Development and validation of web‐based prognostic nomograms for massive hepatocellular carcinoma (≥10 cm): A retrospective study based on the SEER database

Abstract Background and Aims Massive hepatocellular carcinoma (MHCC, a maximum tumor size of at least 10 cm) tends to have a poor prognosis. Therefore, this study aims to construct and validate prognostic nomograms for MHCC. Methods Clinic data of 1292 MHCC patients between 2010 and 2015 were got from the surveillance, epidemiology, and end results (SEER) cancer registration database. The whole set was separated into the training and validation sets at a ratio of 2:1 randomly. Variables, significantly associated with cancer‐specific (CSS) and overall survival (OS) of MHCC were figured out by multivariate Cox regression analysis and were taken to develop nomograms. The concordance index (C‐index), calibration curve, and decision curve analysis (DCA) were taken to validate the predictive abilities and accuracy of the nomograms. Results Race, alpha‐fetoprotein (AFP), grade, combined summary stage, and surgery were identified as independent factors of CSS, and fibrosis score, AFP, grade, combined summary stage, and surgery significantly correlated with OS in the training cohort. They then were taken to construct prognostic nomograms. The constructed model for predicting CSS exhibited satisfactory performance with a C‐index of 0.727 (95% CI: 0.746–0.708) in the training group and 0. 672 (95% CI: 0.703–0.641) in the validation group. Besides, the model for predicting OS of MHCC also showed strong performance both in the training group (C‐index: 0.722, 95% CI: 0.741–0.704) and the validation (C‐index: 0.667, 95% CI: 0.696–0.638) group. All calibration curves and decision curves performed satisfactory predictive accuracy and clinic application values of the nomograms. Conclusion The web‐based nomograms for CSS and OS of MHCC were developed and validated in this study, which prospectively could be tested and may serve as additional tools to assess patient's individualized prognosis and make precise therapeutic selection to improve the poor outcome of MHCC.

T A B L E 1 Basic clinicopathological characteristics in the total, training, and validation sets.

| INTRODUCTION
Hepatocellular carcinoma (HCC) is one of the most common cancers worldwide and the leading cause of tumorrelated mortality in primary liver cancer. 1 The overall survival (OS) of HCC is impacted by several factors such as tumor number, vascular invasion, treatment options feasible, etc. [2][3][4][5] But tumor size is one of the most notable prognostic indicators for HCC. 2,6,7 Patients with huge tumors tend to have an advanced stage and thus leading to poorer prognoses. 7 Even for a patient receiving tumor resection, a huge tumor is associated with early recurrence within 2 years after primary surgery. 7,8 Due to the increasing improvement in public awareness of physical examinations, many HCC patients are now detected and diagnosed at an early stage. However, there is still a considerable proportion of patients who present with large tumors and advanced stages due to a lack of awareness of the importance of health maintenance, especially in China. 9 Massive HCC (MHCC), one of the subtypes of HCC, usually refers to the notion that a patient suffers a maximum tumor size of at least 10 cm. 10,11 It is reported that MHCC accounts for 5.7% of all HCC types. 12 Such subtype is often accompanied by macrovascular invasion, satellite nodules, metastases, and other aggressive characteristics. Besides, MHCC patients usually accept extensive resection, which increases the risk of postoperative liver failure and perioperative mortality. Compared with a tumor <10 cm, the prognosis of a huge tumor after surgery was dissatisfactory [the 5-year OS rate of approximately 30%]. 13,14 Remarkably, some studies suggested that patient with MHCC who was unable to accept surgery had a worse 5-year OS rate of <20%. [15][16][17][18] There was still a lack of enough approaches and studies to discriminate high-risk patients in the MHCC cohort. Under the consciousness of precision medicine, how to effectively intervene in the progression of MHCC deserves clinicians' attention and efforts. Therefore, it is necessary to clarify those risk factors that can independently impact the long-term outcome of MHCC and to take corresponding measures to improve the poor prognosis for MHCC patients.
In this study, we aim to figure out those significantly independent indicators of cancer-specific survival (CSS) and OS for MHCC and to construct web-based prognostic nomograms by using data from the surveillance, epidemiology, and end results (SEER) cancer registration database. We think the constructed predictive models will effectively and conveniently help clinicians provide MHCC patients with appropriately personalized interventions.

| Patient selection
The medical data of a total of 104,173 HCC cases between 2010 and 2015 were extracted from the SEER plus database [18 registries (2000-2018)]. The inclusion criteria in this study were as follows: (I) adult patients (age ≥18 years old); (II) positive cytology or histology evidence of HCC (ICD-O-38170/3 to 8175/3); (III) tumor size ≥10 cm; (IV) survival time ≥1 month. Cases having inadequate clinic data were excluded. Limited by lagging in data updates of the SEER database, we excluded the patients from before 2010 and from 2015 onwards to reduce calculative bias caused by vast missing data of some variables.  (10-20 cm or >20 cm), grade (grade I/II or III/IV), stage according to the 7th AJCC-TNM staging system, combined summary stage (localized, regional or distant)], therapeutic strategies (with or without surgery, with or without radiotherapy, and with or without chemotherapy), survival condition (survival months and vital status). Notably, as for histological grade of differentiation, grade I/II refers to a well or a moderately differentiated tumor while grade III/IV means a poorly differentiated or an undifferentiated tumor.

| Statistical analysis
The medical data of all cases were extracted by the SEER *Stat 8.3.9 software. IBM SPSS Statistics 25.0, R 4.1.3 (www.R-proje ct.org), and MedCalc 19.0 were used for all statistical analyses. The whole set was then divided into the training and validation cohorts at a ratio of 2:1 randomly. Continuous variables are summarized as "mean ± SD" and their differences between groups were compared by the Mann-Whitney U test. Discontinuous variables are expressed as numbers and percentages. The chi-square test or Fisher's exact test were utilized to explore differences in discontinuous variables between groups. The significantly independent factors of CSS and OS, identified by conducting Cox multi-factor analyses, were utilized to construct prognostic nomograms. The concordance index (C-index), calibration curve, and decision curve analysis (DCA) were used to validate the predictive abilities and accuracy; and then, the validation cohort was used to further evaluate the reliability and accuracy of the constructed nomograms. To elaborate on the prognostic differences between subgroups, Kaplan-Meier survival curve analysis and log-rank test were performed. At last, two web applications based on the constructed nomograms for CSS and OS of MHCC were developed and deployed. A p < 0.05 in a two-tailed test was deemed to have statistical significance.

| Survival analyses for CSS and OS
The median survival time of the total set was 8 months (ranges from 1 to 107 months

| Construction and validation of nomogram predicting CSS
A nomogram predicting CSS for MHCC was developed and then deployed the model into an open-access online calculator based on the outcomes of the multivariate analysis (https://sunya tsenm emori alhos ipital.shiny apps.io/ CSS_MHCC/) Figure 1A. The model showed great performance for predicting CSS in the training group (C-index: 0.727, 95% CI: 0.746-0.708) and the validation groups (Cindex: 0. 672, 95% CI: 0.703-0.641). Besides, the calibration curves also exhibited the great predictive ability of the developed model as they were highly consistent with the actual observation Figure 2A-F. Moreover, compared with using sole risk factors to predict CSS, the decision curves for the constructed nomogram revealed the greatest net benefit Figure 3A-F. Based on the above dynamic nomogram, the total cohort was then separated into high-risk and low-risk groups by the median score of 169 after calculating the sum risk scores. The median CSS values were 21 and 4 months in the low-risk and high-risk groups, respectively. And the 5-year CSS rate in the low-risk group of the whole cohort was substantially better than that in the high-risk group (31.4% vs. 2.3%, p < 0.001) ( Figure 4A).

| Construction and validation of nomogram predicting OS
We also developed a nomogram of predicting OS for MHCC according to the outcomes of the multi-factors analysis. We then constructed a web-based calculator to facilitate the use of the nomogram (https://sunya tsenm emori alhos ipital.shiny apps.io/OSMHC C/) ( Figure 1B).
The nomogram also showed satisfactory performance for predicting OS with a C-index of 0.722 (95% CI: 0.741-0.704) in the training group and 0.667 (95% CI: 0.696-0.638) in the validation group. Besides, the calibration curves for the probability of 1-, 3-, and 5-year OS were also highly consistent with the actual observation ( Figure 2G-L). Moreover, decision curve analyses for the constructed nomogram revealed great clinic application values ( Figure 3G-L).
According to the above online nomogram, the total cohort was also separated into high-risk and low-risk groups by the median score of 175 after calculating the sum risk scores. The 5-year OS rate in the high-risk set was poorer than that in the low-risk set (2% vs. 24.7%, p < 0.001) ( Figure 4B) and the median OS values were 18 and 4 months in the low-risk and high-risk sets, respectively.

| Subgroup analyses
The Kaplan-Meier curves suggested that MHCC patients accepting surgery could get better long-term outcomes.
The 5-year CSS rate was 39.4% in the surgery set and 7% in the no-surgery set (p < 0.001). In addition, the OS rate at 5 year in the surgery set was better than that in the nosurgery set (33.9% vs. 4.6%, p < 0.001) ( Figure 4C,D). Moreover, in the early-stage (TNM stage I/II) cohort, the median CSS values in the surgery group and the no surgery group were 73 and 9 months, respectively (p < 0.001). And the median OS values were 62 and 7 months in the surgery and no-surgery groups, respectively (p < 0.001). While in the advanced stage (TNM stage III/IV) cohort, the median CSS and OS values in the surgery group were 22 and 16 months, respectively. But the median CSS value in the no-surgery group was 5 months, and the median OS value was 5 months ( Figure 4E-H).

| DISCUSSION
In this study, we found that the prognosis of MHCC was associated with several independent factors such as preoperative serum AFP status, grade, surgery, etc. According to the above results, we constructed two nomograms harvesting satisfactory predictive abilities to evaluate CSS and OS rates at 1-, 3-, and 5-year for MHCC. It may be helpful to assist in making a precisely prognostic evaluation, therapeutic selection, and individual followup strategy. Although some predictive models evaluating the prognosis of MHCC exist, they were mainly developed based on cases from a single race/region, and did not discuss whether other treatment strategies could impact the prognosis of MHCC or not. 19,20 The SEER database, an authoritative source, provides many cross-race cases with detailed information for studies. Based on data from the database, we constructed and validated the prognostic models for MHCC with more persuasiveness. By clicking the corresponding variables' status on the online operation interface we provided, clinicians can effectively and precisely evaluate the prognosis of MHCC, regardless of patient from White or non-White population.
Nowadays, several therapeutic choices can be selected for HCC, including surgery, radiofrequency ablation, anhydrous alcohol injection, and transhepatic artery chemotherapy and embolization (TACE), etc. [21][22][23][24] But surgery is usually taken as the first-line curative modality for either resectable HCC or as a bridge for future salvage liver transplantation. Certainly, a complete resection of the tumor contributes to improving the poor prognosis of MHCC. Zhu et al 13 retrospectively analyzed the efficacy of surgery for HCC with a tumor larger than 10 cm. Their result showed that the MHCC group could get a 5-year OS rate of 30.3% after primary resection. Another study reported that surgery could contribute to a 5-year OS rate of 44.1% and a 5-year disease-free survival (DFS) rate of 42.2% for MHCC after resection. 25 However, compared with those who accepted surgery, patients receiving nonsurgical treatment have unsatisfactory long-term outcomes. After comparing the efficacy difference of the prognosis of MHCC between surgery and TACE, Min reported that the surgery group showed a higher 5-year OS rate than the TACE group (39.8% vs. 9.7%, p < 0.001). 16 The same results could be found in other studies. 21,22 In our study, multivariate Cox regression analyses figured out that surgery can impact both OS and CSS of MHCC to some extent. Therefore, it was included in our constructed nomograms. In our study, the median CSS and OS values in the no-surgery group were 6 and 5 months, respectively. However, the group successfully accepting surgery had higher 5-year CSS and OS rates (p < 0.001). Even patients at advancedstage can improve their prognosis by getting surgery (the median CSS values: 22 vs. 5 months, p < 0.001). Those results reflect that the patients can gain better prognosis by accepting surgery, if eligible. But limited by the data provided by the SEER database selected, we did not further discuss the specific influences of different surgery methods on the prognosis of MHCC. What's more, we deeply analyzed the influences of chemotherapy and radiotherapy on the prognosis of MHCC by Cox regression analysis. Surprisingly, the results showed that those treatment methods did not significantly impact the long-term outcomes for MHCC patients.
However, surgery for MHCC often means extensive liver resection which may lead to insufficient residual liver volume, increase the risk of postoperative complications, and even death. Some studies reported the morbidity and mortality rates of MHCC after major resection ranging from 10.9% to 43.6% and 4.2% to 18.1%, respectively. [26][27][28][29][30] A retrospective study of 525 huge HCSS reported a postoperative complication rate of 26.4% and a 30-day mortality rate of 2.7%. 28 Even some unconventional procedures can still yield safe short-term outcomes. Jia et al 31 evaluated the safety of anatomic trisegmentectomy for huge HCC of the F I G U R E 3 Decision curve analyses (DCA) of the nomograms for CSS and OS of MHCC. DCA was taken to compare the clinic net benefit between the model and other significant independent factors of CSS/OS in predicting 1-(CSS: A for training set and D for validation set; OS: G for training set and J for validation set), 3-(CSS: B for training set and E for validation set; OS: H for training set and K for validation set), and 5-year (CSS: C for training set and F for validation set; OS: I for training set and L for validation set) CSS and OS rates for MHCC. On the decision curve analysis, the horizontal line represents the assumption that no patient will die, and the oblique line represents the assumption that all patients will die. CSS, cancer-specific survival; MHCC, massive hepatocellular carcinoma; OS, overall survival.

F I G U R E 4
Kaplan-Meier survival analyses to estimate CSS and OS for MHCC stratified by the nomograms (A for CSS, and B for OS) in the whole set, and by surgery in the total set (C for CSS, and D for OS), in the TNM I-II stage set (E for CSS, and F for OS), and in the TNM III-IV stage set (G for CSS, and H for OS). right liver. In their study, 13 patients accepted anatomic trisegmentectomy with complete resection of a tumor but without suffering perioperative morbidity and mortality. Even performed under laparoscopy and accompanied with a higher Iwate difficulty score, patients can still get safe outcomes that 90-day mortality of <1%. 32 Common postoperative complications include ascites, bile leakage, intra-abdominal bleeding, infection, and liver failure. Fei et al 33 constructed a model predicting postoperative liver failure for huge HCC based on preoperative CT radiomics. They figured out that the excision extension, the model for end-stage liver disease (MELD) score, and the radiomics score (Radscore) were significantly risking indicators for the postoperative severe liver failure after major resection. By their model, a surgeon can take enough perioperative prevention measures to decrease the risk of postoperative liver failure. Therefore, although extensive hepatectomy may make MHCC patients vulnerable to various complications and even death, the surgeon can still perform a safe surgery through preoperative comprehensive assessments, intra-operative precise resection, and enhanced recovery after surgery.
Other variables in our models include histological grade, preoperative serum AFP status, etc. Numerous studies showed that they all were associated with the prognosis of HCC. [34][35][36][37] Chen et al 20 figured out that both cirrhosis and AFP were significantly independent factors of OS for HCC. Meanwhile, compared with the non-MHCC set, the MHCC set had a higher level of preoperative AFP. 10 But we did not further discuss whether the discrepancy of elevated AFP could impact patients' prognoses to a different extent or not, due to a lack of concrete values of AFP. Notably, the results of our study suggested that the further increased diameter of the tumor did not exert a significant impact on the prognosis of MHCC. The main reason may be that we just divide the whole set into a 10-20 cm group and a >20 cm group which could be less sensitive for predicting prognoses of MHCC. The pathohistological differentiation of a tumor focus matters a lot to the appraisal of long-term outcomes. A poorly differentiated tumor combined with an elevated AFP level significantly impacts the risk of early recurrence for HCC after surgery. 38 Strikingly, HCC patients along with cirrhosis may endure unsatisfactory survival after orthotopic liver transplantation if diagnosed with a higher histological grade. 39 Therefore, no need to hesitation for clinicians to consider all-around variables that would impact the prognoses of MHCC while making individualized strategies.
Several limitations exist in this study. First, even if the data utilized in this study were extracted from the public database that provided a sea of cross-race HCC cases, the developed nomograms still need further external validation to test their predictive accuracy and performance. Second, limited by the specific variables provided by the SEER database, we did not discuss the associations between specific clinical variables such as etiology, liver function status, vascular invasion, and prognoses of MHCC. Besides, numerous staging systems have been developed and widely employed to evaluate long-term outcome for HCC including the Barcelona Clinic Liver Cancer (BCLC) system and Cancer of the Liver Italian Program (CLIP). 22 However, limited by the SEER database accessibility, we did not conduct an in-depth discussion about the roles of other tumor staging systems on the prognostic evaluation of MHCC compared with our prognostic models. The online predictive models above can serve as convenient but useful tools to help clinicians quickly evaluate the concrete prognosis and make more individualized therapeutic strategies for MHCC patients to get better long-term outcomes.

| CONCLUSION
In summary, we constructed two web-based predictive nomograms for clinical surgeons to precisely evaluate the prognosis of MHCC. Both of the models exhibited satisfactory predictive abilities and clinical application values.

FUNDING INFORMATION ETHICAL APPROVAL STATEMENT
As the study consisted of the retrorespective analysis of anonymous data according to the local ethics committee, a special approval is generally not required.