaths worldwide.1 With the development of image guidance techniques, precision external beam radiation (RT) therapies, including intensity-modulated radiotherapy (IMRT) and stereotactic body radiotherapy (SBRT), are being increasingly used to treat HCC.2,3 Radiotherapy has been used as a palliative to the radical treatment of HCC, depending on the tumor status, liver function, and patient’s general state of health. Recently, increasing and encouraging prospective evidence has supported the clinical application of SBRT with a high local control rate and safety.4–10 SBRT can provide better local control than radiofrequency ablation, with comparable toxicities.11–13 SBRT, as an alternative to conventional bridging therapies, can be safely utilized as a bridge to liver transplantation in patients with HCC, with a similar dropout rate and long-term survival outcomes.14 It is recommended as an alternative radical treatment for medically inoperable patients with HCC of some selected early stages.15–18 In patients with small HCC (≤5 cm), SBRT can even achieve an effect similar to that of radical surgery.19,20 In patients with advanced HCC, SBRT extended survival by 9.3 months compared with sorafenib.21
Several classifications of HCC have been proposed in clinical practice. The Barcelona Clinic Liver Cancer (BCLC) system, endorsed by the American Association for the Study of Liver Disease and the European Association for the Study of the Liver, is used worldwide.22 The eighth edition of the TNM American Joint Committee on Cancer (2017), based on only the tumor characteristics and extent of invasion without liver function, is also a commonly used staging system.23 The China Liver Cancer (CNLC) staging system, including tumor status, number of nodules, liver function, and the Eastern Cooperative Oncology Group (ECOG) performance status score, are widely used in China.2 However, radiotherapy is not recommended as a treatment option in the BCLC guidelines. To the best of our knowledge, there are no clinical staging models based on radiotherapy for HCC. In this study, we aimed to verify the predictive stratification ability of the existing staging systems with radiotherapy for HCC and to compare it with a new modified staging model for the prognostic classification of HCC with radiotherapy-based treatment by separating patients according to different stages and substages and analyzing their long-term overall survival (OS) and time-dependent receiver operating characteristic (ROC) curves.
Methods
Patients and study design
The demographic, clinical, and pathological data of consecutive patients with HCC were reviewed retrospectively. Overall, this multicenter cohort study involved 1,191 patients with HCC treated with radiotherapy. In addition, we divided patients receiving SBRT into a training cohort (Ruikang Hospital, The Fifth Medical Center of PLA General Hospital, Peking University Third Hospital, Changhai Hospital). Others on three-dimensional conformal radiotherapy and IMRT were divided into an external validation cohort (Guangxi Medical University Cancer Hospital, First Affiliated Hospital of Yangtze University, Yulin First People’s Hospital). The study protocol conforms to the ethical guidelines of the Helsinki Declaration revised in 2013 as reflected in a priori approval by the ethics committees of Guangxi Medical University Cancer Hospital and the respective institutions (LW2021008). All subjects provided written informed consent in accordance with the Declaration of Helsinki. The STROBE guideline has been followed during the preparation of the manuscript.
Patients who met all of the following criteria were eligible to participate in this study:
HCC lesions are treated with radiotherapy or combination therapy;
Controllable liver function with or without cirrhosis;
ECOG scores 0–2;
Multiple adjacent lesions that a single RT target area could cover were allowed;
Patients with early-stage HCC that was inoperable or unsuitable for radiofrequency ablation.
Patients with the following were excluded:
Non-HCC confirmed by postoperative pathology or liver metastases;
Uncontrolled ascites and hepatic encephalopathy;
An active gastric or duodenal ulcer or other uncontrolled comorbidities;
A history of severe esophageal varices;
Double cancer;
A history of abdominal radiotherapy.
All included patients were re-evaluated 1 month after radiotherapy and every 3–6 months thereafter. All surviving patients were followed up regularly. Follow-up was performed via telephone calls at 3-month intervals or at outpatient. For patients who developed progressive or recurrent disease, multidisciplinary treatment was selected to treat recurrent HCC based on liver function, number of recurrence nodules, location of the tumor, and intra-, or extra-hepatic metastases.
RT protocol
Contrast-enhanced computed tomography (CT) was performed with the patients in the supine position with both arms raised above the head and vacuum mold immobilization to restrain liver motion. Radical/combined/adjuvant/palliative radiotherapy was performed according to the tumor stage and previous treatment. The final RT dose was determined according to the tumor size, severity of liver dysfunction, and dose-volume constraints of the organ at risk. The general principle was using higher RT doses for patients with small tumors, in an early stage, and with better liver function, and lower doses for patients with larger tumors, in an advanced stage, or with worse liver function.
Three-dimensional (3D) RT and IMRT
The gross tumor volume (GTV) was defined as the primary lesion delineated using the CT simulation image and/or fused diagnostic MRI. The clinical target volume (CTV) was defined as GTV plus a 4–5-mm margin in all directions for intrahepatic tumors. The planning target volume (PTV) was established by adding an asymmetric 0.5–1-cm margin in the cranial-caudal direction and 5 mm axially to the CTV for uncertainties in treatment delivery. A median daily dose of 4.0 (range, 2–7) Gy at median fractions of 15 (range, 6–30) was administered to deliver a median total dose of 52 (range, 36–66) Gy.24–26
SBRT
SBRT was delivered using the CyberKnife system (Accuray, Sunnyvale, CA, USA) with tracking of liver motion using implanted fiducials. Three to four fiducials (diameter, 0.8 mm) were inserted into the tumor tissue or into the area surrounding the tumor under B-ultrasound or CT guidance, 1 week before the CT scan or MRI (slice thickness, 3 mm). The CTV coincided with the GTV. The PTV was defined as the GTV plus 5 mm to account for any setup error and was usually decreased manually when the dose-limiting organs overlapped. The median prescribed dose was 45 (range, 26–55) Gy, delivered with a median of three fractions (1–7 fractions) on consecutive days.10,19,24,27
New stage model
We designed a new staging model for radiotherapy-based treatment by incorporating tumor status and the bilirubin-albumin (ALBI) grade for the prognostic classification of HCC. The ALBI grade was calculated as follows: (log10 bilirubin × 0.66) + (albumin × −0.085). The cut points were as follows: ALBI grade 1: ≤−2.60, ALBI grade 2: more than −2.60 to ≤−1.39, and ALBI grade 3: more than −1.39; where bilirubin is in µmol/L and albumin is in g/L.28 The new modified staging system was established as follows: stage I, solitary nodule without macrovascular invasion or 2–3 nodules no more than 3.0-cm apart, and performance status 0–2 (Ia: ALBI-1 grade; Ib: ALBI-2 or 3 grade); stage II: 2–3 nodules with any one more than 3.0-cm apart, or ≥4 nodules, and performance status 0–2 (IIa: ALBI-1 grade; IIb: ALBI-2 grade); stage III: macrovascular invasion, regional lymph node metastasis or distant metastasis, and performance status 0–2 (IIIa: ALBI-1 grade; IIIb: ALBI-2 grade); stage IV: performance status 3–4 or performance status 0–2 with ALBI-3 grade.
Statistical analysis
Patients with HCC treated with RT were separated according to stages and substage. Cumulative OS was the primary endpoint and was compared using the log-rank test and calculated using the Kaplan-Meier method. OS was evaluated from the date of the first radiotherapy to the patient’s death by any cause, the last date alive, or the last follow-up. Continuous variables were compared using Wilcoxon rank-sum test, whereas categorical variables were compared using Pearson’s chi-squared test. In addition, time-dependent receiver operating characteristic (ROC) curve analysis was performed to compare the discriminatory ability of different staging systems.29 The statistical analysis and data plotting were performed with R version 4.0.2 (2020-06-22). Results with p-values < 0.05 were considered statistically significant.
Results
Patient characteristics
Overall, this multicenter cohort study involved 1,191 patients with HCC after radiotherapy between January 1, 2000, and April 30, 2019. Table 1 shows the baseline characteristics. In addition, the training cohort comprised 658 patients and the external validation cohort comprised 533 patients. Among these patients, 362 (55.0%) in the training cohort died within a median follow-up time of 60.0 (range, 6–100) months, and 397 (77.5%) patients in the external validation cohort died within a median follow-up time of 58.0 (range, 6–120) months.
Table 1Baseline characteristics of the study group
| Baseline characteristic | ALL | Training cohort | Validation cohort | p-value |
|---|
| n | 1,191 | 658 | 533 | |
| Sex | | | | 0.006 |
| female | 145 (12.2%) | 96 (14.6%) | 49 (9.19%) | |
| male | 1,046 (87.8%) | 562 (85.4%) | 484 (90.8%) | |
| Age in years | 52.0 (12.0) | 53.9 (12.3) | 49.6 (11.1) | <0.001 |
| Pretreatment | | | | <0.001 |
| no | 539 (45.3%) | 352 (53.5%) | 187 (35.1%) | |
| yes | 652 (54.7%) | 306 (46.5%) | 346 (64.9%) | |
| Hepatitis B surface antigen | | | | <0.001 |
| negative | 173 (14.5%) | 90 (13.7%) | 83 (15.6%) | |
| positive | 938 (78.8%) | 488 (74.2%) | 450 (84.4%) | |
| unknown | 80 (6.72%) | 80 (12.2%) | 0 (0.00%) | |
| Alpha-fetoprotein in ng/mL | | | | 0.064 |
| 0–8 | 293 (24.6%) | 170 (25.8%) | 123 (23.1%) | |
| 8–200 | 326 (27.4%) | 192 (29.2%) | 134 (25.1%) | |
| >200 | 572 (48.0%) | 296 (45.0%) | 276 (51.8%) | |
| Total bilirubin in µmol/L | 19.3 (31.0) | 19.7 (36.2) | 18.9 (23.2) | 0.643 |
| Direct bilirubin in µmol/L | 9.55 (21.0) | 10.2 (23.7) | 7.91 (10.8) | 0.05 |
| Albumin in g/L | 37.2 (5.04) | 37.2 (5.41) | 37.1 (4.55) | 0.633 |
| ALT in U/L | 45.5 (38.1) | 45.9 (37.4) | 44.6 (39.9) | 0.66 |
| AST in U/L | 45.0 (44.9) | 42.1 (43.3) | 52.3 (48.3) | 0.004 |
| ALP in U/L | 129 (189) | 128 (246) | 130 (71.7) | 0.831 |
| CHE in U/L | 5,314 (2,071) | 5,230 (2,201) | 5,488 (1,762) | 0.08 |
| Urea in mmol/L | 5.06 (5.31) | 5.26 (6.19) | 4.55 (1.27) | 0.006 |
| Creatinine in µmol/L | 81.0 (40.8) | 84.8 (46.6) | 71.1 (14.6) | <0.001 |
| PT in s | 13.1 (1.70) | 13.5 (1.79) | 12.8 (1.51) | <0.001 |
| Ascites | | | | 0.302 |
| no | 968 (81.3%) | 530 (80.5%) | 438 (82.2%) | |
| small | 172 (14.4%) | 103 (15.7%) | 69 (12.9%) | |
| moderate | 51 (4.28%) | 25 (3.80%) | 26 (4.88%) | |
| Child-Pugh | | | | 0.009 |
| A | 999 (83.9%) | 540 (82.1%) | 459 (86.1%) | |
| B | 184 (15.4%) | 110 (16.7%) | 74 (13.9%) | |
| C | 8 (0.67%) | 8 (1.22%) | 0 (0.00%) | |
| ALBI score | −2.39 (0.49) | −2.39 (0.53) | −2.38 (0.44) | 0.599 |
| ALBI grade | | | | 0.003 |
| 1 | 418 (35.1%) | 254 (38.6%) | 164 (30.8%) | |
| 2 | 726 (61.0%) | 373 (56.7%) | 353 (66.2%) | |
| 3 | 47 (3.95%) | 31 (4.71%) | 16 (3.00%) | |
| Tumor size in cm | 6.82 (4.79) | 6.11 (5.29) | 7.69 (3.91) | <0.001 |
| Total dose in Gy | 47.7 (7.55) | 43.2 (4.74) | 53.2 (6.64) | <0.001 |
| Fraction | 9.32 (7.66) | 3.69 (1.02) | 16.3 (6.50) | <0.001 |
| Per dose in Gy | 8.51 (4.89) | 12.3 (2.87) | 3.77 (1.59) | 0 |
| TNM-AJCC stage | | | | <0.001 |
| Ia | 55 (4.62%) | 44 (6.69%) | 11 (2.06%) | |
| Ib | 380 (31.9%) | 225 (34.2%) | 155 (29.1%) | |
| II | 135 (11.3%) | 88 (13.4%) | 47 (8.82%) | |
| IIIa | 134 (11.3%) | 67 (10.2%) | 67 (12.6%) | |
| IIIb | 348 (29.2%) | 161 (24.5%) | 187 (35.1%) | |
| IVa | 104 (8.73%) | 38 (5.78%) | 66 (12.4%) | |
| IVb | 35 (2.94%) | 35 (5.32%) | 0 (0.00%) | |
| BCLC stage | | | | . |
| 0 | 51 (4.28%) | 43 (6.53%) | 8 (1.50%) | |
| A | 352 (29.6%) | 261 (39.7%) | 91 (17.1%) | |
| B | 255 (21.4%) | 111 (16.9%) | 144 (27.0%) | |
| C | 525 (44.1%) | 235 (35.7%) | 290 (54.4%) | |
| D | 8 (0.67%) | 8 (1.22%) | 0 (0.00%) | |
| CNLC stage | | | | . |
| Ia | 259 (21.7%) | 181 (27.5%) | 78 (14.6%) | |
| Ib | 172 (14.4%) | 128 (19.5%) | 44 (8.26%) | |
| II | 267 (22.4%) | 112 (17.0%) | 155 (29.1%) | |
| IIIa | 349 (29.3%) | 158 (24.0%) | 191 (35.8%) | |
| IIIb | 136 (11.4%) | 71 (10.8%) | 65 (12.2%) | |
| IV | 8 (0.67%) | 8 (1.22%) | 0 (0.00%) | |
| New stage | | | | <0.001 |
| I | 430 (36.1%) | 309 (47.0%) | 121 (22.7%) | |
| II | 265 (22.3%) | 112 (17.0%) | 153 (28.7%) | |
| III | 464 (39.0%) | 214 (32.5%) | 250 (46.9%) | |
| IV | 32 (2.69%) | 23 (3.50%) | 9 (1.69%) | |
| New substage | | | | <0.001 |
| Ia | 186 (15.6%) | 153 (23.3%) | 33 (6.19%) | |
| Ib | 244 (20.5%) | 156 (23.7%) | 88 (16.5%) | |
| IIa | 103 (8.65%) | 44 (6.69%) | 59 (11.1%) | |
| IIb | 162 (13.6%) | 68 (10.3%) | 94 (17.6%) | |
| IIIa | 136 (11.4%) | 64 (9.73%) | 72 (13.5%) | |
| IIIb | 328 (27.5%) | 150 (22.8%) | 178 (33.4%) | |
| IV | 32 (2.69%) | 23 (3.50%) | 9 (1.69%) | |
New staging model for the training cohort and external validation cohort
The new staging model showed an excellent ability to discriminate patients according to different stages with four notably different curves and substages with seven notably different curves (Fig. 1A, B). The median OS was 63.0 months in stage I (not reached in Ia and 53.0 months in Ib), 24 months in stage II (28.0 months in IIa and 22.0 months in IIb), 11 months in stage III (18.0 months in IIIa and 9.0 months in IIIb), and 9.0 months in stage IV.
The new staging model also showed an excellent ability to discriminate patients in the external validation cohort with four notably different curves and substages with six notably different curves (Fig. 2A, B). In addition, there was a crossover between stages IIa and IIb (Fig. 2C).
Discriminatory ability of the new staging model for the pretreatment or no-pretreatment cohort
Six hundred fifty-two patients received other treatments before radiotherapy and showed most recurrence and residue, and they were included in the pretreatment group. The remaining 539 patients were included in the no-pretreatment group who received radiotherapy as a first-line treatment. The new staging model also showed an excellent ability to discriminate patients in the pretreatment cohort with four notably different curves and substages with seven notably different curves (Supplementary Fig. 1A, B). The new staging model also showed an excellent ability to discriminate patients in the no-pretreatment cohort with four notably different curves and substages with seven curves with a crossover between stages IIIb and IV (Supplementary Fig. 1C, D).
Discriminatory ability of different staging models
The new staging model may supplement other staging systems, with a better area under the curve of time-dependent ROC than BCLC, TNM, and CNLC staging in the training cohort (Fig. 3A), external validation cohort (Fig. 3B), and the entire cohort (Fig. 3C). The Kaplan-Meier curves showed that existing BCLC, TNM, and CNLC staging system were not complete in differentiating survival outcomes among all stages. BCLC staging could not differentiate stages C to D in selected patients (Supplementary Fig. 2A). TNM staging could not completely distinguish stages IIIb to IV (Supplementary Fig. 2B). CNLC staging could not differentiate among stages IIIa, IIIb, and IV (Supplementary Fig. 2C). BCLC staging could not differentiate stages 0 to A in selected patients (Supplementary Fig. 2D). TNM staging could not completely distinguish stages Ia to Ib or IIIb to IV (Supplementary Fig. 2E). CNLC staging could not differentiate between stages Ia to Ib and IIIa, IIIb to IV (Supplementary Fig. 2F).
Discussion
The American Society for Radiation Oncology (ASTRO) Clinical Practice Guideline guides the definitive management of primary HCC with RT therapy, including standard, preoperative, adjuvant, salvage, and consolidative, and as a bridge to transplant, and palliative management for symptomatic cancers.30 An HCC staging model based on RT data is urgently needed to guide treatment purposes effectively and expected survival. Consequently, incorporating tumor status and the bilirubin-albumin grade, we established a new staging model that divided patients into four-stage groups, with two substages each for stages I and II and III. Based on a multicenter, nationwide cohort involving a large sample set, the new staging system can distinguish the survival prognosis of each stage well when the median survival decreased from stages I to IV.
Pretreatment liver function is essential for patients with HCC receiving radiotherapy, as it is strongly associated with treatment toxicity, survival, and dose selection.10,25,27,31–33 BCLC and CNLC staging systems utilize treatment algorithms based on baseline liver function quantified by the Child-Pugh score. The ALBI is purely quantitative, using only laboratory measures of albumin and bilirubin. Previous studies have shown that ALBI could discriminate liver function reserve and predict OS after radiotherapy more objectively than the Child-Pugh system in western and eastern countries.27,34 Therefore, we established this new staging system by incorporating the liver function index (ALBI grade) for further substaging. The new staging model showed the ability to discriminate substages of HCC stages I, II, and III. It had a better area under the curve of time-dependent ROC than BCLC, TNM, and CNLC staging in the entire cohort.
Radiotherapy has also been used for the radical treatment of small or early-stage HCC. In Asian countries, a larger number of cases and expertise have led to more aggressive treatments, including radiotherapy. Notably, the CNLC guidelines in China provide the most liberal indications for the use of radiotherapy. The BCLC system stratifies patients with HCC into five categories (0, A, B, C, and D).35 The BCLC therapeutic flowchart is not applied routinely in radiotherapy management for HCC. SBRT recently provided better local control in medically inoperable early-stage HCC than transarterial chemoembolization (TACE)15–18,36 or radiofrequency ablation.11–13 Unlike radiofrequency ablation, which has the optimal outcomes for tumors less than 3 cm in a specific location,37 radiotherapy is not limited by the tumor size or location.38 Furthermore, irradiation is a potentially curative treatment option for small HCC (≤5 cm), achieving an effect similar to surgical resection.19,20 In addition, SBRT can be safely utilized as a bridge to liver transplantation in patients with HCC.14 Therefore, we refined HCC staging by merging the BCLC 0 (solitary nodule) and BCLC 0 (2–3 nodules ≤ 3.0 cm) groups into a single stage (stage I). After incorporating the ALBI grade, the new staging system showed a good discriminatory power in stratifying early-stage patients into the Ia and Ib subgroups.
Multiple nodular lesions (2–3 nodules, more than 3.0 cm apart or with ≥ 4 nodules), classified as the new stage II/BCLC-B, exhibit considerable heterogeneity. TACE was recommended to the BCLC-B subpopulation, and the expected survival time was 20 months.22 In real-world settings, curative treatments, including radiotherapy and its combinations, are widely applied, and they have shown promise in well-selected patients.39 In particular, SBRT combined with TACE has been reported to have a synergistic effect with favorable outcomes and disease control.40,41 In the current study, the median OS after SBRT was acceptable at 24 months in stage II (28.0 months in IIa and 22.0 months in IIb) in the selected patients with multiple nodular lesions. However, the selection of irradiation lesions and the optimal treatment strategies for combinations with TACE have to be explored further.
The HCC-associated macrovascular invasion has been considered the bottleneck in HCC treatment. If untreated, a median survival time of 2.7 to 4.0 months has been reported.42,43 In our previous study, radiotherapy-based treatment could extend the median survival to 10.0–15.0 months.24,44 Neither the TNM nor the CNLC classification scheme satisfactorily discriminated OS between macrovascular invasion and extrahepatic metastasis disease, suggesting that they are prognostically similar. TNM staging classified HCC-associated macrovascular invasion as T4 and stage IIIb, but could not completely distinguish stages IIIb and IV. CNLC staging classified HCC-associated macrovascular invasion as stage IIIb, but could not differentiate stages IIIa, IIIb, and IV. Therefore, despite the heterogeneity, the proposed staging system classified HCC-associated macrovascular invasion and extrahepatic metastasis as stage III, incorporating the ALBI grade for substages IIIa and IIIb. This refined staging of HCC also had good discriminatory power in stratifying advanced-stage patients into subgroups. All these advantages may be important considerations in the clinical setting.
Here, we defined patients with stage IV as those with ECOG PS 0–2 combined with ALBI-3 or ECOG PS 3–4. Only then, staging can accommodate all patient types. The median OS was based on patients with ECOG 0–2 with median survival of 9 months. Palliative radiotherapy is still available for some cases of increased ECOG score due to metastasis, such as brain metastasis and bone metastasis.45 Therefore, we empirically included patients with ECOG 3–4 in stage IV with expected survival of less than 9 months. For stage IV patients, the survival time is short, and we recommend careful selection of radiotherapy after fully weighing the potential benefits and risks of radiotherapy. Not all patients are suitable for radiotherapy, and the best supportive treatment is recommended for patients with end-stage in most HCC guides.
Some limitations of our study need to be acknowledged. First, this was a retrospective study with inherent defects; further prospective validation of this new model in patients with HCC is recommended. Second, we did not examine treatment-related mortalities, adverse effects, and morbidities. Third, radiotherapy technology (3D-RT, IMRT, or SBRT), radiotherapy equipment, and radiotherapy experience varied in different centers at different times. Following completion of the total dose of radiotherapy, the subsequent treatment choice of the tumor may further influence the patient’s prognosis, which may further affect the accuracy of the model. Finally, no high-level evidence-based evidence strongly supports radiotherapy as a first-line treatment for HCC. Notably, the new staging model is not intended to replace or challenge existing staging but may supplement other staging systems. We encourage the new staging model to be validated with other treatments.
In conclusion, this new modified staging model has an excellent ability to discriminate patients according to different stages and substages after radiotherapy and may provide an alternative for clinical radiologists.