Primary liver cancer is the sixth most commonly diagnosed cancer and the third leading cause of cancer-related death worldwide, with ~906,000 new cases and deaths in 2020, and hepatocellular carcinoma (HCC) accounts for the vast majority (75–85%) of primary liver cancers (1). Although some risk factors such as hepatitis B virus (HBV) and aflatoxin B1 (AFB1) are declining in many regions due to public health measures, the incidence of liver cancer is still rising, possibly because of the increasing prevalence of obesity and diabetes (2). Due to a variety of diagnostic methods, including ultrasonography, serum alpha-fetoprotein (AFP), computed tomography (CT), and magnetic resonance imaging (MRI) being widely used, a more significant proportion of early HCC patients are diagnosed (3). Besides, a majority of HCC patients are aged between 40 and 60. Hence, the survival prognosis of this group of patients with early HCC remains a significant concern.

Many risk factors are affecting the survival prognosis in patients with HCC, including age (4), race (5), histological grade (6), chemotherapy (7), surgery (8), tumor size (9), and so on. To the best of our knowledge, numerous nomogram prediction models based upon the above factors for HCC have been constructed (10–12), as well as prognostic nomogram for early-stage HCC (13). However, there have been no clinical studies and statistical tools that evaluated the predictive factors for early-stage HCC patients aged between 40 and 60. Considering the earlier the stage and the younger the patient, the better the prognosis and the more meaningful the prediction, we aim to extract data from the Surveillance, Epidemiology, and End Results (SEER) database to develop and validate a practical nomogram for predicting the cancer-specific survival (CSS) of patients with early HCC aged between 40 and 60 by integrating some significant variables so that clinicians can make better decisions.

At present, artificial intelligence has been widely used in human health care. Shubham et al. (14) used deep learning methods to identify glomerulus in human kidney tissue images. Movassagh et al. (15) used a new method to train neural networks. Mohan et al. (16) used machine learning to predict heart disease. Iwendi et al. (17, 18) used machine learning to classify COVID-19 patients and make health predictions. Kumar et al. (19) used neural networks to predict COVID-19. Ngabo et al. (20) use machine learning architectures to tackle urban pandemics. In this study, we used conventional nomogram to construct a new predictive model to predict the survival of middle-aged HCC patients. This prediction model can help HCC patients and doctors to develop clinical and follow-up strategies.

Predictive models for cancer patients have been widely established, including nomograms for predicting survival in HCC patients. Yan et al. (13) constructed a nomogram to predict cancer-specific survival in early hepatocellular patients. Kong et al. (21) constructed a nomogram to predict survival in young patients with hepatocellular carcinoma. Hu et al. (22) developed a nomogram to predict survival in bone metastatic hepatocellular carcinoma patients. Liu et al. (23) constructed a nomogram to predict cancer-specific survival in all patients with hepatocellular carcinoma. Similarly, Ni et al. (24) constructed nomograms to predict survival in all hepatocellular patients. Nevertheless, prognosis and risk of death in HCC patients vary greatly with age. Middle-aged people have a high incidence of hepatocellular carcinoma, and their prognostic survival factors are different from those of other age groups. For example, marriage and race can affect the survival of patients, so a special nomogram is needed to predict patient survival.

Our study extracted information on early-stage HCC patients aged between 40 and 60 from the SEER database. We split the entire cohort into training and validation groups. Differences between the two groups regarding various essential variables such as marriage status, AFP level, grade, stage, and treatments were tested. All tests showed p-values > 0.05, indicating no significant difference between the training and validation groups. Univariate and multivariate Cox analysis found that numerous factors significantly affected CSS, including sex, race, marital status, AFP level, grade, T stage, tumor size, surgery, and chemotherapy. At the same time, other variables such as age and year of diagnosis were not identified as prognostic significance. That is partly because the population we selected is strictly defined, and there was so little difference between patients about age and year of diagnosis.

Social and demographic variables (age, race, and marital status) were identified as prognosticators of HCC. In our study, marital status is a relatively important factor for survival prediction, parallel to many previous studies (13, 25). This is partly because most single, separated, and divorced cancer patients experience more stress and pain than married patients (26). Besides, married patients have better adherence to prescribed treatments than unmarried patients, leading to better cancer control (27). Furthermore, several biological, psychological, and social theories can also explain this phenomenon. A systematic review and meta-analysis indicated significant racial disparities in HCC prognosis in the United States, with Black patients having worse overall survival and Hispanic and Asian patients having better overall survival than White patients (28). Nikita Sandeep Wagle et al. found that Black patients receive curative treatment more minor than White because of social and economic situations (29), which may partly explain this phenomenon. Consistently, we found Black patients having worse CSS, closely followed by the White (P = 0.589), and other races having better CSS (P < 0.001; Table 2).

As for the tumor features (pathological grade, T stage, and tumor size) variables included in the nomogram, the tumor size has the most significant effect for CSS. A retrospective study suggests that tumor size is a vital independent prognostic factor in T2 solitary HCC after curative liver resection (30). In addition, the Milan criteria of liver transplantation strictly limited tumor size, which means that tumors with smaller diameters can be treated in more ways and get better treatment (31). Our results consistently reflected the relationship between tumor size and CSS and quantified its impact on CSS of a specified population compared with previous studies.

There are several kinds of treatment methods for patients with early-stage HCC, including liver resection (LR), ablation, and orthotopic liver transplantation (OLT) (32). Liver transplantation is considered an ideal treatment for selected patients with HCC, as it removes both the tumor and potential cirrhosis, with a 5-year survival rate exceeding 70% (33). Our results once again confirmed its superiority in the treatment of early-stage HCC, and its score is the lowest in the nomogram, which indicated the best outcome. However, the shortage of donor organs has severely limited the utilization of liver transplantation (34). In our results, lobectomy is inferior to liver transplantation but superior to local tumor destruction such as ablation. A systematic review of randomized controlled trials with Meta-analysis and sequential trial analysis found no statistical difference between radiofrequency ablation (RFA) and hepatic resection (HR) in overall survival at 1 and 3 years. Still, RFA is resulted in decreased overall survival compared with HR at 5 years (35). A retrospective study showed that lobectomy has better overall survival and higher cancer-specific survival than thermal ablation (36). Moreover, chemotherapy such as sorafenib and lenvatinib is also an effective treatment for HCC (32). Therefore, it is suggested that surgery (OLT, LR, or RFA) combined with chemotherapy is an ideal choice for early-stage HCC patients aged between 40 and 60.

In our prediction model, the AUC of 1-, 3-, 5-year CSS is 0.747 (95%CI: 72.9–76.5), 77.2 (95%CI: 75.4–78.9), and 78.4 (95%CI: 76.2–80.7), respectively, in the training cohort. In the validation cohort, the AUC of 1-, 3-, 5-year CSS is 0.752 (95%CI: 72.3–78.0), 77.5 (95%CI: 75.0–79.9), and 76.6 (95%CI: 73.5–79.7), respectively (Figure 4). It indicates that our nomogram has relatively high accuracy in predicting CSS of early-stage HCC patients aged between 40 and 60. In addition, we also drew DCA curves to assess the clinical practicability of the nomogram, which suggested that the prediction model has better clinical application value than the traditional TNM staging system and can more accurately predict the 1-, 3-, 5-year CSS of early-stage HCC patients aged between 40 and 60.

However, our study has limitations. First of all, the nomogram was constructed based on the SEER database. It neither necessarily represents the rest of the world nor includes important factors such as alcohol intake, viral infection, cirrhosis, and liver function. Besides, our study was performed by retrospective analysis, which inevitably has a selection bias that is difficult to adjust. Moreover, our prediction model has only been validated internally, and external validation or further prospective studies are necessary. It is suggested that a multicenter study including more liver function variables be conducted.

We explored cancer-specific survival factors in early middle-aged patients with HCC. We found that sex, marriage, race, histological tumor grade, T stage, surgery, chemotherapy, AFP, and tumor size were independent risk factors for patient prognosis. We constructed a new nomogram to predict CSS in patients with early-stage HCC in middle age using these risk factors. This prediction model has good accuracy and reliability, which can help patients and doctors to judge prognosis and clinical decisions.

Publicly available datasets were analyzed in this study. This data can be found here: https://seer.Cancer.gov/.

JT and CW designed the study, revised the article critically, reviewed, and edited the article. CW, JT, and HL collected and analyzed the data. CW drafted the initial manuscript. All authors approved the final manuscript.

This work was supported by a grant from the Science and Technology Department of Sichuan Province, China (Joint Research Fund, No. 2019YFH0056).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.